**Dimensional and Geometrical Quality Enhancement in Additively Manufactured Parts: Systematic Framework and A Case Study**

#### **Natalia Beltrán, David Blanco \*, Braulio José Álvarez, Álvaro Noriega and Pedro Fernández**

Department of Construction and Manufacturing Engineering, University of Oviedo, Pedro Puig Adam St., E.D.O.5, 33203 Gijon, Asturias, Spain; nataliabeltran@uniovi.es (N.B.); braulio@uniovi.es (B.J.Á.); noriegaalvaro@uniovi.es (Á.N.); pedrofa@uniovi.es (P.F.)

**\*** Correspondence: dbf@uniovi.es

Received: 31 October 2019; Accepted: 27 November 2019; Published: 28 November 2019

**Abstract:** In order to compete with traditional manufacturing processes, Additive Manufacturing (AM) should be capable of producing medium to large batches at industrial-degree quality and competitive cost-per-unit. This paper proposes a systematic framework approach to the problem of fulfilling dimensional and geometric requirements for medium batch sizes of AM parts, which has been structured as a three-step optimization methodology. Firstly, specific work characteristics are analyzed so that information is arranged according to an Operation Space (factors that could have an influence upon quality) and a Verification Space (formed by quality indicators and requirements). Standard process configuration leads to characterization of the standard achievable quality. Secondly, controllable factors are analyzed to determine their relative influence upon quality indicators and the optimal process configuration. Thirdly, optimization of part dimensional and/or geometric definition at the design level is performed in order to improve part quality and meet quality requirements. To evaluate the usefulness of the proposed framework under quasi-industrial condition, a case study is presented here which is focused on the dimensional and geometric optimization of surgical-steel tibia resection guides manufactured by Laser-Power Bed Fusion (L-PBF). The results show that the proposed approach allows for part quality improvement to a degree that matches the initial requirements.

**Keywords:** additive manufacturing; quality enhancement; process parameters; design optimization

#### **1. Introduction**

Additive Manufacturing (AM) is defined as "the process of joining materials to make objects from 3D model data, usually layer upon layer, as opposed to subtractive manufacturing fabrication methodologies" according to ISO 17296-1 [1] and ASTM 2792-12 [2]. This definition encompasses a wide variety of processes used to manufacture three-dimensional objects by means of vertical stacking of bi-dimensional layers. Most of the technologies involved have gained maturity in recent years. This has allowed AM applications to evolve from prototype manufacturing to small-batch-size production. Nevertheless, according to Gartner´s hype cycle, consistent adoption of AM in manufacturing operations will still take 5 to 10 years of development [3]. There are many factors that influence AM's difficulties to match the requirements of medium and high batch-size production. Some are related to working volumes and production rates of machines at the current state of development. Others reflect the difficulty of producing parts with similar mechanical behavior to those obtained by traditional manufacturing processes. Finally, dimensional and geometric quality deficiencies of AM parts have also been highlighted as common disadvantages [4,5], which explains their relevance as research subjects during the last decade [6–10]. Quality improvement is a *sine qua non* condition for the generalized industrial adoption of AM processes, since cost-per-unit reduction would not be

enough by itself. Research in the field of dimensional and/or geometric quality improvement of AM parts could be grouped according to three different approaches: error analysis, error prevention and error correction.

Error analysis pays attention to the influence of process parameters on the dimensional or geometrical accuracy of parts. The usual error analysis approach involves comparing the theoretical values of the design parameters and their correspondent values measured upon the manufactured part [11–13]. Consequently, these works provide useful information regarding the expected results of an AM process in terms of quality and accuracy. Research under this approach shows a wide variety of factors influencing the accuracy of AM parts. This is related to the variety of physical principles and implementations available, which lead to specific research solutions for each individual case. These studies also show a lack of uniformity regarding the indicators used to compare the results of modifying process configuration. Researchers sometimes use indicators with no real meaning from an industrial point of view [11,13]. Tests are also frequently carried out upon test specimens and geometries designed *ad hoc* [11,12].

Approaches based on error prevention aim to establish the optimal process configuration to manufacture parts with the highest achievable quality. This goal is commonly based on the analysis of possible error sources and their relative influence upon quality, so that their effects could be minimized by means of parameters adjustment. Once again, the variety of physical principles and configuration parameters used in different AM processes hinders the adoption of a unified methodology for error prevention. Deposition speed or scanning energy are among the factors considered under this approach, but factors related to practical decisions like part location and orientation [6,14] are also used. This last category aims to provide useful recommendations on the best way to place the parts on the working volume. Finally, internal part model parameters, like raster angle or layer thickness [15], are also analyzed. This approach could explore the possibilities of a given technology to improve quality by acting upon influence factors. Nevertheless, this could also be its main limitation, since once the optimal configuration has been determined and errors still exceed tolerance requirements, there is no margin for further improvement.

Finally, error correction approaches act upon accuracy by working on the strategies used to convert 3D geometry into a series of flat layers (slicing), on the generation of material deposition paths and tool trajectories, or directly upon the CAD model. Therefore, this group is formed by those approaches that intend to overpass the limitations derived from the combination of process, technology and geometry to improve part dimensional or geometrical quality. This global objective could be achieved by different solutions. Some works just compensate deviations from the theoretical values [16]; others aim to compensate mechanical errors in the machine [17]; some works elaborate complex models to compensate the influence of different parameters upon the overall quality [18]. In recent years, there has been a tendency to apply machine-learning methods to provide error correction in AM [10,19,20]. Some works focus on compensating in-plane shape deformation [20,21]. Other approaches work with the 3D geometry, mainly compensating thermal deformations that have previously been modelled via finite elements (FE) simulation [22] or by means of virtual manufacturing models [23,24]. These types of works build "predictive" mathematical models that could be based on experimental data [10,20] or build from theoretical models [21,23]. Although Artificial Neural Networks (ANN) are frequently used for building predictive models [10,21,22], alternative mathematical modelling is also used for this purpose (e.g., Gaussian process multi-task learning [20] or particle swarm optimization [25]. The information provided by predictive models could be later used to change input parameters in order to fulfil tolerances.

In sum, the objective of improving dimensional and geometric quality in AM parts is frequently addressed without a recognizable methodology or a standardized procedure, mainly due to the variety of processes and influence factors. This situation is even more pronounced since most of the research has been carried upon "laboratory specimens", neglecting the relevance of industrial tolerances or the problems derived from medium to large batch productions. Additionally, most of the research

has been carried out upon parts designed *ad hoc*, with different levels of complexity, that greatly differ between studies. These artefacts frequently consist of a collection of basic geometries (planes, cylinders, spheres) arranged in one unique part. There are also examples of dimensional quality comparison between biological-type parts (usually bones) in the fields of surgical reconstruction. Organizations like the American National Institute of Standards and Technology (NIST) have even proposed their own round-robin artefacts for establishing repeatability or reproducibility values for a particular AM technology or machine [26]. Accordingly, there are huge differences between quality indicators: some could be considered as artificial indicators, since they are calculated through complex mathematical expressions that ponder a series of individual parameters; others are usually referred to as "volumetric errors" although they lack a standardized physical meaning. Both types of indicators are useful to provide an impression of the overall manufacturing accuracy or for comparison between different process configurations, machines or technologies. Nevertheless, they tend to ignore the fact that dimensional tolerances in industry are limited to Features of Size (FoS) [27], as they are related to fit purposes. Finally, another frequent problem is that error is sometimes assumed to be linear, whereas, in AM processes, there are many factors that could have a non-linear influence upon error, like volumetric shrinkage of thermoplastic in Material Extrusion (ME) processes.

In the present work, a systematic framework for the minimization of dimensional and geometrical errors and tolerance fulfilment in AM parts is presented. This methodology has been specially developed to be applied upon FoS and medium to large production batches. In the following sections, a description of the framework will be presented. The systematic approach is structured in three consecutive steps: Work Analysis, Process Optimization and Design Optimization. The methodology has been evaluated using a case study under quasi-industrial conditions, which is the dimensional and geometric optimization of surgical-steel tibia resection guides manufactured by L-PBF.

#### **2. Systematic Framework Description**

The proposed framework has been designed to be used when production of a given part simultaneously fulfils two conditions:


Therefore, parts with features used for fitting purposes would be candidates for optimization via the proposed approach, whereas parts with features affected only by general tolerances would not be worth of such optimization efforts. Similarly, small batch sizes would not justify the effort of a systematic optimization. In these cases, alternative improvement strategies (e.g., trial-and-error) should be considered. Three consecutive stages have been proposed for the optimization: Work Analysis, Process Optimization and Design Optimization (Figure 1).

**Figure 1.** Steps of proposed approach.

#### *2.1. Work Analysis*

Firstly, a preliminary analysis of the work to be done would be carried out, with the objective of achieving a full description of the problem and evaluating lack-of-quality issues. This implies collecting all the information regarding part, process, production and equipment to elaborate an initial problem statement, defining an operational space and performing an initial quality characterization (Figure 2). Input information should be collected and structured according to three categories—production requirements, design specifications and process characteristics:


**Figure 2.** Work Analysis workflow.

Once all the relevant information has been collected, a statement of the problem should be performed. The objective of this task is to define both an Operation Space and a Verification Space.

• The Operation Space would consist of all those factors that could have an influence upon part quality. This means that every single factor whose modification would presumably affect, to a certain degree, the quality of the part, must be included in the Operation Space. Factors could be subject to modification (controllable factors) or not (non-controllable factors). Controllable factors are those that can be modified according to production decisions. This category includes discrete factors (e.g., selecting "glossy" or "matte" finishing in a Material Jetting process) and continuous factors (e.g., nozzle temperature in ME of thermoplastics) that could adopt many different values within a certain range. On the other hand, non-controllable factors are those that, having an influence upon part quality, should not be modified. This category would include design decisions of requirements (shape, dimensions, material, etc.) that have been set during the design stage. It also should include production decisions (batch size, process, production machine, etc.) that are not subjected to possible modifications. Process parameters could be also considered non-controllable factors when their values have been set according to material or machine supplier recommendations, workshop procedures or workforce experience. Categorization of influence factors into non-controllable and controllable groups is a key task. Most factors undoubtedly belong to one of these groups, but special attention should be paid to those factors that do not have a significant influence upon quality according to previous know-how, since there is a risk of neglecting their influence upon a particular part or feature, despite their actual significance;

• The Verification Space would be formed by Quality Indicators (QI) and Quality Requirements (QR). QI would be used for evaluating the degree of compliance of the tolerances imposed during the design stage. They would usually match FoS quality requirements, like dimensions (diameter of a cylindrical feature or distance between parallel flat surfaces) or geometrical deviation of controlled features (flatness, parallelism, cylindricity or concentricity). Nevertheless, QI could also be defined as relative differences between those parameters and their optimal values (e.g., the difference between the measured diameter and the middle value of the tolerance interval). During the definition of QI, it would also be necessary to define the verification procedure. This means that an inspection plan should be elaborated, including the materials and methods used for verifying each part and calculating actual values for each QI. Finally, QR are defined as the range of acceptable values that each single QI should adopt to enable batch production.

Once information has been structured into the Operating Space and the Verification Space, the objective of this first step is to determine if a standard process configuration could ensure the fulfilment of QR. In order to check this condition, a test set must be manufactured and verified. This implies that staff in charge of improving part quality must decide which process configuration should be used, by setting all controllable factors. This task should be performed taking into account previous experiences, good practices and workforce know-how.

The size of the test set must be determined in order to properly check QR fulfilment and, simultaneously, minimize the number of test specimens to be manufactured. Robustness of this task will increase with the number of replicas, whereas test size is conditioned by experimental cost. Nevertheless, a minimum of two building trays for each particular manufacturing configuration should be demanded, in order to contemplate experimental error. In accordance, calculation of QI values should also be done by means of arithmetic average values of repeated measurements. Staff are encouraged to consult the available literature regarding Design of Experiments (DOE) and Quality Assessment [28].

Manufactured test specimens would then be measured by means of the verification procedure, so that measured values for QI would be calculated and compared with QR. If the results indicate that those requirements would be fulfilled for an acceptable number of parts (defined as the percentage of valid parts per total production), then parts would be considered suitable for batch production and the procedure would finish. If requirements are not appropriately fulfilled, then the strategy continues through its second step. An intermediate situation could also occur when some of the QR are fulfilled but not all of them. In this case, the efforts during Process Optimization should be focused on those requirements that have not been properly fulfilled.

#### *2.2. Process Optimization*

Process Optimization (Figure 3) determines if quality requirements could be fulfilled by just acting upon controllable factors. The complexity of testing the significance on quality of variations in Controllable Factors sharply increases with their number. Checking their influence could be simple when few factors have to be considered, but turns to be extremely complex when the number of factors that should be tested increases. To minimize this problem, factors could be ranked according to their level of significance by means of statistical tools, like Design of Experiments (DOE). This Significance Analysis would be performed as an iterative task to reduce experimental effort, since considering all possible Controllable Factors for DOE could be an inefficient approach when some initial restrictions based on previous know-how have been incorporated into the analysis. Consequently, an initial appreciation of each factor significance could be established based on research papers results, reference books, workforce know-how, etc. This approach would help to fix some factors and reduce experimental effort.

**Figure 3.** Process Optimization workflow.

For instance, it is widely accepted that layer thickness affects geometric quality in ME, since coarse layers increase the staircase effect of sloped surfaces. Consequently, although layer thickness is usually a controllable factor and could be considered during Process Optimization, it can be assumed that increasing layer thickness would not improve geometric quality, and thus this factor should not be included in the optimization step.

Once the number of factors has been initially reduced by means of previous know-how, further decisions should be sustained by experimental testing and supported by statistical analysis. Experimental designs could demand huge experimental effort when a high number of factors are considered. In these cases, the use of fractional factorial designs is widely recommended. Fractional designs would allow for reducing the experimental error by minimizing the number of experiments (and, consequently, the cost of manufacturing and measuring test specimens) by running a fraction of a full factorial design. This approach has the disadvantage of confounding effects of higher-order interaction, but it will be useful to characterize main effects and low-order interactions at a reduced experimental cost. Accordingly, an analysis of variance could provide an ordered list of factors, reflecting their relative influence upon each QI variance. Factors that show no influence upon QI should no longer be considered for Process Optimization. On the other hand, those factors that show a significant influence upon QI results should be ranked according to their relative importance. This procedure could lead directly to an optimized process configuration if only categorical factors (fan on/fan off) have been considered. Nevertheless, if continuous and discrete factors have been included in the DoE, further research could be demanded. In any case, analysis of variance would provide a regression equation that models how QI behaves according to changes in the influence factors. This

equation should be used to optimize a particular QI, whereas multiple response regression optimization methods could be used for simultaneous optimization of multiple Qls.

This optimization effort should lead to a newly optimized process configuration and, once this initial optimization has been established, a new test set should be manufactured in order to check if the QR are fulfilled. A positive result would lead to starting batch production under the optimized configuration, whereas a negative result would lead to revising the significance analysis. If no completely positive result could be achieved, staff should take the decision to finish this step and move onto the Design Optimization step.

#### *2.3. Design Optimization*

Part manufacturing after Process Optimization may still produce features that do not fulfil expected QR. Consequently, part errors (deviations between QI values and optimal QR) should be mathematically modeled and design parameters that could have an influence on the results must be identified. It has to be noted that, in this methodology, design factors are limited to those that could be modified at the CAD definition step. This means that inner-part characteristics like layer thickness or wall thickness are considered process factors, since they are defined at the Computer-Aided Manufacturing (CAM) step.

There are clear differences regarding the level of complexity of error modelling for the non-fulfilment of dimensional QR and geometrical QR:


Additionally, it must be taken into account that, although geometrical re-parameterization could influence dimensional results, it is possible that dimensional optimization would not significantly influence geometrical QR fulfilment. This fact leads to the proposal that geometrical optimization, if necessary, should be carried out before dealing with dimensional optimization.

Once an adequate parameterization has been defined, it would be used to predict the most probable value that each QI would adopt in the final part as a function of controllable factors and design factors. Consequently, a new set of test specimens that include variations in the values of those design parameters that influence QR fulfillment would be manufactured and measured, and QI measures would be used to elaborate a "predictive" model.

There are many mathematical models that could be used at this step, ranging from simple linear regression to complex computing systems (response surfaces, artificial neural networks, etc.). The staff in charge of optimization should evaluate, in each case, which would be the best option for building a predictive model, pondering the required experimental effort and model complexity.

Once the predictive model has been made available, the objective is to define a new mathematical model that answers the inverse question: what should the proper values of design parameters be to obtain a QI as close as possible to the optimal QR? This new model should be known as an "inverse" model. The inverse model would provide new values for design parameters, so that a newly optimized design could be obtained. Ideally, new parts manufactured with this optimized design would be closer to the design theoretical objective, fulfilling the desired QR. Figure 4 contains a graphical explanation of this strategy.

**Figure 4.** Design Optimization strategy.

The proposed workflow for the Design Optimization step is provided in Figure 5.

**Figure 5.** Design Optimization workflow.

#### *2.4. Practical Recommendations*

Once the different steps of the methodology have been described, there are some practical recommendations that should be taken into account:


To evaluate the usefulness of the proposed framework, a case study is presented in the next section.

#### **3. Case Study: Optimization of Surgical-Steel Tibia Resection Guides**

Evaluation case studies must fulfil two conditions: the design must contemplate at least one FoS affected by a dimensional tolerance, and optimization effort must be in accordance with potential batch size. Under both premises, several alternatives were considered until a surgical-steel tibia resection guide was finally selected. This part is a metallic insert used for the guidance of resection instruments during knee arthroplasty, a surgical procedure that is carried out approximately 600,000 times a year in Europe according to EUROSTAT statistical reports [29]. Among different alternatives for resection tool guidance, the one considered here (Figure 6) is a bi-component design, consisting of a polyamide customized alignment part (single-use) and a surgical steel insert (multiple-uses).

**Figure 6.** Parts of a bi-component tibia resection guide.

From a functional point of view, metallic inserts must fit into the PA part and become a single element (no relative movement allowed). At the same time, the theoretical resection plane should be accurately oriented with respect to the alignment features. Guidance is achieved by means of a deep and narrow through slot, defined by two parallel opposite flat surfaces and two slightly tapered lateral sides, resembling an arrow slit, to facilitate instrument handling. This geometry can be obtained through traditional manufacturing processes in two stages. Firstly, basic geometry could be manufactured directly by milling a metal plate or a casting preform; secondly, although external geometry could be completed by milling, the slot is so narrow and deep that it should be manufactured by means of Wire EDM (Electro Discharge Machining) or Sinker EDM. The combined manufacturing costs of these processes could range from 100 to 300 € per unit according to TEKNOS experts. Nevertheless, such geometry could be also manufactured by means of metal AM processes at a competitive cost.

In following sections, the consecutive steps of the proposed methodology applied to such geometry will be explained and discussed.

#### *3.1. Step 1: Work Analysis*

Work analysis starts by gathering together all available information about production requirements, design specifications and process characteristics. Firstly, inserts must be manufactured in a material suitable for biomedical applications, like surgical stainless steel or titanium, by means of a process capable of achieving good quality. This condition led to selecting L-PBF for manufacturing the knee resection inserts included in this work. This process uses the energy of a focused laser beam (typically Nd-YAG) to locally melt metal powder into a solid part [30]. An EOSINT M270 machine has been chosen to manufacture test specimens. This machine has a 250 × 250 mm working area, whereas building height could reach up to 215 mm. A 200 W Yb-Fiber is used alongside with F-Theta precision lens. Layer thickness could range from 20 to 100 μm, whereas other process parameters (scanning speed or effective power consumption) are material-dependent and established according to the indications of the manufacturer. Parts have been manufactured at PRODINTEC Technological Center. The estimated batch size was established at 1000 units per year, and parts must be manufactured according to the design shown in Figure 7. Tolerances were defined by the research team taking into account part functionality and meaningfulness regarding the objectives of this particular case study, but they are not intended to be applicable to other resection guide designs, since present design values for geometric tolerances values could possibly be too restrictive.

**Figure 7.** Part design (**a**) Perspective view; (**b**) Main tolerances in mm.

This part includes two FoS affected by tolerances: external width (nominal value 4.5 mm with a symmetric tolerance of ±0.05 mm) and slot width (nominal value 1.35 mm with an asymmetric tolerance of +0 to +0.05 mm). Internal surfaces of the slot are also affected by a 0.15 mm flatness tolerance, whereas parallelism between those surfaces and the external ones has to be under 0.15 mm. Other part features do not demand specific tolerances, so it can be assumed that usual manufacturing quality would be adequate, since non-compliance of general tolerances related to these features would not critically affect instrument guiding during surgery. Verification will be carried out using a DEA Global Image 09-15-08 Coordinate Measurement Machine (CMM). This machine has been calibrated according to EN 10360-2:2001, with a maximum permissible error in length measurement (MPEE) of 2.2 + 3·*L*/1000 μm (*L* in mm) and a maximum permissible error in probing repeatability (MPEP) of 2.2 μm. PC-DMIS metrology software was used to perform verification operations. Temperature in the laboratory during verification procedures is maintained within 20 ± 2 ◦C. Once information has been gathered and the main features of the problem have been stated, the sequence of tasks continues with the definition of the Operation and Verification Spaces and the subsequent Standard Quality Assessment.

#### 3.1.1. Operation Space

Influence Factors should be grouped in two categories: controllable and non-controllable (Table 1).


**Table 1.** Operation Space.

Since L-PBF has been selected as the most appropriate AM process for this work, and production shall be carried out in an EOS M270, neither process type nor machine could be initially considered as controllable factors. These decisions would also condition subsequent ones, since other possible influence factors, like the type of material, should be accordingly limited to those that the process/machine combination can handle. Therefore, since EOSINT M270 uses proprietary materials, possible materials are reduced to stainless steel PH1 or titanium Ti64. Since weight is not a crucial parameter, SS PH1 has been selected as the construction material, and therefore should be considered as a non-controllable factor. Geometry and dimensions have been set at the design stage. Consequently, all features, geometries and dimensions in the CAD have been also considered as non-controllable factors. Exceptions to this rule are those FoS affected by tolerances since, even when not subjected to modifications at this stage, they could be modified during the Design Optimization stage. Layer thickness in this machine ranges from 20 to 100 μm, so it is a controllable factor. Nevertheless, once layer thickness has been selected, volume rate, scanning speed and effective power are also defined according to material technology specifications. This implies that they cannot be modified by the operator and, consequently, they must be considered as non-controllable factors. This is also applicable to ambient parameters like nitrogen atmosphere (1.5% oxygen), building platform model, or base temperature (40◦). Part orientation has a great influence upon processing time, which usually implies that parts are oriented so that minimum Z travelling is required. Nevertheless, issues regarding manufacturing of slot surfaces were taken into account to avoid excessive overhanging, which would require support structures inside the slot. This concern involved selecting a vertical orientation for the slot and, consequently, orientation has been labelled as a non-controllable factor.

On the other hand, the location of parts within the workspace can be modified by the operator with minimal restrictions (like minimum allowed space between adjacent parts), so it should be considered as a controllable factor. Support structure type is also a controllable factor. Finally, post processing operations could also have an influence upon quality. A support removal operation is unavoidable if support structures are used. Other post-processing operations, like sand blasting or thermal treatments, are optional, and so they should be considered as controllable factors.

#### 3.1.2. Verification Space

QI would be related to part FoS, affected by dimensional and geometric tolerances. They would also be ranked according to their relevance to part functionality.

Accordingly, the distance between parallel surfaces of the slot (*DS*) has been considered as the most relevant QI, since it critically affects resection instrument performance during surgery. An excessive distance would cause noticeable clearance between the slot and the resection instrument, whereas an insufficient distance would make its movement difficult. The second most relevant, Flatness of Slot parallel surfaces (*FSR* for the Rear surface and *FSF* for the Frontal surface) could also have an influence during resection, while Parallelism between these Slot surfaces (*PS*) must be controlled in order to allow uniform behavior of the instrument with independence in its orientation during resection. Finally, Distance between External surfaces (*DE*) has a relatively lower relevance, since its insertion in the alignment part would be favored by PA flexibility. QR have been defined as the acceptable range of values that each QI should adopt. These limits have been established during the design stage, as reflected in Table 2.


**Table 2.** Quality indicators with their correspondent quality requirements sorted by relevance.

#### 3.1.3. Standard Quality Assessment

Standard Quality Assessment implies manufacturing a test set and checking if the QI values measured during verification fulfil QR. In order to manufacture the test set, all controllable factors must be revised, and a Standard Process Configuration defined. This task should be done by taking into account existent know-how, which should include the literature or research works, supplier recommendations or personnel's previous experience. Consequently, layer thickness was set at 20 μm, according to the recommended value for SS PH1. Layer thickness selection determines other variables that, like Volume Rate, fixed at 1.8 mm3/s, are included in the technology files provided by the manufacturer. Part location could also affect quality but, since the objective is to manufacture medium-to-high batches, it is necessary to accommodate the maximum possible number of units in each tray. Accordingly, it was decided that parts within the test set would be distributed along the whole work area. A lightweight supporting structure was selected, since it is the easiest to remove and minimizes removal cost. The same criterion was used to decide that no thermal processing would be applied for the standard configuration. On the other hand, since sand blasting is used to minimize the effect of metallic projections upon surface quality, this post-processing operation was included as part the process. Once the values and alternatives for all these factors have been defined, Process Configuration could be considered complete.

Regarding the Verification Procedure, it has to be noted that QI could be calculated using just four planes adjusted to slot parallel surfaces and external parallel surfaces. To digitize each internal surface, a regular grid with 284 points was used. The distance between adjacent points is 1.4 mm in the same column and 1.6 mm in the same row. Due to slot restricted accessibility, a spherical-end stylus probe with 0.7 mm diameter and 20 mm length was used. In the case of external surfaces, regular grids with 171 digitized points were used. The distance between adjacent points was 2.2 mm in the same column and 2.26 mm in the same row. Complete digitizing of each part in this work (including alignment

routine) was repeated thrice, and, each time, four planes were adjusted to each set of digitized points (Figure 8). Consequently, QI was calculated thrice, and average values were obtained.

**Figure 8.** Verification Procedure. (**a**) Distribution of points on the internal and external surfaces; (**b**) CMM measurement of a test specimen.

The test set used to check part quality under standard process configuration was arranged as a series of sixteen test specimens, manufactured in four independent trays (four units each). These parts were located on the corners of the tray, so that the effect of part location upon QI could be observed through results' analysis. Once the specimens were manufactured, parts were removed by mechanical means and sand blasted before CMM verification at the laboratory. Table 3 collects the average values of the three measurements performed for each QI and each part.


**Table 3.** Standard Quality Assessment results.

Results clearly indicate that QR would not be fulfilled under standard manufacturing conditions. DS average value (1.413 mm) indicates that slots tend to be wider than expected (13 μm wider than the correspondent QR upper limit). Additionally, DS standard deviation (0.040 mm) indicates an unexpectedly high variability. This indicates that the process is not under control and standard configuration would not allow for batch production. Similar conclusions can be derived from the

analysis of the other QR, since none of them are fulfilled under standard conditions. Geometric indicators (PD, FSD and FSF) clearly exceed the desired limits, while simultaneously presenting very high values for the standard deviation. Finally, DE achieves the desired quality in nine out of sixteen parts. Nevertheless, DE standard deviation (0.060 mm) clearly indicates that an unacceptable percentage of parts would not fulfil this condition during batch manufacturing.

Consequently, Standard Quality Assessment reveals that QR are far from being fulfilled. This means that the methodology should move onto the second step: Process Optimization.

#### *3.2. Step 2: Process Optimization*

The objective of Process Optimization is to work exclusively upon process configuration parameters to fulfil quality requirements. This means that the level of significance that variations in controllable factors exert upon variation in QI values must be established. At this stage, controllable factors have been reduced to Layer Thickness, Part Location, Type of Support, Thermal Treatment and Sand Blasting. A DOE considering five parameters could be performed at this point to gain a statistical assessment of each factor's relative influence upon QI. Nevertheless, an analysis of factor influence likeness has been previously performed, to evaluate if the number of factors could be reduced in the first iteration of Process Optimization


As a result of the analysis, only two factors were left for Process Optimization: Type of Support and Thermal Treatment. In order to check if those factors have a real influence upon Quality Indicators, it was decided that two additional trays (Tray 05 and Tray 06) should be manufactured using solid support (instead of a lightweight one) and applying a thermal treatment before releasing parts from the tray. Thermal treatment was intended to release thermal residual stresses of parts and was carried out following the recommendations of the material supplier (EOS) and manufacturer (PRODINTEC), by maintaining the tray at a 482 ◦C during four hours in a Nabertherm oven. Then, the parts and tray were left to cool at room temperature before being taken to a sawing operation. Once each part was released from the tray, it was taken to a Computer Numerical Control (CNC) milling machine to complete support removal. An *ad hoc* designed jig was used to prevent part deformation during milling. When the overall process had finished, parts were verified with the CMM using the same procedure as for trays 01 to 04. Comparisons between values of QI calculated from trays 01 to 04 (lightweight support/no thermal treatment) and 05 to 06 (solid support/thermal treatment) are provided in Figure 9.

**Figure 9.** Comparison of QI measures for trays 01 to 06: (**a**) *DS*; (**b**) *DE*; (**c**) *FSR*; (**d**) *FSF*; (**e**) *PS*.

Results indicate that the variability observed during the first step was related to the factors included in this Process Optimization step, since the QI measured for parts from trays 05 and 06 are clearly more uniform. This result is especially remarkable in the case of QI derived from geometric tolerances, since all the parts in the new trays fulfil QR for *FSR*, *FSF* and *PS*.

In the case of dimensional requirements, none of the new parts fulfil QR for *DS*, whereas only six parts fulfil the requirement for *DE* (although *DE* values are particularly close to the lower acceptable value). Nevertheless, the most important fact about measured dimensions is that variability seems to be significantly lower than that observed during the first step. Standard deviation results for dimensional QI pointed to the possibility of fulfilling QR by means of an optimization of design parameters.

Table 4 provides the measurement results of these eight specimens. The conclusion of this analysis is that the resection guide must be manufactured using solid support structures and that a thermal treatment, like the one described above, must be applied, and both elements have been thereafter incorporated to the Optimized Process Configuration. Manufacturing using this configuration directly allows the fulfilment of geometrical QRs, but it is still clearly unsuccessful in the case of the dimension of the slot and external width.


**Table 4.** Optimized Quality Assessment results.

Reaching this point, no additional QI improvement could be reasonably achieved by means of process parameters without acting upon design. Consequently, the methodology moved to the third step: Design Optimization.

#### *3.3. Step 3: Design Optimization*

Design Optimization can affect both dimensional and geometric tolerances, and the level of complexity required depends on the results observed during the previous stages. In this case study, geometric QR have been fulfilled via Process Optimization, so Design Optimization must focus on dimensional QR. Dimensional optimization implies building a mathematical model capable of accurately predicting the value that each QI would reach, as a function of those controllable factors whose significance has been considered relevant within the scope of the problem. In this case, study of both *DS* and *DE* present low variability after the Process Optimization step (*DS* standard deviation is 9 μm and *DE* standard deviation is 12 μm). This suggests that a unique linear compensation of designed theoretical values for all the parts within a tray could be applied. In the simplest formulation, the average deviation of both QI with respect to correspondent theoretical dimensions could be calculated and the design parameters modified accordingly, assuming linear behavior of results.

Nevertheless, although this could be the optimal approach for *DE*, it would not be equally recommended for *DS*. Quality Requirement for *DE* has a 100 μm range; consequently, uniform compensation should reasonably get most of the parts within QR. However, the Quality Requirement for *DS* has a 50 μm range. In order to achieve further improvements of *DS*, it was decided that the level of significance of remaining controllable factors (part location on the tray, with respect to *X* and *Y* axis) upon *DS* measures variability has to be verified. Although these factors have not shown significance for parts manufactured under Standard Process Configuration, the reduction in variability achieved via Optimized Process Configuration could have modified this circumstance.

Accordingly, a two-level full-factorial 2<sup>2</sup> DOE has been defined. Possible curvature effects have been taken into account by including two central points. This design allows for using data from trays 05 and 06. Part location has been coded according to a virtual *XY* origin ideally placed at the geometric center of the tray.

Parts located on the left side of the tray have been coded as *X* = −1, whereas those on the right side have been coded as *X* = 1. Similarly, parts located closer to the door have been coded as *Y* = −1, whereas those far from the door have been coded as *Y* = 1. Central locations have been coded *X* = 0 and *Y* = 0. Table 5 contains the structure of experiments for the DOE and the measured values for QI.


**Table 5.** Design of experiments (DOE) structure and results.

*DS* results (Table 5) have been processed using Minitab 17 statistical software to obtain variance analysis. Results are reflected in Table 6.


**Table 6.** *DS* Variance Analysis.

Analysis of variance points out that the location along the *X* axis has a significant influence upon *DS* values. Additionally, although the *Y* location appears non-significant, the interaction of *X* and *Y* has also been found to be significant. This means that variance in *DS* could be modelled by considering the location of parts within the manufacturing tray, with respect to both *X* and *Y* axes, since a linear relationship between them could not be discarded. Figure 10 contains an explicative Pareto chart of standardized effects, where the relative significance of factors *A* (*X*), *B* (*Y*) and interaction *A*\**B* (*X*\**Y*) can be observed. Additionally, the interaction plot for *DS* helps to explain the effect of each factor upon *DS*. Parts located to the left tend to have a wider slot than parts located to the right. Parts located closer to machine door are expected to have wider slots than parts located far from the door when the left side of the tray is analyzed, but this behavior is flipped (wider slots for distant parts) when parts manufactured on the right area of the tray are analyzed.

**Figure 10.** *DS* statistical graphs: (**a**) Pareto chart of the standardized effects; (**b**) Interaction plot.

This behavior illustrates the significance of *X\*Y* interaction, and indicates that modelling *DS* variability would have to include both *X* and *Y* locations as parameters.

Additionally, values for center points indicate that a linear relationship between location and DS should be expected, since there is no evidence of curvature. Although there was no need to analyze *X\*Y* significance upon *DE*, this task could be carried out without additional experiments. Consequently, an additional analysis of variance has been performed for *DE* using data from Table 5. Results indicate that neither *X* nor *Y* have any significance on DE variability. This result implies that observed variability cannot be explained by means of part location within tray, so these factors should not be taken into account when building a predictive model for DE. Figure 11 provides a Pareto chart for *DE*.

**Figure 11.** Pareto chart of the standardized effects for *DE*.

In sum, although the part location according to *X* and *Y* axes should be taken into account when predicting *DS* variability (values would differ significantly for different locations), they should not be considered for *DE* (values would be similar, independent of part location). Accordingly, no predictive model is required for *DE*. Instead, a simple linear compensation of the average values of correspondent QI will be used in the present case study.

#### 3.3.1. Predictive and Inverse Models

To elaborate the predictive model for DS, some considerations must be given:


Cross-influence of *DSD* and *DED* has been addressed in this case study by focusing on the predictive model for *DSM*, and calculating *DED* compensation from average values of *DEM* obtained from those parts used to construct the *DSM* model. Since the distribution of *DSM* with part location have previously shown no curvature, a simple polynomial expression has been used for predictive model *DSM* = *f*1(*X*,*Y*, *DSD*) Consequently, *DSM* values should be predicted as a function of *X* location, *Y* location and *DSD*, being *a1*, *a2*, *a3* and *a4* coefficients that minimize the adjustment error of such a function (1).

$$DS\_M = a\_1 + a\_2 \times X + a\_3 \times Y + a\_4 \times DS\_D \tag{1}$$

To calculate these coefficients, a new tray arrangement was defined so that the theoretical distance between slot parallel surfaces could also be taken into account. Consequently, two additional trays were defined and manufactured: the first one includes six parts, four located at the corners of the tray and the other two at the center, with a nominal *DE* of 1.350 mm. The second one follows the same distribution, but the width of the slot has been reduced to 1.250 mm. These limits have been selected according to previous results, which show that *DEM* tends to be noticeably higher than *DED*. Consequently, optimized values for this parameter could be reasonably expected to be smaller than the initial ones and probably within the 1.350 to 1.250 mm range. Parts were thereafter measured and results can be found in Table 7.


**Table 7.** *DS* values used for predictive model construction.

Model coefficients have been calculated by means of a least square iterative method and results are provided in Table 8.



Once the predictive model was defined, an inverse model was constructed. This inverse model should provide an optimized design value for *DS* (*DSO*), so that the correspondent measured value *DSM* of the manufactured part is as close as possible to the theoretical (optimal) value for *DS* (*DST*). This objective could be achieved by means of an optimization problem (2). Note that, in the original design, *DST* and *DSD* were equivalent, whereas, after optimization, each part would have a different *DSO*.

$$\text{minim}error = \min \left( DS\_M - DS\_\Gamma \right)^2 = \min \left( f(X, \mathcal{Y}\_\prime D\mathcal{S}\_\mathcal{D}) - D\mathcal{S}\_\Gamma \right)^2 \tag{2}$$

Consequently, the value that minimizes such functions should be an optimized *DSD* (denoted as *DSO* for clarification purposes). This problem could be efficiently solved by means of an interior-point method, taking advantage of the easily calculable derivatives of the function. In the present work, MATLAB *fmincon* command has been used to determine the optimal design values for *DS*.

Finally, *DE* compensation value was calculated from *DEM* results obtained from Table 7 parts, so that the average value reflects possible variations derived from *DED* modification. Accordingly, a *DEM* average value of −0.043 mm has been calculated and linear compensation should provide an optimized value of 4.532 for *DEO*.

#### 3.3.2. Optimized Design Quality Assessment

To evaluate fulfilment of QR once the design was optimized, an inverse model was used to design a verification tray with nine parts. This tray included six positions that had already been used to elaborate the prediction model and four additional intermediate ones (never used before). According to the predictive and inverse models, the design dimensions of slots were different for every individual part, whereas *DEO* is unique (4.543 mm) for all parts. Table 9 includes design values and measured results.


**Table 9.** Optimized Design Quality Assessment results.

As can be observed, QR have been fulfilled for every manufactured part in both *DE* and *DS*. To illustrate the evolution of QI from the Standard Process Configuration to the last step of the optimization procedures, Figures 12 and 13 provide two comparative histograms of *DSM* and *DEM*.

**Figure 12.** Histograms of *DSM* values: (**a**) Initial; (**b**) Optimized.

**Figure 13.** Histograms of *DEM* values: (**a**) Initial; (**b**) Optimized.

It can be observed that improvement was achieved by two effects: centering the average value of slot width with respect to the limits of the correspondent QR, and reducing initial variability to the extent that an extremely high percentage of parts should fulfil required quality. In fact, standard deviation within the verification tray has been reduced to only 5 μm, whereas its value calculated for the Optimized Process Quality Assessment was 9 μm. QR for DE has also been achieved via Design Optimization but, since the same dimensional compensation has been used, standard deviation presents a similar value (9 μm).

According to these results, batch production could commence once the implications of Process and Design optimization steps have been incorporated into production configuration.

