**Sustainable High-Speed Finishing Turning of Haynes 282 Using Carbide Tools in Dry Conditions**

#### **Antonio Díaz-Álvarez \*, José Díaz-Álvarez, José Luis Cantero and Henar Miguélez**

Department of Mechanical Engineering, University Carlos III of Madrid, 30 Avida. de la Universidad, 28911 Leganés, Madrid, Spain

**\*** Correspondence: andiaza@ing.uc3m.es

Received: 30 July 2019; Accepted: 3 September 2019; Published: 6 September 2019

**Abstract:** Nickel-based superalloys exhibit an exceptional combination of corrosion resistance, enhanced mechanical properties at high temperatures, and thermal stability. The mechanical behavior of nickel-based superalloys depends on the grain size and the precipitation state after aging. Haynes 282 was developed in order to improve the creep behavior, formability, and strain-age cracking of the other commonly used nickel-based superalloys. Nevertheless, taking into account the interest of the industry in the machinability of Haynes 282 because of its great mechanical properties, which is not found in other superalloys like Inconel 718 or Waspaloy, more research on this alloy is necessary. Cutting tools suffer extreme thermomechanical loading because of the high pressure and temperature localized in the cutting zone. The consequence is material adhesion during machining and strong abrasion due to the hard carbides included in the material. The main recommendations for finishing turning in Haynes 282 include the use of carbide tools, low cutting speeds, low depth of pass, and the use of cutting fluids. However, because of the growing interest in sustainable processes and cost reduction, dry machining is considered to be one of the best techniques for material removal. During the machining of Haynes 282, at both the finishing and roughing turning, cemented carbide inserts are most commonly used and are recommended all over the industry. This paper deals with the machining of Haynes 282 by means of coated carbide tools cutting fluids (dry condition). Different cutting speeds and feeds were tested to quantify the cutting forces, quality of surface, wear progression, and end of tool life. Tool life values similar to those obtained with a lubricant under similar conditions in other studies have been obtained for the most favorable conditions in dry environments.

**Keywords:** dry; carbide tool; Haynes 282; finishing turning

#### **1. Introduction**

Turbine components suffer the extreme conditions of thermomechanical loading during their service life. Significant tensile stresses in rotative elements induce fatigue phenomena [1]. The development of new advanced materials and the continuous improvement of the processing routes are required in order to improve the performance of the turbine components [2]. Nickel-based superalloys are widely used in turbine elements because of their excellent mechanical properties at high temperatures and their resistance to corrosion [3]. Being about 50 wt. % of the materials used in these applications [4], Ni alloys are also used in other applications such as pressure vessels, marine equipment, different elements of aircraft engines, and petrochemical plants [5,6]. The excellent mechanical properties of this family of superalloys also include low formability, with different problems during component processing that could affect its service life. New-generation alloys are developed in order to solve these problems. For example, Haynes 282 focuses on the improvement of the weldability and fabricability with a similar creep strength. These combinations of properties are of great interest

for critical steam applications. This superalloy has already been adopted for hot section parts in gas turbines for aircraft and power generation, and it can be a baseline for the further improvement of superalloys [7].

Haynes 282 is highlighted by a high percentage of molybdenum (>6 wt. %), which develops carbide particles at temperatures ranging from 815 to 870 ◦C in a complex cubic structure, with it being more stable at high temperatures [8]. Haynes 282 was developed at the beginning of the 21st century. It is strengthened by the precipitation of the γ phase, which is the L12 ordered structure Ni3 (Al,Ti), and has a coherent relationship with the γ matrix [9]. These γ precipitates, characterized by their size, distribution, morphology, and composition, influence the mechanical properties of the alloy [10].

Despite the renewed interest in Haynes 282, there is a lack of information concerning its machinability. The machining of nickel-based superalloys presents great challenges, mainly because of the high work hardening tendency, structure stability, low thermal conductivity, adhesion of materials in the tool, and carbide particles in its structure [11]. All of the above features result in hard loads and temperatures (up to 1200 ◦C [12]) at the chip–tool interface, resulting in the rapid wear of the tool [13,14], which influences the surface integrity of the piece, generating residual stresses and increments in roughness [15,16]. Elevated temperatures combined with a high chemical affinity between the workpiece and the materials used for the cutting tools promote oxidation and diffusion wear, as well as the adhesion of the work material at the cutting tool area (mostly related to the damage on the tool–rake face) [3]. Moreover, the adhesion and abrasion on the clearance surface normally induce flank wear, chipping, and catastrophic failures [17].

Thus, the selection of the tool is critical during the machining of nickel-based superalloys, requiring elevated wear resistance and hardness, high strength, and chemical stability at elevated temperatures [18]. The industry recommendation for the turning of nickel-based alloys involves the use of ceramic and carbide tools, the latter being used in finishing the turning [4].

Concerning the tool coating physical vapor deposition (PVD), TiAlN, ALTiN, or AlCrTiN are widely used for carbide tools in the turning of nickel-based alloys for improving the competitiveness of carbides as opposed to ceramic tools, because of their lower cost [19]. The TiAlN coating in comparison to the TiN coating decreases the machining forces, whereas it improves tool resistance to flank wear because of its chemical inertness, adhesion resistance, high hardness at elevated temperatures (up to 1000 ◦C), and high oxidation resistance [20,21]. Coated carbide tools are recommended for a medium cutting speed, ranging between 30–70 m/min [22] for the turning of nickel-based alloys, because of its thermomechanical instability [14].