#### **4. Discussion**

The proposed framework allows for combining the advantages of different approaches analyzed in the literature review. Some of the works related to the error analysis and error prevention [6,7,14] could be incorporated under our approach to the Process Optimization stage. In fact, once consolidated, their conclusions on factors' influence upon part quality could be part of an intensive knowledge base to simplify decision-making procedures in AM. Similarly, the mathematical methods proposed for error prediction used in works devoted to error correction [17,18,22] could be easily incorporated into the Design Optimization stage. However, the definition of this framework makes it unnecessary to apply optimization to the whole part, but instead is focused on FoS affected by dimensional tolerances

and their related geometrical tolerances. A global compensation of geometric distortions [22,23] could be researched, to see if it is preferable to our proposal of a specific compensation, but there is a risk of orientating efforts to models that, despite being able to improve overall dimensional and geometric quality, failed to fulfil a specific tolerance. Nevertheless, both approaches should not be considered exclusive, since it is possible to anticipate that, in the future, machine learning and artificial intelligence approaches [10,21] will both be incorporated to machine control systems. The proposed framework would check if specific QR have been fulfilled once optimization procedures have been incorporated into machine technology and, in case embedded optimization rules are still not enough to match tolerances, it would provide a methodology that manufacturers could easily follow. The degree of complexity of the tasks included in the proposed framework is highly dependent on each particular part's design and process characteristics, but it encourages taking advantage of available knowledge to simplify experimental effort. Models capable of accurately predicting dimensional and geometric errors grow in complexity with the number of factors contemplated, so an approach that delays simulation efforts until problem complexity has been reduced (minimizing factors) will probably be more useful to end-use manufacturers in the short term. Further research should be done in order to provide detailed rules on how some decisions should be adopted. Quantitative evaluation of the possible consequences of opting for one process configuration alternative could help staff to reach higher levels of objectivity while minimizing subjective decisions. Moreover, defining a model that describes all available possibilities in terms of statistical analysis and mathematical modelling, while simultaneously helping staff to decide which of these tools should be preferable for a particular situation, would be of great use.

#### **5. Conclusions**

The achieved results support the usefulness of a systematic framework for dimensional and geometric quality enhancement of additively manufactured parts. Work Analysis has permitted a reasonable understanding of the role of different influence factors, grouped according to controllable and non-controllable categories, to define both the Operating Space and the standard process configuration. An initial evaluation of QR fulfilment by means of a verification procedure and a test set provides an idea of the differences between measured QI values and QR objectives. Sorting controllable factors according to their relative influence upon QI values helps to simplify the Operational Space and drives testing to the most promising configurations. This contributes to a reduction in experimental effort and helps save costs and time. A balance between know-how and experimental effort should allow for an improvement in part quality that could eventually make further development unnecessary. Nevertheless, once the process's capabilities have been exhausted, the possibility of working upon the design parameters, or even upon the parameterization model itself, could take part quality to a higher level. The application of this framework to a quasi-industrial case, involving dimensional and geometrical optimization of surgical-steel tibia resection guides, helps explain the proposed workflow in more detail. In fact, L-PBF manufacturing of these inserts has an approximate cost-per-unit of 37 €, according to PRODINTEC, with an approximate production rate of 1.6 units per hour (based on 50 specimen trays). This cost is truly competitive with the conventional manufacturing alternatives, ranging from 100 to 300 €. These results help to reinforce the idea that the proposed framework could contribute to the global objective of AM quality improvement.

**Author Contributions:** Conceptualization, N.B. and D.B. methodology, N.B., D.B. and Á.N.; validation, B.J.Á. and P.F.; investigation, N.B., B.J.Á. and P.F.; data curation, B.J.Á., Á.N and P.F.; writing—original draft N.B., D.B. and B.J.Á.; project administration, D.B.; funding acquisition, D.B.

**Funding:** This research was funded by the Asturian Institute for Economic Development (IDEPA) and ArcelorMittal as part of the RIS3 strategy, grant number SV-PA-15-RIS3-4.

**Acknowledgments:** The authors wish to thank IDEPA and ArcelorMittal for their support and also acknowledge Pablo Suarez from Teknos Biomedical Engineering and Ignacio Dosil from PRODINTEC for their advice.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Mechanical Properties of 3D-Printing Polylactic Acid Parts subjected to Bending Stress and Fatigue Testing**

#### **J. Antonio Travieso-Rodriguez 1,\*, Ramon Jerez-Mesa 2, Jordi Llumà 3, Oriol Traver-Ramos 1, Giovanni Gomez-Gras <sup>4</sup> and Joan Josep Roa Rovira <sup>3</sup>**


Received: 25 October 2019; Accepted: 19 November 2019; Published: 22 November 2019

**Abstract:** This paper aims to analyse the mechanical properties response of polylactic acid (PLA) parts manufactured through fused filament fabrication. The influence of six manufacturing factors (layer height, filament width, fill density, layer orientation, printing velocity, and infill pattern) on the flexural resistance of PLA specimens is studied through an L27 Taguchi experimental array. Different geometries were tested on a four-point bending machine and on a rotating bending machine. From the first experimental phase, an optimal set of parameters deriving in the highest flexural resistance was determined. The results show that layer orientation is the most influential parameter, followed by layer height, filament width, and printing velocity, whereas the fill density and infill pattern show no significant influence. Finally, the fatigue fracture behaviour is evaluated and compared with that of previous studies' results, in order to present a comprehensive study of the mechanical properties of the material under different kind of solicitations.

**Keywords:** additive manufacturing; 3D printing; fused filament fabrication; flexural properties; fatigue; PLA

#### **1. Introduction**

Manufacturing through fused filament fabrication (FFF) or 3D-printing is a phenomenon that has drastically changed the way manufacturing is understood, mainly during the last decade [1]. The interest comes from the clear advantages that this group of technologies presents with respect to traditional manufacturing technologies; that is, great freedom of design and innovation capacities, a stronger connection between design and manufacturing, or the ability to manufacture unique pieces [2]. In addition, additive manufacturing (AM) systems have been easily implemented in domestic or low-scale manufacturing environments as a cheap and easy manufacturing technology.

Regardless of the rapid expansion of AM, the problem related to the identification and prediction of the mechanical behaviour and physical characteristics of the final pieces has been the main handicap for its application in industrial environments or final pieces. This difficulty lies in the fact that the parameters to be defined during the manufacturing process are numerous and interact with one another; and, on the other hand, because of the anisotropy of the material, caused by the high influence of the filament orientations in the manufacturing space [3]. Furthermore, anisotropy also originated thanks to the difference between the bonding forces between strands of the same layer (intralayer) and between layers (interlayer) [4]. For these reasons, the orientation of the layers is a key parameter to be defined when taking into account the work conditions of the piece.

According to Bellehumeur et al. [5], the mechanical resistance of parts is the result of the addition of three factors: the resistance of the filaments, the resistance of the union between filaments of the same layer, and the resistance of the union between layers. The inherent resistance of the filaments mainly depends on the mechanical properties of the raw material and the strength of the joints depends on the cohesion between filaments. This is proportional to the thermal energy of the filaments when they come into contact when being placed. The union is a local sinter in which polymer chains are shared. This process is applicable to all joints, between layer threads of both the same layer and different ones.

The authors Gurrala and Regalla [6], Gray et al. [7], and Zhong et al. [8] agree that the orientation of the layers must be coincident with the directions of the expected service loads to optimize the mechanical properties. In contrast, in compression forces, owing to the buckling effect, the fibres tend to bend. Therefore, the fibres should be oriented perpendicular to the load in this case [9].

This same effect of the orientation of the layers on the mechanical properties of the workpieces, has also been observed in other processes of AM, as in the technology of laminated object manufacturing (LOM), according to Olivier et al. [10]; selective sintering by laser, as reported Ajoku et al. [11]; or stereolithography presented by Quintana et al. [12].

Another parameter with great influence on the mechanical properties is the height of the layer. When the layers have a lower height, the parts show an overall better cohesion between layers, because the contact surface is greater and the empty space between filaments is smaller. This effect improves the transport mechanism of thermal energy, favouring the welding between wires, as found in the work of [9]

On the other hand, the thickness or width of the extruded filament is also a parameter that significantly influences the mechanical behaviour. It has a great impact on the transport mechanisms of thermal energy, which will affect the cohesion of the threads, according to the study proposed by Wang et al. [13].

The printing strategy determines the paths of the machine head in the creation of the piece. Within this context, the printed pieces are composed of two characteristic zones: the contour and infill. The outline is the skin that delimits the piece and corresponds to the outer perimeters. The infill is the one formed by the trajectories that the nozzle follows to fill the empty space that remains inside the contour, as depicted in Figure 1.

**Figure 1.** Section of a piece printed with two types of fill patterns. Left: honeycomb, right: linear.

Generally, in each layer, the contour is first performed followed by the internal filling with the selected printing strategy. Each one provides different mechanical properties. In the present work, the influence of several patterns shall be studied, as well as different infill densities, to assess their impact on the workpiece flexural behaviour.

The printing velocity is also a modifiable parameter. It can be defined for each printing zone, being independent for the contours, fills, and upper and lower layers. The velocity will be a parameter of study in this work since it has influence in the process of melting and solidification of the filaments. In addition, it affects the rate of extruded material.

Considering the aforementioned base of knowledge about FFF, this paper aims to study the influence of the manufacturing parameters on the mechanical properties of pieces made of polylactic acid (PLA) manufactured by FFF. Specifically, the flexural mechanical properties of these parts are evaluated. The results obtained are also compared with the those obtained in a previous study by Gómez-Gras et al. [14] and Jerez-Mesa et al. [15], performed on the same material subjected to a different loading mode. The main novelty delivered by this paper is that it contributes to the enrichment of mechanical behavioural data regarding PLA material. So far, an extensive study about bending properties and their direct comparison to fatigue performance linked to process parameters has not been found in the literature. For this reason, the results presented in this paper complement other results regarding tensile or fatigue properties, presented by authors in previous references, as presented above. The makers and users of FFF machines often ask about the best way to manufacture their parts. The answer should be that printing parameters should be chosen according to the expected part behaviour; this paper contributes to enriching that answer.

#### **2. Materials and Methods**

In this paper, the flexural mechanical properties of PLA are assessed. The influence of the manufacturing parameters in these properties will also be analysed. Therefore, the first experimental stage explained in this paper comprises a series of four-point bending tests performed on prismatic test specimens, following the American Society for Testing and Materials (ASTM) D6272-2 standard [16].

To better understand the influence of the significant parameters, different images of the fractured areas were taken and subsequently analysed. In addition, to complement the fractography, a micro scratch test was performed, which helped to better understand the fracture mechanism of the pieces.

In a second experimental stage, a fatigue Whöler curve generated through flexural fatigue tests was drawn to analyse whether the best conditions obtained in the four-point bending tests also derive in good fatigue properties.

#### *2.1. Four-Point Bending Tests*

#### 2.1.1. Specimens Manufacture

The design of the specimens used in the study was done with SOLIDWORKS® Research Edition 2019 software (Dassault Systèmes, Vélizy-Villacoublay, France) and the models were filleted with Slic3r software (GNU Affero General Public License) [17]. Subsequently, they were manufactured in the domestic 3D printer, Pyramid 3Dstudio XL Single Extruder. Their geometry is shown in Figure 2, with dimensions according to the standard that governs the bending test. All manufactured specimens were submitted to a quality control, in which they were weighed and measured with a calliper. Therefore, they had to be validated before testing from a dimensional and constructive point of view. The resulting lengths, widths, and weights were statistically processed, and those specimens whose descriptors were out of the ±2% were considered not to comply and were immediately discarded.

**Figure 2.** Test specimen's geometry: 80 mm × 10 mm × 4 mm, according to the D6272-02 ASTM standard.

The material used in the manufacture of the specimens, as discussed above, is PLA. It is a biodegradable thermoplastic. The choice of PLA as the study material was based on the fact that it is the most used material in domestic 3D printing. In this case, the selected filament was manufactured by Fillamentum Company from the Czech Republic. It has a diameter of 3 mm and its extrusion temperature is around 210 ◦C. The technical information provided by the manufacturer is indicated in Table 1.


#### 2.1.2. Taguchi Experimental Design

To carry out the four-point bending study, the design of experiments (DOE) technique was used. The design consists of the combination of the printing parameters that are considered most influential in mechanical behaviour. Six parameters are included in the study, and three levels of each one are defined (Table 2). They were selected taking into account the bibliography studied, as well as the experience of previous work of the research group.


**Table 2.** Parameters and levels used in design of experiments (DOE).

Filament width: Determined by the diameters of the extrusion nozzles: 0.3, 0.4, and 0.6 mm. It defines the volume and surface of the extruded threads, as well as the welding surface between filaments (Figure 3A).

**Figure 3.** Schematic representation of the parameters used in the study: (**A**) filament width and layer height, (**B**) infill pattern and fill density.

Layer height: Describes the thickness of each layer and, therefore, the number of layers the printed piece will have. It affects the volume and surface of the threads, as well as the welding between layers. The manufacturing time is inversely proportional to the layer height. Thinner layers imply more layers to print and a longer production time (Figure 3A).

Fill density: Represents the amount of material that is deposited within the contours. It avoids relative movements between contours and gives robustness to the pieces. It also determines the distance between the inner threads and affects material consumption (Figure 3B).

Fill pattern: Defines the trajectories that the nozzle follows to fill the empty space within the contour. Each pattern will create a different interior geometry producing different mechanical behaviours (Figure 3B).

Orientation: The specimens will be printed in the direction of the three coordinate axes: X, Y, and Z, as shown in Figure 4. In this way, the stacking of the layers will be done in three different ways and their behaviour can be studied. Normally, the stacking direction is the most determinant factor in mechanical behaviour [18].

**Figure 4.** The orientation of the layers' stacking, in the manufactured specimens. (**a**) X-axis oriented; (**b**) Y-axis oriented; (**c**) Z-axis oriented.

Printing velocity: It determines the extrusion and deposition of the threads' velocity. The velocity is defined for each part of the piece (inner, external perimeters, inner threads, and so on) to optimize the manufacturing time. In this study, the same velocity was defined for all parts of the piece to homogenize its structure.

In this study, a Taguchi L27 DOE was used. This method has been applied successfully in other studies concerning the mechanical properties of FFF pieces [14]. Table 3 shows an orthogonal matrix with a specific combination of parameters used. The influence of these separately as well as their interaction will be studied.


**Table 3.** Orthogonal matrix of Taguchi L27 for the DOE.

The rest of the parameters that affect the conception of the test specimens remained constant.

#### 2.1.3. Experimental Setup

The tests were carried out on the Microtest EM2/20 universal electromechanical machine, with a capacity of 20 kN, displacement of 300 mm, and a speed range 0–160 mm/min. The force acquisition was performed with a load cell of 500 N and a precision of 0.03 N.

The test consists of placing the specimen of a rectangular cross section over two supports and loading it at two points by means of two loading rollers; each at an equal distance from the adjacent support point. The specimen is bent at a constant speed, until the external fibres break, or until the maximum deformation of the external fibres reaches a 5% elongation. The parameters used in the experiment are described in the D6272-02 ASTM standard; that is, a support span of 64 mm and a load span of 21.3 mm (Figure 5).

**Figure 5.** Diagram of the four-point bending test method, according to the D6272-02 ASTM standard.

The deflection value will be obtained through image processing. High-definition video capture is planned for all tests. That way, the displacement will be obtained through image processing, by following a marker painted on the lower fibre of the specimen. The displacement will be determined to calculate the overall deflection (Figure 6). On the other hand, the force applied by the loading rollers will be measured with a load cell. The objective of data processing is to create the stress–strain curve of the specimens [19]. From the obtained curve, the following results will be extracted: Young's modulus (E), elastic limit (Rp0.2), maximum strength (σmax), and maximum deformation (ε).

**Figure 6.** The installation used to perform the four-point bending tests.

The test method used contemplates two different types, which differ in the test speed according to the behaviour of the test piece.

Type A. Used in test specimens that break with little deflection.

Type B. Used in the test specimens that absorb large deflections during the test.

The Type A test will end when breakage is detected in the outer fibres of the test pieces, and the Type B test will end when specimens break or the deflection D = 10.9 mm, according to measurements of the specimens and the parameters used.

A previous experimental testing was performed to validate the adequacy of the described method. From these experiments, it was detected that specimens printed in the direction of the *Z*-axis do not admit deflection, and present brittle failure, while the specimens printed in the direction of the *X*- and *Y*-axes admit large deflections. The summary of the test types can be seen in Table 4.



#### 2.1.4. Data Analysis

The data analysis was processed by following the steps described as follows:

1. Separation of the frames of the High Definition videos of each test. The camera used registered the image at approximately 60 fps. The tests lasted between 45 s and 2 min, so, in each of the 108 tests, between 2700 and 7200 frames were processed.

2. Calculation of the specimen's deflection through the frames. Position markers were painted on the outer fibre of the specimen, where the maximum deflection occurs, and on the static rollers (Figure 7A). The difference between the final position and the initial one, between the most displaced marker of the specimen and the markers on the static rollers, is considered the maximum deflection (Figure 7B). This analysis was performed through a self-designed MATLAB® code (version 2018) with image processing functions.

The calculation of the stress that is generated in the specimen at each moment by means of Equation (1) is as follows:

$$S = \frac{PL}{bd^2} \tag{1}$$

where


**Figure 7.** Schematic representation of the data collection process during the tests. (**A**) Initial position of markers; (**B**) final position (red crosses) and initial position (green crosses) of the markers.

#### *2.2. Fractography and Scratch Test*

In order to analyse the influence of the parameters that were significant, a SMZ-168 MOTIC stereo microscope was used to observe the fractures surfaces. The most interesting fracture phenomena were photographed with a MOTICAM 2300 camera. Both equipment were manufactured by Motic®, Xiamen, China.

Also, micro scratch tests were conducted in a scratch tester unit (CSM-Instruments, Needham, MA, USA) (Figure 8A) using a sphere-conical diamond indenter with a radius of 200 μm. Tests were done under a linearly increasing load, from 0 to 70 N, at a loading rate of 10 mm·min−<sup>1</sup> and in an interval length of 5 mm, according to the ASTM C1624-05 standard [20]. Figure 8B shows the two different scratches per specimen that were carried out in order to observe the reproducibility of the induced

damage. Furthermore, the micro scratch tests were conducted in the longitudinal and transversal printing direction to observe the main plastic deformation mechanisms induced. Surface damage induced during scratch tests was observed by a desktop scanning electron microscopy (SEM) Phenom XL from ThermoFisher Scientific (Waltham, MA, USA) (Figure 8C).

**Figure 8.** Micro scratch test. (**A**) scratch tester unit; (**B**) specimen; (**C**) scanning electron microscopy (SEM) ThermoFisher Scientific Phenom XL.

#### *2.3. Fatigue Test*

To complete this study, it is proposed to analyse, in a second experimental stage, how cylindrical specimens behave when manufactured through the optimal parameter set found in the previous study, subjected to a rotating fatigue test. This will also allow the comparison with other values previously obtained for the same material using other printing conditions [14].

The rotating bending fatigue test consists of applying a variable bending moment on a cylindrical test piece of known dimensions that rotates on its own axis. In this way, alternative tensile and compressive stresses are generated in the external fibres in each rotation. The test was carried out on printed cylindrical specimens like the one shown in Figure 9. For the fabrication of the fatigue specimens, the same 3D printer was used.

**Figure 9.** Dimensions of the test specimens used in the fatigue test.

#### **3. Results and Discussion**

#### *3.1. Four-Point Bending Test*

Table 5 shows the results, for each printing configuration, of the stress-strain curve as the average results of the five repetitions and their standard deviation.

**Table 5.** Average results and standard deviations of the material properties. E: Young's modulus, Rp0.2: yield strength, σmax: maximum strength, ε: maximum deformation, Std: standard deviation for each property.


An analysis of variance (ANOVA) was performed on the dataset included in the Taguchi experimental array, for each parameter that describes the mechanical behaviour of the evaluated specimens. To validate the statistical relevance of the parameters included in the model, the p-value associated with the ANOVA was compared to a significance level of 5%.

One of the first observations derived from the experimental testing is that specimens printed in the *Z*-axis direction presented fragile failure, as their failure mode was governed by the lower resistance between layers deposited vertically, thus with a lower neck growth area between them. For that reason, the elastic limit (*Rp0.2*) associated with these specimens was by default considered equal to their maximum strength (σ*max*). This approach was necessary to perform the statistical analysis, and allows the brittle behaviour to be included in the statistical analysis.

Alongside the yield limit and the maximum strength, the Young's modulus and maximum deformation were considered as response variables to analyse the influence of the different parameters in the statistical study. The following subsections describe the influence of the different parameters on the considered mechanical properties.

#### 3.1.1. Young's Modulus

As a predictable result, the specimens oriented along the *Z*-axis direction present the lowest rigidity of all, owing to their described brittle behaviour, and thus can be orientation defined as the most influential parameter (Figure 10A). The highest deformation module in the elastic regime is defined by an orientation of the fibres along the *Y*-axis direction, because of the different pattern deposited in this direction with regards to the *X*–axis orientation.

**Figure 10.** Main effects of (**A**) means and (**B**) interactions on Young's modulus.

On the other hand, an increase in the value of Young's modulus occurs when the filament width increases, probably because of the higher inertia of the single filaments that restrict bending. This effect of higher inertia of the surface is also achieved by decreasing the layer height, as it derives in a higher value of Young's modulus. This effect could be related to the fact that porosity is decreased by a lower layer height (and, complementarily, stiffness is increased). Following the same line, the printing velocity proves to increase the stiffness of the specimen as it is lower, probably again by the increase of the overall stiffness.

Of all the tested parameters, both the fill density and the infill pattern had a negligible impact (p-value of the ANOVA test > 5%) and no clear trend, which seems to disagree with the previous analysis. However, it must be considered that the small size of the specimens was derived in a lack of filling, and the geometry was composed basically of boundary layers that have relegated the infill to a second plane in this experimental campaign.

Figure 10B shows that no significant interaction among parameters is observed, as the p-values of them are all greater than 0.05.

#### 3.1.2. Yield Strength

Figure 11 shows the influence of the printing parameters on the elastic limit. Again, the layer height and the infill pattern do not show a significant influence. The effect of the other parameters on the response follows the same pattern as in the case of Young's modulus. The most influential parameter again is the printing orientation. With the *Y*-axis orientation, the highest elastic limit is achieved, while the *Z*-axis orientation shows the lowest one. In addition, with the *X*-axis orientation, an intermediate value is achieved with respect to the other printing orientations. The layer height has an influence somewhat higher than that of the filament width, but in the opposite way; as the layer height decreases or the filament width increases, the elastic limit increases. Although the printing velocity has low relevance, a trend is observed: when the velocity decreases, the elastic limit increases.

When analysing the interactions between the different parameters, it is concluded that there is no significant interaction, as the *p*-values in each case are much higher than 0.05. The same happens for the rest of the parameters. This is positive because it means that the influence of the parameters on the response is independent of each other, at least in the ranges of values analysed.

**Figure 11.** Main effects of means for yield strength.

#### 3.1.3. Maximum Strength

The behaviour of the parameters follow the same pattern as the elastic limit case (Figure 12). The layer orientation is still the parameter with the greatest influence on the mean value, followed by the layer height, filament width, and printing velocity, with less influence. Fill density and infill pattern do not have a statistically significant influence.

**Figure 12.** Main effects of means for maximum strength.

#### 3.1.4. Maximum Deformation

Figure 13 reveals that the only significant parameter is orientation. The *X*-axis and *Y*-axis orientations cause the greatest elongation and the *Z*-axis orientation causes the smallest one. Filament width, layer height, fill density, and printing velocity do not present any pattern or proportionality. The honeycomb fill pattern produces the least effect. Regarding the signal S–N, the only robust parameter is again the orientation.

**Figure 13.** Main effects of means for maximum deformation.

#### 3.1.5. Summary

In Table 6, a summary of the analysis of the influence of each parameter under study on the different mechanical properties studied can be seen. More green checks indicate that the factor is more influential on the response. Three checks indicates that p-value < 0.01, two checks indicate that 0.01 < p-value < 0.04, one check indicates that 0.04 < p-value < 0.05. The red cross is assigned to the parameters that are not statistically significant (p-value > 0.05). The orientation is the most influential parameter in the zone of both the elastic and plastic behaviour of the pieces tested. The layer height and the filament width are also parameters that influence all of the properties studied, except for the maximum deformation. The same thing happens with printing velocity, but to a lesser extent. In Table 7, the optimum levels of each parameter are shown.


**Table 6.** Significance value of the parameters with respect to the answers.


Of all parameters, the lack of influence of infill density deserves a special mention. This observation has already been made by other authors, such as Admed & Susmel (2019) [21] and Andrzejewska et al. (2017) [22]. These authors explain that the mechanical properties of PLA specimens with a 100% infill density depend on three main aspects, namely, the mechanical properties of filaments, the bonding forces between layers, and bonding forces between filaments of the same layer. Decreasing the infill density derives in the loss of bonding strength between filaments of the same layer, regardless of the distance between filaments in the same layer, which is the direct effect of infill density reduction. That is, the effect of changing infill density is more conspicuous when reducing from 100% to any other value, hence the lack of relevance of decreasing it from 75% to 25%. Furthermore, we could add a second fact explaining the lack of influence of infill density on the results, which could be related to the fact that bending specimens are of reduced dimensions, meaning that their mechanical behaviour is governed by their skin and, to a much lesser extent, the infill, which is only comprised by a few layers.

The manufacturing orientation plays a vital role in defining the flexural behaviour of specimens, as stress is normal to the specimen section, and the orientation of the bonding area between filaments shall define the way in which the material processes the stress. This result contrasts with that obtained when the specimens are subjected to fatigue tests, where layer height is the most influential parameter owing to the fact that the limiting factor here is the prevention of crack propagation, and not bearing stress itself [9].

#### *3.2. Fractography and Scratch Test*

It was already noted that the main factor that determines the strength of the specimens is the orientation of the stacking layers. In Figure 14, the outlooks of the fracture section of some printed specimens in the different directions of the coordinate axes are compared. The specimens printed along the *X*-axis direction (Figure 14A) or along the *Y*-axis direction (Figure 14B) have a slight ductile behaviour with high elongation, good flexural strength, and high rigidity. This reaction is caused by the filaments being aligned with the main stress direction. The resistance depends on the strength of the intra-layer bond and the strength of the filaments. The stiffness and flexural strength are slightly higher in the *Y*-axis orientation specimens. The reason is once again the arrangement of the layers. Although the two orientations have the filaments parallel to the direction of the stresses, the specimens printed in the *X*-axis direction can become delaminated between layers when they are bent. The delamination is produced by the breakage of the weak interlayer bonds. The specimen becomes flexible and is unable to withstand the bending stress, although the intralayer bonds remain intact.

**Figure 14.** Breaking section, specimens with orientation in the (**A**) X-, (**B**) Y-, and (**C**) Z-direction.

In Figure 14C, the section of the rupture of a test specimen printed in *Z*-axis orientation is shown. These specimens have fragile behaviour and little deformation, low resistance to bending, and low rigidity. This is caused because the layers are oriented perpendicularly to the stresses generated in the specimen during the bending test. For this reason, failure has occurred in the weak interlayer weld without the affection of filament integrity.

The second most influential factor is the layer height, followed by the filament width. The smaller the layer height and the larger the filament width, the stiffer and more resistant to bending is the test specimen. This is directly related to the compactness of the threads and the welding between threads (Figure 15).

**Figure 15.** (**A**) Test specimen with a layer height of 0.3 mm and filament width of 0.3 mm. Test specimen 9\_2. Microscopic photography. (**B**) Test specimen with a layer height of 0.1 mm and filament width of 0.6 mm. Test specimen 21\_3. Microscopic photography.

Figure 15A shows the extreme case with the maximum layer height of 0.3 mm and minimum filament width of 0.3 mm. With these dimensions, the threads are cylindrical and produce low compaction and weak welding owing to the scarce contact surface between threads. On the other hand, as the layer height decreases and the filament width increases, the threads have a flat shape, with a larger welding surface. Figure 15B shows the optimal case with a minimum layer height of 0.1 mm and maximum filament width of 0.6 mm. In summary, the welding surface of the threads, where the micro-welds are produced between the chains of the polymer deposited at the beginning and those of the filament that is then deposited on it, is determinant in the mechanical behaviour. The greater the welding surface, the greater the rigidity and strength of the piece.

Figure 16 shows the micro scratch test tracks in both the (A) perpendicular and (B) parallel direction to the filaments, on the same piece printed in the X-axis direction shown in Figure 14A. It can be seen how, up to the tested force (70 N), the material deforms ductilely without cracking in the base material, as the indenter moves. It also looks like the burrs produced by the extruder are torn. The fact that there are no disclosures between filaments implies that the adhesion between them in the same layer is enough to resist the efforts applied during the test.

**Figure 16.** Micro scratch test: (**A**) perpendicular to the printing direction; (**B**) parallel to the printing direction.

The graph in Figure 17 shows the results of the micro scratch tests: (A) perpendicular and (B) parallel to the direction of the filaments in the range of test forces. The values of normal force, friction force, penetration depth, residual depth, and friction coefficient are clearly observed. While the value of the friction coefficient measured in the perpendicular test shows oscillations, owing to the abrasive wear of the burrs (see arrows in Figure 17A), in the parallel test, its value remains almost constant. It could be possible to sense that these pieces are not showing remarkable wear adhesive.

On the other hand, these burrs left during the extrusion process form channels on the piece surface. If it is true that this worsens the surface roughness of the pieces, they could be useful for retaining lubricant adhered to the sides of the burr ridges; more taking into account that they do not increase their friction coefficient too much, as shown in Figure 17.

#### *3.3. Fatigue Test*

The parameter that marks the difference between both curves in Figure 18 is the layer height, being 0.1 mm for the results of this study and 0.3 mm for the referenced study [14].

**Figure 17.** Micro scratch test results: (**A**) perpendicular to the printing direction; (**B**) parallel to the printing direction.

**Figure 18.** Wöhler curve for the results obtained in this study and those obtained in the work of [14].

Although the authors of [14] find that layer height is slightly significant—and although it seems that the following assumption holds: the higher the height layer value, the greater the improvement detected regarding resistance—this cannot be assured, as the errors calculated for the multiplicative factor and the exponent in both equations mean that they can be the same.

Therefore, although a dependence on the layer height is insinuated, the current data do not allow it to assert it.

#### **4. Conclusions**

The influence of the layer orientation, layer height, filament width, printing velocity, fill density, and infill pattern on the flexural performance of PLA specimens was studied through a Taguchi DOE. The following conclusions can be extracted:


9. Depending on the mechanical property to enhance, the combination of optimal parameters to use is different.

**Author Contributions:** Conceptualization, J.A.T.-R. and R.J.-M.; Methodology, R.J.-M., G.G.-G. O.T.-R. And J.J.R.R.; Software, J.L. And O.T.-R.; Validation, J.L., J.J.R.R. and R.J.-M.; Formal Analysis, G.G.-G. and R.J.-M.; Investigation, J.A.T.-R., G.G.-G. and O.T.-R.; Resources, J.A.T.-R. and J.L.; Data Curation, J.L. J.J.R.R. and R.J.-M.; Writing-Original Draft Preparation, J.A.T.-R.; Writing-Review & Editing, J.L., G.G.-G. and R.J.-M.; Visualization, J.L. And J.J.R.R.; Supervision, J.A.T.-R. And R.J.-M.; Project Administration, J.A.T.-R., Funding Acquisition, J.A.T.-R., R.J.-M. and J.L.

**Funding:** This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

**Acknowledgments:** J.J. Roa would like to acknowledge the Serra Hunter program of the *Generalitat de Catalunya.*

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


PLA polylactic acid

DOE design of experiments

ANOVA analysis of variance

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **An Experimental and Numerical Analysis of the Compression of Bimetallic Cylinders**

#### **Ana María Camacho 1,\*, Álvaro Rodríguez-Prieto 1, José Manuel Herrero 1, Ana María Aragón 1, Claudio Bernal 1, Cinta Lorenzo-Martin 2, Ángel Yanguas-Gil <sup>2</sup> and Paulo A. F. Martins <sup>3</sup>**


Received: 5 November 2019; Accepted: 5 December 2019; Published: 7 December 2019

**Abstract:** This paper investigates the upsetting of bimetallic cylinders with an aluminum alloy center and a brass ring. The influence of the center-ring shape factor and type of assembly fit (interference and clearance), and the effect of friction on the compression force and ductile damage are comprehensively analyzed by means of a combined numerical-experimental approach. Results showed that the higher the shape factor, the lower the forces required, whereas the effect of friction is especially important for cylinders with the lowest shape factors. The type of assembly fit does not influence the compression force. The accumulated ductile damage in the compression of bimetallic cylinders is higher than in single-material cylinders, and the higher the shape factor, the lower the damage for the same amount of stroke. The highest values of damaged were found to occur at the middle plane, and typically in the ring. Results also showed that an interference fit was more favorable for preventing fracture of the ring than a clearance fit. Microstructural analysis by scanning electron microscopy revealed a good agreement with the finite element predicted distribution of ductile damage.

**Keywords:** metal forming; bi-metallic; cylinders; compression; finite elements; experimentation; microscopy

#### **1. Introduction**

In recent years, there has been a considerable growth in the use of multi-material components due to their advantages over single-material components regarding the possibility of tailoring physical properties, improving stiffness and strength, reducing overall weight, and saving the number of parts and the assembly costs in mechanical systems made of multiple components. Reductions in weight can, for example, be achieved through the combination of materials with lower densities than the original ones. Significant cost savings in electric power systems used in modern hybrid and electric vehicles can also be obtained by combining materials with different electrical and thermal conductivities. Improvements in the surface integrity of components exposed to extreme conditions are also of special interest; namely, in applications subjected to high friction contacts (enhancing the tribological properties) or high corrosion environments, such as those existing in marine and chemical industries [1]. In fact, the range of potential applications of multi-material components is so wide that they can also be found in the production of high denomination coins for security and aesthetic reasons.

The interest in multi-material components is scientifically and socially recognized by its inclusion in different work programs of the EU Horizon 2020, in which the manufacturing of multi-materials by additive manufacturing (for research, transport, customized goods, or biomaterials), and the combination of commercial materials into multi-material components for industrial applications [2] were selected as key research topics. The importance of additive manufacturing is confirmed by the growing number of publications in the field, which are focused on both directed energy deposition (DED) and powder bed fusion (PBF) [3] based techniques. Laser engineering net shaping (LENS) [4–6] is a powder DED based process; laser metal deposition (LMD) [7] is a wire DED based process; and selective laser melting (SLM) [8] is one of the most promising PBF processes for fabricating multi-material components.

Still, there are limitations in the use of additive manufacturing that are similar to those found in the fabrication of multi-material components by welding (e.g., friction stir welding, and laser and explosive welding) [5]. In fact, joining of dissimilar materials suffers from the risk of formation of brittle intermetallic metallurgical structures, and thermal heating-cooling cycles give rise to residual stresses, distortions, and geometric inaccuracies [3,9]. Table 1 summarizes the main problems associated with the production of multi-material components by additive manufacturing, welding, and forming.


**Table 1.** Main limitations in the fabrication of multi-material components by means of additive manufacturing, welding, and metal forming.

<sup>1</sup> X: limitation associated to category of processes.

As seen in Table 1, metal forming successfully overcomes most of the difficulties that are found in the production of multi-material components by additive manufacturing and welding. The main problems are due to formability issues and to the risk of delamination because thermal effects do not play a role in the cold metal forming based process that is considered in this paper.

Despite this, formability studies on multi-material components made from commercial materials by means of metal forming are not very widespread in literature. Studies are mostly limited to bimetallic components made of two different metallic alloys, such as the publications on the extrusion of bi-metallic components [10–12] and on the combination of forming and joining to produce bimetallic bearing bushings [13].

Coin minting of bimetallic disks is probably the most well-known application in the field [14,15] and the technology was recently thrown to a higher level of complexity by the development of new bi-material collection coins with a polymer composite center and a metallic ring to generate innovative aesthetics and incorporate advanced holographic security features [16]. Finite element modelling was utilized to investigate the influence of the initial clearance between the polymer center and the metallic ring on the mechanics of coin minting and performance of the resulting force fit joint.

Other researchers like Essa et al. [17] discussed the possibility of producing bimetallic components or prefroms by upsetting, after concluding that some geometries with a good interfacial contact between the center and the ring can be successfully employed as preforms for further processing. A similar conclusion was made by Misirly et al. [18] after analyzing the open die forging of bimetallic cylinders with steel rings and brass and pure copper centers, and observing that pure copper prevents the formation of cavities at the center-ring interfaces.

A recently publish work by Cetintav et al. [1] on the compression of trimetallic cylinders with aluminum centers and steel, copper, and brass rings, focused on the improvements in mechanical properties and weight reduction that result from the utilization of multi-material components.

More recently, Wernicke et al. [19] developed a new type of hybrid gear made from aluminum and steel to obtain significant weight reductions and locally adapted mechanical properties without the need of performing subsequent heat-treatment processes.

In the meantime, there have also been other investigations in the field aimed at analyzing the deformation mechanics and predicting the compression forces in multi-material components. This is the case of Plancak et al. [20], who developed two special purpose analytical models to calculate the compression force and validated their predictions against experimental tests performed on bimetallic cylinders with centers and rings made from different commercial steels. These models were later improved by Gisbert et al. [21] to include shear friction.

Under these circumstances, this paper aims to analyze the formability of bimetallic cylindrical billets produced by compression by means of a numerical and experimental based investigation. Compression forces and accumulation of ductile damage were analyzed by means of a work plan including different shape factors and two assembly fits between the center and the ring (interference and clearance). Scanning electron microscopy (SEM) observations were included to identify the major defects and to correlate the location of these defects with the finite element predicted distribution of ductile damage after compression.

The problems of delamination included in Table 1 will not be addressed because these are mainly found when the compression forces are not applied perpendicular to the contact surfaces between the different materials to be joined, as in case of extrusion and rolling [17]. This is not the case in the present investigation.

#### **2. Materials and Methods**

#### *2.1. Materials and Experimental Work Plan*

The bimetallic cylindrical test samples utilized in the investigation have an aluminum alloy UNS A92011 center and a brass UNS C38500 ring (Figure 1).

**Figure 1.** Bimetallic cylindrical test samples and notation utilized in the paper.

The aluminum center and the brass rings were machined from commercial rods with 12 and 15 mm diameters, respectively. Both materials were utilized in their as-supplied conditions and their chemical compositions are listed in Table 2.


**Table 2.** Chemical compositions of the aluminum alloy UNS A92011 [22] and brass UNS C38500 [23].

The physical and mechanical properties of both materials are included in Table 3.

**Table 3.** Physical and mechanical properties of the aluminum alloy UNS A92011 [22] and brass UNS C38500 [23].


The density of brass is three times higher than that of the aluminum alloy but its beta metallurgical phase, which is very appropriate for applications with extreme contact pressures, limits its ductility in cold forming. The overall rigidity of brass is also higher than that of the aluminum alloy because the latter has a smaller yield stress and a smaller ultimate tensile strength (UTS), meaning that it requires less energy to be plastically deformed. The experimental work plan is summarized in Table 4 and made use of cylindrical test samples with different height to diameter ratios, H0/d0 (previously designated as the "shape factor") and two different types of assembly fit. The assembly fit (P1i) corresponds to test samples in which the center was mounted into the ring with interference. For this purpose, the center was pushed into the ring using the universal testing machine that was also used in the compression tests. The assembly fit (P2i) corresponds to test samples in which the center was mounted into ring with a clearance of 0.1 mm in order to ensure easy sliding between the two parts.


**Table 4.** Summary of the experimental work plan 1,2.

<sup>1</sup> The dimensional parameters (D0, d0, H0**)** are defined in Figure 1. <sup>2</sup> a,b,c,d,e: denotes the shape factor (H0/d0) of the sample.

Figure 2 shows the bi-metallic cylindrical test samples (notation according to Table 4) before compression. Two samples were prepared for each testing condition.

**Figure 2.** Bimetallic cylindrical test samples before compression. Notation in accordance with Table 4.

#### *2.2. Equipment and Experimental Procedure*

The compression of the bimetallic cylindrical test samples was performed in a universal testing machine Hoytom HM-100kN (Hoytom HM-100kN, Hoytom, S.L., Leioa, Spain) with control software Howin 32 RS (version 3.11, Hoytom, S.L., Leioa, Spain). A precision cut-off machine Mecatome P100 (Mecatome P100, PRESI, Brié et Angonnes, France) was utilized to prepare the test samples for analysis and micrographic observation after compression.

The experimental procedure consisted of the following steps:


#### *2.3. Finite Element Modeling*

Finite element simulations were carried out with the commercial finite element computer program DEFORM 3D. The compression die platens were modelled as rigid objects and the bimetallic cylinders were modelled as an assembly between two plastically deformable objects (center and ring). The center and ring were discretized by means of approximately 11,000 tetrahedral elements. The detail of the initial finite element meshes is provided in Figure 3a.