Traditionally, cutting fluids have been used to lubricate (helping to reduce the friction in the area of contact between the chip and the tool), eliminate the chip from the cutting area, and, above all, to eliminate the heat produced during the process of machining, cooling the tool and the workpiece at the interface. Thus, the use of cutting fluids during the machining process (generally between 10–100 L per minute [23]) has a huge impact on the temperature of the tool, its wear evolution, and life, as well as on the surface finishing of the workpiece (roughness of the surface, generation of residual stresses, etc.). However, the impact of cutting fluids on the environment is significant. Therefore, industrial activities are encouraging manufacturers to implement new green techniques, replacing the use of traditional cutting fluids. Moreover, the use of cutting fluids is both harmful to the environment and very expensive, not only for its acquisition, but also for the costs associated with its recovery and disposal management [24].

Sustainability requirements are leading to the use of new vegetable-based cutting fluids that are sustainable, environmentally friendly, biodegradable, and less toxic, and they are becoming a real alternative to petrol-based cutting fluids [25]. Moreover, cutting fluids are normally applied with flood coolant systems (FC), systems that can account for up to 17% of the total production costs [26] and that sometimes do not reach the area of machining because of the obstruction of the chips. Alternatives for the application of cutting fluids have been developed, such as near-dry-machining (NDM) systems, also known as minimum quantity lubrication (MQL) [27] or minimum quantity

cutting fluids (MQCF) [28]. However, dry machining that avoids the use of cutting fluid would be the best technique, if possible. Cantero et al. [3,29] analyzed the performance of the carbide and PCBN tools in the dry finishing turning of Inconel 718, obtaining a tool life of 29 min and 2 min, respectively, for competitive cutting conditions, confirming the industrial viability of the carbide inserts but not of the PCBNs in the dry finishing of the Inconel 718.

Few papers are available on the topic of Haynes 282 machinability. Suarez et al. [30] carried out an experimental investigation focusing on the effect of lubricant pressure and material heat treatment on the turning of this alloy. A negligible effect for the high-pressure cooling was observed, while the solution annealing large grain solution (LGS) state presented enhanced machinability when compared to the precipitation hardened large grain aged (LGA) state in terms of force levels and tool wear. Díaz-Álvarez et al. [11] studied the performance of a coated carbide tool during the finishing turning of Haynes 282 with a cutting fluid at the conventional pressure, observing that, for all of the cutting conditions, the tool broke because of the fragile fracture of the cutting edge.

There is a lack of research focusing on the machining of Haynes 282. Moreover, the tool wear analysis of the carbide inserts when machining Haynes 282 in a dry environment has not been studied. The present work deals with the finishing turning of the Haynes 282 alloy in dry conditions. Dry machining tests using coated carbide tools were performed under different cutting conditions in order to evaluate the viability of the cutting fluid removal in finishing turning of Haynes 282 with carbide tools. Roughness, cutting forces, and tool wear were quantified in each test. Although the industrial dry machining of Haynes 282 has not yet been applied, in this study, tool life values similar to those obtained with lubricants under similar conditions in other studies have been obtained for the most favorable conditions in dry conditions.

#### **2. Experimental Setup**

#### *2.1. Material Properties and Cutting Tools.*

A Haynes 282 alloy was tested in a round bar with a 90 mm diameter shape, which was manufactured following the AMS5951 specification. The Haynes 282 workpiece was annealed at 1135 ◦C (in the typical range 1121–1149 ◦C) and age hardened according to the following stages: It was heated up to 1283 ◦C, maintained at this temperature for 2 h, and then cooled in air. Afterwards, it was heated up to 1061 ◦C, maintained at this temperature for 8 h, and then cooled in air. The hardness of each specimen tested was quantified at different points, obtaining values that varied between 42.2 and 43.5 HRc. Each element percentage of the Haynes 282 that was tested in the present paper is summarized in Table 1.


**Table 1.** Haynes 282 chemical composition [11].

A carbide tool (CW, TS200 grade) with a multilayer coating of TiAl/TiAlN, provided by SECO (SECO tools, Fagersta, Sweden), were used for turning tests. These coated carbide tools are especially recommended for finishing turning of Nickel superalloy. Insert presents a tip and honing radius of 0.4 mm and 25 μm respectively, tip angle equal to 80◦, rake angle of 16◦ and a relief angle of 7◦. The cutting tool with the code CCMT 09T304F1 was fixed in a tool holder type SCLCR 2525M09JET provided by SECO.

#### *2.2. Experimental Setup and Instrumentation*

Haynes 282 turning tests were carried out in a lathe Pinacho Smart turn 6/165 (Pinacho, Castejón del Puente, Spain) equipped with a Kistler 9257B dynamometer (Kistler, Winterthur, Switzerland) for the cutting force measurement (Figure 1).

During the development of the turning tests and at the end of each pass, a rounded surface remained because of the effect of the tool tip radius. Therefore, because of the consequent increase of material needing to be removed in that zone in the next pass, which did not allow for a continuous cut, a sudden increment of undeformed chip cross-section was caused [29]. The finishing operation was characterized by small cutting depths, so this increase in material as a result of the tool tip radius at the end of the pass led to a significant increase in the cutting forces, hence influencing the tool wear. To avoid this phenomenon, a second tool was attached in the tool holder in the lathe (see Figure 1) in order to remove this zone once the cutting force had been stabilized and measured using the tested tool.

**Figure 1.** Instrumentation and setup.

The tool wear level was periodically evaluated during the turning tests for each cutting condition, tested by means of obtaining images from a stereo microscope Optika SZR (Optika, Ponteranica BG, Italy). Also, a scanning electron microscopy (SEM) Philips XL-30 (Philips, Eindhoven, Netherlands) with an EDSDX4i system was used to analyze the wear evolution. At the same time, the surface finish of the workpiece was evaluated by means of the surface roughness through a Mitutoyo model SJ-201 (Mitutoyo, Kawasaki, Japan) rugosimeter, obtaining the mean of nine measurements as the representative roughness value.

All of the cutting tests in this study were carried out without any type of coolant by analyzing the finishing turning of Haynes 282 under dry conditions.

As knowledge of the machining of the Haynes 282 alloy at an industrial level is poor, tool manufacturers do not include the relevant information for the selection of the cutting parameters for its process. Nevertheless, in the bibliography, there are general recommendations establishing the ranges for the cutting speed (30–35 m/min), feed rate (0.1–0.18 mm/rev), and depth (1 mm) [31]. Moreover, Díaz-Álvarez et al. [11] investigated the machining of Haynes 282 with carbide tools under a conventional pressure coolant using cutting speeds between 50–90 m/min, feeds between 0.1–0.15 mm/rev, and a depth of pass of 0.25 mm, obtaining a maximum tool life of 33 min. Thus, the cutting parameters selected for the present study are summarized in Table 2.


**Table 2.** Cutting parameters for the turning tests.

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

#### *3.1. Cutting Forces Analysis*

The evolution of the cutting forces—cutting force (Fc), feed force (Ff), and back force (Fp)—were recorded for each preformed test using a frequency of acquisition of 100 Hz. To guarantee the repeatability of the results, each test was performed twice, obtaining deviations lower than 5% with respect to the mean value. Thus, the average values have been used for the subsequent analyses. For the sake of simplicity, in the following analysis, the specific force components (kc, kf, and kp) have been defined as the each of the cutting forces over the undeformed cross section of the chip. In the subsequent points, the results of each component are compared with the observed tool wear damage (notch, chipping, flank, and built up edge).

#### 3.1.1. Fresh Tools Results for the Specific Cutting Force

In Figure 2, the obtained results for each component plus the resultant specific cutting force (*kr*) quantified at the first stages of each of the tests through fresh tools are represented. For the series of cutting parameters that were studied, the results of the specific cutting force (*kc*) ranged from 3580 N/mm<sup>2</sup> (case: *Vc* = 90 m/min and feed = 0.15 mm/rev) to 4200 N/mm2 (case: *Vc* = 50 m/min and feed = 0.1 mm/rev). The values of the resultant cutting forces that take into account all of the cutting forces components range from 4330 N/mm<sup>2</sup> (case: *Vc* = 90 m/min and feed = 0.15 mm/rev) to 5700 N/mm2 (case: *Vc* = 90 m/min and feed = 0.1 mm/rev).

Cutting Speed vs. Specific Cutting Forces

• For the lowest feed (0.1 mm/rev) used, the specific cutting force (*kc*) was not significantly affected by the cutting speed for the studied range. However, the rest of the components increased by up to 26% for the specific feed force (*kf*), and up to 100% for the specific back force (*kp*) when the cutting speed was increased from 50 m/min to 90 m/min. This behavior was not observed for the feed equal to 0.15 mm/rev, whereas the cutting speed was increased from 50 m/min to 90 m/min, the values of the specific cutting forces were decreased by up to 57%, 70%, and 30% for the specific cutting force (*kc*), the specific feed force (*kf*), and the specific back force (*kp*), respectively. By increasing the cutting speed, the temperature of the material to be cut rose, so that it softened, thus requiring lower cutting forces. At the same time, increasing the cutting speed also increased the strain rate by increasing the resistance of the material to be cut. For a feed of 0.1 mm/rev, it was observed that, because of the higher proportion of chip sections with high levels of deformation, when increasing the cutting speed, the specific cutting forces increased because of the strain hardening effect; however, for the feed value of 0.15 mm/rev, the softening effect of the material, because of the increment of the cutting speed, was the predominant effect.

Feed vs. Specific Cutting Forces

• Regarding the specific cutting force component induced by the feed, reductions of up to 13% for the specific cutting force (*kc*), up to 39% for the specific feed force (*kf*), and up to 38% for the specific back force (*kp*) were recorded with increments on the feed from 0.1 mm/rev to 0.15 mm/rev. The specific cutting force component induced by the feed was as expected. For the lowest feed, as the proportion of material subjected to large deformation (along the cutting edge) was higher, the specific cutting force results were also higher; this tendency can also be verified through the resultant specific cutting force (*kr*).

#### 3.1.2. Specific Cutting Forces Evolution During Haynes 282 Turning

The specific force progression and the resultant specific force for the different components with the cutting time are represented in Figure 3. For all of the cutting conditions, all of the specific cutting force components increased with the time of use of the tool. However, *kc* presented an increasing linear trend for the tool life, while the growth of the *kf*- and *kp*-specific forces showed other trends in all of the cases being highlighted in two regions, as follows: the first one with a linear growth and the second one with a more pronounced increment.