The center and ring materials (aluminum alloy UNS A92011 and brass UNS C38500) were assumed to be isotropic and their flow curves (true stress–true strain curves) are disclosed in Figure 4.

Friction was modelled by means of the law of constant friction. As explained by Essa et al. [17], it is not possible to accurately define the frictional conditions prevailing at the center-ring contact interface. But previous research in multi-material upsetting [1,17,18], also lead to the conclusion that variations of the friction factor in the range of 0 to 0.5 do not influence the overall deformation of multi-material components.

**Figure 3.** Finite element modelling of the compression of bimetallic cylinders: (**a**) Detail of the initial mesh; (**b**) Identification of the two paths (path 1 in green and path 2 in red) that will be later utilized in the presentation to analyze ductile damage.

**Figure 4.** Flow curves of the aluminum alloy UNS A92011 and brass UNS C3850.

Therefore, taking into consideration a previous study performed by the authors [24], which points to the same above-mentioned conclusion, it was decided to use a friction factor equal to 0.08 along the center-ring contact interface. The same study was utilized to define a value of 0.12 at the contact interfaces between the specimen and the upper and lower die platens.

The accumulation of ductile damage *D* was modelled by means of the Cockcroft–Latham criterion [25]. According to this criterion, fracture is supposed to occur when the accumulated ductile damage reaches a critical value *Dcrit*, for a given temperature and strain rate loading condition

$$D\_{crit} = \int\_0^{\overline{\pi}\_f} \sigma^\* d\overline{\pi}\_\prime \tag{1}$$

where σ*\** is the maximum principal stress, ε is the equivalent strain, and ε*<sup>f</sup>* is the equivalent strain at fracture.

The ductile damage distributions included in this paper are based on a normalized version of the Cockcroft–Latham criterion (1),

$$C = \int\_0^{\overline{\varepsilon}} \frac{\sigma^\*}{\overline{\sigma}} d\overline{\varepsilon} \,\tag{2}$$

where σ is the effective stress. The values of strain and stress were calculated at the centre of each tetrahedral element, and therefore, the values of the normalized accumulated ductile damage *C* were also accumulated at the centre of the elements.

Damage distribution along the two paths shown in Figure 3b were calculated during post-processing of results. Path 1 was taken at the contact interface between the deformed cylinders and lower die platen, and path 2 was taken from the middle plane of the cylinders after compression.

#### *2.4. Scanning Electron Microscopy (SEM)*

Microstructural observation and analysis of the test samples with interference (P1a to P1e) were carried out with a high-resolution scanning electron microscope of the Center for Nanoscale Materials (CNM) at Argonne National Laboratory. The equipment utilized was a Hitachi S-4700-II (Hitachi, Krefeld, Germany), with an electron dispersive spectroscopic (EDS) detector, Bruker XFlash 6160 (Bruker, Billerica, MA, USA).

P1a to P1e samples were microstructurally characterized because damage was observed but fracture did not occur. This approach allows one to assess the formability of this multi-material sample prepared with interference, so more useful information was obtained from the defects found in the microstructural characterization.

The results obtained from these observations were compared with the finite element predictions of accumulated ductile damage. Further validation of the finite element computations was performed by comparing the numerical and experimental force-displacement evolutions. This is displayed in the following section.

#### **3. Results and Discussion**

#### *3.1. Compression Forces*

Figure 5 shows the bimetallic cylindrical test samples after compression. As seen, the influence of the assembly fit only provides visible differences for the samples with a shape factor H0/d0 = 2 because of cracking in the samples where the center was mounted in the ring with clearance.

**Figure 5.** Bimetallic cylindrical test samples after compression. Failure by cracking is observed in sample P2e (Table 4).

The lack of visible differences in the other test samples with smaller shape factors H0/d0 was further confirmed by the finite element predicted evolution of force with displacement shown in Figure 6. In fact, the force-displacement evolution is only sensitive to the shape factors H0/d0, as in case of single-material (solid) cylinders—the higher the shape factor, the lower the compression forces. In other words, there is no influence of the type of assembly fit on the force-displacement evolution for the test samples with shape factors H0/d0 = 1, 1.25, 1.5, and 1.75.

**Figure 6.** Finite element predicted evolution of the force with displacement for the compression of bimetallic cylindrical test samples with different shape factors H0/d0 (Pa: 1.00, Pb: 1.25, Pc: 1.50, Pd: 1.75).

Figure 7a shows a comparison between the experimental and finite element predicted force-displacement evolution for the entire set of test samples included in Table 4.

The first conclusion to be taken from these results is the lack of influence of the type of assembly fit on the experimental evolution of the force with displacement, as it had been previously observed in finite elements.

The second conclusion is that the reason why finite elements are not able to predict the drop in force after cracking of the two test samples P2e is because modelling did not take crack propagation into consideration.

The third conclusion is that the overall agreement between experimental and numerical prediction of the force-displacement evolution improves as the shape factor H0/d0 increases. In the compression of single-material (solid) cylinders, this discrepancy was attributed to the influence of friction, which becomes more important and leads to more significant deviations, as the cylinders reduce their height—typically when the shape factor goes below 0.5 [26,27].

This type of influence was also observed in the compression of bimetallic cylinders, especially for test samples P1a-P2a and P1b-P2b with the lowest shape factors. Improvements of the numerical estimates would require tuning the friction factor for each shape factor H0/d0 in order to match the experimental results. This was not carried out because the actual differences between numerical and experimental results were considered not relevant for the overall aims and objective of the investigation.

The finite element distribution of effective strain after 3.5 mm displacement of the upper die platen is shown in Figure 7b. Effective strain values were obtained at the center of each tetrahedral element and interpolated between old (distorted) and new meshes during remeshing procedures [28]. As seen, the effective strain values are higher for small shape factors H0/d0 and the distribution is more homogeneous for high shape factors H0/d0. This result is interesting because it goes against the expected conclusion that cracks would be triggered in test sample P2e because its overall level of effective strain and its overall level of inhomogeneity would be the highest.

**Figure 7.** (**a**) Experimental and finite element predicted evolution of force with displacement for the entire set of test cases included in Table 4; (**b**) finite element predicted evolution of effective strain after 3.5 mm displacement of the upper die platen.

#### *3.2. Ductile Damage*

Figure 8 shows the cylindrical test samples cut along their axial cross-sectional planes, after compression. As seen, some of the samples with clearance fit (P2i) showed a permanent joint between the center and the ring after the compression—one sample for each shape factor. On the contrary, two of the samples with interference fit (P1a and P1c) showed separation between the center and the ring after compression. The latter result was attributed to the appearance of internal voids at the contact interface, as reported by Cetintav et al. [1], who previously observed the existence of such voids for height reductions of 30%.

**Figure 8.** Cross section of bimetallic cylindrical test samples mounted with (**a**) interference fit (P1i) and (**b**) clearance fit (P2i), after compression.

Because Essa et al. [17], also claimed the occurrence of voids when the ratio between the center and the ring diameters was higher than 0.6, it is not possible to claim a general design rule for obtaining bimetallic cylinders with permanent joints between the center and the ring after compression.

The distribution of accumulated ductile damage in both bimetallic and single-material cylinders made of the aluminum alloy UNS A92011 is disclosed in Figure 9. As seen, damage is higher in bimetallic cylinders and a discontinuity is also observed in the center-ring contact interface.

The accumulated ductile damage along paths 1 and 2 (Figure 3) is plotted in Figure 10. Results show that the higher the shape factor, the lower the damage in the center, especially for sample Pe, where damage is almost negligible due to the limited amount of inhomogeneous material flow (small amount of barreling).

Moreover, the mostly damaged region was found to occur at the middle plane of the outer ring surface, as previously claimed by Silva et al. [29]. The only exception is sample Pa, with the lowest shape factor, in which the highest value of damage, and therefore, the most critically damaged region, were also found at the intersection between the center-ring interface and the die platens (refer to path 1 in Figure 10). As expected, the cylinder center does not experience a significant amount of damage due to a nearly homogeneous material flow.

Now, focusing our attention of the test samples with the highest shape factor (samples Pe), one concludes that cracking in both samples P2e after a displacement of 3.7 mm is not compatible with the fact that the Pe samples are those presenting the smallest amounts of accumulated ductile damage. In fact, the finite element predicted damage was below 0.07 (Figure 11), and therefore, is more compatible with the absence of cracking observed in samples P1e than with the existence of cracks in samples P2e.

**Figure 9.** Finite element distribution of accumulated ductile damage in (**a**) bimetallic cylindrical test samples and (**b**) single-material cylindrical test samples made from the aluminum alloy UNS A92011 after 3.5 mm displacement of the upper die platen.

**Figure 10.** Finite element accumulated ductile damage as a function of the radial distance from the symmetry axis for paths 1 and 2 (Figure 4), after 3.5 mm displacement of the upper die platen: (**a**) Pa; (**b**) Pb; (**c**) Pc; (**d**) Pd; (**e**) Pe.

Despite the above-mentioned contradictory results, cracking in Figure 11 can only be explained by a combination between the maximum accumulated damage at the outer ring and at the intersection between the contact interface and the die platens. This explanation is not straightforwardly evident, but the experimental results allow concluding that the type of assembly fit (P1 versus P2) plays a key role on the development of cracks. In particular, interference fit is more favorable to preventing cracking of the ring.

**Figure 11.** Cross section of the sample P2e showing failure by cracking and the corresponding finite element prediction of ductile damage (after 3.7 mm displacement of the upper die platen).

#### *3.3. Microstructural Observations*

Figure 12 shows different defects (voids, cracks, and microcracks) found in the center and ring in the tests samples with interference fit (P1i). The surface observed corresponds to the middle plane because it is the most damaged region, as described in Section 3.2.

Table 5 exhibits a comparison between the defects detected by SEM and the finite element predictions of ductile damage. The comparison is very good.

Finally, Figure 13 shows, graphically, the locations of major presence of defects (maximum damage) observed by SEM and the damage location range predicted by the finite element analysis.

**Figure 12.** *Cont.*

Crack nucleation

**Figure 12.** *Cont.*

**Figure 12.** Microstructural observations in the center, and ring of the test samples' interference fit (P1i).


**Table 5.** Comparison between SEM observations and finite element predictions of ductile damage.

**Figure 13.** Representation of the locations with major presences of defects (maximum damage) observed by SEM, and the damage location range predicted by the finite element analysis. (**a**) Center; (**b**) ring.

#### **4. Conclusions and Future Work**

This paper looked at the compression of bimetallic cylinders from a combined damage and microstructural point of view. The cylinders were made from an aluminum alloy UNS A92011 center and a brass UNS C38500 ring with various height to diameter ratios ("shape factor ratios") and difference assembly fit tolerances (by interference and clearance). Ductile damage predictions were obtained from finite element modelling with a commercial finite element (FE) program, whereas the microstructural observations were carried out with a scanning electron microscope (SEM).

The comparison between the experimental and numerical predicted forces showed that the shape factor ratio influences the force-displacement evolution in a similar way to what is commonly found in the compression of single-material cylinders. Differences between experimental and numerical results were more significant for test samples P1a-P2a and P1b-P2b, having the lowest shape factors, due to variations in friction for large amounts of height reduction. In contrast, the type of assembly fit did not influence the overall force-displacement evolutions.

SEM observations of voids, microcracks, and cracks revealed a general good agreement with the finite element estimates of ductile damage. In particular, SEM observations detected voids, microcracks, and cracks in specific areas of the maximum predicted damage. Numerical simulations also showed that ductile damage is higher in bimetallic cylinders than in single-material cylinders made from the aluminum alloy UNS A92011, and that there is a discontinuity in ductile damage distribution at the contact interface between the center and the core.

The overall results allow concluding that the higher the shape factor, the lower the damage, for the same amount of displacement of the upper die platen. This is due to differences in material flow inhomogeneity that favor the test samples with larger shape factors, but it is in clear contradiction to the fact that cracking was only found in test samples P2e having the largest shape factors. The explanation for this discrepancy was attributed to the type of assembly fit. In particular, results show that mounting the center in the ring with an interference fit prevents the occurrence of cracking. This is the most favorable condition for application in multi-material forming by compression. The amount of interference-fit needs to be addressed in future work, together with conducting complementary microstructural analysis in order to find relations between microstructure and material flow.

**Author Contributions:** Conceptualization, A.M.C., C.B., and Á.R.-P.; formal analysis, A.M.C., J.M.H., Á.R.-P., A.M.A., and C.L.-M.; funding acquisition, A.M.C., Á.R.-P., and Á.Y.-G.; investigation, A.M.C., Á.R.-P., J.M.H., and C.B.; methodology, A.M.C., J.M.H., and Á.R.-P.; project administration, A.M.C.; resources, A.M.C., C.L.-M., and Á.Y.-G.; supervision, A.M.C. and Á.R.-P.; validation, A.M.C. and Á.R.-P.; writing—original draft, A.M.C. and Á.R.-P.; writing—review and editing, A.M.C., Á.R.-P., Á.Y.-G., and P.A.F.M.

**Funding:** This research was funded by the Annual Grants Call of the E.T.S.I.I. of UNED through the projects of references 2014-ICF04, 2015-ICF04, and 2019-ICF04. A mobility grant for junior researchers was also granted by MES (Manufacturing Engineering Society) to Álvaro Rodríguez-Prieto.

**Acknowledgments:** The authors would like to take this opportunity to thank the Research Group of the UNED "Industrial Production and Manufacturing Engineering (IPME)" for the support provided during the development of this work. We also acknowledge to Center for Nanoscale Materials (CNM), supported by the US Department of Energy, Office of Science and Office of Basic Energy Sciences under contract number DE-AC02-06CH11357. Paulo Martins would like to thank the support of by Fundação para a Ciência e a Tecnologia of Portugal and IDMEC under LAETA-UID/EMS/50022/2019.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Study of the Degree of Cure through Thermal Analysis and Raman Spectroscopy in Composite-Forming Processes**

#### **Juan A. García-Manrique \*, Bernabé Marí, Amparo Ribes-Greus, Llúcia Monreal, Roberto Teruel, Llanos Gascón, Juan A. Sans and Julia Marí-Guaita**

Institute of Design for Manufacturing and Automated Production, Universitat Politècnica de València (UPV), Camí de Vera s/n, 46022 València, Spain; bmari@fis.upv.es (B.M.); aribes@ter.upv.es (A.R.-G.); lmonreal@mat.upv.es (L.M.); r.teruel@upvnet.upv.es (R.T.); llgascon@mat.upv.es (L.G.); juasant2@upvnet.upv.es (J.A.S.); juliasetze@gmail.com (J.M.-G.) **\*** Correspondence: jugarcia@mcm.upv.es; Tel.: +34-963-877-622

Received: 8 November 2019; Accepted: 28 November 2019; Published: 2 December 2019

**Abstract:** The curing of composite materials is one of the parameters that most affects their mechanical behavior. The inspection methods used do not always allow a correct characterization of the curing state of the thermosetting resins. In this work, Raman spectroscopy technology is used for measuring the degree of cure. The results are compared with conventional thermal gravimetric analysis (TGA), differential scanning calorimetry (DSC), and scanning electron microscope (SEM). Carbon fiber specimens manufactured with technologies out of autoclave (OoA) have been used, with an epoxy system Prepreg System, SE 84LV. The results obtained with Raman technology show that it is possible to verify the degree of polymerization, and the information is complementary from classical thermal characterization techniques such as TGA and DSC; thus, it is possible to have greater control in curing and improving the quality of the manufactured parts.

**Keywords:** prepeg; carbon fiber; Raman spectroscopy

#### **1. Introduction**

There is currently a great demand from the aerospace, wind, nautical and automobile industries for the development of new high-performance composites. The cure characterization of the parts affects both the forming processes and the repair–maintenance processes. Cure time is generally calculated very conservatively to ensure complete curing of the part before removing it from the mold. However, this practice greatly slows manufacturing times, and in some cases, it damages the part. The "in situ" control techniques of the mold filling and cure monitoring provide enough information to reduce the injection and curing times. Biomedical applications are used to increase the thickness of the part's fillers; however, a high degree of cure could be difficult to achieve, reducing its mechanical properties and the biocompatibility [1].

The manufacturing processes of composite materials can be divided into two large families: the wet layup and the dry layup processes (see Figure 1). Wet processes are those where the resin and the fabric have been mixed and fabrics already impregnated with the resin are supplied. The chemical polymerization reaction is "frozen" by the low supply temperature (−20 ◦C) and its reaction will be activated by heat. These materials are formulated with an excess of resin that must be evacuated during the fiber-compaction phase. When the resin is heated, its viscosity decreases and can flow through the thickness of the part. Thanks to this resin displacement through the thickness, the layers that form the laminate will be joined together. Autoclave, mechanical press, or atmospheric pressure can be used for compaction. This paper analyzes the forming processes compacted only with atmospheric pressure, which are known as VBO (vacuum bag only) or OoA (out of autoclave). These processes are those that have an easier processing; however, sometimes they are very expensive due to the large amount of labor they need, the cost of raw materials, etc.

On the other hand, regarding the dry layup forming processes, the impregnation phase of the fabrics takes place inside the mold. Therefore, the manufacturing process has a filling phase, a melt front advance, injection points, vents, etc. Depending on the type of injection or the stiffness of the countermold, it will define the different forming processes that make up the family of Liquid Composite Molding (LCM). These processes have great advantages over wet processes because they allow manufacturing with the same quality but at a significantly lower cost. Depending on the resin used, they may need an oven curing stage where the resin cures. However, these require a greater knowledge regarding the behavior of materials and manufacturing techniques than wet processes.

**Figure 1.** General classification of composite manufacturing process.

The objective of any composite manufacturing process is to impregnate the fabric with resin in the most efficient way and evacuate all the pores or voids that reduce the mechanical properties of the parts. In this context, the simulation of the formation and displacement of the pores has been an important research topic in recent years. An extensive review on this topic can be found in (Pillai, 2004 [2]; Park and Lee, 2011 [3]). Autoclave curing processes (OoA, out of autoclave) and vacuum bag-only compression (VBO, vacuum bag only) processes are often considered equivalent, but there are some differences between them. OoA processes, as the name implies, are all processes that heal without using an autoclave, but it is possible to apply a compaction pressure as large as necessary, for example, using a hot press for curing. However, concerning compaction processes with vacuum bag only, the compaction pressure is limited to atmospheric pressure. In both cases, the cost reduction with respect to autoclave manufacturing is important, and production times can be greatly reduced. In both cases, the manufacture of components out of autoclave with the same quality as those obtained with the autoclave is a major technological challenge. In this context, it is necessary to know the behavior of the materials during the forming process, and numerical models adapted to the different material scales are needed. In all forming processes, either wet (prepregs) or dry (LCM), numerical methods that include capillary effects and/or resin–air interface modeling should be used. The most advanced models in these techniques are based on numerical schemes by biphasic finite elements [4] that use the Stokes or Darcy's flows models adapted to the biphasic behavior.

Nowadays, these models cannot solve the full geometry of the part on a macro scale, so reduced schemes are used for its numerical calculation. The equations to be solved are very similar for dry path processes where the flow to be studied is in-plane XY; as for the wet path, processes where the flow to be studied is basically through the thickness Z. The most traditional manufacturing process of high-performance composite parts is prepregs. These materials are made of fiber and resin and are supplied in sheets. To achieve the maximum mechanical properties, the laminate must undergo a compaction phase and a heat phase for resin curing. The material compaction is necessary to achieve the maximum fiber volumetric fraction, and in addition, the excess of resin with which the prepregs are manufactured must be evacuated. This excess of resin flows in through the thickness direction and is necessary to get the bond between each layer of the laminate. Maximum compaction is achieved using expensive autoclaves that are only accessible to certain high value-added industry such as aerospace or racing vehicles.

The prepreg forming process (OoA) basically consists of three stages: layup, vacuum closing, and curing. First, the prepregs are cut in the required number of layers and sizes to obtain the thickness and shape of the part. Preimpregnated layers are stacked manually or using automated techniques (automatic tape layup). Intermediate compaction steps are performed to improve the layers compaction by applying vacuum at room temperature (see Figure 2). Once the part has reached the required thickness, it is compressed under vacuum, and the curing cycle begins. During the compaction phase, physical phenomena occur that are normally neglected. These phenomena include capillarity, drainage, inhibition, or pore mobility [5].

**Figure 2.** Prepreg technology.

Therefore, the resin flows through the fabric and fills both the spaces between the tows and inside of the tows. The distances or gaps between the tows are around tenths of a millimeter, while the spaces inside the tows are 10–15 μm. Then, the problem needs a double-scale approximation, including the meso and the micro scale. Many authors have tried to solve this problem by modifying the permeability [6] and the subsequent treatment of flow equations as a traditional LCM problem. There is a difference of several magnitudes of order between the value of the permeability inside the tows (micro scale) and the space between the tows (macro or meso scale). As a result, the numerical simulation of the double-scale process is extremely complex. There are simplified models that estimate the saturation of the fabrics, but the work done so far does not consider many of the physical phenomena that occur inside the fabrics, resulting in them being inadequate for most applications of industrial interest. In this sense, it is worth highlighting the work carried out by Professor Veronic Michaud of the Ecole Polytechnique de Lausanne [7], who proposed a numerical model for the behavior of air permeability during the curing phase. Other researchers proposed the phenomenological characterization of the process, such as the research group of Professor S.G. Advani at the University of Delaware (USA) [8]. This approach requires the experimental data to have quantitative and qualitative results [9].

Composite manufacturers in the aerospace industry and other large industries, such as the automobile and wind industries, are looking for manufacturing processes out of autoclave (OoA) that can achieve 2% content in pores with less expensive and more efficient equipment. Autoclaves are used when the highest quality is needed in the final part, with a pore content of less than 2% and high glass transition temperatures (Tg). Aerospace autoclaves normally operate at 120 to 230 degrees Celsius within a nitrogen atmosphere at 7 bar pressure. It has been shown that processing with a preimpregnated vacuum bag only produces high porosity composite laminates, due to the air and moisture trapped in layers and between layers, which cannot be evacuated, and the lower (atmospheric) pressure cannot sufficiently compact. Moisture in prepreg can cause pores when processed only with the vacuum bag, but when processed in an autoclave, the higher pressure causes moisture to condense, suppressing pore growth. The first preimpregnated OoA was designed in the early 1990s for initial curing at low temperature (approximately 60 ◦C), followed by post-curing at high temperature (approximately 110 ◦C) [10].

The thermosetting resins used in the manufacture of high-performance composite (HPC) are initially processed in a liquid state, and by an exothermic curing reaction the polymer chains are crosslinked, forming covalent bonds of high hardness and chemical stability. When these properties are mixed with the appropriate fiber, they become an excellent composite material. The composite materials analyzed in this work are composed of a high modulus epoxy resin and long carbon fiber (HPC). The curing of this type of resin can be carried out by chemical methods, thermal methods, or a combination of both. The most commonly used method is the last one, since it allows greater control over the quality of the curing and reduces the manufacturing time. In all curing processes, not only the time and temperatures, but also the reaction rate must be controlled. There is an optimum in terms of cure temperature, which should coincide with the time when the reaction rate is higher. A higher temperature causes degradation of the material, rather than an increase in the reaction rate. The classic methods of resin characterization are differential scanning calorimetry (DSC) and dynamic mechanical analysis (DMA). These methods provide excellent information on the glass transition temperature, the heat generated by the chemical reaction, the cure rate, and the degree of cure. However, all these methods have the disadvantage of being destructive methods, and therefore can only be applied on a laboratory scale.

The cure reaction as a function of the manufacturing variables is a challenge for the analysis and design of processing operations. In addition, the physical properties of composite materials strongly depend on their microstructure and are directly related with the failure of the fibers or delaminations. Calorimetry will be used for the macrokinetic analysis of cure reactions. The cure reaction of epoxy resin will be analyzed also by means of thermal characterization using differential scanning calorimetry (DSC) and thermogravimetry (TGA). The vibrational properties of the composites were studied by Raman spectroscopy. The degree of cure of three samples of composite material with different levels of cure has been measured. The Raman spectroscopy vibrational method has been used and compared with those from thermal gravimetric analysis (TGA) and differential scanning calorimetry (DSC). It has been observed that complementary information can be obtained from classical trials, which allows progress toward non-destructive quality control in CFRP composites.

#### **2. Materials and Methods**

The coupons manufacturing begins with the consolidation stage. At this stage, the trapped air between the prepreg sheets is removed by vacuum sealing with a bag. This vacuum is applied for approximately 15 min at room temperature. Due to the stiffness of the prepreg layers, each of these consolidations are applied to 3–5 layers, depending on the total thickness of the part. In any case, the first layer is consolidated individually to ensure a good surface finish.

The SE 84LV is used for the manufacture of the coupons and can be cured at low temperature (85 ◦C) or, for faster molding of components, at 120 ◦C. Once the entire compaction sequence has been carried out, the mold is introduced into the oven (in-house manufacture) where the viscosity is reduced. The compaction pressure generates the flow of the resin, which mainly flows through the thickness of the part, and the resin excess is trapped in the absorption blanket. The polymerization reaction is activated by heat, and the oven must be programmed to control the resin reaction rate. The oven must control the temperature to avoid significant gradients inside the mold; therefore, the temperature rise ramp must be slow (approximately 1 degree/min). The oven temperature should be held for 10 h at 85 ◦C, and the tolerance must not exceed ±5 ◦C. Once the part is totally cured, quality control is recommended to ensure that no air has been trapped inside the part; the methods used are ultrasonic, stereography, thermography, etc.

A thermogravimetric study (TGA) of the samples has been carried out in order to simulate and optimize the percentage of volumetric fraction for the resin, fiber, and residue. Once the process conditions have been determined, the samples are introduced into a muffle furnace and the non-isothermal thermogravimetric multi-rate experiments are performed. Mettler-Toledo TGA/SDTA 851 (Mettler-Toledo, Columbus, OH, USA) equipment has been used. The samples weigh 10–11 mg and are heated in an alumina holder with a capacity of 70 μL. A heating curve from 30 ◦C to 1000 ◦C has been programmed with a rate of 10◦/min in a controlled flow of oxygen atmosphere (50 mL/min) to simulate the real manufacturing conditions. Each experiment has been repeated three times, and finally, the characterization was carried out with the software STAR® 9.10 from Mettler-Toledo. In accordance with ASTM D3171-11 [4], "Standard methods for constituent content of composite materials", the weight percentage of each constituent is obtained from the following expressions:

$$P1\left(\%\right) = \frac{P\_1}{P\_0} \cdot 100\tag{1}$$

$$P2\left(\%\right) = \frac{P\_2}{P\_0} \cdot 100\tag{2}$$

$$P\mathfrak{3}(\%) = \frac{P\_3}{P\_0} \cdot 100\tag{3}$$

$$\text{Residue}\left(\%\right) = \text{P1 -- P2 -- P3}\tag{4}$$

$$P\_0 = \text{initial mass} \tag{5}$$

$$P\_1 = \text{final mass} \tag{6}$$

$$P\_3 = \text{carbon mass}\tag{7}$$

$$P\_4 = \text{resin mass} \tag{8}$$

The degree of resin cure of the specimens has been measured by differential scanning calorimetry for later comparison with the results obtained from Raman spectroscopy. A Mettler Toledo DSC 822 DSC (Mettler-Toledo, Columbus, OH, USA) with samples of 4–6 mg weight was used and sealed in aluminium pans with a capacity for 40 μL. The programmed heating curve was from 25 to 300 ◦C with a heating rate of 10 ◦C/min for dynamic DSC scanning. The total heat of the reaction, HT, is obtained from the total area in the heat generation versus the time line graph. The degree of cure and the rate of the degree of cure will be determined in the isothermal scanning experiment. The measurements were made at temperatures from 125 to 145 ◦C with increments of 5 ◦C and a rate of heating of 10 degrees/min.

The isothermal curing curve is calculated from the total area enclosed by these exothermic curves. The samples are cooled and heated again (from 25 to 300 ◦C at 10 ◦C/min) after each isothermal

scan to obtain the residual heat of the reaction, RH. This value is obtained from the area under the exothermic peak in the resulting curve. To carry out Raman spectroscopy tests, a backscattering geometry by a confocal HORIBA Jobin-Yvon LabRam high-resolution micro Raman spectrometer (HORIBA Jobin-Yvon, NJ, USA) has been used. The test characteristics were 1200 grooves/mm grating at a 100-μm slit and 50× objective, in combination with a thermoelectrically cooled multichannel CCD (charge-coupled device) detector (spectral resolution below 3 cm<sup>−</sup>1). A solid-state laser with a power of 50 mW emitting at 532.12 nm has been used.

#### **3. Results and Discussion**

The degree of cure of three composite samples with different characteristics has been analyzed. Sample 1 is a non-cured sample, which is used as the reference. It is a carbon prepreg sheet with any degree of cure, corresponding to a sample at the beginning of the manufacturing process. Sample 2 has been semi-cured out of autoclave, and Sample 3 was completely cured also in OoA. Figure 3 shows the FESEM micrographs for the three studied samples with different degrees of cure. The effect of the cure time is evident in these micrographs, the resins covering the carbon fibers disappear with the evolution of the cure reaction. According to Figure 3, the diameter of carbon fibers was found to be about 6 μm.

**Figure 3.** FESEM micrographs of the samples with different degree of cure: (**a**) Non-cured, (**b**) Semi-cured, (**c**) Cured.

#### *3.1. Derivative Thermogravimetric (DTG)*

Derivative thermogravimetric (DTG) is a type of thermal analysis in which the rate of material weight changes upon heating is plotted against temperature. It is used to simplify reading the weight versus temperature thermogram peaks, which occur close together. The results of the DTG analysis for the semi-cured (Sample 2) and cured (Sample 3) samples are shown in Figures 4 and 5, respectively.

In both figures, three peaks corresponding to three phases were observed by the dynamic thermogravimetric studies in oxidative conditions, as specified in Table 1.

**Figure 4.** Derivative thermogravimetric (DTG) for semi-cured composite samples (Sample 2).

**Figure 5.** DTG for cured composite samples (Sample 3).

**Table 1.** Thermal degradation temperature. Samples 2 and 3.


It can be noticed that in Figure 4, the first phase starts at about 425 ◦C, the second phase reaches a maximum rate near 550 ◦C, and finally, the third phase is near 790 ◦C. The third phase corresponds with the degradation of the residue. For both samples 2 and 3, the initial (lost) masses are shown in Table 2 and the mass losses are presented in Table 3.

**Table 2.** Initial mass for Samples 2 and 3.



**Table 3.** Mass loss in the tests of Samples 2 and 3.

Table 4 displays the fraction percentages for Samples 2 and 3, after applying the cure phases from Equations (1)–(4).


P2 5 20 3 22 47 2 P3 58 43 61 41 16 63 Residue 2 1 2 121

**Table 4.** Fraction percentages for Samples 2 and 3.

The analysis of the decomposition process of the composites suggests that the early degradation step is due to the removal of the most volatile components and low molecular mass species, with a rate of about 35% for the three replicas of both samples. The second stage of the thermal decomposition would correspond to the scission of the polymer network, and the later decomposition stage would be related to the thermal degradation of the carbon fibers [11]. Table 5 displays the average and standard deviation values for the four studied fractions. The high standard deviations found for fractions P2 and P3 indicate a non-uniform curing process. This behavior is in good agreement with the composite internal double scale, as represented in Figure 6. The gap between fibers inside and between tows are around 5 μm and 500 μm respectively, and then the cure kinetic is different.

**Figure 6.** Micro and meso spaces between and inside tows.


**Table 5.** Standard deviation and average for Samples 2 and 3.

Differential scanning calorimetry (DSC) was used to characterize the degree of cure of the epoxy carbon composites. Figure 7 shows the exothermic heat curves of the epoxy system for Sample 1 at a heating rate of 10 ◦C/min. The area under the curve, which is calculated from the extrapolated baseline at the end of the reaction, was used to assess the total heat of the cure, giving a value of HT = 135 J/g. In general, the curing reaction of a thermosetting resin can spread over a wide temperature range. However, considering the sensitivity and response limitation to heat changes, the isothermal DSC analysis is usually run in a moderate curing range of temperature [12]. According to the dynamic curing DSC curves displayed in Figure 6, the drop on the weight is a Ti = 125 ◦C and it reaches a minimum at Tpeak = 145 ◦C. This means that the temperature for optimum curing is in the range 125–145 ◦C.

**Figure 7.** DSC measurements for Sample 1: exothermic heat curves at a rate of 10 ◦C/min.

The isothermal degree of cure was obtained from isothermal DSC curves (Figure 8). As expected, when the curing temperature increased, the time required to reach the minimum value became shorter. The isothermal heat of cure, ΔHi, was derived from the total area under the curve. The residual heat of reaction (ΔHR) is obtained by a cooling and heating process for each Tiso value. The values of ΔHi and ΔHR are shown in Table 6 at the Tiso temperature.

**Figure 8.** Isothermal DSC curves for composite samples.



Then, the degree of cure at Tiso can be calculated through two equivalent expressions:

$$
\alpha\_{\rm Hi} = \frac{\Delta \rm H\_i}{\Delta \rm H\_T} \tag{9}
$$

$$
\alpha\_{\rm Hi} = \frac{\Delta \mathcal{H}\_{\rm i}}{\Delta \mathcal{H}\_{\rm T}}.\tag{10}
$$

Table 7 shows the results obtained for the degree of cure, α, for each. As can be seen, there is no clear trend on the behavior of reaction heats, which suggests that a non-uniform process is occurring. This circumstance has also been observed in the TGA experiments, so it is interpreted that the information of both results is complementary and can help the interpretation of physical phenomena to be modeled with the cure equations.

**Table 7.** Values of DOC for Sample 2.


#### *3.2. Raman Spectroscopy*

Raman scattering measurements of the three samples (1, 2, and 3) were performed under the same conditions: two accumulations of 120 s of exposure time and a neutral 0.6 density filter. The use of an attenuator has been revealed to be necessary to avoid the radiation damage on the composite samples. The Raman spectrum shown for each sample is obtained by calculating the average of the three points of one fiber after checking that several fibers gave similar results. In order to obtain the optimal conditions to perform the measurements, different combinations of exposure time and attenuators have been studied. Once the tests are finished, no significant changes are observed in the surface of the samples measured by Raman spectroscopy.

Figure 9 shows the Raman spectra for the three composite fiber samples. The Raman spectra have been represented vertically to clarify their visualization. All samples have a peak located around 1352, 1585 and 1620 cm−1. The peak located at 1585 cm−<sup>1</sup> is a consequence of ordered or graphitic carbon (known as G band), while the peaks located at 1352 cm−<sup>1</sup> (D band) and 1620 cm−<sup>1</sup> (D' band) are assigned to disordered carbon atoms, which is usually explained by the double-resonance Raman mechanism in carbon [12,13]. With the present spectral resolution of our Raman spectrometer (below 3 cm<sup>−</sup>1), the D' band appears as a shoulder of the G band and cannot be separated. Therefore, only the values of the D and G bands were used in our calculations.

**Figure 9.** Raman spectra of three analyzed composite samples as a function of their degree of cure. The Raman spectra have been vertically shifted to improve the comparison among them.

The ratio between the intensities of both D and G bands gives us an idea of the crystalline order of the sample. The crystalline size, La, is derived from the Knight formula [14,15],

$$\mathrm{L\_a(nm)} = \left(2.4 \times 10^{-10}\right) \,\mathrm{A}^4 \left(\frac{\mathrm{I\_D}}{\mathrm{I\_G}}\right)^{-1} \tag{11}$$

where λ is the laser line wavelength in nanometer units, and ID and IG are the intensities of the D and G bands, respectively [16,17]. The values for the crystalline order and crystallite sizes for the composite samples with different cure degrees are displayed in Table 8.

**Table 8.** Crystalline order and crystallite size derived from the Knight formula.


The results of Raman spectroscopy indicate that the crystalline order increases in the cured samples, while the crystallite size decreases with the cure. These results are consistent with the results obtained with DTG studies. As can be seen, the degree of cure is directly related to an increase in temperature peaks, as shown in DTG studies, which results in a rise of the crystalline order.

#### **4. Conclusions**

Thermal analysis and Raman spectroscopy technologies provide relevant information for the modeling of the kinetic behaviour of the resins in composite materials. The control of the degree of curing of composite parts is fundamental for the optimization of the mechanical properties. The degree of polymerization of carbon–epoxy composites was investigated through several techniques such as FESEM, TGA, DSC, and Raman spectroscopy. FESEM micrographs reveal that the surface of the carbon fibers is better defined as the cure time becomes longer, due to the homogeneous redistribution of covering resins. TGA and DSC experiments confirm that the thermal characteristics of cured samples depend on the applied cure process. Raman spectroscopy offers an assessment of the crystalline order and crystallite sizes through the ratio between the intensities of D and G bands, which corresponds to the disordered and ordered phases of graphitic carbon, respectively. The results show that both Raman spectroscopy and thermal analysis are useful and complementary techniques for evaluating the degree of cure in composite materials. The technologies studied in this article are easily extrapolated to other materials or different applications, such as biomaterial for tissue engineering or biocomposites. Nowadays, many biotissues are manufactured with thermoset resins that need proper curing to achieve maximum mechanical properties. The problems associated with correct curing such as shrinkage or surface defects can be easily improved with these techniques.

**Author Contributions:** Conceptualization, J.A.G.-M. and L.G.; methodology, A.R.-G., L.M. and R.T.; validation, all authors; formal analysis, B.M., J.A.S. and J.M.-G.; writing—Review and editing, all authors.

**Funding:** This research was funded by Spanish Ministerio de Ciencia, Innovación y Universidades through the grants ENE2016-77798-C4-2-R, FIS2017-83295-P and RED2018-102612-T and CENTRO DE EXCELENCIA EN NANOFIBRAS LEITAT CHILE, 13CEI2-21839. J.A.S. also acknowledges Ramón y Cajal Fellowship for financial support (RYC-2015-17482).

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Deformation-Assisted Joining of Sheets to Tubes by Annular Sheet Squeezing**

#### **Luis M. Alves, Rafael M. Afonso, Frederico L.R. Silva and Paulo A.F. Martins \***

Instituto de Engenharia Mecânica, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1049-001 Lisboa, Portugal; luisalves@tecnico.ulisboa.pt (L.M.A.); rafael.afonso@tecnico.ulisboa.pt (R.M.A.); frederico.rocha.e.silva@tecnico.ulisboa.pt (F.L.R.S.)

**\*** Correspondence: pmartins@tecnico.ulisboa.pt; Tel.: +351-21-841-9006

Received: 28 October 2019; Accepted: 25 November 2019; Published: 26 November 2019

**Abstract:** This paper is built upon the deformation-assisted joining of sheets to tubes, away from the tube ends, by means of a new process developed by the authors. The process is based on mechanical joining by means of form-fit joints that are obtained by annular squeezing (compression) of the sheet surfaces adjacent to the tubes. The concept is different from the fixing of sheets to tubes by applying direct loading on the tubes, as is currently done in existing deformation-assisted joining solutions. The process is carried out at room temperature and its development is a contribution towards ecological and sustainable manufacturing practices due to savings in material and energy consumption and to easier end-of-life disassembly and recycling when compared to alternative processes based on fastening, riveting, welding and adhesive bonding. The paper is focused on the main process parameters and special emphasis is put on sheet thickness, squeezing depth, and cross-section recess length of the punches. The presentation is supported by experimentation and finite element modelling, and results show that appropriate process parameters should ensure a compromise between the geometry of the mechanical interlocking and the pull-out strength of the new sheet–tube connections.

**Keywords:** joining; forming; sheet–tube connections; experimentation; modelling and simulation

#### **1. Introduction**

In recent years, there has been a growing utilization of deformation-assisted joining processes driven by an increasing demand of assembling lightweight components. Deformation-assisted joining processes are classified into three distinct groups according to their operating principles:


Some of the above-mentioned processes have been extensively investigated for joining sheets and tubes made from similar or dissimilar materials. The state-of-the-art reviews by Mori et al. [12] and Groche et al. [13] provide detailed information on the most significant developments and applications of deformation-assisted joining processes to connect sheets and tubes.

Despite the progresses in sheet-to-sheet and tube-to-tube connections, the joining of sheets-to-tubes is still preferentially accomplished by means of conventional fastening and welding technologies. Little progress has been made in the application of deformation-assisted joining to sheet–tube connections, aside from recent developments involving the utilization of form-fit mechanical joints based on compression beads formed by local buckling [14], on the combination of tube compression beads with flaring [15], and on the combination of partial sheet-bulk compression of tubes with upsetting [16] or flaring [17] (Figure 1a–c).

**Figure 1.** Deformation-assisted joining of sheets to tubes by mechanical joining. (**a**) Form-fit joints produced by tube compression beads formed by local buckling. (**b**) Form-fit joints produced by a combination of tube compression beads and flaring. (**c**) Form-fit joints produced by a combination of partial sheet-bulk compression of tubes and flaring. (**d**) Force-fit joints produced by the pressure that remains after elastic recovery from magnetic loading.

The utilization of force-fit mechanical joints that exclusively rely on the pressure that remains on the contact interfaces after elastic recovery from mechanical [18], hydraulic [5], or magnetic loading [6] is not a feasible solution for the fixing of sheets to tubes due to the limited contact surfaces provided by the small thickness sheets that are commonly used in the industry (Figure 1d). Similar constraints apply to new developments for the joining of thick blocks (or plates) to tubes by compressing the upper end of a connecting block flange [19,20] because of the need to produce the connecting flange by machining.

Regarding the connection of sheets to tubes, which is the main objective of this paper, it is worth mentioning that all the previously mentioned deformation-assisted joining processes based on mechanical form-fit joints [14–17] involve multiple operations. This undercuts their advantages in material consumption, energy requirements, and end-of-life recycling with disadvantages related to efficiency and costs that are better ensured by conventional technologies based on fastening, welding, and adhesive bonding. Besides the limitations resulting from existing processes based on mechanical form-fit joints being carried out in multiple operations, there is also the risk of failure by cracking in the compression beads produced by local buckling. This reduces their applicability to tubes with low fracture toughness [21].

All of the above-mentioned problems prompted the authors to develop a new process for fixing sheets to tubes at room temperature that is based on an entirely new mechanical joining concept in which loading is applied on the sheet surface instead of the tube itself [22]. The process is schematically shown in Figure 2, and consists of squeezing (compressing) the annular surface of the sheet adjacent to the tube in order to ensure the material from the sheet flows inwards and is shaped as a form-fit joint with good mechanical interlocking between the sheet and the tube.

**Figure 2.** A representation of the deformation-assisted joining of sheets to tubes by annular sheet squeezing at the open and closed positions [22]. A photograph of a longitudinal cross section of a test specimen is enclosed.

The new joining process is performed in a single punch stroke and requires both the sheet and the tube to have some degree of ductility to plastically shape the form-fit joint. The optimum operating conditions require the material of the tube to have a lower elastic modulus than that of the sheet, so that the pressure remains on the contact interface after producing the form-fit joint as a result of the more pronounced elastic recovery of the tube in the direction of the sheet. Otherwise, the resulting form-fit joint may end up slightly loose as in the case of sheet–tube connections made from dissimilar materials (e.g., metals and polymers or composites) with very different elastic modulus. Another requirement of the new proposed joining process is the necessity of the sheet strength being similar or higher to that of the tube to allow for easy shaping of the inner tube bead that is needed to produce the form-fit mechanical joint.

Finally, it is worth mentioning that the new joining process circumvents the previously-mentioned difficulties resulting from the utilization of tubular materials with low fracture toughness.

This paper is focused on the main process parameters of deformation-assisted joining of sheets to tubes by annular sheet squeezing, hereafter designated as 'mechanical joining of sheets to tubes'. Special emphasis is put on the sheet thickness *ts*, squeezing depth *d*, and cross-section recess length *l* of the punches due to their influence on the inner tube bead shape of the form-fit joint and on the quality and performance of the mechanical joint between the sheet and the tube. The presentation includes results from experimentation and finite element modelling, and from destructive pull-out tests that were carried out to determine the maximum force that the new joints can withstand before failing.

#### **2. Materials and Methods**

#### *2.1. Materials and Flow Curves*

The work on the mechanical joining of sheets to tubes by annular sheet squeezing was carried out in two different aluminum alloys. The sheets were made of aluminum AA5754-H111 with a 5 mm thickness and the tubes were made of aluminum AA6063-T6 with an outer radius of 16 mm and a 1.5 mm wall thickness.

The flow curves of the two materials were obtained by combining tensile and stack compression tests [23] performed in a hydraulic testing machine with a cross-head speed of 5 mm/min. The tensile tests allowed characterization of the stress response of the materials for small values of strain, whereas the stack compression tests were utilized to determine the stress response for larger values of strain, beyond plastic instability in tension, following a procedure similar to that of Silva et al. [4]. The flow curves of the AA5754-H111 sheets and AA6063-T6 tubes are shown in Figure 3.

**Figure 3.** Flow curves of the two aluminum alloys utilized in the experiments.

The reason for using two materials with similar flow curves and nearly identical elasticity moduli (with differences within the range of 68 to 68.9 GPa) was to allow for the study of the performance of the form-fit joints alone and independently of the interfacial pressure, which develops on the sheet–tube contact surface in case of form-fit joints made from materials with a different elasticity modulus.

#### *2.2. Experimental Tests*

In their original paper on the mechanical joining of sheets to tubes by annular sheet squeezing [22], the authors put emphasis on the influence of the cross-section recess length *l* of the punch on plastic material flow inside the sheet thickness. The squeezing depth *d* was kept constant and a deformation-zone geometry parameter Δ = *ts*/*l*, defined as the ratio of the sheet thickness *ts* to the cross-section recess length *l*, was introduced to characterize material flow and identify the deformation modes associated with acceptable and unacceptable joints.

In a subsequent paper [24], the authors focused on the complementary work plan by analyzing the influence of the squeezing depth *d* and keeping the cross-section recess length *l* of the punch at a fixed value. The investigation allowed the characterization of the physics behind material separation at the cross-section recess corner of the punch and to reach a better understanding of the influence of the squeezing depth *d* on the pull-out destructive strength of the joints.

The work plan giving support to this paper takes the combined influence of *d* and *l* into account. The goal is to understand how changes in both variables at the same time will influence material flow

and the overall pull-out performance of the joints so that a procedure can be reached to determine the combination of squeezing depth *d* and cross-section recess length *l* that is capable of ensuring the best form-fit joint for a supplied set of geometries and materials.

Such a procedure has not been addressed in previous papers and is of paramount importance when producing form-fit joints in dissimilar materials in which the strength of the tube is similar or slightly higher than that of the sheet.

The overall methodology utilized in the experimental tests consisted on the following steps:


**Table 1.** The experiments on the mechanical joining of sheets to tubes by annular sheet squeezing. Notation is in accordance with Figure 2.


Table 1 presents a summary of the experiments on mechanical joining together with the major geometrical specifications of the sheets and tubes that were utilized in the investigation. The influence of the sheet thickness *ts* was taken into consideration by extending the experimental work to sheets with a smaller thickness (*ts* = 2.5 mm) than the nominal supplied thickness *ts* = 5 mm.

Figure 4 presents a schematic representation of the experimental setup that was utilized to evaluate the performance of the new form-fit joints. The joints were subjected to destructive pull-out tests in which the sheet was detached from the tube, and the objective was the determination of the maximum force *F* that the joints can withstand before failing.

#### *2.3. Finite Element Modelling*

The in-house finite element computer program I-form was utilized to analyze the mechanical joining of sheets to tubes by annular sheet squeezing. The computer program is based on the finite element flow formulation [25], and its implementation follows the extension of the rigid-plastic Markov's principle [26] of minimum plastic work to include incompressibility and contact between deformable bodies,

$$\Pi = \int\_{V} \overline{\sigma} \,\overset{\circ}{\varepsilon} \,dV + \frac{1}{2} K \int\_{V} \dot{\varepsilon}\_{v}^{2} \,dV - \int\_{S\_{T}} T\_{i} \mu\_{i} \,dS + \int\_{S\_{f}} \left(\int\_{0}^{|\mu\_{i}|} \pi\_{f} d\mu\_{i}\right) dS + \frac{1}{2} K\_{1} \sum\_{\varepsilon=1}^{N\_{\varepsilon}} \left(g\_{n}^{\varepsilon}\right)^{2} + \frac{1}{2} K\_{2} \sum\_{\varepsilon=1}^{N\_{\varepsilon}} \left(g\_{\varepsilon}^{\varepsilon}\right)^{2} \tag{1}$$

In the first term of Functional (1), the symbols <sup>σ</sup> and . ε denote the effective stress and the effective strain rate, respectively,

$$
\overline{\sigma} = \sqrt{\frac{3}{2} \sigma\_{ij}' \sigma\_{ij}'} \dot{\overline{\varepsilon}} = \sqrt{\frac{2}{3} \dot{\varepsilon}\_{ij} \dot{\varepsilon}\_{ij}} \tag{2}
$$

where σ *ij* is the deviatoric stress tensor and . ε*ij* is the strain rate tensor.

In the second term, the symbol . ε*<sup>v</sup>* is the volumetric strain rate, given by

$$
\dot{\varepsilon}\_{\upsilon} = \delta\_{ij} \dot{\varepsilon}\_{ij} \tag{3}
$$

where δ*ij* is the Kronecker delta and *K* is a large positive number utilized to impose incompressibility in volume *V* by means of a penalty factor.

**Figure 4.** A representation of the experimental pull-out destructive setup before (left) and during testing (right).

The third term of Functional (1) makes use of the surface tractions *Ti* and velocities *ui* on the surface *ST*, whereas the fourth term takes care of the frictional effects along the contact interface *Sf* between the sheet and tube with the tools. In this work, tools were assumed as rigid bodies, and τ*<sup>f</sup>* and *ur* denote the friction shear stress and relative sliding velocity of the sheet and tube. The friction shear stress was modelled according to the law of constant friction,

$$
\pi\_f = m \, k \tag{4}
$$

where *m* is the friction factor between the sheets and tubes, taken as 0.1 after checking the predicted forces that best matched the experimental results. The symbol *k* denotes the flow shear stress.

The fifth and sixth terms account for the contact between the sheet and tube modelled as deformable bodies along their contact interfaces defined by means of *Nc* pairs extracted from the faces of the elements that were utilized in their discretization. The symbols *g<sup>c</sup> <sup>n</sup>* and *g<sup>c</sup> <sup>t</sup>* denote the normal and tangential gap velocities in the contact pairs, which are penalized by large numbers *K*<sup>1</sup> and *K*<sup>2</sup> to avoid penetration. A more detailed look into the numerical implementation of Functional (1) in the finite element computer program I-form can be found in reference [27].

Figure 5 shows in detail the finite element model before and after the sheet is mechanically joined to the tube. The finite element models utilized in the numerical modelling of the process made use of rotational symmetry conditions and required discretization of the longitudinal cross-section of the sheets and tubes by means of approximately 20,000 and 800 quadrilateral elements, respectively. The sheet and the tube were modelled as deformable bodies, whereas the tools were modelled as rigid objects and discretized by means of linear contact-friction elements. Several remeshings were carried out to avoid excessive element distortion during annular sheet squeezing.

**Figure 5.** Element meshes before and after fixing the sheet to the tube using mechanical joining by annular sheet squeezing (*l* = 2 mm and *d* = 2 mm and *ts* = 5 mm).

#### **3. Results**

Figure 6 shows the finite element computed reduction of the inner tube radius *R* = (*r*<sup>0</sup> − *rb*)/*r*<sup>0</sup> as a function of the cross-section recess length *l* of the punch for four different values of the squeezing depth *d*. Experimental measurements of *R* for a squeezing depth *d* = 2 mm and a cross recess length *l* = 2 mm are included to assess the validity and reliability of the finite element estimates.

The first conclusion derived from Figure 6 is that small values of *d* lead to small amounts of material being displaced against the tube, and therefore, to the development of less pronounced inner tube bead geometries of the form-fit joints. In contrast, large values of *d* give rise to large amounts of material being squeezed against the tube and to more pronounced inner tube bead geometries of the form-fit joints. This is graphically shown in the two schemes that are included in Figure 6.

**Figure 6.** Reduction of the inner tube radius *R* as a function of the cross-section recess length *l* of the punch for different values of the squeezing depth *d*.

As seen in Figure 6, a typical *R*(*l*) evolution passes through a peak corresponding to the cross-section recess length *l* that provides the maximum reduction *R*max of the inner tube radius for a given squeezing depth *d*. The dashed curve passing through all these peaks defines the correlation between *l* and *d* that maximizes the inner tube bead geometry of the form-fit joints.

To the left of the dashed curve, there is a decrease in the amount of sheet material being squeezed as the cross-section recess length *l* diminishes. This leads to smaller values of *R* and to the development of less pronounced inner tube bead geometries of the form-fit joints. At the limit, there will be no form-fit joint.

Contrary to what one would expect, to the right of the dashed curve, the *R*(*l*) evolutions should increase were it not for the squeezed sheet material starting to moving outwards instead of inwards.

This last conclusion is confirmed by the computed evolution of the normalized radial velocity *vr*/*vp*, where *vp* is the vertical punch velocity, shown in Figure 7b. As seen in case of *d* = 2 mm, when the cross-section recess length of the punch increases from *l* = 2 mm to *l* = 2.5 mm, there is a shift of the neutral point (*NP*), corresponding to the transition between inward and outward material flow towards the left corner of the punch, meaning that more squeezed sheet material will start flowing outwards (refer to the black arrows included in Figure 7b).

Another consequence of the squeezed sheet material starting to flow outwards is the occurrence of bending. Bending gives rise to form-fit joints with a lack of perpendicularity between the sheet and the tube, as shown in the bottom experimental and numerically predicted cross-sections that are included in Figure 7a.

**Figure 7.** Influence of the cross-section recess length *l* of the punch on the form-fit joints for a squeezing depth *d* = 2 mm. (**a**) Experimental cross-sections of the form-fit joints; (**b**) Finite element predicted cross-sections of the form-fit joints with the distribution of normalized radial velocity *vr*/*vp*.

#### **4. Discussion**

#### *4.1. Pull-Out Destructive Forces*

The results obtained in Section 3 showed that both the cross-section recess length *l* and the squeezing depth *d* of the punch play a key role in the geometry of the inner tube bead of the form-fit joints.

One question that naturally arises from Figure 6 is whether the correlation between *l* and *d* that maximizes the inner tube bead geometry of the form-fit joints is capable of ensuring the maximum pull-out forces that the joints can safely withstand before failing.

To answer this question, the authors took the form-fit joint with *l* = 2 mm and *d* = 2 mm lying close to the correlation line that maximizes the inner tube bead geometry and compared its pull-out destructive force with those obtained for other joints that were obtained by varying *d* along the dotted vertical line of Figure 6.

The results from experimental tests and numerical simulations are shown in Figure 8 and allow to conclude that the maximum pull-out destructive force *F* is not obtained for the process operating conditions that maximize the inner tube bead geometry of the form-fit joints.

**Figure 8.** Influence of the squeezing depth *d* on the pull-out destructive force *F* of sheet–tube connections obtained with a punch having a cross-section recess length *l* = 2 mm.

Whilst at a first glance, this result may seem easy to explain because large values of *d* should give rise to larger inner tube beads of the form-fit joints, there is an additional phenomenon that also needs to be considered. Otherwise, it is not possible to fully understand the experimental and numerical results included in Figure 8, because the increase of the pull-out force *F* with *d* is only monotonic up to a peak, after which the force drops sharply.

The additional phenomenon that needs to be considered for understanding the evolution of the pull-out force *F* with *d* is the failure mechanism. In particular, the change in mechanism when the active sheet thickness *ta* left below the cross-section recess length of the punch (refer to the inset included in Figure 8) becomes very small.

Taking, for example, the pull-out force vs. displacement evolutions for the form-fit joints with *d* = 3 mm and *d* = 4 mm retrieved from Figure 8, it is easy to observe two totally different separation mechanisms (Figure 9a). One mechanism, leading to higher forces, is similar to tube extrusion while the other mechanism, leading to lower forces and a sharp drop at the end, is related to shearing.

**Figure 9.** Pull-out tests. (**a**) Experimental evolution of the force with displacement for destructive pull-out tests performed in two different form-fit joints that were produced with a punch having a cross-section recess length *l* = 2 mm. (**b**) Photographs of the two different tests specimens after failure by extrusion (left) and shearing (right).

In case of the first mechanism, observed in the pull-out test of the form-fit joint with *d* = 3 mm, the sheet acts as a floating die and the tube is forced to plastically deform in order to reduce its inner radius from *r*<sup>0</sup> to *rb*. The maximum pull-out force *F* attained in the destructive test is approximately equal to 12 kN.

The second mechanism, observed in the pull-out test of the form-fit joint with *d* = 4 mm, is typical of shearing along the small active sheet thickness *ta* that was left below the cross-section recess length of the punch. The maximum pull-out force *F* associated to this mechanism is smaller than that of tube extrusion.

The photographs included in Figure 9b illustrate the differences between the two above-mentioned failure mechanisms. The conclusion to be taken from these tests is that the design of a joint to be produced by a punch having a cross-section length *l* = 2 mm must consider a squeezing depth *d* = 3 − 3.5 mm to ensure high pull-out destructive forces.

In connection to this, it is worth mentioning that the concept of maximum pull-out force utilized by the authors refers to the force that prevents collapse by avoiding detachment of the two components and not to the force that is needed to produce the first relative movement between the sheet and the tube.

#### *4.2. Sheet Thickness*

The other question one may have about the mechanical joining of sheets to tubes by annular sheet squeezing is whether the process works for smaller sheet thicknesses than that utilized in the previous sections. To answer this question authors decided to produce sheet–tube joints using the same tube geometry but reducing the sheet thickness from *ts* = 5 mm to *ts* = 2.5 mm. The photograph included in Figure 10a shows that the process is still feasible for sheets with smaller thicknesses.

(**b**)

**Figure 10.** Mechanical joining of sheets to tubes by annular sheet squeezing using a punch with a cross section recess length *l* = 2 mm and two different thicknesses *ts* = 2.5 mm and *ts* = 5 mm. (**a**) Photographs of sheets and tubes after being joined. (**b**) Finite element predicted cross section at the end of stroke for the two test cases shown in (**a**) using a squeezing depth *d* = 1 mm (left) and *d* = 3.5 mm (right).

In view of the above, it is interesting for the readers to know which values of the cross-section recess length *l* and squeezing depth *d* of the punch were utilized to produce the form-fit joint with a sheet thickness *ts* = 2.5 mm. For this purpose, it is important to observe in Figure 8 that the maximum pull-out force of a form-fit joint produced with a punch having a cross-section recess length *l* = 2 mm is obtained for a squeezing depth *d* = 3.5 mm. Larger values of *d*, leading to active sheet thicknesses *ta* < 1.5 mm, provide smaller pull-out forces because the failure mechanism changes from extrusion to shearing.

So, in order to use the same punch (*l* = 2 mm) for comparison purposes, and prevent failure by shearing in case *ta* < 1.5 mm, it was decided to use a squeezing depth *d* = 1 mm for obtaining the form-fit joint with a sheet thickness *ts* = 2.5 mm.

Figure 10b shows a detail of the computed finite element cross sections for the form-fit joints obtained for the two different sheet thicknesses. As seen, the magnitude of bending is significant in both cases, namely in the sheet with smaller thickness due to its lower stiffness. However, the phenomenon can be easily avoided by diminishing the gap between the cross-section recess length and the remaining flat surface of the punch so that the total amount of bending is limited.

#### **5. Conclusions**

Deformation-assisted joining of sheets to tubes by annular sheet squeezing is based on mechanical joining by means of form-fit joints. The amount of sheet material to be squeezed and the final shape and volume of the inner tube beads of the form-fit joints is controlled by the cross-section recess length *l* of the punch, the squeezing depth *d* and by the geometry of the punch and sheet, namely the sheet thickness *ts*.

By varying these parameters, it is possible to change the plastic flow of the squeezed sheet material from a predominantly inward into a combination of inward and outward. In particular, it is possible to define a combination of parameters capable of ensuring maximum shapes and volumes of the form-fit joints for given values of *d* or *l*. However, the maximum shapes and volumes of the form-fit joints do not necessarily provide the maximum pull-out destructive forces because there is a minimum active sheet thickness *ta* below which the failure mechanism changes from extrusion to shearing and the forces drop significantly.

In connection to this, it is also shown that the new mechanical joining process is not limited to thick sheets because what needs to be fulfilled is a combination of parameters capable of ensuring a sound mechanical joint for values of the active sheet thickness *ta* above the critical threshold of failure by shearing.

**Author Contributions:** L.M.A. and R.M.A. designed and fabricated the tools. L.M.A., R.M.A. and F.L.R.S. performed the experimental and numerical simulation work. L.M.A and P.A.F.M. developed the finite element computer program and analyzed the results. P.A.F.M. supervised the overall research work and wrote the article with the collaboration of all the other authors.

**Funding:** This research was funded by Fundação para a Ciência e a Tecnologia of Portugal under LAETA - UID/EMS/50022/2019 and PDTC/EMS-TEC/0626/2014.

**Acknowledgments:** The authors would like to acknowledge the support provided by IDMEC and Fundação para a Ciência e a Tecnologia of Portugal.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **On The Influence of Rotary Dresser Geometry on Wear Evolution and Grinding Process**

#### **Leire Godino 1,\*, Jorge Alvarez 2, Arkaitz Muñoz <sup>1</sup> and Iñigo Pombo <sup>1</sup>**


Received: 22 October 2019; Accepted: 20 November 2019; Published: 22 November 2019

**Abstract:** Dressing is a critical issue for optimizing the grinding process. Dresser tool and dresser parameters must be designed according to the grinding wheel material, shape, or even the dimensional and geometrical tolerances of the workpiece and its surface roughness. Likewise, one of the problematic issues of dressers is the wear that they suffer. In order to tackle this issue, the present work characterized the wear of two rotary dressers by analysing the wear behaviour depending on the pit radius of the dressers while studying the influence of the wear on ground surfaces. This work showed that the rotary dresser with a higher pit radius presents wear that is approximately 28% higher than the dresser with a half pit radius.

**Keywords:** OD grinding; CBN; dressing; rotary dresser; wear; CVD diamond

#### **1. Introduction**

The grinding process is a very important machining process characterized by the high added value of the ground parts. Its capacity for manufacturing advanced materials of poor machinability while achieving high-quality surfaces is made possible due to the combination of grinding wheel design, new advances in abrasives and an optimal dressing process. In recent years, the use of super-abrasive grinding wheels, primarily CBN (cubic boron nitride), has become increasingly extensive in the industry due to their thermal stability and the absence of chemical affinity with ferrous materials. Moreover, CBN grinding wheels are suitable when high removal rates are required and are used in creep feed grinding and high-speed grinding applications. Due to the different bond used with CBN grinding wheels, vitrified grinding wheels present more advantages than the other resin bond, metallic bond or electroplated grinding wheels. These advantages are primarily related to the ability to transport the coolant, the efficiency of chip removal and the capacity for dressing using rotatory dressing tools. Moreover, CBN grinding wheels are not the only widely used wheel in industrial applications, with conventional grinding wheels also being used due to their versatility and low cost in comparison with super-abrasives. The latest advances make them more competitive with regard to abrasive grain shape and crystalline structure.

One of the most important aspects of the grinding process is dressing, which is used in order to regenerate the abrasive capacity of the grinding wheel and the initial shape of the abrasive tool. Among the various types of dressers, these can be classified as static and rotary. The present work was focused on the rotary dressing tool, which consists of a cylindrical body with a single layer of inserted or deposited diamond particles. Different designs of rotary dressers can be found depending on the shape and size of the diamonds and their location, as well as their bonding chemical compositions. The main advantage of using this tool is the high dressing speed that can be achieved in comparison with a stationary single or multiple diamond tool, along with the possibility of generating a complex profile on the grinding wheel surface [1]. In addition, the wear level suffered by the dresser is relatively

low [2]. Taking this fact into account, these types of dressers are used for dressing CBN and diamond grinding wheels with vitreous bond and conventional profile grinding wheels. The main advantages of rotary dressing tools are the high dressing speed and the possibility of generating complex profiles in the grinding wheel [1]. Moreover, this type of dressing tool provides the process with a high level of repeatability and precision, reducing processing costs and the number of rejected parts.

However, the complexity of the process is high, particularly when compared with the one carried out using a stationary diamond tool. In this regard, specific works focused on the influence of particular dressing parameters on the characteristics of the wheel surface can be found in the specialized literature. In [3], the first attempts were made to analyse the influence of dressing parameters on the wheel performance when using roller diamond tools. The Schmitt diagram is currently used in the industry to design dressing processes. In [2], a more in-depth analysis was conducted on the state of the art regarding grinding wheel conditioning, including the rotary dressing process. This work shows that a considerable amount of research has focused on the influence of the dressing parameters on grinding wheel performance, including the influence of the radial feed of the dressing process on the roughness of the wheel surface, theoretical pathways of dressing grains, influence of dressing speed ratios on the radial dressing force and effective grinding wheel roughness or the influence of the depth of dressing cut on specific normal grinding force. In addition, some studies on diamond tool wear were included, particularly [4] and the studies of Linke and Klocke [5,6], which focused on analysing diamond tool wear for stationary dressers.

Recent works regarding the analysis of the rotary dressing process are worth mentioning. For instance, in [7], the authors analysed the performance of various diamond tools under the same dressing conditions for small grinding wheels for internal grinding. In [8], a new type of rotary diamond tool was presented. In this case, the geometry and density of abrasive grits was completely monitored. These authors evaluated the performance of the new dresser against conventional dressers in terms of force and wear of the abrasives. The results were positive and the forces for the new dresser were found to be approximately 50% lower than that of the classic dressers, while the wear was also improved, although no quantitative values were reported. In [9], an analysis was presented regarding the performance of a roller dresser for the case of the micro-grinding of a titanium alloy. The authors focused their study on analysing the influence of the overlap ratio (Ud) on the ground surface quality. The main conclusion reached by the authors was that extremely high values of overlap ratios were suitable for achieving high surface quality in micro-grinding processes. Finally, Palmer et al. [10] analysed the characteristics of the dressed wheel surface when using roller dressers.

Although one of the most relevant issues in dressing is the wear of the dresser, very little information can be found in the specialized literature regarding the wear of rotary dressers. This fact is very important when using this kind of dressing tool for two main reasons. The first is related to the characteristics of the dressed surface. In the case of rotary dressing tools, excessive wear of the dressing tool implies changes in the tool geometry and hence the tool sharpness parameter. This could imply changes in ad, or even in Ud, and in the characteristics of the dressed surface. In Figure 1, the influence of a worn dressing tool on a plane grinding wheel is shown, with a smaller ad being achieved for the worn rotary dresser.

**Figure 1.** Influence of the worn rotary dressing tool on the wheel surface.

The second reason is related to the geometry of the dressed wheel for non-plane profiles in the wheel. Depending on the geometry of the dressing tool, a dressing path is programmed in the machine. Once the rotary diamond tool has lost its shape due to wear after several dressing passes, either the dressing path or the rotary tool must be changed. If the wear suffered by the rotary diamond tool is not controlled, several ground parts could be rejected. Figure 2 shows the effect of grinding a profile wheel with a worn rotary dressing tool. In the left part of the image, the theoretical wheel profile is shown in green, and also points to the programmed path of the rotary dresser. In contrast, in the right part of the image, the path of the worn rotary dressing tool is shown. In this case, the real wheel profile is going to be larger than theoretical profile. Thus, when the part is ground, of the removed material is higher than programmed ones. This effect leads to rejection of the part.

**Figure 2.** Influence of the worn rotary dressing tool on the wheel profile.

In order to tackle these problems, the present work constitutes a preliminary approach toward characterizing the wear of rotary dressers. From research point of view, there is not a unique parameter to define the wear of rotary dressers. In contrast, the volumetric wear of grinding wheels is defined by G-ratio and the wear of stationary dressers is quantified using dressing wear ratio, Gd, applied by Shi and Akemon [11] for stationary blade diamond tools. Therefore, the aim of the present work was to define a wear parameter to quantify the wear suffered by the rotary dressing tool. A new parameter, termed "wear parameter, Wd", was presented. This parameter allowed the characterization of rotary dresser wear in order to be comparable to the stationary blade dressers. To this end, one of the objectives was to develop a systematic methodology for analysing and characterizing the wear suffered by a rotary dressing tool. The proposed methodology included the development of specific software (in Python) to measure the rotary diamond tool wear, the proposal of a parameter to measure such wear and an analysis of the grinding wheel behaviour, paying particular attention to the consumed power and surface roughness. This methodology was used to analyse the wear suffered by different geometry rotary dressing tools when dressing plane profile CBN grinding wheels with a vitreous bond.

For this purpose, the employed experimental setup and methodology was first presented. Second, the results are analysed and the Wd is defined. Finally, the main conclusions drawn from this work is presented.

#### **2. Materials and Methods**

This study examined the wear of rotary dressing tools and its influence on ground workpieces. To this end, dressing and grinding tests were combined on a cylindrical grinding machine (DanobatGroup, Elgoibar, Spain). The study consisted of the analysis of two rotary dressers, varying the diamond pattern and the dresser geometry, i.e., the tip radius. RIG 52035 and RIG 52034 were manufactured by TYROLIT. For simplicity, the two rotary dressers are referred to as RIG 35 and RIG 34, respectively. Table 1 lists the main characteristics of the rotary dressers. In both cases, CVD (cultivated diamond) diamonds were inserted. In the case of RIG35, the interlayer was positioned. In contrast, in RIG34 dresser, the CVD diamonds were aligned. Regarding to the geometry of the rotary dressers, RIG35 presented a pit radius of 0.5 mm, whereas the pit radius of RIG34 was 0.25 mm. As previously mentioned, rotary dresser geometry is essential when profile grinding wheels are dressed. Any variation in the rotary dresser geometry is copied on the wheel surface.

In the present work, a CBN grinding wheel was used to conduct dressing and grinding tests. In order to distinguish between the influence of the wear on the ground workpiece surface and the grinding wheel shape, a straight grinding wheel was used. Thus, the influence of dresser wear on the wheel shape was not taken into account in this analysis, and only ground surface quality was analysed. The nomenclature of the used grinding wheel was CBN170N100V (UNESA S.L, Hernani, Spain), which was a medium hard wheel, presenting a vitreous bond and high density of abrasive grains. The external diameter of the wheel was Ø450 mm, the width was 10.3 mm, and the grain size was approximately Ø170 μm, corresponding to a finishing grinding wheel. Furthermore, plunge grinding tests were carried out on a cylindrical workpiece of hardened steel (AISI 52100). The medium hardness of the workpiece was approximately 54 HRC, while the external diameter of the parts was Ø80 mm.


**Table 1.** Characteristics of rotary dressing tools.

The present study involved dressing and grinding, the analysis of the wear suffered by the rotary dresser, and the influence of wear on the grinding wheel surface and hence on ground workpiece surfaces. Accumulative dressing tests and plunge grinding tests were conducted using the same cylindrical grinding machine, *DANOBAT FG600S* (© DanobatGroup, Elgoibar, Spain), as shown in Figure 3. Moreover, CBN grinding wheel was used at a cutting speed of 50 m/s, and a water-based coolant with a concentration of 3.2% was used at a pressure of 13 bar. Furthermore, to conduct a complete analysis of the process during both dressing and grinding, real power consumption was measured using a *Load Control UPC* (© Load Controls Incorporated, Sturbridge, MA, USA) power meter, and a *USB-6008* data acquisition card from *National Instruments*. Additionally, in order to quantify the influence of dresser geometry on wear evolution and the effect of wear on the subsequent grinding process, a new methodology was developed, which is described in the following section. The complete approach was validated through experimental grinding tests under industrial grinding conditions, as detailed below.

**Figure 3.** Dressing and grinding test set up.

Table 2 lists the dressing and grinding parameters. Moreover, the two dresser tools presented a similar dressing overlap ratio in order to compare the influence of the pit radius in the wear of the rotary dresser and in the ground surface. The overlap ratio is the relation between the effective width of the dressing tool and the feed per wheel revolution (Ud = bd/fd). A high value of dressing overlap ratio leads to a smooth grinding wheel surface, but the grinding forces and the specific energy are high. In contrast, a low value of dressing overlap ratio generates a sharper surface, with more cutting edges, thereby decreasing grinding forces. The range of values for the dressing process is Ud = 2–20 [12]. For the present study, a smooth wheel surface was required. Thus, the dressing overlap ratio was Ud = 2.64 for RIG 35 and Ud = 2.017 for RIG 34. Additionally, the smoothness of the grinding wheel surface depends not only on the dressing overlap ratio, but also on the dressing sharpness ratio, which is defined as the relation between the dressing depth and the effective width of the dresser (γ<sup>d</sup> = ad/bd). This parameter represents the influence of the dresser shape on the wheel surface. In the present study, in both cases rotary dressers were studied, varying the pit radius. Thus, γ<sup>d</sup> = 0.05 for RIG 35 and γ<sup>d</sup> = 0.07 for RIG 34.



Accumulative dressing tests were carried out for each dressing tool, removing a total of 49,415 mm<sup>3</sup> of wheel volume. The first step was to measure the new surface of the rotary dresser. The topography of new rotary dresser was characterized using a Confocal microscope Leica DCM3D® (Leica microsystems AG, Wetzlar, Germany). Once the starting surface was characterized, the dressing test was carried out. Moreover, in Table 2, fine-dressing parameters are built. The CBN grinding wheel was continuously dressed, removing 12,354 mm3 of abrasive material. Immediately after the dressing process, the plunge grinding test was carried out. The specific rate of material removal during the grinding process was Q'w = 5 mm3/mm s. After both the dressing and grinding tests, the worn surface of the rotary dressing tool and the ground surface were analysed. First, the dresser topography was measured again using a confocal microscope. The measurement after each dressing tests allowed for analysing the evolution

of the wear during a complete test. Second, in order to observe the effect of dressing on the ground workpiece, the roughness of the ground surface is measured using a portable surface roughness tester (Taylor Hobson, Leicester, United Kingdom).

This step was completed a total of four times, following the same methodology, with 12,354 mm3 of abrasive material being removed at each step. Table 3 details the range of workpiece material removed at each step. The complete test was finished after dressing a total of 49,415 mm3. Similarly, during both dressing and grinding tests, the power consumption was measured in order to analyse the influence of dressing with a worn rotary dressing tool on the efficiency of the process. The last step involved the analysis of the data obtained during the tests. Power was readily analysed, while the topography data had to be processed and analysed to characterize the wear, which was a complex process. Thus, in the present work, a methodology for quantifying dresser wear was proposed, which is detailed hereafter.



#### *A Methodology for Quantifying Wear in Rotary Dressers*

As a first step, it was necessary to accurately establish the geometry of the brand-new rotary dresser. To this end, specific tooling was designed for the dresser to take measurements on a Leica DCM3D® confocal microscope (Leica microsystems AG, Wetzlar, Germany), as shown in Figure 4. The tooling system, together with the first measurement of the dresser surface, were necessary to set the references required to quantify the wear parameters.

**Figure 4.** Initial characterization of the geometry of the rotary dresser on the confocal microscope.

Using this equipment, dresser topographies were obtained. Each state of wear was measured in four different zones along the profile of the disk, separated by 90 degrees. This is shown in the first image of Figure 5. The complete measured area was 2.546 <sup>×</sup> 8.477 mm<sup>2</sup> and 2.808 mm in height, with a height resolution of 12 μm. The blue light was used in order to avoid dresser surface brightness. Three-dimensional (3D) profiles were then extracted, as seen in the second image. Profile comparison was carried out by slicing the 3D geometry and obtaining 2D curves. Five curves were obtained for each 3D profile. For this purpose, the topography layer of the LeicaMap® (Leica microsystems AG, Wetzlar, Germany) was used, as shown in the third and fourth images of Figure 5. At this stage, it is important to note that the reference must be set at the diamond and not at the bonding, since the latter will suffer more pronounced wear. Therefore, the intensity layer (shown in the third image of Figure 5) of the data was used as a reference because, on this layer, the infiltrated diamonds can be clearly observed and slices were made to coincide with CVD diamonds. Thus, only the wear of the diamond was taken into account, avoiding the influence of the bond. Moreover, this layer used the same scale as that used by the topography layer from which the 2D profiles were extracted. It is important to note

that the rotary dresser profiles could not be obtained using a stylus profilometer due to the shape of the dresser surface and the abrasive surface. Therefore, a confocal microscope is the best option to analyse this kind of surface.

**Figure 5.** Rotary dresser measurement to characterize the wear.

Once the 2D profiles are available, it is possible to compare different states of wear of the dresser. Various geometrical parameters can be selected for comparison. In this work, two auxiliary parameters (namely worn area and contact length) and one main wear indicator (wear height, hd) were defined. To obtain these parameters, profiles at different wear stages must be overlapped while maintaining stable references. In order to do so, a Python app was developed.

The first utility developed to assist profile comparison was used for profile smoothing. Any possible measurement defect, or even the presence of noise on the signal, was filtered using a low band pass filter. After some experimentation, the order of the filter was set at 6, the sample frequency at 15 Hz, and the cut-off frequency at 1.5 Hz. The second step involved overlapping the profiles at different wear stages. Errors were removed using a best-fitting technique on the profiles. In order to apply the best-fitting technique, reference points were set. These points did not suffer wear during dressing because they were not in contact with the grinding wheel. In Figure 6a, it is shown that points 1 and 2 are the reference points of corresponding new and worn profiles. Moreover, in Figure 6b, the two overlapping profiles show the wear suffered by the rotary dresser tool.

**Figure 6.** (**a**) Incorrect overlapping of profiles 1 and 2; (**b**) solution after applying best-fitting technique.

The contact area was defined to correspond to the limits set by the points where deviations were more important in value. Subsequently, the worn area and the contact zone can be effectively quantified, as shown in Figure 7a. The worn area can later be used to estimate the total volume of dresser worn during the operation. Further, maximum and average values of the wear height parameter hd can be obtained at this stage, as shown in Figure 7b.

**Figure 7.** (**a**) Contact area and zone of the dresser where wear concentrates; (**b**) definition of wear height hd.

#### **3. Results and Discussion**

In this section, the wear of the rotary dressers with CVD infiltrated diamonds was analysed. First, the influence of dresser geometry on the wear was studied, after which the influence of dresser wear on the surface quality of the workpiece after dressing was assessed. The majority of works that have analysed dresser wear used the wear volume (Vd), the wear of removed grinding wheel (Vw), and thus, the dressing wear ratio (Gd), in order to quantify the wear of the dressers, to compare different wear states, and to determine dresser life. Almost all the studies have been conducted for stationary dressers, namely single point or blade diamond tools [11]. In contrast, in the present work, dressers of differing geometry were analysed, and these parameters were not appropriate for making the comparison. Therefore, wear height was used due to the dissimilarities in pit radius and hence the wear volume of both the studied rotary dressers, as Figure 8a shows. For a given hd, Vd can be approximately half the value for RIG 34 in comparison with RIG 35. Moreover, the height is the parameter that affects dressing precision and, as a consequence, the quality of the grinding process. If the rotary dresser decreases in diameter due to the height wear, the wheel does not achieve the designed dimensions and shape, obtaining larger dimensions. The error is translated to the workpiece during grinding, as removing more than the corresponding material would cause the workpiece to be rejected. Thus, it was necessary to analyse the evolution of hd during the process.