Cutting Speed vs. Specific Cutting Forces


regions, with a drastic increase in the second one, whereas for the *kf* component, this increase was not so evident. Therefore, all of the components of the cutting speed exhibited a slight increase during the first region, with the flank wear progression moderated by means of a light chipping. However, the components of force *kf* and *kp* suffered suspected growth during the second region because of a great deterioration of the cutting edge through the notch and more intense chipping in this final stage.

• Cutting speed of 90 m/min: As in the previous cases, there were two clearly differentiated regions of specific forces of growth, the main difference between them being the use time of the tool, in which the trend change appeared much smaller than for the lower speeds. These two growth zones were also related by a moderate growth of flank wear together with chipping, until the chipping was dominant, progressing in quick increment of the specific force components *kf* and *kp*.

Feed vs. Specific Cutting Forces

• For all of the conditions analyzed, a remarkable influence of the feed on the evolution of the components of the specific cutting force were not found during the turning tests.

Near the end of the tool life, values up to 10 times of those obtained with a fresh tool were obtained for the specific back force, whereas values up to 7 and 2.5 times were obtained for the feed and specific cutting forces respectively, when compared with the ones obtained for the fresh tool. Therefore, especially for lower cutting speeds, the evolution of the specific back force could be a suitable indicator of tool wear progression. The value of the resultant specific cutting force included in Figure 3 can be used as a more stable variable to evaluate the wear state of the tool.

**Figure 3.** *Cont*.

**Figure 3.** Specific cutting force evolutions with the times for the different cutting conditions tested. (**a**) *Vc* = 50 m/min and f = 0.1 mm/rev; (**b**) *Vc* = 50 m/min and f = 0.15 mm/rev; (**c**) *Vc* = 70 m/min and f = 0.1 mm/rev; (**d**) *Vc* = 70 m/min and f = 0.15 mm/rev; (**e**) *Vc* = 90 m/min and f = 0.1 mm/rev; (**f**) *Vc* = 90 m/min and f = 0.15 mm/rev.

#### *3.2. Analysis of Wear and Tool Life*

During each experiment, the test tool wear progression was checked, with the main wear modes identified being the notch, chipping, flank, and built-up edge (BUE). In order to quantify the wear for all of the cutting conditions analyzed, the tool wear was periodically studied within each test. Tools reached the end of tool life by means of the breakage of the cutting edge, or through the end of tool life criterion, established by means of a notch or flank wear larger than 0.4 mm; however, only one cutting condition reached a value of flank wear close to 0.4 mm. The value of 0.4 mm for the notch or flank wear was established by attending to the behavior of the tools, and for values of flank wear close to 0.4 mm, high increments of the cutting forces and a rapid growth of the chipping wear leading to the catastrophic failure of the cutting edge were observed (see Figure 3, *Vc* = 50 m/min and a feed of 0.1 mm/rev to check the rapid increments of the cutting forces when the value of the flank wear reached values close to 0.4 mm).

Although both BUE and the adhesion of the material were observed for all of the cutting conditions analyzed, as can be seen in Figure 4, they have not supposed a significant influence on the tool life.

Chipping, together with notch wear, were predominant along the entire cutting edge of the tools at the beginning of the performed tests. The wear progression was similar, regardless of the cutting conditions. Chipping became larger, being filled by material (BUE). Furthermore, the area of the notch that grew throughout the tests was clearly differentiated, and, at the same time, the flank wear progressed. It was found that for higher cutting speeds, the chipping progressed more rapidly, exposing more of the flank surface (causing the flank wear to grow much faster than with lower cutting speeds, where chipping was not so aggressive).

The progression of chipping wear is enhanced with the increment of the cutting speed, and, to a lesser extent, by increasing the feed, causing a reduction in the tool life and leading to the final catastrophic breakage of the cutting edge for all of the cases analyzed. Increasing the feed results in obtaining higher forces and a more unstable cut because of the increase of material that is to be removed in each pass, thus, favoring the appearance of fragile breaks in the cutting edge. Increasing the cutting speed is related to an increase in the temperature at the cutting area [12]. This increase in temperature favors the adhesion of materials in the tool (and the consequent chipping) through a reduction of the tool material strength.

In Table 3, both the tool life values by means of cutting time, the machined surface per time (S*mach.t*) and the machined surface per cutting edge (S*edge*), quantified through Equations (1) and (2), respectively, have been summarized for all of the cases analyzed [11].

$$\mathbf{S}\_{\text{mach.t}} = V\_{\text{c}} \cdot \mathbf{f} \cdot \mathbf{1}000 / 60 \tag{1}$$

$$\mathbf{S}\_{\text{adj};\varepsilon} = \mathbf{S}\_{\text{match};t} \cdot \mathbf{T} \cdot \Theta \,\tag{2}$$

where *Smach.t* is the machined surface per unit time (mm2/s), *Sedge* is the machined surface per edge (mm2), *Vc* is the cutting speed (m/min), f is the feed (mm/rev), and T is the tool life (min).

**Figure 4.** *Cont*.

**Figure 4.** Scanning electron microscopy (SEM) images at the end of the tool life for the different conditions tested. *Vc* = 50 m/min and *f* = 0.1 mm/rev: (**a**) relief and (**b**) rake surface view. *Vc* = 50 m/min and *f* = 0.15 mm/rev: (**c**) relief and (**d**) rake surface view. *Vc* = 90 m/min and *f* = 0.1 mm/rev: (**e**) relief and (**f**) rake surface view. *Vc* = 90 m/min *f* = 0.15 mm/rev: (**g**) relief and (**h**) rake surface view. BUE—built up edge.


**Table 3.** Tool life, machined surface per unit time (mm2/s), and machined surface per edge (mm2) for the different cutting conditions tested.

For the lowest cutting speed (50 m/min), values of 30.1 and 21 min of tool life for 0.1 mm/rev and 0.15 mm/rev feeds, respectively, were obtained when reaching the point of the highest level of flank extension (almost 0.4 mm), which is near to the end of tool life criterion that has been established (Figure 4a–d). The end of tool life by means of cutting-edge breakage was reached because of predominant chipping.

Wear due to chipping appears during tests at cutting speeds of 70 m/min, compared with the cutting speed of 50 m/min, reaching the end of tool life through a breakage of the cutting edge. Thus, tool life values of 4.5 and 2.1 min for feeds of 0.1 and 0.15 mm/rev, respectively, were obtained during turning tests at 70 m/min.

In the turning tests at 90 m/min cutting speed, wear due to chipping appeared at the first stages, with a fast progression up to cutting edge breakage at the end of tool life (Figure 4e–h). As expected, the lowest tool life values were reported at 90 m/min, obtaining 2.3 and 1.9 min for feeds of 0.15 and 0.1 mm/rev, respectively.

As shown in Table 3, tool life values of 1.9 (*V*<sup>c</sup> = 90 m/min) up to 30.1 min (*V*<sup>c</sup> = 50 m/min) were obtained during the dry turning tests on Haynes 282. As mentioned above, increasing the cutting speed increased the adhesion of the material in the tool and therefore the chipping wear. Thus, the best results in terms of tool life time were those obtained for both low cutting speeds and feeds. The tool life obtained for the cutting speed of 50 m/min and 0.1 mm/rev feed was very close to those obtained by the authors in analogous tests, where a conventional pressure coolant was used [11]. However, a great influence of the cutting speed in the tool life has been found, decreasing its life up to 85% when the cutting speed increases from 50 to 70 for a feed of 0.1 mm/rev, and up to 90% for a feed of 0.15 mm/rev.

The machined surface per cutting edge (known as an indicator for tool industrial performance) at 50 m/min cutting speed was similar for both feeds (0.1 and 0.15 mm/rev), whereas, because of the short tool life derived from increasing the cutting speed from 70 m/min to 90 m/min, there was no significant variation in the machined surface per cutting edge. It is necessary to highlight the important result obtained in terms of the mechanized surface and tool life for the less aggressive tool parameters (50 m/min and 0.1 mm/rev) in dry conditions, these being very similar to those obtained at the conventional coolant pressure [11]. This result makes the use of this type of tool suitable for the finishing the machining of Haynes 282 under dry conditions, which, until today, was done with cutting fluid.

#### *3.3. Analysis of Surface Quality*

The surface roughness progression was evaluated at different stages during the development of the tests. The surface quality was measured three times at three different zones over the machined surfaces in terms of the average roughness (*Ra*). Thus, the maximum value of these measured values for each stage were taken as the value of the roughness for each condition tested (Figure 5).

**Figure 5.** Roughness evolution at the machined surface for all of the cutting conditions tested: (**a**) *V*c = 50 m/min, (**b**) *V*c = 70 m/min, (**c**) *V*c = 90 m/min.

During the first stage of tests, with fresh tools and no significant wear, the values of roughness within the range of 0.7 and 2.5 μm were obtained. It should be noted that the *Ra* values were reduced with the wear of the tool for a cutting speed of 50 m/min, which is related to the type of wear found, with the flank for this cutting speed evolving progressively, reaching values close to 0.4 mm, causing an artificial increase in the tip radius, resulting in lower values of *Ra*. However, for higher cutting speeds, the chipping was dominant, as the beginning caused the original honing of the cutting edge, which was not so defined. The authors obtained similar results during the finishing turning of Inconel 718 [32].

The best roughness values were those obtained for the 50 m/min cutting speed and 0.1 mm/rev of the feed. This phenomenon is related to the lower chipping obtained at the beginning for the lowest cutting speed, because a more linear trend was observed in the roughness progression.

On the contrary, for cutting speeds of 70 and 90 m/min, where chipping wear affects the tool more severely from the first moments of the test, it has not been possible to establish a clear trend in the roughness progression.

The feed shows a clear influence on the roughness values obtained, with it being generally greater for higher feeds, regardless of the cutting speed. This result agrees with that which is theoretically expected from the application of Equation (3) [33], namely,

$$R\_a = 0.0321 \text{\textdegree } / r\_c \tag{3}$$

where *f* is the feed (mm/tooth)*,* and *rc* is the tool nose radius (mm).

#### **4. Conclusions**

This work dealt with the sustainable finishing turning of Haynes 282 by means of coated carbide tools without cutting fluids (dry condition). Different cutting speeds and feeds for Ni-based alloys were tested in order to quantify the cutting forces, the quality of surface, the wear progression, and the end of tool life. The main contributions of the analysis are summarized below.