**Figure 8.** (**a**) Dresser comparison and (**b**) wear height evolution during a complete test.

Figure 8b represents the evolution of an accumulative dresser wear with the workpiece removed from the material. A linear increase in wear was shown with Vw for both rotary dressers. RIG35 presented a slightly higher slope than RIG34, achieving higher wear at the end of the tests. When 49,415 mm3 of grinding wheel was removed, the rotary dresser RIG 35 presented a level of wear that was 28% higher than RIG 34. Thus, RIG 35, which had a greater pit radius, showed a higher tendency towardswear than RIG 34. In this regard, it is necessary to highlight that if the comparison had been made through Vd, the wear difference between RIG 34 and RIG 35 would have been 145%, and also higher for RIG35. This result indicates that the volumetric parameters are valid for comparing dressers with the same geometry when the compared volume is equivalent. Thus, the parameter for determining the wear of two rotary dressers that differ in geometry—in this case, different pit radius or even diameter—must be the wear height.

Moreover, dressing volumetric ratio, i.e., the relation between the removed volume of the wheel and the worn volume of the rotary dressing tool (Gd = Vw/Vd), is also used to determine dressing process efficiency in the case of stationary dressers. However, as mentioned, this is not a valid parameter for comparing rotary dressers of differing geometry. In this regard, the present work measured efficiency through the relation between removed volume of the wheel and the height of the wear of the rotary dressing tool, defining the dressing wear parameter (Wd = Vw/hd). High values of Wd imply that the rotary dresser suffers less wear when dressing a higher quantity of wheel volume.

Regarding the case studied here, in Figure 9, the rotary dresser with lower pit radius, RIG34, presents a higher value of Wd than RIG35. Thus, when using a rotary dresser with a half pit radius, the Wd was approximately 31% higher in the case under study. Thus, RIG 34 was more efficient than RIG 35, increasing dresser life and hence becoming a more economic process due to the high cost of CVD rotary dressers. With the wear values obtained, both hd and Wd, it can be confirmed that the rotary dresser with smaller pit radius presented less wear in height, as well as higher dressing efficiency for the studied dressing and grinding parameters.

**Figure 9.** Dressing wear parameters.

Figure 10 plots the mean value of power consumption during grinding after each dressing step. To analyse the power, only the three last steps of the test, from 24,707 mm3 until the end of the test, were taken into account. Thus, the first contact, in which the dresser wear presented a transitory behaviour, was avoided. Thus, from 24,707 mm3 to the end of the test, power consumption during grinding decreased for both rotary dressers. However, when comparing the two rotary dressers, power consumption tended to decrease more markedly after dressing with RIG35 in comparison with RIG34. If the grinding wheel was dressed using the RIG35, the power consumption during grinding was approximately 18% higher than if the wheel was dressed using the RIG34 rotary dresser. This could be due to the influence of the loss of sharpness ratio on the wheel surface.

**Figure 10.** Power consumption during grinding.

The initial sharpness ratio was γ<sup>d</sup> = 0.05 for RIG35 and γ<sup>d</sup> = 0.07 for RIG34. Thus, RIG 34 was sharper, leading to a smoother wheel surface with more active (but shallower) cutting edges, as shown in the upper part of Figure 11a. When dressers were worn, the sharpness ratio decreased. The resulting wheel surface had fewer (and deeper) cutting edges than at the beginning of the tests. This helps to remove material during grinding, consuming less power. However, this is not the case for any dresser wear state, that is, it only occurs if the dresser is slightly worn. Analysing Figure 9a and taking into account the shape of rotary dressers, for the case studied here, RIG34 presented lower power consumption during grinding because the dressing process generated a greater quantity of (more shallow) cutting edges. Likewise, for RIG 35, in the last state it can be observed that the power consumption was similar. From this point, the power increased because the wear of the dresser was too high.

**Figure 11.** (**a**) Influence of sharpness ratio and the shape of the rotary dresser on the wheel surface, (**b**) roughness of workpiece after grinding and (**c**) the real Ra profile corresponding to 0.26 μm.

Finally, the influence of dressing on the quality of the workpiece surface was analysed. To this end, the roughness of ground workpieces was measured. In Figure 11c, the real profile generated by the rotary dresser tool RIG 35 is shown. During a complete test, higher Ra was achieved for RIG 34 than for RIG35. Thus, RIG35 led to smoother ground surfaces with the used parameters. If the evolution of the roughness during the test is analysed, different behaviour is shown in both cases, as displayed in Figure 11b. For RIG35, first, slightly lower values of Ra were measured (approximately 0.26 μm), and 0.3 μm were observed by the end of the test. In contrast, RIG34 did not present a tendency toward roughness, and the values varied from 0.35 to 0.4. In the last studied state, the difference in roughness between the two surfaces was lower than 16%. In any case, the values of Ra obtained were lower than 0.4 μm. Thus, a good surface quality was achieved despite the wear of rotary dressers.

#### **4. Conclusions**

After conducting a complete analysis of worn dressers and studying the influence of worn dressers on the ground workpiece surface, the obtained results were carefully analysed. From a discussion of the results, the following conclusions can be drawn:


These results suggest that for the range of dresser life studied here, the wear suffered was around 60 μm in height for RIG35 and 47 μm for RIG34, while the roughness of the ground workpiece did not undergo any significant changes. Moreover, it is necessary to bear in mind that this analysis was carried out for a plane grinding wheel, so the influence of the geometry lost in the wheel profile was not analysed. In this preliminary approach, only the plane grinding wheels were tested. Thus, it will be of interest to tackle the problem of non-plane grinding wheels in future works.

**Author Contributions:** Conceptualization, J.A.; Data curation, A.M.; Funding acquisition, I.P.; Investigation, I.P.; Methodology, L.G.; Software, A.M.; Supervision, J.A.; Writing—original draft, L.G.; Writing—review & editing, I.P.

**Funding:** The authors gratefully acknowledge the funding support received from the Spanish Ministry of Economy and Competitiveness and the FEDER operation program for funding the project "Scientific models and machine-tool advanced sensing techniques for efficient machining of precision components of Low Pressure Turbines" (DPI2017-82239-P).

**Acknowledgments:** The authors gratefully acknowledge the Basque Digital Innovation Hub (BDIH). Experiments have been carried out in facilities of the Digital Grinding Innovation Hub, part of the Basque Digital Innovation Hub (BDIH) initiative of the Basque Government.

**Conflicts of Interest:** The authors declare no conflict of interest. The founding sponsors had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Repairing Hybrid Mg–Al–Mg Components Using Sustainable Cooling Systems**

#### **David Blanco 1,**†**, Eva María Rubio 1,\*, Marta María Marín <sup>1</sup> and Joao Paulo Davim <sup>2</sup>**


Received: 5 December 2019; Accepted: 13 January 2020; Published: 15 January 2020

**Abstract:** This paper focused on the maintenance or repair of holes made using hybrid Mg–Al–Mg components by drilling, using two sustainable cooling techniques (dry machining and cold compressed air) and taking surface roughness on the inside of the holes as the response variable. The novelty of the work is in proving that the repair operations of the multi-material components (magnesium–aluminum–magnesium) and the parts made of aluminum and magnesium (separately) but assembled to form a higher component can be done simultaneously, thus reducing the time and cost of the assembly and disassembly of this type of component. The study is based on a design of experiments (DOE) defined as a product of a full factorial 2<sup>3</sup> and a block of two factors (3 <sup>×</sup> 2). Based on our findings, we propose that the analyzed operations are feasible under sustainable conditions and, in particular, under dry machining. Also, the results depend on the machining order.

**Keywords:** hybrid components; light alloys; magnesium; aluminum; drilling; dry machining; cold compressed air; lubrication and cooling systems; arithmetical mean roughness; *Ra*; average maximum height; *Rz*; repair and maintenance operations

#### **1. Introduction**

Today, energy efficiency and sustainability play an increasingly important role in the development of new materials and their applications, especially in the transport industry, due to the pollution generated by the vehicles or their different parts in many stages of their life cycle.

Therefore, in order to reduce such contamination, it is necessary to approach the problem in a global way, analyzing all the factors that can have an influence during the manufacturing, the use, the maintenance, the repair, and the recycling of each part.

Although new forms of energy are being investigated to propel vehicles, these new alternative energies need to be greatly improved to match the level of development achieved with fossil fuels.

Until these new kinds of sustainable energy become competitive, in the transport sector, it is still essential to decrease the weight of vehicles to reduce the quantity of fossil fuels used, and, as a consequence, the pollution associated with its consumption. This is particularly true for the aeronautic and aerospace sectors.

To achieve this goal, research focuses on the use of the combination of different materials to create new multi-material components with better global properties than those of the original individual materials. These materials are called hybrid components, and the number of studies associated with them grew exponentially in recent years. Rubio and collaborators [1] affirmed that, among the possible combinations of materials used to form hybrid components with structural uses, the metal–polymer and metal–metal combinations are two of the most used, with aluminum alloys being most used to form these types of hybrid materials. Some of the main applications of these combination materials in the automotive sectors are car sleeper roofs, door structures, car fronts, tubular components of the exhaust system, and tubular components of the suspension system. In the aeronautic sector, the main applications are in the fuselage of airplanes and helicopters, the wings and rotors, and the other control surfaces (ailerons, flaps, spoilers, aerodynamic brakes, slats, and horizontal and vertical stabilizers).

The use of lightweight structural materials is widespread in various industries, particularly the aeronautics and automotive industries, since the weight of aircraft and cars is directly related to their consumption and pollution. A 10% reduction in the weight of an automobile can lead to improvements in fuel efficiency (by 8%), acceleration, and braking performance, and can reduce CO and hydrocarbon (HC) emissions by 4.5%, and NOx emissions by 8.8% [2]. In the civil aircraft industry, the weight of a Boeing 747–400 is approximately 183,500 kg, and the estimated fuel savings for an airplane is, over short distances, 117–134 kg of kerosene per kilogram of reduced weight and, over long distances, 172–212 kg of kerosene per kilogram of reduced weight [3].

Within the industries mentioned above, light alloys are used widely thanks to their excellent weight/mechanical strength ratio. Also, in recent years, combinations of light alloys (hybrid components) began to be used. Among them, it is possible to highlight the combination of aluminum and magnesium alloys. Both types of alloys can be combined with each other or with other lightweight and resistant materials to create hybrid materials, thereby extending the boundaries of the material's property space [4–8]. Also, magnesium and aluminum present other interesting advantages in relation to sustainability; for example, both are easy to recycle [9,10].

Regarding magnesium, several recycling options exist. Firstly, when the scrap is of sufficient quality, it can be reprocessed to obtain parts with good specifications. Secondly, magnesium scrap can be used to produce other metallic materials. Finally, magnesium can be recycled for use as a raw material in the production of fertilizers [11].

Aluminum has 95% recyclability at the end of its useful life. Also, recycling saves 95% of the energy used in its initial production. Moreover, according to the Environmental Product Declarations (EPD), it was shown that aluminum contains an average of 39% recycled aluminum and 61% primary aluminum. In addition to its recyclability, aluminum has an excellent carbon footprint. The EPD states that, for its different types (anodized and lacquered, with and without a thermal break), the values of CO2 range between 10.3 and 11.8 kg of CO2 per kg of aluminum, and recycling can achieve exemptions of between 3.0 and 4.0 kg of CO2 per kg of aluminum.

The primary non-renewable energy used in the manufacture of aluminum products is another indicator of the impact they generate. As with the previous rate, considering all aluminum types included in the EPD, the primary energy is 420 MJ and 324 MJ per kg of aluminum. Therefore, aluminum recycling provides energy savings of between 49 and 38 MJ [9].

On the other hand, the geometric, dimensional shape and surface requirements are very strict in the aeronautics or aerospace sectors, which makes these parts types expensive and sometimes difficult or even impossible to keep stock of them ready for when it is necessary to maintain or repair damaged parts. Therefore, it is important to guarantee that it is possible to carry out efficient and sustainable repair or maintenance operations, thereby extending the lifetime of these parts and improving sustainability [12–14].

As the density of magnesium (1740 kg/m3) is the lowest among structural metals (being two-thirds that of aluminum and one-quarter that of steel), its alloys are interesting candidates for combination with heavier alloys to reduce weight. However, magnesium has low machinability and high flammability (especially as powder or chips). In fact, the magnesium flame temperature and its alloys can reach 3100 ◦C and, once the fire starts, it is difficult to extinguish since there is continuous combustion of nitrogen, carbon dioxide, and water [15]. Also, molten magnesium reacts violently with water. For these reasons, it is necessary to study the behavior of magnesium when it is mechanized along with other materials (forming hybrid components), testing if such combinations are suitable for manufacturing, repair, and maintenance. For example, steel produces sparks during machining at

cutting speeds between 200 to 300 m/min [16], and this can be extremely dangerous if magnesium is present.

Therefore, when magnesium-containing hybrid components are going to be mechanized, it is necessary to take certain security measures regarding the lubricants or coolant systems employed. Depending on the material or materials with which magnesium is combined, different lubrication/cooling techniques can be used [17,18] both individually (dry machining [19–31], minimum quantity lubrication (MQL) [30–42], solid lubrication [43–47], cryogenic cooling [48–55], gaseous cooling [56–61], nanofluids [62–66], and sustainable cutting fluids [67–72]) and in combination [73–76]. Some of these techniques were tested in several previous works with the intention of better describing the behavior of the individual materials (especially aluminum [19–23], titanium [30,31,39], and magnesium [16,28,77–104]). From the point of view of the optimization of the costs of the process and its sustainability, the ideal would be (1) to be able to completely eliminate lubricants or coolants and carry out dry machining, (2) to test more recently developed techniques (such as machining with minimum quantity of lubricant, cold compressed air or cryogenic refrigerants), and (3) to develop new lubricants or refrigerants compatible with magnesium.

The machining of this type of hybrid component results in an increase in process instability due to the different properties of the materials that form them and their particular cutting characteristics [12]. This makes it necessary to determine the best cutting parameters for each combination of materials, especially when strict design requirements must be reached. Although the literature contains an important number of experimental works with regard the machining of hybrid components, only some of them addressed hybrid components based on magnesium. Most studies focused on machining processes [105–113] and tried to find the optimal combination of machining conditions and lubrication/cooling systems by means of experimental tests, taking the surface roughness required in a particular industry sector or application as a response variable. Others dealt with friction and wear between contact materials [114,115], the effects of pre-treatments on the adhesion of hybrid materials [116], innovative techniques for forming these types of components [117,118], or identification of some of the most prominent issues.

The scope of this work is to prove that repair and maintenance operations can fix holes made in pieces of hybrid components based on magnesium and aluminum, not only in an efficient way, but also sustainably. Therefore, considering the above, this experimental study focused on the drilling of hybrid Mg–Al–Mg components. For decades, aluminum and magnesium were used (separately) in the aeronautical sector due to their good weight/mechanical properties. Thus, given that the density of magnesium is two-thirds that of aluminum, it was hypothesized that they could be used together, reducing weight by replacing aluminum with magnesium where possible.

The drilling process is used regularly in this sector and, thus, it was well analyzed; for only the assembly of the wing to the fuselage, thousands of holes are required [119,120].

The repair and maintenance operations were selected as they represent an additional challenge versus those of manufacturing, since, as previously mentioned, the lack of parts in aeronautical stock means having to do the repair in the shortest possible time to reduce the costs associated with the downtime of the aircraft. Also, two cooling systems (dry machining and cold compressed air) were tested to analyze the sustainability of the process. Surface roughness was chosen as a response variable as it is one of the most widespread in the literature, thereby allowing a better contrast of the results obtained; the required values for the sector are also standardized (0.8 μm < *Ra* < 1.6 μm) [121].

The novelty of the work is in proving that the repair operations of multi-material components (magnesium–aluminum–magnesium) and the parts made of aluminum and magnesium (separately) but assembled to form a higher component can be repaired simultaneously. This approach saves time and reduces cost.

#### **2. Methodology**

As this work is part of a broader research project that involves different geometries, material combinations, cutting conditions, tools, and lubrication/cooling systems, the methodology is similar to that followed in other previous works [105–108] and is based on the guidelines given by Montgomery [122].

#### *2.1. Pre-Experimental Planning*

Here, we report the findings of an experimental study addressing the repair or maintenance of holes made in parts of Mg–Al–Mg hybrid components using sustainable cooling systems. The study focuses on the aeronautical sector. As the response variable, we chose surface roughness since it is commonly used as a reference of quality in aeronautical components and is also used to evaluate the efficiency of the machining processes; thus, there are several works in the literature with which to make comparisons.

To determine the factors, levels, and range of their values, it should be noted that these are repair operations; hence, the depth of cut must be as small as possible to maintain the dimensional design requirements. On the other hand, since these are two non-ferrous alloys with similar machinability characteristics, it would be sufficient to test a unique type of tool. Moreover, since two cooling systems (dry machining and cold compressed air) were tested, and only a single pre-drilled part with eight holes through the Mg–Al–Mg combination was available, it was necessary to adapt the remaining factors and levels to the number of holes. Therefore, we decided to use the feed rate and the spindle speed (and two levels for each) as factors.

Additionally, as it was thought that the surface of the drilled holes could be damaged, not only by the cutting process but also due to friction caused by the chips inside of them (due to the accumulation along the mechanized length), and seeing that a similar factor was taken into account in other works [19–26], it was decided to include two additional factors related to the location relative to the insert (with three levels, one for each one of the stacks: Mg–Al–Mg) and related to the location relative to the specimen, that is, each hole (with two levels, one at the entry of the holes and one at the exit holes), where measurement of the surface roughness was taken.

#### *2.2. Experimental Design*

Considering everything explained in the pre-experimental planning, the depth of cut and the type of tool do not affect the design of experiments (DOE) since they only have one level. For the feed rate, *f* (mm/rev), the spindle speed, *N* (rpm), and the type of cooling system, *C*, two levels were taken for each, i.e., (*f1*, *f2*), (*N1*, *N2*), and (*C1*, *C2*), respectively. In the same way, for the additional factors where the surface roughness was measured, i.e., location relative to the insert, *LRI*, and location relative to the specimen, *LRS*, three (*LRI1*, *LRI2*, *LRI3*) and two (*LRS1*, *LRS2*) levels were taken, respectively. Table 1 describes the factors and levels selected for this experimental analysis.


The surface roughness was taken as the response variable, and the average roughness values (*Ra*) and the average maximum height (*Rz*) were measured in the different zones defined by the factors location relative to the insert (*LRI*) and location relative to the specimen (*LRS*). The factors and levels

are shown in Table 1, and a DOE, as a product of a full factorial 23 and a block of two factors (3 <sup>×</sup> 2), was defined with a total of eight experimental re-drills and 24 measurements of the surface roughness, which provided a total of 48 values (24 of *Ra* and 24 of *Rz*). Also, the design was randomized to reduce the influence of non-considered variables [122] (Table 2).

**Table 2.** Experimental design: product of a full factorial 23 and a block of two factors (3 <sup>×</sup> 2).


*\* f* (mm/rev); *\*\* N* (rpm).

#### *2.3. Performing the Experiment*

Before carrying out the re-drilling tests, it was necessary to collect the specimens of the hybrid parts, the tools, and the cooling systems, as well as introduce the parameter values into the machine tools and establish cutting conditions and data collecting protocols. Next, the machining operations were carried out and, finally, photographs and videos of the trials were taken for subsequent analysis.

#### *2.4. Statistical Analysis of the Data*

Once the machining process was finished, the arithmetical mean roughness (*Ra*) and average maximum height (*Rz*) were measured. The data were statistically analyzed, including an analysis of variance (ANOVA) to identify the influential factors for surface roughness variation and the interactions among them.

#### *2.5. Conclusions*

The main conclusions extracted from the descriptive analysis of the obtained results and their statistical analysis were established.

#### **3. Applications and Results**

#### *3.1. Materials*

The hybrid component specimen was made of three parallelepiped plates of dimensions 50 × 50 × 15 mm pre-drilled with eight holes of 8 mm in diameter. The three plates were mechanically fixed so that they could be easily disassembled to take the measurements inside the holes. The plates placed above and below were of magnesium alloy (UNS M11917), and the other (placed between them) was of aluminum alloy (UNS A92024). The chemical composition of both materials is given in Table 3.


**Table 3.** Chemical composition of the materials used for the manufacturing specimens.

These materials were selected because the authors had previous experience in their machining, both independently [12–14,16–23,25,28–32,77–79,82–94] and together [105–108]. Also, there were interesting works of other researchers in the literature, thus allowing comparisons to be drawn [15,24,26,27,80,81,95–104].

#### *3.2. Tools*

As the target of this study was to analyze the feasibility of carrying out the repair and maintenance operations in an efficient and sustainable way, a single level for the depth of cut factor, *d*, was taken (*d* = 0.5 mm, using of a 9-mm-diameter drill bit). On the other hand, keeping the depth of cut at a low level also helps to keep the cutting temperature low and, therefore, to keep the magnesium temperature far from its ignition temperature.

The tools used in the trials were helical drill bits of high performance. They were made of a high-speed-steel (known as Cobalt Steel, HSSE, or HSS-E), obtained by powder metallurgy (PM) (Figure 1). The tools, with reference HSS-E-PM A1 1257, were purchased from Garant (Hoffmann Iberia, San Fernando de Henares, Madrid, Spain).

**Figure 1.** Helical drill bits HSS-E-PM A1 1257 manufactured by Garant [123].

Their dimensions were 9 mm of diameter, 81 mm of helical length, and 131 mm of total length. In addition, their special geometry allowed self-centering and optimal chip evacuation.

#### *3.3. Machines and Equipment*

The trials were carried out in a Tongtai TMV510 machining center (Tongai Machine &Tool Co., Luzhu Dist, Kaohsiung City, Taiwan) equipped with a Control Numeric Computer (CNC) Fanuc (Fanuc Iberia, Castelldefels, Barcelona, Spain) (Figure 2a). A drilling cycle was programmed that made the tool penetrate 10 mm, and then return to evacuate the generated chips. The same sequence was repeated until the tool crossed the entire width of the piece formed by the three stacks of Mg–Al–Mg. This was done so that the accumulated chips inside the holes did not scratch the surface or stop or hinder the tool inside the piece. Cold compressed air (CCA) was used as the cooling system, implementing a Vortec Cold Air Gun (Vortec, Cincinnati, Ohio, USA) (Figure 2b). The roughness measurements were taken using a Mitutoyo Surftest SJ 401 roughness tester (Figure 2c) with the following settings: measuring range, 800 μm; resolution, 0.000125 μm; transverse length, 25 mm; cut off, 0.8 mm; scan rate, 4 mm (*N* = 5); the standard ISO 1997 [124] was used.

*Materials* **2020**, *13*, 393

**Figure 2.** (**a**) Tongtai TMV510 machining center; (**b**) details of the Vortec Cold Air Gun during the trials; (**c**) Mitutoyo Surftest SJ 401 roughness tester.

#### *3.4. Experimental Tests*

The design of experiments, the materials, tools, machines, and equipment used in the trials, and the parameter value ranges are given in Table 4.



The locations of the measurement zones of the surface roughness are shown in Figure 3. *LRI1* denotes the first magnesium plate, *LRI2* denotes the aluminum plate, and *LRI3* denotes the second (and last) magnesium plate. *LRS* took into account the location of the measuring zone inside each hole after re-drilling it, and its levels were defined as *LRS1* (the specimen entry zone) and *LRS2* (the specimen exit zone). Figure 4 provides a graphical summary of the experimental set-up.

**Figure 3.** The concrete locations of the measurement zones of the surface roughness: (**a**) location relative to the insert, *LRI;* (**b**) location relative to the specimen, *LRS*.

**Figure 4.** The experimental set-up.

#### *3.5. Analysis and Discussion of the Results*

After performing the eight re-drilling tests, surface roughness measurements were made in each hole in the three plates (entry and exit zones). The values of the arithmetical mean roughness (*Ra*) and the average maximum height (*Rz*) were calculated (in micrometers) and are given in Table 5.

**Table 5.** The arithmetical mean roughness (*Ra*) and the average maximum height (*Rz*) obtained during the measurement tests.



**Table 5.** *Cont.*

Initially, a descriptive method was used to analyze the *Ra* and *Rz* values. The obtained results are separated into Tables 6 and 7. Tables 6 and 7 give the values of *Ra* and *Rz*, respectively, in each plate (both in the entry and exit zones of the holes).

**Table 6.** Values of *Ra* in each plate at the entry and at the exit zones of the holes.



**Table 7.** Values of *Rz* in each plate at the entry and exit zones of the holes.

From the *Ra* and *Rz* values given in Tables 6 and 7, the graphics of Figure 5 were drawn. Figure 5 shows the normal distribution of *Ra* (left column) and *Rz* (right column) with respect to feed rate, *f* (mm/rev), (a) *Ra* and (b) *Rz*; spindle speed, *N* (rpm), (c) *Ra* and (d) *Rz*; type of cooling system, *C*, (e) *Ra* and (f) *Rz*; location relative to the insert, *LRI*, (g) *Ra* and (h) *Rz*; and location relative to the specimen, *LRS*, (i) *Ra* and (j) *Rz*.

**Figure 5.** *Cont.*

**Figure 5.** Normal distribution of *Ra* (μm) and *Rz* (μm), respectively, with respect to (**a**,**b**) feed rate, *f* (mm/rev); (**c**,**d**) spindle speed, *N* (rpm); (**e**,**f**) type of cooling system, *C*; (**g**,**h**) location regarding insert, *LRI*; (**i**,**j**) location regarding specimen, *LRS*.

Taking into account that a good behavior of the results is considered when the obtained values are concentrated in the interval [0.8 μm; 1.6 μm] given by the standard [121], a first approach to the analysis can be made by observing data collected in Tables 6 and 7 and graphics from Figure 5. Thus, it was possible to affirm that *Ra* and *Rz* had a similar behavior for the cutting parameters; however, they were perhaps slightly better for high feed rates (*f* = 0.10 mm/rev) and dry machining and they were very similar for both tested values of the spindle speed (perhaps slightly better for low values *N* = 500 rpm). Regarding the location relative to the insert, in both magnesium plates, *Ra* and *Rz* were better than in the aluminum plate. Also, when comparing the results of the first and the last magnesium plates, the results were lower for the latter (*LRI3*). However, *LRI1* was considered as better since the surface roughness values were closer to the standard values used in the aeronautic sector (between 0.8 μm and 1.6 μm) [121]. Finally, regarding the location relative to the specimen, the results were lower at the entry of the holes than at the exit.

The values from Tables 6 and 7 are plotted in Figures 6 and 7, respectively, revealing possible combinations of parameters that could be used for repairing hybrid parts by re-drilling. Figure 8 plots the *Ra* and *Rz* values obtained in the trials and collected in Table 5. All of them were inside of the usual upper and lower limits given in the chart of conversion relations between *Ra* and *Rz*, according to DIN 47 [125].

**Figure 6.** Graphic representation of the *Ra* values.

**Figure 7.** Graphic representation of the *Rz* values.

**Figure 8.** The relationship between *Ra* and *Rz* values.

Observing the *Ra* and *Rz* values in Figures 6 and 7, we see that, for the aluminum plate, the roughness values were higher than for the magnesium plate, especially at the exit of the holes and when using cold compressed air as the cooling system. Therefore, it seems reasonable to select aluminum as the more critical material when establishing the cutting parameters. As the results were better at the entry of the holes, it would perhaps be possible to improve the results by modifying the geometry of hybrid component, for example, by searching for the adequate proportions of the thicknesses of the combined materials. On the other hand, in the second plate of magnesium, most of the surface roughness values (except the obtained one for *f* = 0.05 mm/rev, *N* = 1200 rpm, and *C* = CCA) were lower than those established in the design requirement standards. Therefore, roughness values might be improved by drilling halfway, turning the hybrid component, and then continuing to drill from the opposite side.

Reviewing the surface roughness values at the entry of the aluminum holes (Table 6 column *LRI2*, *LRS1*), it can be seen that test numbers 2, 3, 5, and 7 presented values within the range of the values given by the standard (0.8 μm < *Ra* < 1.6 μm) [121]. Among them, test numbers 2 and 3 were within the same range for the first magnesium plate; test number 5 could be an option if the magnesium plates were about half as thick, since the *LRI1* had values within such an interval; test number 7 indicated room for improvement because the roughness value of the aluminum was close to the lower limit of the roughness required in the aeronautic sector (0.8 μm). In fact, comparing the roughness values obtained in test number 7 with those from test number 8, we propose that a new parameter combination is possible by selecting a spindle speed equal to 1000 rpm or near this value; this would also increase the feed speed, decrease the machining time, and, consequently, improve the efficiency of the process.

In addition, an ANOVA was performed to identify the factors that influence the variation of the response variables, *Ra* and *Rz*. To apply an ANOVA, it is necessary that the variables meet three conditions: (1) each data group must be independent, (2) the results obtained for each group must follow a normal distribution (although a breach of this assumption is supported when the distribution is symmetric), and (3) the variances of each data group must not differ significantly (homoscedasticity).

Using the data extracted directly from the experiment, the *Ra* and *Rz* values did not follow a normal distribution (Shapiro–Wilk test *p*-value < 0.05). Therefore, the data were processed using logarithmic transformation, maintaining its order but softening the effect of outliers.

By this approach, normally distributed Ln*Ra* and Ln*Rz* values (Shapiro–Wilk test *p*-value > 0.05) were obtained (Figure 9). In addition, the condition of homoscedasticity was also fulfilled (Levene statistic, *p*-value > 0.05), and independent data groups had a similar number of cases (Table 8). In the analysis, interactions up to the third order were considered, and successive iterations were performed until all values were significant. In each iteration, the statistically less significant effect was excluded if it had a *p*-value greater than 0.05. Tables 9 and 10 give the outcome of the first and the last ANOVA over *LnRa*, and Tables 11 and 12 collect the outcome of the first and the last ANOVA over *LnRz*, respectively.

**Figure 9.** Probability plots: (**a**) *Ra* and Rz; (**b**) Ln*Ra* and Ln*Rz*.

**Table 8.** Homogeneity test of variances for factors *f* and *N* and response variables *LnRa* and *LnRz*.



**Table 9.** Outcome of the first iteration for the ANOVA over Ln*Ra*.

\* DF, degrees of freedom.


**Table 10.** Outcome of the last iteration for the ANOVA over Ln*Ra*.

\* DF, degrees of freedom.

**Table 11.** Outcome of the first iteration for the ANOVA over Ln*Rz*.


\* DF, degrees of freedom.



\* DF, degrees of freedom.

Taking into account the results shown in Tables 10 and 12, we conclude that the most influential factors, in both cases, were the location relative to the insert (*LRI*) and the type of cooling system (*C*), and that there were no interactions among factors with influence.

Considering the re-drilling surface roughness variability of hybrid Mg–Al–Mg components explained by the statistically significant effect obtained from the ANOVA, the percentage of variability attributed to each factor is shown in Table 13, and the contribution of each effect was obtained as the percentage of the sum of squares values of each significant effect relative to the sum of squares of all significant effects.


**Table 13.** Percentage variability of the statistically significant effects obtained from ANOVA.

#### **4. Conclusions**

This work focused on the maintenance and repair of holes made of hybrid Mg–Al–Mg components by drilling, using two sustainable cooling techniques: dry machining and cold compressed air. The aeronautic and aerospace sectors were selected as relevant applications. In such sectors, the pieces have strict design requirements for surface roughness (0.8 μm < *Ra* < 1.6 μm). *Ra* and *Rz* were taken as response variables. From our analyses, we propose that *Ra* and *Rz* values have similar behaviors and they exhibit the following characteristics:


In addition, from the ANOVA analysis, we found that the factors that influence the response variables *Ra* and *Rz* are location relative to the insert and type of cooling system, with percentages of influence of 72.6% and 27.4%, respectively, for Ra, and 79.2% and 20.8%, respectively, for *Rz*.

With this work, we showed that it is possible to simultaneously repair magnesium–aluminum–magnesium multi-material components and parts made of aluminum and magnesium (separately) but assembled to form a higher component using sustainable cooling systems (dry machining). This approach reduces the time and cost associated with the assembly and disassembly of these types of components during maintenance or repair.

In conclusion, we propose three ways to optimize (or at least improve) the process: (1) using different parameters values (for example, higher values of the spindle speed that increase the efficiency of the process); (2) designing a hybrid component with new proportions of the thicknesses of the materials combined; (3) applying other drilling sequences (e.g., firstly drilling halfway and then turning the part and drilling from the opposite side).

**Author Contributions:** E.M.R., M.M.M., and D.B. contributed to the conceptualization, methodology, and formal analysis. D.B. performed the investigation. E.M.R. and M.M.M. managed the project resources. E.M.R., M.M.M., and D.B. prepared the original draft of the manuscript. E.M.R., M.M.M., D.B., and J.P.D. reviewed and edited

the manuscript. E.M.R., M.M.M., D.B., and J.P.D. contributed to data visualization. E.M.R., M.M.M., and J.P.D. supervised the study. E.M.R. and M.M.M. were responsible for the funding acquisition and project administration. All authors read and agreed to the published version of the manuscript.

**Funding:** This work was partly funded by grants from the Ministerio de Ciencia, Innovación, y Universidades, and the Industrial Engineering School-UNED (RTI2018-102215-B-I00, REF2019-ICF05, and REF2019-ICF08), Spain.

**Acknowledgments:** The authors thank the Industrial Production and Manufacturing Engineering (IPME) Research Group and the Industrial Engineering School-UNED (Projects REF2019-ICF05 and REF2019-ICF08).

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **An Approach to Sustainable Metrics Definition and Evaluation for Green Manufacturing in Material Removal Processes**

#### **César Ayabaca 1,2,\* and Carlos Vila <sup>1</sup>**


Received: 31 October 2019; Accepted: 9 January 2020; Published: 14 January 2020

**Abstract:** Material removal technologies should be thoroughly analyzed not only to optimize operations but also to minimize the different waste emissions and obtain cleaner production centers. The study of environmental sustainability in manufacturing processes, which is rapidly gaining importance, requires activity modeling with material and resource inputs and outputs and, most importantly, the definition of a balanced scorecard with suitable indicators for different levels, including the operational level. This paper proposes a metrics deployment approach for the different stages of the product life cycle, including a conceptual framework of high-level indicators and the definition of machining process indicators from different perspectives. This set of metrics enables methodological measurement and analysis and integrates the results into aggregated indicators that can be considered for continuous improvement strategies. This approach was validated by five case studies of experimental testing of the sustainability indicators in material removal operations. The results helped to confirm or modify the approach and to adjust the parameter definitions to optimize the initial sustainability objectives.

**Keywords:** green manufacturing; sustainability metrics; cleaner product life cycle; material removal processes

#### **1. Introduction**

The analysis of industrial manufacturing processes from the sustainability point of view started during the early 1980s in order to meet the sustainable development concept that arose in the 1970s as a result of a general worry about the global environment due to pollution and the consumption of energy and raw materials. In 1983, the United Nation (UN)'s World Commission on the Environment and Development, known as the Brundtland Commission, prepared a formal report entitled "Our Common Future", in which the concept of sustainable development was defined as "development that meets the current needs of people without compromising the ability of future generations to meet theirs" [1]. Global movements and policies were generated in order to find common strategies that could be applied worldwide [2,3].

Subsequently, in the area of product manufacturing, many research projects were initiated to generate proposals for process improvement and the optimization of the consumption of resources and raw materials [4–6] as well as rules and regulations for waste management for cleaner production methods [7]. At the UN summit in 2015, the 17 goals for sustainable development for 2015 to 2030 were laid down, including goal number 9 (Industry, Innovation, and Infrastructure) which is focused on building resilient infrastructures, promoting inclusive and sustainable industrialization, and fostering innovation [8]. Among others, we could list the most important international initiatives that encourage a better understanding of sustainability:


Since the end of the last century, researchers, technicians, managers, and environmentalists have recommended three economic, environmental, and social dimensions for evaluating economic, environmental, and social aspects, which are known as sustainability dimensions [12].

We must underline the contribution of Zackrisson et al. [13], who studied the relationship between performance measurement systems and sustainability. Although the research was done in Swedish manufacturing companies, they found that at the shop floor level, 90% of the indicators have at least an indirect relationship with one or more of the economic, environmental, or social dimensions, while 26% of the indicators are indirectly related to the environmental dimension.

In materials and manufacturing engineering research topics, indicators are being proposed to measure the sustainability of industrial processes. Reich-Weiser et al. defined a general set of metrics for sustainable manufacturing [14], while Shuaib et al. [15] and Jayal et al. [16] designed a framework and a model for sustainable manufacturing, respectively. Singh et al. developed an expert system for the performance evaluation of small and medium enterprises [17]. With a greater focus on machining technologies, we find the work of Rajurkar et al., which explored how to ensure sustainability and optimize non-traditional machining processes [18].

Since machining technologies pollute and consume energy and raw materials, it is understandable that more research actions are needed. This work is part of the research into the design of green manufacturing activities within the product life cycle, focusing on machining technologies. This work will propose a framework and a list of indicators that will help to get data and information from manufacturing activities as part of the Life Cycle Assessment. The paper is structured into six sections. The Section 2 reviews the state of the art of green manufacturing applied to machining and presents our vision of a green manufacturing activities model. The Section 3 describes the framework for defining the metrics from the point of view of materials, parts, and processes during the product life cycle. The Section 4 describes the validation experiments, while the conclusions and future work are presented in the last two sections.

#### **2. State of the Art in Green Machining Operations**

#### *2.1. Analysis of Previous Works of Sustainability in Industrial Manufacturing Processes*

The aim of this work was to carry out several experiments to calculate different types of sustainable indicators with different materials, processes, and machining centres. Many studies have been carried out on sustainable manufacturing and, in the literature, we can find many focused on specific disciplines, such as machining technologies. Peralta et al. [19] showed the trends emerging in the last 15 years in the sustainability of machining processes from the point of view of the triple bottom line and the three general dimensions: economic, ecological, and equity. Bhanot et al. [20] presented a statistically validated study that proposed a comprehensive sustainability framework for the manufacturing domain to strengthen the enablers and mitigate barriers based on the responses of researchers and industry professionals.

Eastwood et al. [21] developed a sustainable assessment methodology to both improve the accuracy of the existing approaches in identifying the sustainability impact of a product and to assist manufacturing decision makers using unit process modeling and life cycle inventory techniques. The proposed methodology can quantify sustainability metrics by aggregating information from the

process level, where various metrics require different aggregation methods, from the manufacturing process to the manufacturing system level.

The work of Garretson et al. [22] facilitated standards development efforts by harmonizing the terms used to describe production processes. A set of 47 terms focusing on process characterization and describing sustainable production was generated, although terms unique to individual production processes were omitted. The terms were organized into six categories to define the overall concepts: Scope, Boundary, Material, Measurement, Model, and Flow. Then, definitions of the terms were derived from: (a) the literature in sustainable manufacturing and chemical and process industries, (b) process characterization and planning, (c) organization standards, and life cycle assessment and management. The reported terms and definitions are not unique to sustainable production.

Helleno et al. [23] proposed a conceptual method of integrating a new group of sustainability indicators into the Value Stream Mapping (VSM) tool to assess manufacturing processes. The method was applied in three case studies, and the results demonstrated that the proposed method identified different levels of manufacturing process sustainability and thus enabled the development of improved scenarios.

Kluczek [24] introduced an "improvement scenario" in a company producing heating devices, between existing and new processes, based on an approach that can be applied to perform the sustainability assessment of manufacturing processes, requiring less detailed data, time, and expert knowledge but still providing a company-level analysis.

Latif et al. [25] developed an interactive model to determine the sustainability index based on user responses; the model is able to provide suggestions to improve sustainability, as well as carbon footprint reduction, and can assist industry to identify its shortcomings in achieving sustainability, can determine the carbon footprint reduction potential, and can compare the sustainability index as a benchmark measure.

Moldavska and Welo [26] analyzed the different definitions of sustainable manufacturing (SM) and identified the current understanding using an inductive content analysis of definitions published in a variety of academic journals. It is proposed that the findings can serve as a foundation for the development of a common language in both the research field and industrial practice.

In the literature reviewed, Winroth et al. [27] compiled the existing Sustainable Framework of Indicators, in which sustainability assessments can be carried out at different levels within an organization.

The hierarchical dimension of the activities (global, national, corporate, or factory), and the functional dimension (product, supplier, production, logistics, and customer) are shown in Figure 1a. This analysis allows us to select the production indicators at the factory level, at which in each industrial process there are proposals for the measurement of sustainability, among which manufacturing companies use the key performance indicators (KPIs) for the control and monitoring of their processes for continuous improvement.

**Figure 1.** Dimensions selection sustainability indicators modeling. (**a**) Criteria for Metrics Selection [27]. (**b**) Generic Process Flow Diagram [28].

For example, Linke et al. [28] proposed a generic process diagram: resources (raw material, energy, auxiliary materials, etc.) and enablers (the machine, the worker, the tools, etc.) are considered as inputs to the process, as shown in Figure 1b.

The process can be evaluated through performance parameters, indicators, or metrics, which can be related not only to outputs but also to resources, enablers, and inputs in order to give feedback during the manufacturing phase of the product life cycle. The final quality of the manufactured part or the assembled product will be achieved by controlling activities but also by obtaining data and information from the metrics. The waste output generated from the process will be in the form of physical material or environmental pollution.

From the perspective of manufacturing throughout the product life cycle, industrial processes can be analyzed by considering stages such as the extraction of raw materials, manufactured materials, product manufacturing, shipping, distribution, use, recycling, and the final disposal of the product. It is important to emphasize that a certain percentage of the recycled material can be integrated back into the material manufacturing process, while the rest must be used in other applications.

Vila et al. [29] proposed a framework for defining a structured set of metrics that are customizable for operations in different manufacturing technologies. Although the research work was applied to AISI 1018 material turning operations in order to analyze the surface integrity of the part, the contribution established the relationships between the machining parameters of the turning process and the final properties of the manufactured part, such as roughness, microhardness, and other parameters. This was the first attempt to link general sustainable metrics with technological-related metrics. For this reason, several perspectives were proposed. Table 1 shows Product, Process, and Resources (PPR) perspectives and some activity indicators that can be defined during the product life cycle. High-level general indicators aligned with sustainable objectives are initially created for each activity.



Note: Objectives symbol's meaning maximize; minimize.

With this general view of manufacturing sustainable metrics, the next step was to explore previous works in order to define a sustainable scorecard indicator for machining or material removal techniques.

#### *2.2. Dimensions of the Sustainability Metrics for Machining Processes*

In the review of the state of the art, we can find some contributions that define sustainable metrics and show how important it is to define them at different levels of the product life cycle. One of the most interesting works was done by Bhanot et al. [30], in which they analyzed the complex interdependences between parameters that affect the result of a metal cutting process when seeking sustainable objectives. From the literature and from other previous works, we can highlight the manufacturing aspects to measure.

In manufacturing, it is critical to guarantee the competitiveness of each activity from many points of view, and a balanced optimization between the economic, environmental, and social dimensions must be obtained. Therefore, it is necessary to define not only the technological metrics, but also the sustainability metrics aligned with the manufacturing process. For example, for each dimension, we can list some aggregated metrics:


The indicators in the economic and environmental dimensions can be defined through analytic expressions, and they can be quantitatively evaluated using data mining and process calculations. In the social dimension, the indicators are mainly evaluated qualitatively. Some of these indicators are introduced in Table 2. According to the product life cycle phase and for different PPR perspectives—Process, for example—we can find technical metrics associated with the economic dimension (Material Removal Rate), the environment dimension (Cutting Temperature), or the social dimension (Worker Skills).


**Table 2.** Generic indicators for sustainable machining. PPR: Product, Process, and Resources.

The appropriate selection of sustainable indicators allows for the diagnosis of continuous improvement plans in industrial processes, especially in manufacturing processes. However, it is still difficult to define social metrics in manufacturing activities, as Ayabaca and Vila presented in their work and, moreover, to analyze the data [31].

Bhanot et al. [32] presented a study on a machining group in which the interdependencies of different sustainable machining parameters were examined in the context of milling and turning processes.

In order to ensure competitiveness in the manufacturing field, there must be a balance between the economic, environmental, and social dimensions. Gupta et al. [33] presented an experimental investigation that compared empirical and experimental results, which was complemented by a desirability optimization technique, to study the impact on cutting forces, surface roughness, tool wear, surface topography, microhardness, and surface chemical composition in turning the aerospace material titanium (grade-2) alloy, considering Minimum Quantity Lubrication (MQL) conditions.

Hegab et al. [34] developed and discussed a sustainability assessment algorithm for machining processes. The four life cycle stages (pre-manufacturing, manufacturing, use, and post-use) are included in the proposed algorithm. Energy consumption, machining costs, waste management, environmental impact, and personal health and safety are used to express the overall sustainability assessment index. Kadam et al. [35] analyzed the surface integrity in high-speed machining of Inconel 718, and the results show that a good surface finish and residual stresses in compressive regimes can be ensured in the high-speed machining range with low MRR in a water-vapor machining environment, this also being feasible at high MRR in dry cutting.

Benedicto et al. [36] presented a comprehensive analysis of the use of cutting fluids and their main alternatives in machining, focusing on the economic, environmental, and technical dimensions. Zhao et al. [37] reviewed a critical assessment of energy consumption in a machining system at the process, machine, and system levels. Machine tool power demands in different machine states with different components were also discussed, and the predictive methods of energy consumption at different levels were summarized. Energy consumption reduction strategies to achieve sustainable manufacturing were also discussed.

Abbas et al. [38] presented an extensive study of the effectiveness of using different cooling and lubrication techniques when turning AISI 1045 steel. Three multi-objective optimization models were employed to select the optimal cutting conditions. The results offer a clear guideline for selecting the optimal cutting conditions based on different scenarios: MQL nanofluid compared to dry and flood approaches.

Ali et al. [39] found that the tool path strategy has a significant influence on the end outcomes of face milling, considering the surface topography with respect to different cutter path strategies and the optimal cutting strategy for the material Al 2024. Li et al. [40] evaluated the cutting performance of cutting tools in the high-speed machining (HSM) of AISI 4340 by using tools coated with TiN/TiCN/TiAlN multi-coating, TiAlN + TiN coating, TiCN + NbC coating, and AlTiN coating, respectively. A TiN/TiCN/TiAlN multi-coated tool is the most suitable for the high-speed milling of AISI 4340 due to the lower cutting force, lower cutting temperature, and high diffusion resistance of the material. Gupta et al. [41] discussed the features of two innovative techniques for machining an Inconel-800 superalloy by plain turning while considering some critical parameters, reducing the amount of cutting fluid while using sustainable methods. Near dry machining (NDM) will be possible and will solve the problem of chemical components in the fluids being harmful to human health.

In recent research, Gamage et al. [42] used a Taguchi design of experiments and analysis of variance (ANOVA) to identify the significant parameters that optimize the process energy consumption of wire electro-discharge machining (WEDM) of the superalloys Inconel-718 and Ti64Al4V. The results indicate that the preferred parameters to minimize the specific energy consumption are workpiece thickness, wire material, wire diameter, and pulse-OFF time. The reduction of carbon emissions corresponds to the non-working energy consumption of the machines, which is also calculated.

Gunda et al. [43] presented a novel technique for the generation of machining techniques—namely, high-pressure minimum quantity solid lubricant (HP-MQSL) and an experimental setup, with an aim of improving process performance and eliminating the use of cutting fluids in machining operations.

Lu and Jawahir [44] presented a sustainability evaluation methodology for manufacturing processes based on cryogenic machining processes which involves a metrics-based Process Sustainability Index (ProcSI) evaluation. This helps to decide the best cutting conditions from the sustainable manufacturing viewpoint.

Pusavec et al. [45] presented an experimental study of the sustainable high-performance machining of Inconel 718 with the development of performance-based predictive models for dry, near-dry (MQL), cryogenic, and cryolubrication (cryogenic þ near-dry) machining processes using the response surface methodology (RSM). The models developed in the first part of the paper are used in the second part for process evaluation and optimization, to determine the optimum machining conditions for an overall process performance improvement.

Goindi and Sarkar [46] presented a review of all aspects of dry machining, including the sustainability aspects of machining, especially focusing on three research objectives: (1) identifying the areas where dry machining has been successfully adopted and where it has not been possible to do so, (2) reporting on the research work carried out and various alternative solutions provided by the researchers in the area of dry machining, and (3) finding gaps in the current knowledge and suggesting some directions for further work to make dry machining more sustainable, profitable, and adaptable to product manufacturing. Shin et al. [47] presented a component-based energy-modeling methodology to implement the online optimization needed for real-time control in a milling machine. Models that can predict energy up to the tool path level at specific machining configurations are called component models.

Um et al. [48] proposed an approach for deriving an energy estimation model from general key performance indicators of the sustainability of machine tools in the laser welding process of an automotive assembly line and the milling process of an aircraft part manufacturer. ANOVA and RSM are widely used for optimizing cutting parameter tools. Zhang et al. [49] proposed using the Pareto diagram to calculate multi-objective optimization, although this is difficult when there are more than two objectives. This proposal lists and characterizes all the 128 scenarios of sustainable machining operations, considering seven objectives that include energy, cost, time, power, shear force, tool life, and surface finish. The results show that all the scenarios can be converted into a simple objective situation that has a single solution or a set of contradictory bi-objective cases that can be represented on a simple Pareto front.

The use and storage of the calculated indicators, together with modern systems of data acquisition and information management in real time, will strengthen the implementation of advanced manufacturing systems, or what is called Industry 4.0. Activities such as team maintenance and specific service requirements can be planned and adjusted in real time. Gao et al. [50] reviewed the historical development of prognosis theories and techniques and projected their future growth in the emerging cloud infrastructure.

#### *2.3. Sustainable Metrics for Manufacturing Processes*

The first contribution of the research after the review of the state of the art is shown in Figure 2, which gives the Sustainability Approach to Manufacturing Industrial Processes and summarizes the stages of the product life cycle: inputs, enablers, manufactured parts, and waste. This framework was outlined after a deep analysis of manufacturing industrial processes and activities. The activities model was sketched using ICAM Definition (IDEF) methodologies [51] and Unified Modeling Language (UML). Figure 2 presents the top level from the hierarchical model that is later deployed in different layers with more detail of activities and customized for different manufacturing technologies.

The activity model includes, for a manufacturing process, the inputs, which can be either raw material or a geometrical preform from previous manufacturing technology. The input of controls depends on the manufacturing technique, and it is represented here as a Manufacturing Process Group Technique (MPGT). For each manufacturing technology group (i), we can find different variations or techniques (j). The evaluation of the technological process parameters, according to the manufacturing process plan (control), will assure the quality of the final product or the quality of an individual part.

The controls parameters will depend on the technological process that is applied, while the performance parameters, which can be also introduced, define the process efficiency.

**Figure 2.** Activity model of a generic industrial manufacturing processes.

The model describes the general enablers and resources that are grouped into tools and tooling, energy consumption, machine tools, and human resources. For each enabler and resource, we can analyze the different issues and therefore define specific metrics. For example, the model shows countable metrics such as the machine-consumed energy or uncountable metrics such as works formation.

Finally, an important issue is to define the economic impact considering all the activities, inputs, controls, resources, enablers, and outputs. The costs of part manufacturing can be calculated considering the material costs (CM), the cost of tools (CT), the costs of the process (CP), and the costs of waste management (CW).

For the final output, the analysis of all the metrics defined for inputs, controls, enablers, and resources, for each product life cycle phase and from different perspectives, will help to establish whether the result meets the technical, functional, and sustainability requirements.

With this activity model, the next step is to validate it for a manufacturing process and technology that, in our research, will be material removal or machining processes and technologies.

#### *2.4. Sustainability Metrics for Machining Operations*

At the factory level, a manufacturing technology or technique requires one to define correctly the process plan so that the processes, equipment, people, etc., in the production system perform a specific function. Manufacturing process engineering requires a clear and complete description of the information associated with the process and the exchange of data to be of such a quality that forecasts of the results can be obtained. The modeling of systems and processes is expected to standardize the process—what is done, what is controlled, what resources are required—and define the products generated.

For the improvement of industrial quality, a knowledge of industrial processes is required. In this case, we will analyze the process in which it is defined: function, inputs, outputs, resources, and controls that allow measuring performance, as well as the emissions generated, which are evaluated in the context of sustainability.

The second contribution of this research is the matching of the previous general model in a specific model for machining processes and technologies and the definition of metrics in each phase for different machining activities and detailed operations. The activity model for machining is shown in Figure 3, and it represents the material inputs, cutting tool preparation, and the different resources used for machining: cutting fluids, compressed air, energy consumption, facilities consumption, machine tool use, and repayment and human resources.

**Figure 3.** Activity model for machining operations and metrics definition.

For machining processes and technologies, the activity model was defined considering the following basic issues:


product. Apart from these, we introduce indicators that can be evaluated from the sustainability perspective, considering economic, environmental, and social dimensions.


Nevertheless, this activity model is not enough if we want to design a balanced scorecard that includes sustainability indicators. It is obvious that a product's design will have a great influence on how it is manufactured and what materials, processes, and systems are used, as one of the specialists in green manufacturing, Dornfeld [52], indicated. These contributions to sustainability indicators provide the basis for acquiring data during the product life cycle. We propose four main phases for the product life cycle, which include design, manufacturing, use, and end of life, as shown in Figure 4, in order to locate the proposed activity model.

**Figure 4.** Main product life cycle phases where the activity model is positioned.

The approach to these four phases will allow us to define indicators in each one from the PPR perspective

1. Design. This phase includes raw material management and product design and development stages. To design indicators, we consider not only materials flowing from mining but also from recycled products and cause–effect actions on next phases in engineering activities.


Supply chains and materials transport are sometimes considered a separate life cycle phase, but from our point of view, transport and distribution happen throughout the product's life cycle. The complete supply chain is also an integral part of the product life cycle, as these supply chains must produce, deliver, and collect a finished good for use or at the end of its life.

With the activity model for machining and the product life cycle framework, we define the sustainability metrics in our research.

#### **3. A Framework for Sustainability Machining Metrics**

The proposal includes the definition of grouped indicators from the PPR perspective. In this work, we will not include resource indicators, and the Product perspective will be subdivided into Materials and Parts. The name of the grouped indicator will start with the PPR name (Material, Part, or Process) followed by another name related to what we want to measure, as shown in Figure 5.

**Figure 5.** Machining indicators from different PPR perspectives along the product life cycle.

In order to describe the indicators, we have divided the proposal into phases. Table 3 shows how the specific information is organized in this section, and the following tables contain detailed information about the indicators per phase and per PPR perspective.


**Table 3.** General and specific information on sustainable machining indicators.

The definition of the indicators can be used to create a balanced scorecard aligned with a company's sustainability strategy. The first approach to specific indicators is shown in Tables 4–8, and they include both quantitative and qualitative indicators that can be used all along the product life cycle and its name and a short description.

For the first product life cycle phase, Design, the design considerations are strengthened by the search for new materials that have high performances for parts and have allowed a wide range of manufacturing options. High-performance materials and new mechanical and chemical characteristics are incorporated into the databases in product life cycle management (PLM) platforms and help engineers to make the right decision. Table 4 shows the general definitions for the Design phase.


**Table 4.** Design phase indicators definition.

In machining processes, the relationship between the material, geometry, cutting tool, and machine tool opens the field to the research in manufacturing process optimization and new materials. The operating conditions (process parameters for machining) depend on the efficient performance of these four elements. Table 5 describes, for the manufacturing phase, the indicators for the machining process, including turning, milling, drilling, and boring operations, among others.

Machine tools can be manual, automated, or numerical control-driven. Today, most of them are ready for Industry 4.0 connection, and CAD/CAE/CAM applications are associated with shop floor cells and provide data to PLM platforms. The manufacturer's recommended tool parameters (controls) are based on extensive studies of the process, part material, part geometry, and tool performances. With appropriate sustainable indicators, we can improve them. For example, the energy consumption

of the process and its emissions metrics will require experimental measurements on the shop floor to establish the process indicators and online data collection.



Machined parts used as machine components or consumer product components must meet the design specifications, in which the consumables and the consumption of energy sources for their operation must be considered. Table 6 shows the definitions of metrics from this perspective.



The end of life of a machined part can be postponed by maintenance and repairs, which may include processes for the recovery of dimensional tolerances, for which an analysis of surface integrity may be necessary. Table 7 shows the main definitions.

The specific information regarding some indicators is shown in Table 8. The table shows the indicator, the simplified name, acronym, units, goal, and the possible source of the information. The indicator units depend on the specific variables that are measured, and the objective may be seeking the maximum ( ) or the minimum (). The sources of information can be standards or databases of materials, machines, tools, and consumables, which can be taken as a reference for the analyzed process.




Note: maximize; minimize; @ various information sources; # number of.

#### **4. Experimental Development, Results, and Discussion**

In order to validate the metrics definition, a set of experiments were designed. The experiments had the objective of acquiring data for some indicators and to analyze how variations on manufacturing process plan parameters could affect the sustainable indicators.

The experiments were included in a green product life cycle management initiative and, for example, advanced Computer Aided Design and Manufacturing (CAD/CAM) tools were used to prepare the design of experiments for different indicators. It should also be noted that the proposal managed high-level and low-level indicators and indicators in different phases that can be validated on the shop floor. For example, machining time can be determined through the design phase indicators with CAM applications, and the real operation time can be measured in the machine tool numerical control and then compared.

The experiments are summarized in Table 9 and, for each one, the indicators from the PPR perspective are evaluated. For each experiment, a basic machining technology was tested with previous machining simulation arrangements.



Note: maximize; minimize.

In the following subsections, the five experiments are briefly described to show how to obtain data for the defined sustainable indicators.

#### *4.1. Test #1. Material Removal Rate (MRR) and Machining Time*

The objective of this experiment was to set the minimum processing time and the highest possible MRR in a planning process. The design of the experiments considered cutting directions of 0◦, 45◦, and 90◦, and the material was AISI1045 in the Gentiger machining tool (Taichung City, Taiwan), with the cutting tool Mitsubishi VPX300R 4004SA32SA and LOGU1207080PNER-M (MP6120) inserts (Tsukuba, Japan). The cutting diameter Ø = 40 mm, the number of flutes Zc = 4, and the main cutting angle K = 90◦. The operation was done in all cases without cutting fluids.

The CAD/CAM Inventor HSM 2019 application was used to find the 27 possible combinations of the Taguchi method to get the combination of four parameters (ABCD) that reaches the highest MRR and the minimum processing time. In this experiment, A is the direction of the cutting trajectory pass *pd*, B is the depth of cut *ap*, C is the cutting speed *vc*, and D is the feed rate per tooth *fz*. The time is expressed in min:s.

The indicator, shown in Table 9, revealed that the highest MRR material removal rate obtained for the roughing operations was MRR = 10.1 cm3/min. In this operation, the control parameters were *pd* = (0◦ or 45◦ or 90◦), *ap* = 1.2, *vc* = 1671, and *fz* = (0.10 or 0.12 or 0.08).

For the finishing machining operation, the highest material removal rate obtained was MRR = 6.1 cm3/min. In this operation, the control parameters were *pd* = (0◦ or 45◦ or 90◦); *ap* = 0.8; *vc* = 1671; and *fz* = (0.12 or 0.08 or 0.10).

Finally, the minimum machining time obtained for the finishing operation was *tmin* = 2:25 min:s. The process parameters were *pd* = 0◦, *ap* = 0.8, *vc* = 1671, and *fz* = 0.12.

#### *4.2. Test #2 Machining Strategies on Concave and Convex Surfaces*

The objective of this experiment was to find the machining strategy with the shortest machining time among the possible options of the cutting trajectories strategy. The surface geometry was specially designed for the test, and the machined material was AISI1045.

The experimental setup used a high-performance Gentiger machining center, and the cutting tools for surface machining were the Mitsubishi Ball Mill VQ4SVBR0600 (cutting diameter Ø 6 mm) for roughing surface milling and the Mitsubishi VQ4SVBR0300 (cutting diameter Ø 3 mm) for the finishing operations from Tsukuba production centre (Japan). Both were done with the recommended cutting fluids, which encouraged us to optimize its use for sustainability reasons. All the experiments were prepared with the CAM application of the 3DEXPERIENCE 2019 platform. In this case, 21 different options were achieved.

Several evaluations were made on the part that had a concave and convex surface. The test analyzed different options for tool path generation and trajectory strategies in three-dimensional surface milling with all the possible combinations and simulations [53]. For example, with the *style pocketing* surface machining option and the *back and forth* strategy, it was found that the total operation time was reduced by 10%, while only a 3% time reduction could be achieved for the rest of the options compared to the longest one.

In this experiment, we could obtain qualitative information about the cutting trajectory strategy and quantitative information about the operation time and, therefore, the energy consumption.

#### *4.3. Test #3. Roughness, Microhardness, Plastic Deformation*

The objective of this experiment was to find the effect of the machining parameters on the surface quality of the part, which was a quality requirement. The part was a machine axis, and the main machining operations were turning. The part material was AISI1018, and the machine tool was the ROMI numerical control lathe (Sao Paulo, Brasil).

The turning operation cutting tool used was the SANDVIK DNMG 15 06 08-PM4325, and the operations were done with cutting fluids. The machining microprocess plan with trajectory strategies was prepared with the CAD/CAM application SolidCAM 2018 [29].

Mathematical relationships were found to predict these properties and recommendations for the use of different parameters. This test used experimental data from a turning process to determine the influence of the machining parameters in the surface quality, such as the depth of cut, yield strength, plastic deformation, and roughness. The analytical indicator helped us to decide which combination of parameters can be accepted or rejected according to the requirements and dimensional tolerances of the part. Experimental equations were obtained for the selected process, machine, material, and tool.

#### *4.4. Test #4: Roughness and Power Consumption*

The objective of this experiment was to reproduce the machining process plan to build the same part in two different machining centers (A and B) that were geographically distributed. The experiment was reproduced with exactly the same machining process parameters, and the aim was to determine whether the minimum roughness and the minimum power consumption would be obtained with the same parameters [54].

The experiment was carried out in a high-performance Gentiger Machining Center (Taichung City, Taiwan) (A) and in a high performance Deckel Maho Machining Center (Pfronten, Germany) (B).

The part material was AISI1045, and the cutting tool was the Mitsubishi VPX300R 4004SA32SA, with LOGU1207080PNER-M inserts (MP6120). The cutting diameter Ø = 40 mm, the number of flutes Zc = 4, and the main cutting angle K = 90◦. The operation was dry machining, without cutting fluids.

Although both experiments had the same machining cutting parameters, it was discovered that the machine has an important influence on the result and, therefore, on the micro process plan.

**Power consumption**. In Machine #A, the minimum power consumption was 2.79 kWh, with the conditions of a pass direction of 90◦, a cutting depth of 1.0 mm, a cutting speed of 180 m/min, and a feed per tooth of 0.1 mm/tooth. In Machine #B, the minimum power consumption was 4.88 kWh, with the cutting conditions being a pass direction of 0◦, a cutting depth of 0.8 mm, a cutting speed of 140 m/min, and a feed per tooth of 0.08 mm/tooth.

**Roughness**. In Machine #A, the minimum value of Ra = 0.55 μm, with the cutting conditions of a pass direction of 0◦, a cutting depth of 0.8 mm, a cutting speed of 210 m/min, and a feed per tooth of 0.12 mm/tooth. In Machine #B, the minimum value of Ra = 0.83 μm, with the cutting conditions of a pass direction of 45 ◦, a cutting depth of 0.8 mm, a cutting speed of 210 m/min, and a feed per tooth of 0.08 mm/tooth.

The conclusion was that although we had twin machine tools with similar performances, we have to slightly customize the process plan to reach the indicator objective.

#### *4.5. Test #5. Social Dimension Analysis*

For this experiment, the evaluation of the social dimension in machining was the main objective. The experiment was carried out on a shop floor with numerical control machine tools that perform machining processes on various parts at the same time.

An assessment questionnaire was designed and completed by workers, shop floor officers, and middle managers, obtaining the minimum number of answers to validate the method.

The indicators were calculated by the grey relational theory. There were 16 indexes analyzed: Worker Productivity, Relations with Other Workers, Worker's Skill Level, Flexibility of Job Rotation, Punctuality, Top Management Support on Various Issues, Job Satisfaction, Conducive Working Environment, Awareness of Sustainable Manufacturing Initiatives, Technological Upgrades, Financial Support (loans, etc.), Required Product Quality, and Waste Management [31].

The obtained results when evaluating the sustainability indicators in the social dimension, after the application of the Plan, Do, Check, Act (PDCA) continuous improvement cycle were 9.21% higher than the initial evaluation after the implementation of the improvements.

Some improvements were implemented after the analysis of the initial evaluation and the final evaluation of this sustainability dimension, which is one of the most difficult to get information and data analysis for. Fortunately, it helped to implement objective indicators on the machining shop floor.

#### **5. Conclusions**

This paper's contributions can be highlighted in three main areas of interest that have been presented, from the indicators' definition to the shop floor in machining operations.

The first one is the general activity model of industrial manufacturing processes that can be deployed in more detailed activities to identify indicators, where needed, for the manufacturing phase of the product life cycle management.

The second one is the customization of the activity model for material removal and machining processes. In this model, we detected the manufacturing machining process inputs, controls, resources, and enablers in order to define the general, technical, and sustainability indicators. These indicators

can be defined for different product life cycle phases in an organized way, which is why we defined the PPR life cycle phases matrix.

The different experiments carried out provided skills, data, and information about the applied indicators in the manufacturing and materials engineering discipline. They gave real case studies for validating the metrics' definition.

Apart from the manufacturing phase metrics' definition, the use of manufacturing authoring applications within a PLM platform in the design stage can simulate the manufacturing process and help to predict its behavior under the required conditions. Computer-aided manufacturing simulation software has different simulation levels that can help to define and validate machining strategies or manufacturing cell activities to ensure good product quality, achieving sustainable strategies.

In the comparative study of the two machining centers, the lowest roughness and the lowest energy consumption were obtained with different machining parameters. The experiment was carried out on the same material and the same tool, and it was determined that sustainability indicators must be established for each machining center.

When considering surface roughness and the power consumed as the variables to find the best cutting conditions, it was determined that each machining center has its own operating parameters for these conditions. These parameters are related to each other, and the value depends on the material selected, the tool, the machine center, and the lubrication.

The manufacturing parameters that can be tested with virtual manufacturing help to minimize the iterative process when fixing the indicators' objective values. In other cases, it will be more difficult, and we will need to do shop floor measures, as shown in the last experiment.

Finally, sustainability indicators should be evaluated in the product design stage for the best results, and the characteristics of the manufactured part can best be predicted by including the sustainability criteria in the product life cycle management (PLM) platforms.

#### **6. Future Work**

The research plans aligned with this proposal, and we learned multiple lessons, including several key actions. Firstly, we suggest a proposal to incorporate indicators' reports into product life cycle management platforms, in which the sustainability alternatives proposed in the design stage can be evaluated, as part of digital twins' implementation for Industry 4.0 demonstrators. Secondly, further research should carry out experiments to determine the influence of mooring in milling processes and its influence on the quality of the part and the sustainability indicators. Finally, future research should focus on developing a system for evaluating sustainability indicators that can quantify the increase in the indicators when the product is improved, while the options or alternatives are analyzed in the design and manufacturing stages.

**Author Contributions:** C.A. conceptualization and writing—review and editing, analyzed the results, and prepared the draft manuscript. C.V. supervised the research, reviewed the analysis of the results, and wrote the final manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Escuela Politécnica Nacional (Ecuador) Research Project: PIS 16-15, the Universitat Politècnica de València UPV (Spain) and the Carolina Foundation (Spanish Government Scholarships) Call 2017.

**Acknowledgments:** The authors would like to express their gratitude for the support provided by the Escuela Politécnica Nacional (Ecuador), the Universitat Politècnica de València (Spain) and the Carolina Foundation (Spanish Government) with the corresponding grants.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Sustainable Lubrication Methods for the Machining of Titanium Alloys: An Overview**

**Enrique García-Martínez 1,2, Valentín Miguel 1,2,\*, Alberto Martínez-Martínez 2, María Carmen Manjabacas 1,2 and Juana Coello 1,2**


Received: 29 October 2019; Accepted: 19 November 2019; Published: 22 November 2019

**Abstract:** Titanium is one of the most interesting materials in modern manufacturing thanks to its good mechanical properties and light weight. These features make it very attractive for use in the aeronautical and aerospace industries. Important alloys, such as Ti6Al4V, are extensively used. Nevertheless, titanium alloys present several problems in machining processes. Their machinability is poor, affected by low thermal conductivity, which generates very high cutting temperatures and thermal gradients in the cutting tool. Lubricants and cutting fluids have traditionally been used to solve this problem. However, this option is unsustainable as such lubricants represent a risk to the environment and to the health of the operator due to their different chemical components. Therefore, novel, sustainable and green lubrication techniques are necessary. Dry machining is the most sustainable option. Nevertheless, difficult-to-machine materials like titanium alloys cannot be machined under these conditions, leading to very high cutting temperatures and excessive tool wear. This study is intended to describe, analyse and review the non-traditional lubrication techniques developed in turning, drilling and milling processes since 2015, including minimum quantity of lubricant, cryogenic lubrication, minimum quantity of cooling lubrication or high-pressure coolant. The aim is to provide a general overview of the recent advances in each technique for the main machining processes.

**Keywords:** titanium alloys; sustainable lubrication; cryogenic lubrication; MQL

#### **1. Introduction**

Because of their mechanical properties, titanium and titanium alloys are one of the most commonly used materials in manufacturing processes in certain key industries such as the aeronautical, aerospace and medical ones. These alloys exhibit high strength while also being very light [1], and also have very high wear and corrosion resistance. These unique characteristics, along with their possessing the highest strength to weight ratio [2], make these alloys very attractive for such industries. Titanium and titanium alloys offer the best mechanical features possible for applications in which low weight is required.

Nevertheless, pure titanium and titanium alloys present several problems for machining processes, which mean they are classified as difficult-to-machine materials. Their low modulus of elasticity and extreme strength at high temperature [3], and inferior thermal conductivity generate long ductile chips and relatively large contact length between chip and cutting tool in machining processes. Thus, very high temperatures are reached and aggressive thermal gradients appear in the cutting tool. Their low thermal conductivity and high heat capacity play a critical role in the heat dissipation process [4], which finally causes tool wear and a rapid reduction in tool life.

Furthermore, the plastic deformation of the material, friction and high chemical affinity of the titanium and the cutting tool materials produce built up edge (BUE), affecting the geometry of the tool and impairing the surface integrity of the final machined part.

The combination of all these conditions reduces the machinability of titanium alloys, the improvement of which is one of the greatest challenges recently addressed by a large number of researchers. Such researchers are striving to provide new machining strategies for this material, as well as to determine optimal cutting conditions in the machining processes.

Ti6Al4V is the most significant titanium alloy from the point of view of industrial applications. Table 1 shows the distribution of the most frequently used titanium alloys in the literature we have reviewed. Ti6Al4V alloy is estimated to account for 50% of global titanium metal production, and 80% of this corresponds to the aerospace and medical industries [2]. Ti6Al4V is an *α* − *β* titanium alloy, in which the amount of beta stabilizer added is about 4–6%. These *α* − *β* alloys can obtain different mechanical properties by heat treatment and their characteristics are optimal for application with warm temperatures (400 ◦C) [5].


**Table 1.** Proportion and composition of titanium and titanium alloys in the reviewed literature.

Ti553 alloy is a relatively modern alloy, which is gaining prominence in applications in the automotive, chemical and medical industries because of its excellent properties, such as its good strength-to-weight ratio and high hardness at heating state [4]. The machinability of this alloy has not been extensively studied in comparison with Ti6Al4V, and thus the optimal conditions for the machining of this material are still far from determined.

Titanium aluminide (TiAl) is an intermetallic chemical compound of titanium and aluminium as base metals and other elements in small proportions [6]. This material achieves superior properties compared to traditional titanium alloys, such as excellent heat, corrosion and oxidation resistance. Moreover, it is very light, thanks to its high proportion of aluminium. However, it exhibits poor ductility and low fracture toughness [5].

TC17 is an alloy with high toughness with some applications in gas turbine engine components. Althought Ti6Al7Nb alloy has similar properties as Ti6Al4V, no general applications are found with it.

#### *1.1. Traditional Lubrication Approach in Titanium Machining Processes*

Flood lubrication with abundant quantities of lubricant has traditionally been the focus of attempts to overcome the machinability problems of titanium alloys [7]. The lubricant, coolant and cutting fluids (CFs) industrially used are estimated to account for about 17% of the total manufacturing costs of the final part [8], while the cutting tool costs are only 4% of total machining costs [9]. The cost factor, along with the environmental and health risks associated with the use of these cutting fluids, are the main reasons for attempting to reduce their use. Cutting fluids used as cooling agents contain environmentally hazardous and harmful chemical elements [6] and may cause diseases, such as respiratory problems, asthma and cancer.

The functions of the cutting fluids include lubrication and cooling, which can be carried out in parallel. In short, the objectives pursued by the application of these fluids is to improve the dissipation of the heat generated in the cutting area, which is very high in the machining of titanium alloys (up to 1000 ◦C), and to reduce the friction between the chip and the cutting tool.

The better or worse performance of the cutting fluid depends on the machining process, as well as the cutting conditions and the fluid characteristics, such as density, viscosity and specific heat. According to Lin et al. [10], the properties of the cutting fluid may affect the cutting conditions when turning Ti6Al4V, depending on the lubrication method. Three main types of cutting fluids exists: mineral, semi-synthetic and synthetic. Synthetic and semi-synthetic fluids are aqueous based fluids in which the good heat conduction of water combined with the oil properties enhances the performance of the cutting fluid.

#### *1.2. Need for Sustainability*

As previously explained, cutting fluids are damaging for the environment and human health, apart from involving high additional costs in the machining processes. There has been some efforts for using sustainable cutting fluids based on vegetable oils like sesam, coconut, sunflower, palm and others [11]. Some applications of this kind of fluids increases the cutting tool life a 170% in drilling of materials difficult to cut as AISI 316 steel. Coconut oil improves significantly the efficiency of machining processes and with the palm oil better results for Ti6Al4V alloy have been obtained compared to traditional flood lubrication [11,12]. Nevertheless the sustainability of this kind of fluids are controversial nowadays, specially palm oil.

Thus, the next qualitative advance in cutting fluids consists of the use of synthetic fluids made out of a liquid vegetable base and the addition of nanoparticles as Al2O3, MoS2, diamond and graphene [11]. The synthetics fluids reduce the cutting forces and temperature and improve the surface finishing after machining. Moreover, some of these components act in an efficient way in applications in which friction at high pressure appears [13]. Nevertheless the cost of these lubricants is much higher than traditional ones [12].

For these reasons, in recent years, different lubrication and cooling techniques have been developed with the goal of reducing the use of cutting fluids and improving the machinability of titanium alloys under environmentally-friendly conditions [7,12,14–16]. Figure 1 shows the evolution of the number of papers related to the principal sustainable lubrication techniques.

**Figure 1.** Evolution of principal sustainable machining techniques for titanium alloys.

These modern techniques provide different solutions for the cooling and lubrication problem, reducing the quantity of lubricant, as in the case of the minimum quantity of lubricant technique (MQL), or substituting the harmful lubricant for another green substance that can provide the cooling action, as in the case of cryogenic lubrication with liquid nitrogen, LN2. This paper reviews the

main sustainable techniques, such as minimum quantity of lubricant (MQL), cryogenic lubrication, minimum quantity cooling lubrication (MQCL) and high-pressure cooling (HPC), based on the recent literature since 2015.

Although dry machining can be considered a sustainable technique due to the lack of any lubricant, it cannot be applied in titanium and titanium alloy in most cases because of the high temperatures reached, which cause excessive tool wear and considerably reduce the quality of the machined parts.

The minimum quantity lubrication method (MQL) involves the atomization of cutting fluid droplets mixed in the air and directed at the cutting interface of the tool. The MQL technique is comparatively less hazardous for the environment than flood lubrication because it consumes very little toxic and non-biodegradable cutting fluid and needs less energy to pump the atomized fluid at higher flow rates and pressure [14].

The MQL strategy limits the cutting temperature by reducing the friction force tool-chip during the cutting process, thanks to the effective lubrication directly applied on the rake face of the tool. Compared to traditional flood lubrication, MQL requires much less than 1 L/h of lubricant [17] which is a great decrease compared to over 100 L/h adopted flood lubrication [12], although specific equipment is needed to atomize the lubricant droplets. In addition, vegetable-based oils are extensively used in this lubrication method [8].

Although air contributes to refrigerating the cutting interface and to dissipating heat by using the convective heat transfer mechanism, it has been proven for difficult-to-machine materials, like titanium alloys and nickel-based materials, that the MQL technique is not completely suitable because of its low heat transfer capacity [10] and its role depends on the specific machining process and cutting conditions. Typically, the cooling action of the air is not sufficient to reduce the cutting temperature, being the main drawback of MQL lubrication. Moreover, Mathew et al. [6] demonstrated a low efficiency of the MQL lubrication for deep hole drilling. However, it has been extensively demonstrated that MQL method is able to reduce cutting forces and to improve tool life compared to dry machining, and, depending on cutting conditions, compared to traditional flood lubrication [1]. Mathew et al. [6] reported MQL improvement on Ti aluminide drilling compared to flood lubrication based on cutting forces thanks to better penetration of the lubricant when eliminated chips obstruct its passage. Benjamin et al. [17] explain that the MQL method is able to overcome the formation of a vapor blanket in the cutting zone, which inhibits the effective lubrication effect of flood cooling in operations with low machinability materials.

As a result of the low quantity of lubricant applied in this technique, metallic chips generated during the process are almost dry and can be easily recycled [1].

Definitively, the MQL method solves the friction problem between the chip and the cutting tool by means of effective lubrication with a low quantity of lubricant compared to flood lubrication, which makes it a sustainable, environmentally friendly lubrication technique. Nevertheless, MQL does not enhance the cooling effect efficiently and is unable to rapidly dissipate the heat generated during the machining process.

The MQL approach has been developed to improve the cooling action. Many authors have demonstrated the improvement of mixing oil with water as a mode of transport and cooling source [10]. This technique, called Oil on Water (OoW) MQL, uses water droplets to transport a small quantity of lubricant. Specific equipment is required.

With this improvement, lubricant oil performs the lubrication effect, reducing the friction coefficient between chip and tool, while water, when evaporated, provides a cooling effect, which is more effective than supplied air flow. Briefly, when the water and oil droplet reach the objective surface, water evaporates, decreasing the cutting temperature, and oil remains, forming the lubricant film that reduces the friction force [14].

Another variant whose purpose is to improve heat dissipation in the MQL method is the minimum quantity of cooling lubrication technique (MQCL) [15]. Figure 2 shows the schematic diagrams for MQL and MQCL set ups. In this method, MQL is combined with sub-zero cooling, provided by low temperature air to enhance heat transfer from the tool-chip interface [17]. This system uses temperatures below 0 ◦C, but not as low as those achieved in cryogenic lubrication. For both MQL and MQCL procedures, the droplet size of the lubricant must be controlled and always greater than 5–10 μm. If the particles are smaller than that value, they can remain in the air and might cause healthy problems in the machine laborer [15,18]. Pervaiz et al. [8] performed their investigation on turning of Ti6AL5V titanium alloy by using vegetable-oil based lubricant at a flow rate of 60–100 mL/h and supplied air at −4 ◦C. They found that this MQCL cooling strategy is a successful substitute for conventional flood cooling methods, mostly at high cutting speeds.

Benjamin et al. [17] carried out an investigation on milling of Ti6Al4V alloy under the MQCL approach, using refined palm oil as lubricant at a flow rate of 350 mL/h and a cold temperature of −10 ◦C. They found surface roughness improvements compared to the MQL technique.