**Author Contributions:** Conceptualization, J.D.-Á. and J.L.C.; data curation, J.D.-Á. and A.D.-Á.; formal analysis, J.D.-Á., A.D.-Á., and J.L.C.; funding acquisition, H.M. and J.L.C.; investigation, J.D.-Á. and A.D.-Á.; project

administration, H.M. and J.L.C.; resources, J.L.C.; supervision, H.M. and J.L.C.; validation, J.D.-Á., A.D.-Á., H.M., and J.L.C.; visualization, A.D.-Á.; writing (original draft), J.D.-Á. and A.D.-Á.; writing (review and editing), A.D.-Á., J.D.-Á., and H.M.

**Funding:** This research was funded by the Ministry of Economy, Industry, and Competitiveness, and the FEDER program, grant number DPI2017-89197-C2-1-R.

**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* **On the Surface Quality of CFRTP**/**Steel Hybrid Structures Machined by AWJM**

**Fermin Bañon 1,\*, Bartolome Simonet 2, Alejandro Sambruno 1, Moises Batista <sup>1</sup> and Jorge Salguero <sup>1</sup>**


Received: 23 June 2020; Accepted: 18 July 2020; Published: 21 July 2020

**Abstract:** The joining of dissimilar materials in a hybrid structure is a line of research of great interest at present. Nevertheless, the machining of materials with different machinability requires specific processes capable of minimizing defectology in both materials and achieving a correct surface finish in terms of functional performance. In this article, abrasive water jet machining of a hybrid carbon fiber-reinforced thermoplastics (CFRTP)/Steel structure and the generated surface finish are studied. A parametric study in two stacking configurations (CFRTP/Steel and Steel/CFRTP) has been established in order to determine the range of cutting parameters that generates the lowest values in terms of arithmetic mean roughness (*Ra*) and maximum profile height (*Rz*). The percentage contribution of each cutting parameter has been identified through an ANOVA analysis for each material and stacking configuration. A combination of 420 MPa hydraulic pressure with an abrasive mass flow of 385 g/min and a travel speed of 50 mm/min offers the lowest *Ra* and *Rz* values in the CFRTP/Steel configuration. The stacking order is a determining factor, obtaining a better surface quality in a CFRTP/Steel stack. Finally, a series of contour diagrams relating surface quality to machining conditions have been obtained.

**Keywords:** AWJM (abrasive water jet machining); CFRTP (carbon fiber-reinforced thermoplastics); hybrid structure; surface quality; Ra; Rz; C/TPU (carbon/thermoplastic polyurethane)

#### **1. Introduction**

Carbon fiber-reinforced thermoplastics (CFRTP) composites have an excellent weight-to-mechanical property ratio and high impact and corrosion resistance [1]. These are very interesting materials due to their ability to be remolded after curing, adopting new geometries and being of great interest for mass production [2]. Applications such as chassis in the automotive sector or the development of a lighter fuselage with better fatigue resistance developed by the company STELIA are an example of the current interest in these materials. In addition, in comparison with thermosets (CFRP), they have shorter production times and the possibility of storing the matrix at room temperature which reduces the final costs [3]. Within the wide range of thermoplastic polymers, thermoplastic polyurethane (TPU) can achieve high performance in service. The shaping of this matrix together with carbon fibers results in a flexible compound that can be adapted to various uses [4].

In order to increase the performance of these materials, current research is focused on their bonding with metal alloys in the form of hybrid structures through laser welding or friction stir welding processes [5]. These materials are essential elements in structural applications in the industry nowadays due to their mechanical properties, lightness, and corrosion resistance. Nevertheless, they must be joined to metallic elements to obtain a more robust structure that combines the performance of both materials in the form of a hybrid design [6]. Furthermore, in terms of production, the manufacture and subsequent machining of both materials at the same time means a reduction in operating times. A kind of hybrid structure of great interest is the union of CFRTP with a steel in order to obtain a structural element of high performance. This turns them into elements of great interest for the automotive sector where weight reduction is required and, at the same time, their ability to produce them in mass [2,7].

Their combination with a structural steel allows to obtain a hybrid structure of excellent performance and lightness minimizing energy consumption and CO2 emissions. Nevertheless, the quality of the interlayer of these materials on applying an adhesive or thermal bonding process has been studied due to the formation of thermal defects or the formation of bubbles due to poor surface preparation of the steel [8,9]. In addition, according to the selected process, the thermoplastic matrix of the CFRTP can be used as an integrating element of the hybrid structure affecting the final thickness [10].

In addition, to achieve a final geometry, specific machining processes are required due to the anisotropic behavior of these materials, as well as the low glass transition temperature of some thermoplastics [11]. Inside conventional processes, such as drilling or milling, wear caused to the cutting edges increases the final costs and reduces the efficiency of these processes. Processes such as milling generate a smooth and clean surface with Rz values close to 9 μm and Ra values close to 2 μm [12,13]. This is in line with the results obtained in the conventional machining of thermoset composite materials, where results below 3.2 μm are required due to aeronautical tolerances [14].