Nevertheless, the optimal conditions of this technique are still under development; Maruda et al. [19] establishes the threshold of useful conditions in terms of droplets diameters and emulsion mass flow and state that the number of droplets that arrives to the part surface are influenced by the volumetric air flow, and the device-part distance. Anyway, some research is being done regarding to improve the performance index of this kind of lubrication, i.e., the use of extreme pressure-antiwear additives [20].

**Figure 2.** Schematics of MQL and MQCL systems.

The cryogenic lubrication method has been extensively studied in recent years. In cryogenic lubrication, liquid nitrogen LN2 is usually supplied by means of small-diameter nozzles at a temperature range between −194 ◦C and −200 ◦C [16]. The significant cooling action produced is the greatest advantage of this strategy, compared to the MQL method. Cryogenic lubrication is able to effectively control the cutting temperature, reducing the heat generated in the cutting zone, which represents a substantive improvement on the machining of titanium alloys thanks to the high thermal gradients that appear in the machining operations.

Although this technique has been widely studied in the literature on turning, milling and drilling processes, it is not clear under what conditions the improvements are achieved and whether it is economically profitable, due to the expensive cost of liquid nitrogen [15]. The price of liquid nitrogen is a limiting factor for the application of this lubrication technique since large quantities of nitrogen are required, as well as special equipment. Moreover, in many industries, the savings that can be achieved by increasing the life of the cutting insert or reducing the cutting oil is not sufficient compared to the cost of liquid nitrogen. Nonetheless, LN2 is a green lubricant, being harmless for the environment and human health, which makes this technique attractive. A variant of the deployment of liquid nitrogen is the use of dry ice or supercritical CO2 [14]. The differences lie in the temperatures. While liquid nitrogen boils at −178 ◦C, carbon dioxide boils at −78 ◦C [21]. In addition, the preservation methods are different. LN2 must be kept at atmospheric pressure and low temperatures in insulated containers, needing specific equipment, while CO2 needs to be preserved at room temperature and high pressure, about 55 bar.

The cooling effect of CO2 is due to the expansion of the gas when it comes out through the nozzle. A pressure drop takes place, resulting in a phase change into liquid and solid due to the Houle-Thompson effect, which promotes the cooling of the cutting area [22].

Shokrani et al. [2] found that cryogenic lubrication with LN2 on the milling of the Ti6Al4V alloy reduced surface roughness by 40% in comparison to flood cooling, while tool life incremented almost three times. Nevertheless, Dix et al. [23] found, using the Finite Element Method (FEM), that cryogenic cooling produced high torque and force in the axial direction on drilling in comparison to dry machining.

Bordin et al. [24], reported that cryogenic cooling is the optimal solution and most suitable method for machining titanium parts in case of medical applications, because it minimizes the further severe cleaning requirements thanks to the lack of oil lubricant.

High-pressure cooling (HPC) is a technique whose principle is not based on the reduction of the quantity of lubricant, but on its efficient use, through effective penetration in the contact zone between the tool and the chip and in the flank face. In comparison with flood cooling, in which the average flow rate of lubricant is about 1.7 L/min [12], HPC utilizes a greater quantity of coolant. Ezugwu et al. [25] performed their study by using flow rates between 18.5 L/min and 24 L/min, while Mia et al. [26] utilized a flow rate of 6 L/min at a pressure of 8 MPa. For these reasons, HPC should not be classified as a green, environmentally friendly lubrication technique.

Laser-assisted machining is a thermally assisted machining process in which the method of operation is notably different from cryogenic lubrication and which has been gaining popularity in recent years [27]. In this modern technique, the specific area of the workpiece is preheated before the cutting process to reduce the flow stress and enable chip formation. This method, effective for machining processes of difficult-to-cut materials, such as titanium alloys, requires specific equipment and high precision in the control of laser variables [28]. The power of the laser and its movement velocity are critical parameters to promote improvements in the reduction of the cutting force. Nevertheless, diffusion can be increased thanks to the temperature reached by the material (about 500 ◦C). Therefore, the wear processes involved can be accelerated. In addition, thermal shocks can be produced by the temporal delay existing between the material overheated and the arrival of the cutting tool. Tool damage would appear if the laser variables are not properlly controlled [29].

#### **2. Literature Review Methodology**

This study is strictly focused on the machining processes of titanium and titanium alloys, considering the importance of these materials in current industrial manufacturing processes. The main objective of this review is to analyse the state of the art of the principal modern lubrication techniques that have been widely developed over approximately the last 20 years.

The literature includes many reviews on these lubrication techniques. However, they either give a limited explanation of each method or focus on only one of them, analysing a number of studies on one machining process.

The purpose of this paper is to provide firstly a general overview of each technique, and then a detailed analysis of each one, addressing the principal machining processes (turning, milling and drilling), based on the most novel works found in the published literature, and always with regard to titanium and titanium alloys. For this reason, this work review s from 2015 to 2019. It has been conducted using the Web of Science search engine.

Machining has been used as the general topic for the research. Terms such as MQL, HPC, cryogenic, MQCL, or the same terms written in full have been included in the search for papers related to each technique and filtering from the year 2015. All the papers reviewed are classified in Table 2, according to the type of process, material used, technique evaluated, number of citations and impact of the journal. The number of papers for each technique is shown in Figure 3.

**Figure 3.** Number of papers on different techniques. Minimum quantity of lubricant (MQL), cryogenics (Cryo), high pressure coolant (HPC), minimum quantity of cooling lubrication (MQCL).

Cryogenic and minimum quantity of lubricant are the two most extensively studied techniques, while high-pressure cooling is only used in turning processes. Other techniques, such as dry machining, ultrasonic vibration assisted machining, machining with specific inserts or electrostatic high-velocity solid lubricant machining have been classified together.

A statistical analysis of these papers has been developed with the aim of obtaining the main parameters studied for each of the principal machining processes, such as cutting forces, surface integrity, cutting temperature, tool wear, chip morphology, friction, specific energy or power consumption. The principal variables studied in each paper on drilling, milling and turning are shown in Tables 3–5.

Tables 3–5 reveal that the most commonly researched aspects are cutting forces, tool wear and surface integrity of the machined parts, while parameters such as friction coefficient or power consumption are rarely analysed, Figure 4. Cutting temperature is an important variable that is the subject of less research than cutting forces or tool wear due to the difficulty of its being accurately measured in the cutting area.

With the aim of providing a more detailed overview of the state of the art, the highest-ranked papers on drilling, milling and turning have been filtered, firstly taken into account the number of citations and the impact index of the journal. In this way, a number around 15 papers have been managed for turning and milling processes; for drilling, all papers have been carefully analysed as the few number of total papers published. After that, a second filtering step has been applied based on a detailed reading, selecting a balanced quantity of papers about each main lubrication technique. This led to take into account eight manuscripts for turning and milling processes.

For the selection of these papers, a content analysis has been made that considers the influence of the lubrication process on the technological capacity of the processes in terms of surface roughness, cutting forces, temperature and tool life. In addition, it has been evaluated qualitatively the discussion and conclusions carried out by the own authors in comparison with traditional lubrication.

The main strenghts of the review methodology applied herein are the relevant and complete classification of involved papers based on the machining process, the lubrication technique used and the different variables analysed and researched by the authors. This permits to locate in an easy way the points of interest of each paper to be considered as background for specific researchings. Likewise, based on some representative papers, the different lubrication techniques and their effect have been carefully analysed providing the grounded fundamentals of them.


**Table 2.** Classification of reviewed literature (June 2019).

#### **Table 3.** Main variables studied in papers on drilling.



**Table 4.** Main variables studied in papers on milling.

**Table 5.** Main variables studied in papers on turning.


Nevertheless, the methodology is conditioned to the use of a single search engine that, althought is the most important one, might set aside some significant papers. Besides, sometimes, the keywords in the papers could not be well identified according to our searching criteria.

**Figure 4.** Main criteria mentioned in papers on turning, milling and drilling.

#### *2.1. Sustainability in Milling Processes*

A detailed review of the most significant sustainable techniques in milling processes was conducted. The papers selected are shown in Table 6.

With this selection of milling research papers, a general overview can be obtained about the state of art of each of the main sustainable techniques.

Shokrani et al. [2] investigated the application of cryogenic lubrication with liquid nitrogen LN2 in the milling process of Ti6Al4V alloy, compared to dry and flood lubrication. They performed the experimental investigation by supplying cryogenic nitrogen around the cutting tool at −197 ◦C, 1.5 bar of pressure and with a flow rate of 0.4 L/min, but without submerging the workpiece in LN2. For the experiments, they used a TiN-TiAlN coated solid carbide end mill and tested three cutting velocities (30 m/min, 115 m/min and 200 m/min), three feed rates (0.03 mm/tooth, 0.065 mm/tooth, and 0.1 mm/tooth) and three depths of cut (1 mm, 3 mm and 5 mm).

This study examined the potential of the cryogenic technique to improve surface roughness and tool life, thanks to decelerating thermally induced wear, while reducing power and energy consumption. The authors found that, on average, surface roughness was 30.42% lower in comparison with dry and flood machining. They also found that the environment (dry, flood or cryogenic) has a 21.5% contribution on final surface roughness and a 73% contribution in power consumption.

Park et al. [1] obtained poor results for the application of only cryogenic LN2, both by the internal and external method. The cutting force at the beginning of the process was lower than for flood lubrication, but a significant increase was obtained at the final pass. In addition, the criterion of tool rupture was achieved in both cases. The reduction of tool life is explained by the lack of effective lubrication due to the deep axial depth of cut with a large contact area between the tool and chip, causing excessive adhesion on the tool.



These authors also investigated the end milling process of Ti6Al4V alloy, comparing different techniques, such as flood coolant lubrication, MQL mixture with Hexagonal Boron Nitride nano-particles and the combination of internal cryogenic and Nano-MQL. They conducted the experiments at cutting speeds of 72 m/min for flood lubrication and 86 m/min for the other techniques, feed rate of 0.1 mm/tooth and depth of cut of 24.5 mm. They used a tool coated with Aluminium Chromium Nitride (AlCrN).

They found that the application of Nano-MQL always reduces the cutting forces in comparison with flood cooling, obtaining the best results for the combination of Nano-MQL and internal cryogenic. In addition, tool life was enhanced by up to 32%, the reasons being the effective lubrication supplied by Nano-MQL and the reduction of cutting temperature thanks to cryogenic lubrication.

In another study, Park et al. [9] analysed face milling and end milling of Ti6Al4V alloy under dry, flood and MQL. They used an uncoated insert and the conditions for the face milling experiments were cutting speed of 47.7 m/min, 76.4 m/min, 100 m/min and 120 m/min, feed rate of 0.15 mm/rev and depth of cut of 2 mm. For the end milling experiments, the cutting speeds were 72 m/min and 90 m/min, the feed rate was 0.1 mm/rev and the depth of cut was 1.5 mm. They tested two MQL methods: the application of a vegetable oil at the tool-chip interface at 5 bars and 3 mL/min and MQL mixture with exfoliated graphite nano-platelets (MQLN). They showed that MQLN is most effective at a high cutting speed because nano-particles play the role of a lubricant reducing the friction between the tool and work material.

They found MQL reduced cutting force compared to dry and flood machining in all cases, obtaining the best performance in combination with cryogenic conditions.

Similar results were obtained by Tapoglou et al. [22]. These authors reported that the best performing cryogenic method was CO2 in combination with the MQL technique, being better than MQL and the combination of CO2 and air when they used a single insert. When they used five inserts, they found that at 70 m/min and 80 m/min, the combination of MQL and cryogenic CO2 improved tool life by up to 29% and 32%, respectively, compared to MQL. Nevertheless, the application of cryogenic CO2 at maximum velocity reduced tool life in comparison with the MQL technique, thus not being an efficient lubrication method.

Similarly, Benjamin et al. [17] focused their study on comparing MQL and MQL with sub-zero cooling on end milling of Ti6Al4V alloy. They used refined palm oil as lubricant with a flow rate of 350 mL/h and a pressure of 2 bar. The cold stream was supplied by a Vortex tube at −10 ◦C and 6 bar. The cutting conditions were cutting speed between 90 m/min and 150 m/min and feed rate from 0.025 mm/rev to 0.075 mm/rev.

They found that lower temperatures were achieved under the MQCL technique in comparison with MQL with cold air, which increased the viscosity of the palm oil, improving its lubricating properties.

In addition, with MQCL, the tool life was increased from 11.06 min to 15.9 min and flank wear decreased by 19%.

Park et al. [9] examined what increase in flow rate is needed in the MQL technique as cutting velocity increases. They showed that an optimal flow rate exists for each cutting condition that improves tool life.

Celik et al. [34] studied the influence of cryogenic treatment of WC-Co end mills for machining Ti6Al4V alloy. They analysed the behaviour of coated and uncoated cryogenically treated end mills over 12 h, 24 h and 36 h, developing the experimental tests under dry condition. They found that 36 h cryogenically treatment reduced the friction coefficient and friction forces. In the milling process, an increment of cutting forces was demonstrated in each pass due to toll wear. Increasing the cryogenic treatment times from 12 to 36 h positively affected the coated and uncoated tools, except for the AlTiN sample, in which the coating substrate interfacial adhesion bond was weakened.

AlCrN coated tools with 36 h of cryogenic treatment showed the best performance for both tool wear and cutting forces, while AlTiN coated tools exhibited lower performance with the treatment. In conclusion, it can be said that cryogenic treatment is not an effective method to improve milling process performance when working with an AlTiN coated tool.

Bermhingham et al. [27] studied tool life and wear in laser assisted milling of Ti6Al4V. They performed their face milling experiments under dry conditions, MQL, flood coolant, LAM and a combination of LAM with MQL with the aim of analysing the influence of the environment in tool life. Using a PVD coated (TiN) mill, they tested three laser power levels (50 W, 100 W and 150 W) to analyse the influence of overheating the workpiece material.

They found that laser-assisted milling increased tool life at some cutting speeds and decreased it at others. LAM at high power (150 W) was found to produce the lowest tool life of any test, which means that overheating reduces tool performance due to thermal shock. Compared to the rest of lubrication techniques, laser-assisted milling generated an improvement in tool life at low velocity. For the combination of LAM with MQL, no failure appeared at 69 m/min after 28 min of testing. It was verified that for high cutting velocities, laser-assisted milling has a detrimental effect on tool life due to thermal shock, while the combination of LAM with MQL is capable of improving tool life as the cooling action of MQL slows the rate of thermal wear processes.

Sim et al. [29] analysed the optimal laser power and inclination angle for laser-assisted milling of Ti6Al4V alloy. Using FEM, they studied the influence of the milling rotation angle in relation to the inclination angle of the workpiece surface with respect to the horizontal surface. They found that for inclination angles greater than 30◦, the preheated temperature decreased in line with the increase of the inclination angle in the tool path until 75◦, from which a new increment took place, meaning that the minimum value occurs for 75◦.

In the experimental tests, they found that the cutting force decreased depending on the increase in the path inclination angle. They showed that the preheating temperature decreased as the tool path inclination angle increased, but the preheating temperature increased when the tool path inclination angle was 75◦ or more, which corresponds to the simulated results.

#### *2.2. Sustainability in Turning Processes*

There now follows a detailed review of the most important sustainable techniques in milling processes. The papers selected are shown in Table 7.



This selection of papers on turning provides a general overview on the state of the art of each principal sustainable technique.

Mia et al. [3] performed their experimental study on turning of Ti6Al4V alloy by applying high-pressure coolant technique on the flank and rake surfaces of the tool. They compared the difference in surface roughness and tool wear between machining at dry and HPC conditions, using a coated carbide tool (TiN, WC and Co).

They found that double jet action significantly improved the heat transfer because the coolant is able to reach the point of highest temperature by overcoming the chip obstruction and removing it from the cutting zone. Using ANOVA analysis, they found that the environment lubrication has an 18% influence on the final surface roughness. HPC reduced the contact length between chip and tool and inhibited the probability of BUE formation, improving the tribological interaction. Similar results were obtained by Busch et al. [21].

Nevertheless, although HPC decreases the surface roughness at high cutting speed, at low cutting speed the effect was contrary, due to the sliding of chips over the tool surfaces. If coolant output pressure is not sufficient to break the chip and cutting velocity is low, the adhesiveness of the chip is enhanced.

Pervaiz et al. [8] studied the effect of minimum quantity cooling lubrication (MQCL) on Ti6Al4V turning. They used a vegetable oil-based MQL system at flow rates between 60 mL/h and 100 mL/h. The cooling action was added by pressurized air at 0.5 MPa and −4 ◦C and an uncoated cutting insert was used.

It was found that the application of different oil flow rates under MQCL had no impact on surface roughness. At 90 m/min and lower feed levels, the cutting force showed a decreasing trend with the increment of lubricant flow rate due to effective lubrication. Nevertheless, at the highest feed level, a higher flow rate increased the cutting force. Pervaiz et al. suggested that increased cooling of the workpiece helps maintain material hardness without thermal softening. Similar conclusions were reached by Lin et al. [10] when they analysed the turning process of Ti6Al4V under MQCL conditions. They found that although the lowest air temperature of −26 ◦C provided a better cooling effect by reducing the cutting temperature, the lowest cutting forces and surface roughness were obtained for −16 ◦C of temperature. This is due to the reduction of Ti6Al4V temperature that enharden the material. However, both temperatures provided better results than for dry and wet machining conditions.

Sartori et al. [48] conducted their research on semi-finishing turning of Ti6Al4V ELI using cooled gaseous nitrogen, comparing its effects to wet condition and cryogenic liquid nitrogen lubrication. The cutting parameters were cutting speed of 80 m/min, feed rate of 0.2 mm/rev and depth of cut of 0.25 mm. A TiAlN coated tungsten carbide tool was used. Gaseous nitrogen was applied at 2.5 bar in a range of temperatures between 0 ◦C and −150 ◦C. For the wet condition, a water emulsion with 5% of semi-synthetic cutting fluid was utilized. For cryogenic tests, liquid nitrogen was supplied at 15 bar and −196 ◦C.

They found that LN2 lubrication increased flank wear by 30% with respect to the wet condition, while cooled N2 always improved tool life. The best results took place at −100 ◦C, with a reduction of flank wear of 26% and 43% with respect to wet and LN2 cooling. Less difference was found with an N2 temperature between 0 ◦C and −100 ◦C, which does not justify the cost involved in reducing the temperature. Crater wear was considerably reduced by N2 at −75 ◦C, while the greatest reduction took place at −150 ◦C. The lowest tool wear was found at −100 ◦C, with a reduction of 43% compared to LN2, for which the thermal power required to achieve a low temperature is almost 10 times the thermal power required by gaseous nitrogen. However, the improvements were already evident from −50 ◦C.

Liquid nitrogen was found to ensure the best surface roughness of the machined part. However, N2 conditions at −100 ◦C and −150 ◦C were also close to this result.

Nevertheless, good results were obtained by Krishnamurthy et al. [49] in their study. They found that cryogenic treatment allowed a reduction of 300 N in cutting forces in comparison with dry machining, which corresponds to a 25% reduction. During the cryogenic treatment, chip segments became susceptible to fracture, which was detected by the segmentation of the chip. Using a Charpy V-notch impact test, they found that less energy is needed to break Ti6Al4V alloy at cryogenic conditions. Nevertheless, in these conditions, the shear strain of the machined piece was lower, which is an indicator of the high flow stress of Ti6Al4V at this temperature.

Krishnamurthy et al. [49] performed turning of Ti6Al4V alloy by cryogenic cooling and using ethanol in metal removal fluid (MRF). They performed turning tests under different conditions, such as dry machining, flooded machining using water-based an MRF mixture with vegetable oil and flooded machining using the same MRF with an addition of 10% of volume of ethanol, both with a flow rate of 8.3 mL/s. Finally, they performed dry machining after having immersed Ti6Al4V workpieces in a bath of LN2 for 20 min.

Applying ethanol, a further reduction of 65% in cutting forces was generated and the best roughness surface was obtained, thanks to a reduction in the coefficient of friction at the tool-workpiece interface.

Mia et al. [45], in other different study, evaluated surface roughness and cutting forces in the turning of Ti6Al4V under cryogenic lubrication, applied at flank and rake faces and using two cutting inserts (SNMM 120408 and SNMG 1240408).

Using an ANOVA, they found that, under cryogenic conditions, the greatest variation in cutting force came from the cutting velocity, followed by the feed rate, while the tool showed an insignificant impact. For the surface roughness, the variations came from the cutting tool, SNMM insert, associated with lower cutting forces, producing the lowest surface roughness with a reduction of 50% in comparison with the other insert.

Gupta et al. [47] studied the turning process of Ti grade 2 under MQL conditions in comparison with dry machining. By using Box-Behnkens response surface methodology and ANOVA, they obtained the most influent parameters in the machining process. The turning tests were conducted under a cutting velocity from 200 m/min to 300 m/min, feed rate from 0.1 mm/rev to 0.2 mm/rev, depth of cut of 1 mm and approach angle from 60◦ to 90◦, by using a CBN insert, coated with TiN. For MQL conditions, the flow rate was fixed at 300 mL/h, with an output pressure of 5 bar.

They verified that with the MQL lubrication condition, lower cutting temperatures were achieved due to effective lubrication, producing lower forces than for dry cutting, for which the no lubrication condition resulted in greater chip-tool contact length. Gupta et al. found that an increase of the cutting speed from 200 m/min to 250 m/min was beneficial and cutting forces, surface roughness and tool wear were reduced thanks to an initial increase in cutting temperature, which is able to soften the workpiece. However, the increment of cutting velocity from 250 m/min to 300 m/min, generated a negative effect on cutting forces, tool wear and surface roughness because MQL lubrication does not sufficient cooling capacity to evacuate heat.

Lin et al. [10] analysed the turning process of Ti6Al4V under the oils on water MQL and the MQCL approach. Oils on water (OoW) is a lubrication method in which a thin layer of oil is transported by droplets of water, providing effective lubrication and cooling effect. They tested two forms of spraying the droplets, from the interior of the cutting tool with a specific insert (IOoW) and from the outside of the cutting tool (EOoW). Based on chip morphology, cutting temperatures, forces, surface roughness and tool wear, they analysed the differences between OoW techniques and MQCL, for which MQL was mixed with cryogenic air between −16 ◦C and −26 ◦C.

Under EOoW conditions, the effects of three different spraying locations (rake face, flank face, and rake and flank faces) were studied. Under IOoW experiments, the effects of small (1.2 L/h) and large (2.4 L/h) amounts of water were studied. Two different lubricants were used, fatty alcohol and synthetic ester, with different lubricity and cooling abilities. The turning tests were conducted under a cutting speed of 70 m/min, 90 m/min and 110 m/min, feed rate of 0.25 mm/rev and depth of cut of 1 mm, using coated carbide inserts.

From these EOoW turning tests, they found the lowest forces for flank face spraying location because the chips were much shorter due to the air direction of the shingle. Nevertheless, the highest temperature was achieved for this configuration due to the lack of lubricant on the rake face. The lowest temperature was obtained from the double air direction. EOoW on the flank face provided the best surface roughness thanks to better chip breaking.

In the case of internal oil on water, it was found that the larger amount of lubrication reduced cutting forces and cutting temperatures thanks to effective cooling capacity. However, poorer surface roughness and greater tool wear was obtained in comparison with the lower quantity of lubricant. Lin et al. suggested the explanation that when a small amount of water is used, the mixture of oil and water takes the desirable configuration, oil forms a thin layer around the water droplet and provides an effective lubrication. In the case of using a large amount of oil, the configuration is inverse. Water forms a thin layer around the oil droplets, reducing the lubrication properties.

In addition, Lin et al. reported that lubricant properties had an influence on the IOoW method but not on the EOoW method.

#### *2.3. Sustainability in Drilling Processes*

A detailed review of the main sustainable techniques in milling processes was conducted. The papers selected are shown in Table 8.



This selection of turning papers provides a general overview on the state of art of each principal sustainable technique.

Perçin et al. [30] studied the effect of MQL and cryogenic lubrication on micro-drilling processes of Ti6Al4V. They compared cutting parameters, such as cutting forces, torque or tool wear, by developing micro-drilling tests under dry, wet, MQL and cryogenic conditions. The experiments were conducted under five different spindle speeds and five feed rates. The depth of the holes was established as a fixed 3 mm. They used a 700 μm diameter uncoated tungsten carbide micro-drill. For the MQL test, lubricant was supplied under low pressure, lower than 3 bar, because higher pressures broke the tool.

They found that, under cryogenic conditions, the cutting force was always greater than for dry and wet conditions, while under MQL, for the greatest feed rates, the force was lower than for the dry condition. The improvement was also seen for the lowest cutting speed in comparison with dry cutting.

The use of lubrication reduced cutting force and torque. At the beginning of the hole, torque was higher for wet and MQL lubrication because the lubricant made it more difficult for the chip to be released. However, with the cutting time, wet and MQL conditions provided better performance in the drilling process due to effective lubrication reducing the torque.

Nevertheless, the best quality for surface roughness was obtained for both wet and MQL lubrication. The results for the cryogenic conditions were worse than those for the dry condition. In addition, they found that the increment of cutting speed reduced surface micro-hardness due to thermal softening. Cryogenic lubrication conditions reduced softening, obtaining greater hardness. For this reason, cryogenic lubrication provided the greatest surface micro-hardness. It also ensured the minimum tool wear, while MQL generated less wear compared to dry cutting due to less abrasion.

Mathew et al. [32] investigated the drilling process of titanium aluminide under MQL lubrication. The experiments were conducted under a cutting speed of 72 m/min and feed rate of 0.01 mm/rev with a depth of cut of 10 mm (low aspect ratio) and 37.5 mm (high aspect ratio). A 4 mm diameter solid carbide drill (TiAlN coated) was used.

They found that in the high aspect ratio (HAR) method, in the absence of coolant for dry cutting, the temperature generated on the workpiece and the cutting tool was significantly higher than for MQL lubrication. This means that MQL provided effective lubrication to evacuate the heat.

Built-up edge was formed in both cases, but, for the MQL condition, there was found to be less BUE formation due to the reduction of chip adhesion on the tool. During LAR tests, the probability of BUE forming is reduced. Mathew et al. analysed tool wear based on tool roughness after 4 holes, comparing it to the initial roughness, reporting that MQL lubrication reduces tool wear significantly, providing as good results as wet machining. Similar results were presented by Nandgaonkar et al. [31]. They developed their experimental investigation on drilling Ti6Al4V alloy under oil on water MQL approach by using a 8 mm diameter solid carbide twist drill coated with TiAlN, finding that 43 holes of 20 mm depth could be made under dry conditions, while, under oil on water MQL conditions, 55 holes could be completed. This means that tool life was improved by 27% thanks to effective lubrication which reduces chip-tool contact length and cutting temperature. The authors reported that tool wear was reduced by 66% under oil on water MQL conditions.

In another study in the same experimental conditions [6], Mathew et al. analysed the evolution of cutting force and torque on drilling titanium aluminide at different aspect ratios. They found that for low aspect ratio (LAR), MQL lubrication provided a reduction in cutting force compared to wet lubrication and dry machining. Torque under MQL and wet conditions was always lower and more stable than under dry drilling.

However, different results were obtained for HAR drilling. In this case, the thrust force achieved under wet conditions was lower than for MQL lubrication, mainly at the end of the drilling process. The authors explained that for HAR drilling the air pressure is not sufficient to evacuate the chip, in addition to the amount of lubricant being insufficient to ensure the evacuation of heat, which causes higher tool wear.

In conclusion, these papers propose MQL as an efficient solution for drilling titanium aluminide at low aspect ratio, while it must be optimized for high aspect ratio drilling, for which the performance index is lower than for wet lubrication.

#### **3. Conclusions**

Titanium and titanium alloys are widely used in industry for aeronautical and biomedical applications, thanks to their good mechanical properties and low density. Ti6Al4V is the most important alloy, representing 50% of titanium production, although in recent years interest has been growing in titanium aluminides. Based on the literature review, the following conclusions can be drawn:


Finally, it has been found that the most widely studied criteria for the analysis of machining processes are cutting forces, surface roughness and tool wear, but important phenomena, such as friction, have not been sufficiently studied. For this reason, future research should focus its efforts on the characterisation of the friction phenomenon between the tool and the chip, which is closely related to cutting process performance.

In addition, the analysis should be extended to other materials that are gaining prominence in applications within the industrial sector, such as titanium aluminides, for which the machining processes have been the subject of limited study, especially under environmentally sustainable working conditions.

**Author Contributions:** This paper is to establish the starting point of the doctoral thesis of E.G.-M. under the supervision of V.M. All authors are active members of the research group Science and Engineering of Materials at the Castilla-La Mancha University in Spain. E.G.-M., J.C., A.M.-M. and M.C.M. have done the literature research and have selected and classified the papers according to the aims persecuted in the manuscript by a first reading of them. E.G.-M. has analysed the selected papers in detail and has written the manuscript in a continuous discussion process with the rest of the authors. V.M. has directed and planning the manuscript, and has supervised the writing of the paper and the relevance of the references considered in it. Conceptualization, V.M.; methodology, V.M., M.C.M., E.G.-M.; formal analysis, V.M., M.C.M., J.C., A.M.-M. and E.G.-M.; investigation, V.M., E.G.-M.; resources, V.M., A.M.-M., M.C.M., J.C.; data curation, E.G.-M.; writing—original draft preparation, V.M., A.M.-M., E.G.-M., J.C., M.C.M.; writing—review and editing, V.M. and E.G.-M.; visualization, E.G.-M.; supervision, V.M.; project administration, V.M.; funding acquisition V.M., A.M.-M., J.C., M.C.M.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Study of Drilling Process by Cooling Compressed Air in Reinforced Polyether-Ether-Ketone**

#### **Rosario Domingo \*, Beatriz de Agustina and Marta María Marín**

Department of Construction and Manufacturing Engineering, Universidad Nacional de Educación a Distancia (UNED), C/Juan del Rosal 12, E–28040 Madrid, Spain; bdeagustina@ind.uned.es (B.d.A.); mmarin@ind.uned.es (M.M.M.)

**\*** Correspondence: rdomingo@ind.uned.es; Tel.: +34-91-398-6455

Received: 16 December 2019; Accepted: 20 April 2020; Published: 22 April 2020

**Abstract:** This study is focused on the application of a cooling compressed air system in drilling processes; this environmentally friendly technique allows removing material at very low temperatures, approximately up to −22 ◦C in the cutting area. The main goals are to find the most improve cutting conditions with less energy consumption, for the drilling of reinforced polyether-ether-ketone with glass fiber at 30% (PEEK-GF30) with cooling compressed air by a Ranque-Hilsch vortex tube, and to find a balance between environmental conditions and adequate process performance. Drilling tests were carried out on plates of PEEK-GF30 to analyze the influence of cutting parameters and environmental temperature (–22, 0 and 22 ◦C) on variables such as thrust forces, energy and material removed rate by the use of statistical methods; analysis of variance, analysis of means, response surface, and desirability function were employed to identify the optimum region that provides the most improved values of the aforementioned variables. Drill bit diameter was also analyzed to determine the quality of drilled holes. During the drilling processes, force signals were detected by a piezoelectric dynamometer connected to multichannel amplifier and a pyrometer was used to control the temperature. The diameters of the drilled holes were measured by a coordinate measuring machine. Cooling compressed air can be considered an adequate technique to improve the results from an environmental and efficient perspective; in particular, the maximum desirability function was found at a spindle speed of 7000 rpm, a feedrate of 1 mm/rev and a temperature close to −22 ◦C.

**Keywords:** drilling; cooling compressed air; thrust force; energy; material removed rate; PEEK-GF30; multi-response optimization; sustainable manufacturing

#### **1. Introduction**

Over the past decade, within manufacturing industries there has been an increasing interest in researching techniques that allow the performance of processes with high efficiency under sustainable environments. In fact, different systems have been applied in manufacturing facilities to identify the most improved parameters to combine both objectives; in some cases, the effects on the organization of production system by the implementation of lean techniques [1] and in others, the effects on the manufacturing process where the improved selection of parameters can reduce CO2 emissions [2]. The machining is the most used manufacturing process in the industry due to its versatility [3], so studies in this field can be relevant. In this context, several methods have been applied to achieve both objectives, efficiency and sustainability, to the most possible extent. One of them is dry machining or machining without the use of any cutting fluid. Krolczyk et al. [4] have carried out an extensive literature review on four ecological methods in machining of difficult to-cut metals, in particular dry machining, minimum quantity lubrication (MQL)/minimum quantity cooling lubrication (MQCL), cryogenic cooling, high-pressure cooling and biodegradable oils; they found that dry machining is the most sustainable procedure with respect to others, despite the high temperature generated during the

machining operation, in particular in operations such as drilling; besides, these authors pointed out the non-use of cryogenic cooling in industry despite the benefits of this procedure; this is the combination of high productivity as well as low cost and energy. Sen et al. [5], in their literature review focussed on metal cutting, collected results from life cycle assessment (LCA) models for different cooling techniques, showing that MQL and dry machining have the minimum negative effects respect to techniques such as flood lubrication, MQL, cryogenic cooling with CO2, cryogenic cooling with LN2 and cryogenic cooling with MQL. However, machining under such dry conditions causes excessive temperature rise at the interface between workpiece and tool and, in general, an increase in strains. These undesirable effects have been determined by the analysis of forces and strains [6] or considering the influence of tribology [7]; in both cases finite element models (FEMs) were used in the orthogonal cut of titanium alloys, so the application of new cooling techniques can be explored. Cooling technologies [8], such as cryogenic machining with liquid nitrogen and the application of cooling compressed air to the cutting zone during the machining process by means of a Ranque-Hilsch vortex tube have been identified as possible environmentally friendly procedures. These cooling systems are particularly interesting under dry cutting conditions; in this way, cutting fluids are avoided. Goindi and Sarkar [9] identified cooling compressed air obtained by vortex tube separation as a method to analyze into sustainable machining, as a variant of dry machining and also as an element to consider within the MQL. Moreover, the use of cooling compressed air systems in machining processes requires less investment than cryogenic machining and they are easier to employ on an industrial scale. In addition, although the MQL procedure is convenient from the point of view of sustainability, the use of lubricants could not be suitable in polymeric materials due to its possible absorption by the composite.

In fact, the application of cooling compressed air during the machining process is being studied and some researches can show its suitability as environmentally friendly procedure. Jozi´c et al. [10] developed an experimental study, in which several machining aspects were analyzed in the milling process of a certain steel with this procedure at −34 ◦C under dry cutting conditions and also with the use of cutting fluid; it was found the most improved solution (less surface roughness, cutting forces, flank wear and more volume of removed material) in case of the application of cooling air during the process. In addition, regarding the milling processes, Perri et al. [11] analyzed the effect of cooling air on the milling tool, using simulation models by FEM and experimental contrast and verifying important difference of temperatures reached respect to the procedure carried out without cooling air. On the other hand, Nor Khairusshima et al. [12] studied the quality and the tool wear in the milling of the carbon fiber reinforced plastic with cooling air at −10 ◦C; they pointed out that the tool wear and delamination factor were improved at high cutting speeds. Domingo et al. [13] analyzed the effects of cooling air on reinforced and unreinforced polyamides during the tapping, finding that an adequate procedure at −18 ◦C provided improved results with respect to the values of forces, torques and power reached during the process.

As shown in the aforementioned literature, the cooling compressed air has been also applied in the machining of composites with polymeric matrix and different reinforcements as carbon fiber [12] or glass fiber [13], and an improved performance has been reported. Nevertheless, studies towards the drilling of reinforced polyether-ether-ketone with 30% glass fiber (PEEK-GF30) have not been found in the scientific literature. The novelty in this study is the analysis of its behavior during drilling processes, the most important process in the assembly as the previous operation to riveting and tapping. This material can withstand very low temperatures maintaining a stable behavior [14], therefore the machining of this material with cooling compressed air can be adequate as its mechanical and chemical properties are not expected to modify significantly during the process. Due to its extended use in the industry, its machinability has been studied, finding that with an adequate selection of cutting parameters and tools, it is a material that can be widely used in industry [15].

This environmentally friendly procedure can be contrasted evaluating the main variables of drilling under specific cutting conditions. Variables such as thrust forces should be low to improve process stability, at least in drilling operations; the literature shows that a homogeneous variation of thrust forces does not exist with respect to cutting conditions, at least in different types of composites such as glass fiber reinforced PEEK [16], fibre metal laminates [17], magnesium matrix based silicon carbide and graphene nanoplatelets (Mg/SiC/GNPs), hybrid magnesium matrix composite [18] or wood-based composite of medium density fiberboard [19]. Nevertheless, the energy consumed during the process could be a key factor to stablish the most sustainable procedure, and also the material removed rate (MRR) in order to evaluate the process efficiency. In the study carried out by Davim and Reis [20], the precision of the hole was considered as a variable to measure surface quality, so the diameter of the hole is another variable to study. Although the surface roughness is another variable that determines the hole quality, its influence in inner holes depends on the hole functionality; in this study, the evaluation of roughness surface it is irrelevant as an internal threading operation is expected to approach; its evaluation is important in other further operations such as riveting, however this characteristic can be corrected in other finishing operations, even with the same tool. Thus, the main variable to be accepted as an adequate dimensional accuracy of the holes is their dimension, and moreover it can be influenced by the chip evacuation and thermal expansion of the matrix occurring during the cutting process due to the lack of lubricant, as can be seen in materials based on multilayer metallic and/or composite stacks such as comprising titanium stacks, carbon fibre reinforced plastics (CFRPs) and aluminium [21], functionally graded composite, carbon/epoxy and glass/epoxy composite [22] or carbon fiber-reinforced plastic composite [23]. The reduction of the temperature in the cutting area could avoid this effect.

Taking into account the above-mentioned points, the objectives of this paper are the following: (i) to find the cutting conditions more environmentally friendly, with less energy consumption, in the drilling of reinforced PEEK with glass fiber at 30% with cooling compressed air by vortex tube, and (ii) to establish a balance between environmental conditions and adequate process performance.

#### **2. Materials and Methods**

The methodology used in this work combines experimental and statistical procedures.

#### *2.1. Experimental Procedure*

Drilling tests were carried out in a CNC vertical machining center, Manga Tongtai TMV-510 (Kaohsiung Hsien, Taiwan). The drilling operations were performed on plates of PEEK-GF30 (Gapi, Bergamo, Italy), with thickness of 6.5 mm, and a coefficient of linear thermal expansion of <sup>30</sup> <sup>×</sup> <sup>10</sup>−<sup>6</sup> <sup>m</sup>/m·K).

The drill bits employed were provided by FMT Tooling Systems Company (Trofa, Portugal); they had a diameter of 6 mm and are made of solid carbide with coating of zirconium oxide. Their material and geometrical characteristics are described in Table 1. Note that, as the point angle of drill was 140◦, the corresponding cutting length was 7.592 mm. The selection of cutting tool was made based on the results of the analysis of variables, such as, for example, the energy or CO2 emissions [24] and preliminary tests. In drilling processes, the use of coating of zirconium oxide provides a good behavior of tool respect to the tool wear, the cutting forces and the chip flow [25]. A design of experiments (DOE) was used to optimize the resources. In particular, a three-level factorial design with three replicates was employed to analyze the influence of the three factors: feedrate, spindle speed and environmental temperature, on the response variables: thrust force, energy required for the drilling operation, material removed rate and drill bit diameter. This was, in total, 81 drilling tests. Later on, in the Section 2.2 statistical procedure, a more detailed description of the statistical analysis method is included.


Values of spindle speed (*N*), from 5000 to 7000 rpm and values of feedrate (*f*) from 0.5 to 1 mm/rev were applied. The drilling tests were carried out at temperatures (*T*) from −22 to 22 ◦C. The speed and the feed were chosen taking into account previous tests, information exchanged with the tool supplier, characteristics of the drill bits and high performance. Besides, similar ranges can be found in the literature [15]. Note that the feedrate was very high and this fact allowed quickly increasing the MRR. Firstly, drilling tests were performed at the room temperature that was 22 ◦C and secondly, drilling tests were carried out with the cooling compressed air system at 0 and −22 ◦C; thus, three specimens were used, one for each temperature. For the measurement of the temperature, an infrared pyrometer, Optris, was used; in this way, the temperature was monitored along the drilling process. The cooling was achieved by means of a Ranque–Hilsch vortex tube with two outlets of cold fluid, Dual Nozzle CAG Vortec model. This cooling technology allows obtaining the most improved outcomes when the nozzle number is 2, due to a more homogeneous distribution of cooling [26]. In other machining processes, with regard to this technology it is underlined that cutting forces and the power required in the titanium turning could be reduced with the application of vortex tube [27]. Similar findings were obtained by other studies with respect to flank wear and surface roughness towards the machining of aluminum alloys [28].