Nevertheless, although very low values are obtained, the machining temperatures deteriorate the thermoplastic matrix and cause delaminations in regions in which the reinforcement is left unprotected [12]. In addition, the fact of machining two materials of different machinability requires the use of specific cutting geometries and complex machining strategies with change of cutting parameters in the interlayer. This means an increase in operating time and costs.

On the contrary, within nonconventional technologies, abrasive water jet machining (AWJM) has proved to be a very effective technology for machining this type of material [13,15,16]. It is a flexible process, capable of achieving high material removal rates and low cutting forces and machining different materials at the same time. In addition, due to the nature of the cutting process, the temperatures reached are very low, which minimizes thermal defects [17]. Furthermore, it is a clean and environmentally friendly technology, a fundamental aspect within the field of "Green machining," and does not generate suspended particles that could affect the health of the operators. This technology offers advantages such as the recovery of abrasive particles after machining, which can be reused after treatment, and no harmful gases are generated [18]. Another important point is the retention of particles of the machined material in the pool pit, especially in composite materials, preventing them from remaining in suspension, avoiding the exposure of the operators to a harmful atmosphere. The cutting tool is water, which can be reused after machining, and abrasive particles, which can be recovered later and treated to be reused.

Nevertheless, there is little literature on abrasive water jet machining of dissimilar material stacks. Most focus on machining FMLs or CFRP/Titanium stacks due to their relevance within the aeronautical industry [19–23]. These studies are focused on the influence of the parameters that govern abrasive water jet machining on the surface quality generated, as well as on the defectology associated with this process such as the taper angle. Although there is literature on water jet machining of thermoplastic composites [24] and steel alloy [25], little information exists on machining these materials in the form of a hybrid structure.

A crucial aspect in the machining processes is the surface quality obtained. The divergence of the water jet during machining generates a reduction in the kinetic energy that results in regions of different surface quality. Thus, in the initial moments of machining, the overlapping of the abrasive particles generates a highly eroded zone known as IDR (initial damage region) [26]. Subsequently, the convergence of the jet generates a stable zone in which a homogeneous surface is obtained, known as

the SCR (smooth cutting region) [27]. Finally, due to the reduction of the kinetic energy and cutting capacity of the water jet, the surface generated has a region with very high irregularities in the form of grooves known as RCR (rough cutting region) [28,29].

In this sense, the stacking order of the hybrid structure has a fundamental role to play in the development of these regions. Depending on which material receives the first impact of the water jet, the final quality obtained in the second material will be conditioned by the difference in its machinability [23].

For this reason, minimizing water jet divergence during machining is a key aspect for achieving an acceptable surface quality. Thus, the correct selection of cutting parameters according to the order of stacking is essential to achieve this objective [30]. In studies carried out, the difference in results obtained between the composite material and the metal alloy when they are machined is also highlighted. This is due to the fact that the composite material is anisotropic and the water jet is able to eliminate the matrix generating greater irregularities or the formation of delaminations. In contrast, the isotropy of the metal alloy allows a more stable and smoother cut, but is more affected by the effect of the abrasive particles in the initial zone (IDR).

Within the cutting parameters, traverse speed and hydraulic pressure seem to be the most relevant [31]. Increases in traverse speed lead to increased water jet divergence, especially in the second material, significantly raising the roughness in the material [25]. Pahuja et al. [30] also explains the importance of traverse speed in water jet machining of a hybrid CFRP/Ti structure. Here, by increasing the speed from 1 to 10 mm/min, the Ra values increase by 14% for the titanium alloy and 260% for the composite material. On the other hand, an increase in hydraulic pressure increases the jet's machining capacity, allowing for an improvement in surface quality and obtaining *Ra* values that are very close to each other [22].

Nevertheless, the combination of the different machinability between materials, their stacking order, and the lack of knowledge about a range of cutting parameters capable of machining this type of hybrid structures require further studies. Due to this, this article proposes a parametric study in abrasive water jet machining of a hybrid CFRTP/Steel structure. The surface quality in terms of *Ra* and *Rz* has been evaluated in each material for the two stacking configurations (CFRTP/Steel and Steel/CFRTP). The difference in results between the two configurations has been evaluated, as well as a range of cutting parameters that improve surface quality has been determined. The machining of dissimilar materials in the form of a hybrid structure generates a difference in surface quality between the two elements that must be minimized or eliminated. Nevertheless, the study of the surface quality in both materials separately has been the main objective of this work. Thus, a range of cutting parameters that improves the surface quality has been determined. In order to obtain a homogeneous cut, a ratio between the results obtained for the composite material and the metal alloy has been established to identify which variation in cutting parameters generates the greatest difference in surface quality. Finally, an ANOVA analysis and a set of contour diagrams using predictive mathematical models have been obtained for the most relevant roughness parameter.

#### **2. Methodology**

#### *2.1. Materials*

This article focuses on the machining of a hybrid structure in order to evaluate the surface quality obtained. The materials selected to obtain this structure were a thermoplastic composite material reinforced with carbon fiber (Twill 200 g/m2) and a steel alloy S275. The main characteristics of the composite material are shown in Table 1. Reinforced thermoplastic laminates were produced by hot compression molding.

The CFRTP used has a thermoplastic polyurethane matrix with a melting temperature of 145◦ and the final thickness of 2.1 mm and is composed of 7 layers in 0◦ and 90◦ orientations.