For the application of the cooling compressed air, it was necessary that the tube received the compressed air with a pressure of 0.8 MPa. To achieve temperatures of 0 and −22 ◦C, the compressed air had to flow for 8 and 10 min respectively, along vortex tube, before impinging on the drill bit; this step was very important to avoid the formation of ice on the drill bit during the drilling process. The nozzles of vortex tube were positioned at a distance of 30 mm, approximately, to achieve these temperatures.

The thrust force and the torque on the drilling direction were calculated directly during the monitoring of the drilling process. A piezoelectric dynamometer type Kistler 9257B connected to multichannel amplifier type Kistler 5070A was used (see Figure 1). The data were processed by DasyLab software (version 9.0, Measurement Computer, Norton, USA) [29]. The torques were taken to calculate the energy according to Equation (1), derivated of the expression used by Li et al. in the drilling of titanium alloys [30].

$$E = \int\_{l} Ft \times dl + \int\_{l} \left(\frac{2 \times \pi \times To}{f}\right)\_{z} \times dl \,, \tag{1}$$

where *E* is the energy required to drill a hole in J, *Ft* is the thrust force in N, *l* is the drilling length in m, *To* is the torque respect Z axis in N·m, and *f* is the feedrate respect Z axis, in m/rev.

**Figure 1.** Machining process and data collection: (**a**) Scheme of assembly to drilling process and data capture; (**b**) Outlet of cold fluid after 8 minutes, with ice in the exterior.

In the drilling process, the material removed rate, in mm3/s, is identified according to Equation (2) [31],

$$MRR = \left(\pi \times D i^2 \times F\right) / \left(4 \times 60\right) \tag{2}$$

where *Di* is the drill bit diameter in mm, *F* is the feedrate in mm/min.

To control the diameters quality, measurements were taken on 24 points along the circumference of the hole (as considered a suitable number [32]) by a coordinate measuring machine, Mitutoyo BX 303. The least square circle method was used to calculate the diameter measurements. Each measurement was repeated three times. The procedure of the measurement method was developed according to ISO 4291 standard [33]. The surface points were measured using a ball point stylus with a diameter of 1.6 mm. The data obtained were fitted by a Gaussian filter with 50% cut off. The outcomes were treated by Geopak-Win MCC software. Figure 2a shows a scheme of the location of center and circle of measured point after applying least square circle technique. Figure 2b shows the coordinate measurement machine.

**Figure 2.** Measurement process of diameters: (**a**) Scheme of determination of center and circle of least square; (**b**) Coordinate measuring machine, during the measurement process of diameters.

#### *2.2. Statistical Procedure*

The statistical procedure focused on the application of response surface methodology when several responses should be simultaneously optimized, through a multi-objective method, in particular the desirability function, which allows finding a common objective for variables with different sub-objectives.

The experimental data collected were statistically analyzed, in particular, by the Response Surface with three-level factorial design, 3 × 3 in this case, and using Statgraphics software [34]. The Analysis of Variance (ANOVA) was carried out for the variables thrust forces, energy and MRR, meanwhile the diameter was used to evaluate the quality of the holes. With ANOVA analysis, the significant factors can be identified in a particular confidence interval [33]. The ANOVA of each variable is shown by tables, where the sum of squares or variance of the observations, the degrees of freedom (Df), the mean square, the F-ratio obtained from Fisher–Snedecor distribution and its probability associated (P-value) are represented; values of means squares are calculated dividing the sum of squares by its associated degrees of freedom. In this case, the confidence interval considered was 90%, adequate in manufacturing environments; therefore, a P-value less 0.1 denoted that the factor was significant using an ANOVA analysis. Although a 95% confidence interval is usually common in ANOVA studies, the election of 90% allows increasing the range of significance in the energy variable, which is dependent of other two variables, as it can be observed in Equation (1). Moreover, from a statistical perspective, this interval is adequate [35]. The percentage of contribution of the significant factors to variability was determined by dividing the sum of squares for the factor by sum of squares total. An analysis of means (ANOM) was also carried out to determine if there are differences between the means of each variable for the values considered; this analysis was performed through Tukey-honestly significant difference (HSD) test, which allows identifying what means are different at a confidence interval of 95% [36].

*Materials* **2020**, *13*, 1965

Once an ANOVA study and an ANOM study were developed, regression models were defined based on the surface response [36], and finally, the desirability function was defined. The desirability function, *di*, was defined by Derringer and Suich [37], with different expressions according to the objective of the study variable. In this paper, there are, a variable to maximize, the MRR, and two variables to minimize, the thrust force and the energy. When the variable must be maximized, Equation (3) represents this option [37]:

$$d\_i(\mathfrak{g}\_i(\mathbf{x})) = \begin{cases} 0 & \text{if } \mathfrak{g}\_i(\mathbf{x}) < L\_i \\ (\mathfrak{g}\_i - L\_i / \mathcal{U}\_i - L\_i)^s & \text{if } L\_i \le \mathfrak{g}\_i(\mathbf{x}) \le \mathcal{U}\_i \\ 1 & \text{if } \mathfrak{g}\_i(\mathbf{x}) > \mathcal{U}\_i \end{cases},\tag{3}$$

where *yˆi* is *i*th estimated response of the variable (thrust force, energy, MRR), *Li* is lower acceptable value, *Ui* is the upper acceptable value and *s* is the weight. Equation (4) represents the desirability function when the response must be minimized, being *t* the weight [37].

$$\mathrm{Ad}\_{\mathrm{i}}(\mathbb{\hat{y}\_{i}(\mathbf{x})}) = \begin{cases} 1 & \text{if } \mathbb{\hat{y}\_{i}(\mathbf{x})} < \mathrm{L}\_{\mathrm{i}} \\ \left(\mathbb{U}\_{\mathrm{i}} - \mathbb{\hat{y}\_{\mathrm{i}}}/\mathrm{U}\_{\mathrm{i}} - \mathrm{L}\_{\mathrm{i}}\right)^{\mathrm{t}} \text{if } \mathrm{L}\_{\mathrm{i}} \le \mathbb{\hat{y}\_{i}(\mathbf{x})} \le \mathrm{U}\_{\mathrm{i}} \\ 0 & \text{if } \quad \mathfrak{Y}\_{i}(\mathbf{x}) > \mathrm{U}\_{\mathrm{i}} \end{cases},\tag{4}$$

Therefore *d (yˆi*(x)) takes, uniquely, a value between 0 and 1 when it transforms a response into free-scale. The global desirability (*D*) can be determined through a weighted geometric mean by Equation (5) [37]:

$$\mathbf{D} = \left(\prod\_{\mathbf{i}=1}^{n} \mathbf{d}\_{\mathbf{i}}^{\mathbf{i}}\right)^{1/\sum \mathbf{r}}\,,\tag{5}$$

where *n* is the number of variables and *ri* is the impact value. Thus, *D* takes also takes values between 0 and 1, and its optimum consists to maximize it. The values of the weights and impact can be fixed by the authors. In case of using Statgraphics software, the values can vary between 1 and 5. This range allows to establish the usual values in manufacturing environments, in particular in machining processes, e.g., in tapping operations [13] or in turning operations [38].

#### **3. Results and Discussion**

In Table 2, the experimental results obtained are detailed: thrust forces, energy, MRR and input diameter of holes. The values are the result of calculating the means of three values because the reproducibility of the measurements was very high. Outcomes from forces or energy show the influence of the feedrate, in a way that at higher feedrates, a reduction of the values of these variables were obtained, compared to other experimental data [16], which could be explained by the fact that the flute helix angle of tools improves chip evacuation. A low variability can be observed in data from input diameters; in all tests except in test 21, diameters slightly lower than the nominal drill bit (6 mm) were obtained. The reason could be the improved chip evacuation and the lack of thermal expansion during the drilling tests.


**Table 2.** Experimental results.

#### *3.1. Analysis of Experimental Results*

This subsection is developed to explore the experimental results; the influence of the considered cutting factors (feedrate, spindle speed and environmental temperature) are analyzed.

#### 3.1.1. Thrust Forces

The ANOVA analysis for thrust forces (see Table 3) shows that the significant main factors were environmental temperature (T), spindle speed (N) and feedrate (f), and the significant interactions are T2, T <sup>×</sup> N, N2 and T<sup>2</sup> <sup>×</sup> N in a confidence level of 90%. The most influential main factor was the spindle speed, with a contribution of 24.82%, as was expected due to the influence of the incidence of the drill bit on the plate. As can be seen, the contribution of temperature as main factor (first or second order) was 25.31% and taking into account its interactions with spindle speed, T <sup>×</sup> N and T2 <sup>×</sup> N, their contribution reached 49.12%. Thus, the temperature influence was clear; in fact, in Table 2 it can be observed, as less desirable results were obtained at temperatures close to 0 ◦C; similar results were approached in the tapping of PA66-GF30 as shown in the study carried out by Domingo et al. [13].


**Table 3.** Analysis of Variance for *Ft*.

On the other hand, the ANOM study and the Tukey-HSD test reveal that the pairwise means were significantly different from each other at 95% confidence level, with respect to feedrate (statistically significant differences between pairwise means, 0.5–0.75: −7.67, 0.5–1: −19.19 and 0.5–1: −11.53), spindle speed (statistically significant differences between pairwise means, 5000−6000: 2.71, 5000−7000: 27.53, and 6000−7000: 24.82), and enviromental temperature (statistically significant differences between pairwise means: −22–0: –47.28, −22–22: −33.65, and 0–22: 14.23). This denotes that the thrust force showed different behavior in each of the selected cutting speeds, feedrates and temperatures.

According to the above, thrust forces obtained in drilling processes depended strongly on the enviromental temperature.

#### 3.1.2. Energy

From the ANOVA analysis for energy (Table 4), the factors N2, f2, T <sup>×</sup> N <sup>×</sup> f, T <sup>×</sup> f <sup>2</sup> and N <sup>×</sup> f 2 (not significant) were eliminated in order to achieve a more improved adjustment between the factors. Feedrate and T<sup>2</sup> erre the significant factors at 90% confidence level, with an influence of 9.19% and 6.78%, respectively. Although the effect of these factors was lower than those considered for thrust force, the environmental temperature and cutting conditions reappeared.


**Table 4.** Analysis of variance for *E*.

In this case, from Tukey-HSD test it was obtained that the pairwise means were significantly different from each others at 95% confidence level, with respect to feedrate (statistically significant differences between pairwise means, 0.5–0.75: 4.10, 0.5–1: 6.98 and 0.5–1: 2.86), spindle speed (statistically significant differences between pairwise means, 5000–6000: −0.56, 5000–7000: 0.08, and 6000-7000: 0.67), and temperature (statistically significant differences between pairwise means: −22–0: −3.88, −22–22: −2.97, and 0–22: 0.91). This implies that the energy showed a different behaviour in each of the selected spindle speeds, feedrates and temperatures.

#### 3.1.3. Material Removed Rate

The ANOVA analysis for MRR can be seen in Table 5. Although the data from different temperatures were included in the study, the significant factors were N, f, N <sup>×</sup> f and f2, being the most influential factor on the feedrate (f) with a percentage of 44.9%, as was expected despite f2 having a very low influence. Note that the feedrate considered was very high.


**Table 5.** Analysis of variance for material removed rate (*MRR)*.

From Tukey-HSD test it was obtained that the pairwise means were significantly different from each other at 95% confidence level, with respect to feedrate (statistically significant differences between pairwise means, 0.5–0.75: −706.85, 0.5–1: −1431.71 and 0.5–1: −706.86) and spindle speed (statistically significant differences between pairwise means, 5000–6000: −353.43, 5000–7000: −706.86, and 6000−7000: −353.43). Obviously, there were no statistically significant differences between pairwise means with respect to temperature because the MRR was independent of it, with differences as: −22–0: 0, −22–22: 0, and 0–22: 0. This implies that the energy showed a different behaviour in each of the selected spindle speeds and feedrates.

#### 3.1.4. Input Diameter

As has been mentioned before, the measurements of input diameters were carried out as quality control evaluation; if measurements were correct the experimental tests could be considered valid. Otherwise, the tests could not be taken out and the experimental data would have to be discarded. In Table 2 can be seen that the greater difference between diameters was 0.093 mm (0.013 mm at 22 ◦C, 0.070 mm at 0 ◦C and 0.054 mm at −22 ◦C), being adequate cooling temperaturesfor different spindle speeds at feedrates of 0.5 mm/rev (Figure 3a), 0.75 mm/rev (Figure 3b) and 1 mm/rev (Figure 3c). In these figures, larger undersized holes could be observed at 0 ◦C with spindle speed of 5000 and

6000 rpm. However, the difference between them was very low, and it did not seem relevant. Perri et al. [11] found that the effect of the cooling system and the flow air on the displacement of the tool centre point was less intense; this can explain the data obtained in the diameters, and the undersized holes obtained.

**Figure 3.** Input diameter executed at different temperatures and spindle speed, and with: (**a**) feedrate of 0.5 mm/rev; (**b**) feedrate of 0.75 mm/rev; (**c**) feedrate of 1 mm/rev.

#### *3.2. Response Surface*

Taking into account, uniquely, the significant main factors and interactions, and with a coefficient of determination, R2, superior to 70% in all of the cases, the regression equations, Equations (6)–(8), are the following:

$$\begin{aligned} \text{Ft} &= 83.727 - 5.662 \times T - 0.0415 \times N + 223.276 \times f - 0.455 \times T^2 + \\ &+ 0.00193 \times T \times N + 0.0000059 \times N^2 + 0.0 \times T^2 \times N,\end{aligned} \tag{6}$$

$$E = -129.475 + 195.038 \times f - 0.01164 \times T^2 \,\text{.}\tag{7}$$

$$MRR = 0.05 - 0.14 \times f + 0.47124 \times N \times f + 0.08 \times f^2,\tag{8}$$

The effect of the temperature on thrust forces can be seen in Figure 4. The influence of spindle speed and the feedrate was lower at room temperature (Figure 4a) than at 0 ◦C (Figure 4b). Finally, it is remarkable that, at −22 ◦C (Figure 4c), the thrust forces were lower and less dependent of cutting conditions and the lowest values were reached. Figure 4 plots the values of Equation (6), also, the evolution of Ft and its trend, with continuous values. The evolution of forces at 0 ◦C deserves special mention, that showed the force decreased with the increase of spindle speed; the friction seemed to increase at this temperature, though this effect was diminished at higher spindle speeds; in this case, the feedrate more convenient is the lowest value. However, at −22 ◦C, the friction phenomenon did not seem to be affected, which could be explained because of greater influence of the air pressure at low temperature. Note that, as mentioned in Section 1, PEEK-GF30 maintains a stable behavior at

low temperatures [14], which is very important because it prevents the modification of the material properties at these temperatures.

**Figure 4.** Estimation of response surface for thrust forces: (**a**) Temperature of 22 ◦C; (**b**) Temperature of 0 ◦C; (**c**) Temperature of −22 ◦C.

The estimation of response surface for energy can be seen in Figure 5. At 22 ◦C (Figure 5a) and 0 ◦C (Figure 5b) the distribution of the energy obtained was more uniform. At −22 ◦C (Figure 5c), the highest spindle speeds provided a similar behavior to that in thrust forces at this temperature, despite the greater influence of the torque in the calculation of the energy consumed during drilling [16]. The explanation can be the same as found for the forces, given the dependence of the cutting conditions along the cutting length during drilling (see Equation (1)). To clarify this point, in Figure 5 the values of Equation (7), the evolution of energy and its trend are plotted. Dry machining (at 22 ◦C) is an option with more energy consumption, maybe because there is a heating in the cutting area.

**Figure 5.** Estimation of response surface for energy: (**a**) Temperature of 22 ◦C; (**b**) Temperature of 0 ◦C; (**c**) Temperature of −22 ◦C.

Figure 6 shows the estimation of response surface for MRR, with the clear and known influence of cutting conditions. This representation of Equation (8) allows the representation of the evolution of MRR, its increase with the cutting conditions, as expected, and its independence of temperature, as can be observed in Equations (2) and (8). In Figure 6, from the last equation MRR values are plotted respect to cutting conditions, spindle speed and feedrate.

**Figure 6.** Estimation of response surface for MRR, at 22 ◦C, 0 ◦C and −22 ◦C.

#### *3.3. Multiple Response Surface Optimization*

The multiple response surface optimization is represented in Table 6; in this Table, the goal of each variable and the parameters values are shown, lower values of variables (*Li*), upper values of variables (*Ui*), weights considered in the desirability function (*s* and *t*), and impact. The values taken for weights and impact avoid that some variables influenced more than others in the results, and besides they are used to optimize the manufacturing processes, in particular in machining operations [13,38]. As can be seen in Table 6, the weights and the impact were the same for each variable, 1 for weights and 3 for impact; in this manner, all variables were compensated. The lower and upper values of each variable were taken from Table 2; the Ft values corresponded to −22 ◦C, 7000 rpm and 0.75 mm/rev (lowest values) and 0 ◦C, 5000 rpm and 1 mm/rev (uppest value); the Energy values corresponded to −22 ◦C, 6000 rpm and 1 mm/rev (lowest value) and 22 ◦C, 7000 rpm and 0.5 mm/rev (uppest value); and MRR corresponded to any temperature, and 5000 rpm and 0.5 mm/rev (lowest value) and 7000 rpm and 1 mm/rev (uppest value). This indicates the importance of the application of a multi-objective method to seek a common objective when the goals are different and when the minimum and maximum values of each variable correspond to different cutting conditions.


**Table 6.** Values considered in the desirability function.

Figure 7 shows the values of desirability function for thrust force, energy, MRR and Ft-E-MRR; while that the maximum value (1) was achieved for thrust forces and MRR, the minimum value was for the energy, that obviously it was noted in the final value for the combined variable Ft-E-MRR. While with Ft and MRR, the maximum desirability was achieved, the energy showed a value of 0.88. This value can be considered normal due to the dependence of different variables on its calculation (see Equation (1)).

**Figure 7.** Desirability function.

The optimization allowed finding the maximum desirability at −22 ◦C, feed of 1 mm/rev and 7000 rpm; under these conditions, the optimum values were 22.93 N for thrust force, 4.2 J for energy and 3298 mm3/s for MRR. Figure 8 shows the contours of estimated response surface. From this graphic it can be observed that the area of major desirability was located at a temperature of −22 ◦C and at high spindle speed, considering the thrust force, the energy and the MRR. Note that the optimal parameters allowed machining at high cutting conditions at −22 ◦C. This possibility can increase the potential use of machining at low temperatures by avoiding thermal expansion of the matrix of composite materials. This balance combines objectives of sustainability and efficiency.

**Figure 8.** Contours of estimated response surface for f = 1 mm/rev.

#### **4. Conclusions**

The drilling of reinforced PEEK plates with drills of coating zirconium oxide and diameter of 6 mm was analyzed at different cutting conditions and environmental temperatures, employing cooling compressed air by a Ranque–Hilsch vortex tube. Experimental data from spindle speed, feedrate, temperature and input diameter of holes were measured, and a statistical study was developed.

The Response Surface methodology with three-level factorial design was applied to optimize, simultaneously, several responses through the desirability function, in order to find the significant factors (spindle, speed, feed rate and temperature) and their interactions on the variables such as thrust force, energy and MRR, the relationships between factors and variables, and the value of desirability function. Thus, the response surface methodology should be optimized.

From the results of this experimental and statistical study, conclusions can be summarized as:


Future researches can address lower temperatures, avoiding temperatures close to 0 ◦C which do not improve the performance with respect to dry conditions at room temperature. Moreover, in future developments, the application of cryogenic drilling on this material can be considered due to the improved outcomes obtained at low temperatures. In addition, the tool wear is proposed to be a factor to analyze.

**Author Contributions:** Conceptualization, R.D.; methodology, R.D.; validation, B.d.A.; formal analysis, R.D., B.d.A. and M.M.M.; investigation, R.D., B.d.A. and M.M.M.; resources, R.D. and M.M.M.; writing—original draft preparation, R.D. and B.d.A.; writing—review and editing, R.D., B.d.A. and M.M.M.; funding acquisition, R.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Spanish Ministry of Science, Innovation and Universities, DPI2014-58007-R and RTI2018-102215-B-I00.

**Acknowledgments:** The authors also thank the College of Industrial Engineers of UNED for supporting through 2019-ICF08 and 2019-ICF03 projects.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Reusing Discarded Ballast Waste in Ecological Cements**

#### **Santiago Yagüe García \* and Cristina González Gaya**

ETS Ingenieros Industriales, Universidad Nacional de Educación a Distancia (UNED), C/Juan del Rosal, 12, 28040 Madrid, Spain; cggaya@ind.uned.es

**\*** Correspondence: syague14@alumno.uned.es

Received: 13 October 2019; Accepted: 21 November 2019; Published: 25 November 2019

**Abstract:** Numerous waste streams can be employed in different cement production processes, and the inclusion of pozzolans will, moreover, permit the manufacture of concrete with improved hydraulic properties. Pozzolanic materials can be added to Ordinary Portland Cement (OPC) in the range of 10%–20% by mass of cement. One such example is the phyllosilicate kaolinite (K), and its calcined derivative metakaolin (MK), incorporated in international cement manufacturing standards, due to its high reactivity and utility as a pozzolan. In the present paper, discarded ballast classed as Construction and Demolition Waste (C&DW) is reused as a pozzolanic material. Various techniques are used to characterize its chemical, mineralogical, and morphological properties, alongside its mechanical properties, such as compressive and flexural strength. Discarded ballast in substitution of cement at levels of 10% and 20% produced type II or IV pozzolanic cements that yielded satisfactory test results.

**Keywords:** cements; ballast waste; cornubianite; mechanical properties; Spain

#### **1. Introduction**

The cement industry is one of the main contributors of greenhouse gas emissions, such as CO, CO2, and NO. In this sector, new strategies are now prioritized in many countries that are trying to surpass the European Union (EU) targets set for 2020, which aspire to 20% lower emission levels than in 1990. Natural additions to cement have been used for the improvement of its properties since the days of antiquity. One of the best known is the addition of pozzalonic material consisting of a type of volcanic pumice, purportedly found around the town of Pozzuoli (Italy) that causes certain hydraulic characteristics in cement, hence the name of pozzalonic cements [1].

A pozzolan material is understood to be a material that generally consists of silica and alumina that, in itself, has no cementitious value when mixed with water. However, when finely ground and in the presence of water, it reacts with portlandite (Ca(OH)2), originating from the hydration of silicates present in the clinker, generating compounds with cementing properties [2]. The normalized artificial additions that present pozzolanic behavior are industrial by-products, such as silica fume and fly ash—the natural pozzolans—and calcined bauxite. The non-normalized materials include paper sludge, calcined sugar cane bagasse ash, and rice husk ash, among others.

Various natural materials that might function as pozzolans have been tested in cement research, and various artificial additions have more recently been undergoing trials, especially industrial waste materials including slags, silica fume, fly ash, etc., which are usually dumped in large volumes in landfill sites [3]. The utility of their addition to cement is two-fold: the elimination of waste and the enhancement of the cement. These additions are understood as technological solutions, designed to improve both the performance of high-strength cements and cement behavior against aggressive environmental agents [4]. Hence, the incorporation of these additions to cement can be seen as an

attempt to reduce manufacturing costs, while simultaneously searching for materials that are more respectful towards the environment.

In the current panorama of waste recycling, construction and demolition waste (C&DW) assumes fundamental importance, because it constitutes one of the main waste streams within the EU. The recycling/reutilization rates of C&DW in the EU vary significantly between countries, fluctuating between 5% in Portugal, and 90% in the Netherlands and Estonia. On average, the recycling rate in the 27 countries of the EC is 55% [5].

According to data from the Plan Nacional de Residuos de Construcción y Demolición 2008–2015 (National C&DW Plan in Spain), 40 million tons of C&DW are produced annually, which is equivalent to over 2 kg per person, per day—a higher rate than domestic rubbish. From data provided by the Environment Ministry, 2.5 million tons were recycled in 2018. It implies a C&DW recycling rate of 5.1%—a figure well below the European average—implying that the main means of disposing of these wastes continues to be in regulated or unregulated landfill sites in Spain, as opposed to their recycling or reuse.

The publication, in February 2008, of Royal Decree (RD) 105/2008, in regulation of the production and management of C&DW (BOE of 13 February), implies growing environmental concerns and interest in this matter, that national government and the autonomous regions have been expressing for numerous years. Prior to this point, there was the 2002–2006 Plan Nacional de Residuos Urbanos (PNRU) [National Urban Waste Plan] and, more specific to the type of wastes that are of interest here, the Plan Nacional de Residuos de Construcción y Demolición 2001–2006 (PNRCD) [National C&DW Plan] (BOE 12 July 2001), which approached the treatment, recovery, and recycling of these wastes. These national plans were substituted by the Plan Nacional Integrado de Residuos (PNIR) [National Integrated Waste Plan] for the period 2008–2015 (BOE 26 February 2009), within which, under Point nº 12, it covered the second National C&DW Plan. Among the plans' objectives was the controlled collection and proper management of 95% of C&DW by 2011, the reduction or reuse of 15% of C&DW by 2011, the recycling of 40% of C&DW by 2011, and the exploitation of 70% of all waste construction material packaging, as from 2010.

Among other aspects, the RD 105/2008 introduced the obligation of including a C&DW management study in a building or construction work design project, which had to contain, as a minimum, an estimation of the amount of waste that could be produced, as well as measures for risk prevention, management process, and the exploitation of those waste streams.

Added to the current situation is an increasing demand for fines linked to the environmental restrictions on the exploitation of new quarries that have led to proposals for the reuse of certain waste streams as alternative raw materials. Different studies on the viability of valorizing granite waste and other ornamental rocks have been consulted. In this case, it is the waste from hornfels considered [6–9].

The renewal of a train track is carried out when the ballast does not meet the precise specifications related to wear and alteration of the rocks, which depends on their nature and is controlled by convoys that partially remove and replace the ballast. This control is performed when needed, according to the auscultator train.

In the present work, the addition of a construction waste stream from used ballast is studied for its use as a pozzolan. The need to activate this residue has been assessed using the pozzolanicity test, and the mechanical properties of the mixtures have been obtained by the partial replacement in the Ordinary Portland Cement (OPC) of the waste considered.

#### **2. Materials**

The aim was to obtain C&DW from discarded ballast that had been replaced by new ballast. To do so, the process of ballast wear was simulated using an accelerated method, which consisted of wearing down new ballast in a ball mill and collecting the fines. Aggregate materials were employed in this work, with coarse granulometry (gravel) from CANTERAS and CONSTRUCCIONES S.A. (CYCASA), at Aldeavieja, Ávila (Spain). The ballast was taken from the same batches supplied by

the quarry for the renewal of the stretch of rail from PB Río Duero - Est. Valladolid C.G., on the Madrid—Segovia—Valladolid High Speed Rail Line.

The sample under study (C) is a hornfels (metamorphic rock) with granoblastic texture, although in many examples the regional schistosity can also be seen. Hornfels is dark rock with a matt shine and opaque colors.

An Ordinary Portland Cement (OPC) type CEM I 42.5 R cement was used by Italcementi group.

#### **3. Methods**

The Los Angeles abrasion test [10] was used to monitor wear, in accordance with the conditions specified in annex C of standard UNE ES 13450 [11].

The ballast fragments were worn down in a steel ball mill to obtain the waste product, that was then transported to a laboratory for oven drying for 24 h at 110 ◦C, to a constant weight (through the removal of humidity).

In accordance with the consulted literature [12,13], the calcination process was followed once the waste had been dried in an electric oven at 600 ◦C for 2 h (sample CC), with a view to establish the pozzolanic potential of this addition and to improve the activation potential of the waste.

To determine the pozzolanic activity, an accelerated method was used that consisted of placing 1 g of the sample solution in contact with 75 mL of a saturated solution of calcium hydroxide (17.68 mM/L) to 40 ◦C over 1, 7, 14, 28, and 90 days. The solution underwent vacuum filtration in a Buchner funnel at each age under study. The concentration of calcium ions expressed as calcium oxide or fixed lime was determined in the filtrate by the assay detailed in standard UNE—EN 196-5 [14,15].

The specific surface of the waste was studied using the BET method, through isothermal absorption of nitrogen, and the distribution of particle size (Sympatec Helos 12LA laser diffraction spectrometer, Sympatec, Clausthal-Zellerfeld, Germany).

The solid waste was analyzed through X-ray Fluorescence Spectroscopy (XRF) (to analyze chemical composition, using a Bruker S8 Tiger XDR spectrometer (Bruker, Fremon, CA, USA) with Spectra Plus Quant Express software v1.0.0.13), X-ray Diffraction (XRD) (to analyze mineralogical composition with a SIEMENS D-5000 diffractometer (Anton Parra, Madrid, Spain), working between 3 and 60 degrees with a sweep velocity of 2 degrees per minute), and scanning electron microscopy/energy dispersive X-ray spectroscopy (SEM/EDX) (providing surface data at a microscopic level and surface analysis, using a PHILIPS XL 30 flexible scanning electron microscope (Philips, Leuven Belgié) with a wolfram filament, a BIO-RAD SC 502 disk type sputter target, and an EDAX Energy Dispersive X-ray spectrometer with a DX4i silica/lithium detector and analyzer for chemical analysis, Philips, Leuven, Belgie).

Tests have been carried out on the samples for their resistance to compression and bending according to the UNE EN 196-1 standard [16] (compressive and flexural strength).

#### **4. Results and Discussion**

The particle size distribution of the discarded ballast waste was studied, showing a bimodal distribution with two maximum particle sizes from 6 to 15 μm.

BET surface determination provided a value of 1.32 m2/g, similar to the value obtained for fly ash of 1.40 m2/g, lower than the value for ceramic tiling of 3.00 m2/g [17], very much lower than fly ash at 20 m2/g [18], and higher than the values close to 0.98 m2/g for ladle furnace slag [18].

The result of the chemical analysis by XRF of the original sample (C) of OPC cement is provided in Table 1.


**Table 1.** Chemical analysis by X-ray Fluorescence Spectroscopy (XRF) from C sample and Ordinary Portland Cement (OPC) cement.

In addition, concentrations of zirconium, copper, chrome, cobalt, nickel, strontium, vanadium, zinc, and lead were detected in quantities that were not in excess of 50 ppm. The high content of both silica and aluminum points to good pozzolanic activity.

In turn, the mineralogical composition of the discarded ballast waste, obtained by XRD, indicated the presence of quartz, potassium feldspar, plagioclase (soda-lime feldspar), biotite, and clay-type minerals, such as kaolinite and chlorite with scarce little muscovite, as well as small quantities of hematite, as shown in Figure 1. The components of the cement were tricalcium aluminate, belite, alite, calcite, and ferrite phases.

**Figure 1.** X-ray diffraction by the ballast waste (B = biotite; K = kaolinite; Cl = chlorite; Q = quartz; F = K feldspar; P = Ca, Na feldspar; H = hematite).

#### *4.1. Calcination of Ballast Waste*

The waste material was calcined at a temperature of 600 ◦C for 2 h, in stove at a constant heat to increase its pozzolanic activity. The increase of this activity was due to the loss of structural water in the clay (kaolinite) minerals and in the phyllosilicates (biotite, chlorite).

The result of the mineralogical analysis of the calcined sample (CC) obviously affected the dehydroxylation of the kaolinite, which changed into amorphous metakaolinite, and had an incipient effect on the phyllosilicate structure, which started to lose hydroxyl groups. All other mineral components remained unchanged in the diffractogram.

#### *4.2. Sample Pozzolanicity*

The pozzolanicity test on the initial sample and the sample that had been thermally activated at 600 ◦C/2 h is shown in Figure 2. The improved behavior of sample C and the initial waste is shown, in all cases, except for at 28 days. Thermal activation of the waste was therefore not considered necessary and, henceforth, the initial sample was the only material used in the tests. The differences are so small that thermal activation is not required.

**Figure 2.** Measures of pozzolanicity of the initial discarded ballast waste (C) and the thermally calcined waste (CC).

#### *4.3. Preparation of Mortars with Discarded Ballast Waste*

Standardized CEN sand was used with a granulometry of between 1 and 0.08 mm, which met the requirements specified in standard UNE EN 196-1 [16]. The cement was a CEM 1 42.5 R-type OPC, the composition of which is shown in Table 1 [19].

A mixture of the above components, ballast waste, and Portland cement was used to prepare a mixture that guaranteed the homogeneity of the corresponding mixtures. The cements were differentiated by the substitution of either 10% or 20% by weight of OPC for discarded ballast waste, in an attempt to design type II/A (6/20%) and IV/A (11–35%) cements, in accordance with standard UNE EN 197-1 [14].

#### *4.4. Mechanical Behavior*

Compressive strength and flexural tests were performed in accordance with standard UNE EN 196-1 [16]. Prismatic mortar specimens were prepared, measuring 4 cm × 4 cm × 16 cm, with a sand/cement ratio and water/cement ratio of 3/1 and 3/2, respectively. At 24 h after their manufacture, the specimens were demolded and cured, at a temperature of 20 ± 1 ◦C and a relative humidity of 100%, up until failure.

#### 4.4.1. Mechanical Strength under Compression

All the cements under study presented compressive strengths of over 10 MPa after two days of curing, and greater than or equal to 42.5 MPa over the following 28 days, thereby complying with the mechanical specifications contained in standard UNE EN 197-1 [14] for cements classed as high-strength (42.5 MPa).

From the graph shown in Figure 3, it may be seen that the incorporation of discarded ballast waste in the mortars in no way modified the existing logarithmic tendency (although the trend seems linear, the adaptation to a logarithmic equation meets R2 better) between compressive strength and curing time, regardless of the substitution level (10% or 20%). Correlation coefficients higher than 0.90 were obtained for R2.

**Figure 3.** Variation of the compressive strength of the different mortars.

The following expression was used to arrive that figure:

$$\text{y}\_{\text{OPC}} = 6.18 \,\ln(\text{x})\_\prime + 41.64 \,\text{, con } \text{R}^2 = 0.945^\prime$$

while the two levels of substitution were calculated as follows:

y10%<sup>+</sup>90%OPC = 6.39 ln(x) + 34.88, with R<sup>2</sup> = 0.984 y20%<sup>+</sup>80%OPC = 6.24 ln(x) + 28.23, with R<sup>2</sup> = 0.992

It was also noted that the incorporation of the discarded ballast waste implied a significant improvement in performance as the percentage substitution level increased. The weakened performance was close to 11% at 10% C + 90% OPC and around 23% at 20% C + 80% OPC, with regard to the standard OPC specimen, considering a time of 28 days of curing. This tendency to lose strength coincides with the trend noted in conventional mortars by Ramos et al. [20].

In addition, this behavior coincides with the behavior described by Frías et al. [21] and Vardhan et al. [22] in their studies on slate and marble quarry sludge, respectively, in the manufacture of new cements. An addition of 20% seems to improve the conditions of compressive strength compared to one of 10%.

The incorporation of different proportions of SiO2 and Al2O3 improved the structure of the pores, since the ballast waste affects the hydration of the cement and is responsible for the mechanical properties [23]. Fine pores are generated when the Si/Al ratio is high, which increases contact points and resistance. This mechanism is similar to secondary hydration and requires long ages to make it happen.

#### 4.4.2. Flexotraction Strength

Regarding the flexotraction strength of the mortars containing substitutions of ballast waste, a similar tendency was detected to the one observed for compressive strength, with a loss of strength of 10% and 17%, after 28 days of curing in the mixtures with substitution levels of 10% and 20%, respectively, as seen in Figure 4. In this case, the addition of 10% was more favorable in terms of flexotraction than the addition of 20% over time. It can be attributed to the delayed onset of the pozzolanic reaction. It is known that when the pozzolanic reaction begins, the amount of Ca (OH)2 decreases and the microstructure improves, with densification exceeding longer ages [24,25].

**Figure 4.** Variation of flexotraction strength of the different mortars.

#### *4.5. Total Porosity and Pore-Size Distribution*

Table 2 shows the values of total porosity and the average pore size in the mortars with substitutions of ballast waste at 2 and at 90 days of curing. It is understood that the incorporation of ballast waste generates a slight increase in the total porosity of the mortars with substitutions of 10 and 20%, with a slightly higher increase at substitution levels of 10%, but with values very close to 9% with respect to OPC.


**Table 2.** Values of the total porosity and the pore diameter average at 2 and 90 days of curing for the mortars studied.

Table 2, likewise, shows the evolution of the average pore size, with a refinement of the system of pores as the cement hydration process progresses, showing a smaller average size with the age of curing at the higher substitution level (OPC, 10% and 20% of substitution), with respect to the mortars at 2 days.

The above is clear from the SEM images of the mortars under study. Accordingly, SEM-EDAX observations of the grain edges of the mortar specimens (4 cm × 4 cm × 16 cm) with substitution levels of ballast waste at 0%, 10% and 20%, cured over 90 days, have shown that when the cement has no substitution—as shown in Figure 5A—the inter-grain contact is formed by calcium silicate hydrate (CSH) gels and ettringite fibers that give it the slightly porous appearance that can been seen in Figure 5B. In turn, as demonstrated in Figure 5C,D, the inter-grain contact was not observed to be so well defined in the cement with no ballast waste. Nevertheless, as shown in Figure 5E–H, the tendency at levels of substitution of 10% were similar at 20%, although less acute, thus the inter-grain contacts were better defined and substitution levels at 20% were therefore considered unnecessary. The consideration of the results of the previous tests means that the addition is directed towards 10%, which would lead it to be considered a cement type II/A (6/20%) instead of the budget type IV/A (11–35%) in accordance with standard UNE EN 197-1 [14].

**Figure 5.** OPC Mortar specimen with no ballast waste: (**A**) magnified image of inter-grain contact; (**B**) grain edge; (**C**) magnified image of inter-grain contact of mortar specimen 20% + 80% OPC; (**D**) grain edge of mortar specimens 10% + 90% OPC; (**E**,**G**) magnified image of inter-grain contact; (**F**,**H**) grain edge.

#### **5. Conclusions**

Discarded ballast waste has been used as a pozzolanic addition in cement, to reuse the C&DW instead of sending it to landfill. This type of waste participates in the Circular Economy due to its high

amount of silicates, and, when using C&DW, it will be considered as a secondary raw material, as the displacement to waste plants or landfills is unnecessary when taking advantage of its reuse through its management in situ, with consequent economic benefits.

The use of this waste in the same type of facilities eliminates the participation of the necessary track material in its renewal.

The waste does not need any type of activation, neither thermal nor chemical, and can be directly used in mortar mixtures.

The importance of quantifying the addition to the OPC has been investigated at 10% and 20% for ballast waste. Both additions have given good results, especially 10%, for design type II/A (6/20%) cement. In the case of opting for the 20% addition, a type IV/A (11–35%) cement would be obtained, giving good results over time.

The variety of rocks used as ballast warrants further study, looking toward the consideration of materials other than hornfels.

**Author Contributions:** S.Y.G. contributed materials and performed the experiments and wrote the paper. C.G.G. designed the experiments and analyzed the data.

**Funding:** This research received no external funding.

**Acknowledgments:** This paper is based on the ongoing activities that form part of the lead author's Ph.D. thesis, under preparation at the International Doctorate School of the National University of Distance Education (Spain); the authors therefore wish to express their gratitude for the support from that institution. They are also grateful for the assistance kindly provided by Hormigones y Morteros del Río (Aldeavieja, Spain), the Instituto Eduardo Torroja de Madrid, and UAM.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