**Table 1.** Mechanical properties of thermoplastic composite material (carbon fiber-reinforced thermoplastics, CFRTP).

On the other hand, the thickness of the steel was 3 mm, obtaining a final thickness of the hybrid structure of 5.1 mm. This carbon steel is a structural type with wide applications in the industrial sector due to its mechanical properties, and it is of great interest to combine it with a composite material in order to obtain a hybrid structure. Its main characteristics are shown in Table 2.

**Table 2.** Characteristics of S275 steel.


The bonding between these materials was carried out by thermoforming in a hot plate press with the aim of obtaining a continuous and quality bond to avoid the formation of delaminations in the interlayer and to relate possible defects to the machining conditions. The characteristics of the thermoplastic matrix allow the matrix itself to be used as an integrating element between both materials by changing from a solid to a liquid state when its glass transition temperature is exceeded. Subsequently, the matrix expands and impregnates the surface of the steel alloy to generate a constant bond after it has cooled down. To ensure a quality bond, the steel surface is modified by sand-blasting, using a pressure of 5 bar, 630 μm corundum particles, and an impact distance of 100 mm. Because of it, a surface free energy value of 50 mJ/m<sup>2</sup> [32] was obtained.

#### *2.2. Abrasive Water Jet Machining (AWJM)*

The equipment used consisted of a water jet cutting machine (TCI Cutting, BP-C 3020, Valencia, Spain). The nozzle of the machine had a diameter of 0.8 mm, an orifice diameter of 0.3 mm, and a nozzle length of 94.7 mm. The AWJM machine was equipped with an ultrahigh capacity pump (KMT, 158 Streamline PRO-2 60, Bad Nauheim, Germany). All trials were carried out by 120 mesh Indian Garnet abrasive particles.

Three cutting parameters were modified according to the literature consulted. Three levels of hydraulic pressure (P), traverse speed (TS), and abrasive mass flow (AMF) were established (Table 3). At the same time, in order to obtain a greater robustness and repeatability in the results obtained, each combination of cutting parameters were carried out twice, obtaining a total of 54 tests. Due to its importance in the conservation of kinetic energy of the water jet, the jet-piece distance was set at 3 mm.



The tests consisted of slots with a machining length of 30 mm and a gap between cuts of 6 mm. In order to guarantee a constant flow and traverse speed, the cuts were started 15 mm before the beginning of the material (Figure 1).

On the other hand, the order of the materials during machining is a key parameter in the final quality. Because of this, each test has been performed in two stacking configurations, CFRTP/Steel, and Steel/CFRTP.

**Figure 1.** Abrasive water jet machining of a hybrid structure Steel/carbon fiber-reinforced thermoplastics (CFRTP).

#### *2.3. Surface Quality Evaluation*

The surface finish after machining processes is a key parameter in determining the functional performance of the geometry obtained. The importance of the surface of the parts in the functional behavior of the latter is relevant when considering that it is through their surfaces that contact is established between them, being the main basis of most mechanical functions [33].

Surface quality in abrasive water jet machining is a key factor. The loss of kinetic energy of the water jet in combination with poor selection of cutting parameters produces large variations. This results in a first region that is highly affected by abrasive particles and a final zone of high roughness due to the formation of grooves. This in combination with the fact that dissimilar materials are being machined at the same time makes it an essential parameter to study.

Due to this, the surface quality has been evaluated in terms of arithmetic mean roughness (*Ra*) and maximum profile height (*Rz*). The anisotropy of the composite material and the possible formation of defectology associated with the loss of thermoplastic matrix requires the study of several parameters that provide more complete and real information about the surface obtained [34]. Three measurements were made at three height levels in each material (Figure 2). Each roughness profile was measured at three different levels, i.e., at 25%, 50%, and 75% of the thickness of each material. The goal was to determine the presence of the three characteristic regions in abrasive water jet machining in terms of surface quality: IDR (initial damage region), SCR (smooth cutting region), and RCR (rough cutting region). The measurements were made in a perpendicular direction to the grooves generated by the water jet.

**Figure 2.** Graphical representation of surface quality evaluation.

A roughness-meter (Mahr Perthometer PGK 120, Göttingen, Germany) was used. The surface quality evaluation was carried out following ISO 4288:1999 standard. A cut-off of 2.5 mm was established for a total evaluation length of 12.25 mm. Stylus with 2 μm tip radius and 90◦ tip angle was used for the measurements, reference M-250 from Mahr.

Finally, the surface generated after the surface modification was evaluated by visual inspection using a scanning electron microscope (Hitachi, VP-SEM SU1510, Schaumburg, IL, USA).

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

#### *3.1. CFRTP*/*Steel*

This section shows the results obtained in the first CFRTP/Steel configuration in order to determine the influence of the cutting parameters. The surface quality obtained in the thermoplastic composite material in terms of Ra is shown in Figure 3. For pressures of 250 MPa, two tendencies are observed when increasing the abrasive flow. When the traverse speed is 50 mm/min, high values of abrasive increase the roughness. This may be due to an excess amount of particles impacting the surface. In combination with reduced pressure, the jet does not have enough energy to obtain a clean cut (Figure 4). Thus, the intercollisions of the abrasive particles reduce its cutting capacity, producing a more eroded zone [35].

**Figure 3.** CFRTP surface quality results in terms of arithmetic mean roughness (*Ra*) as a function of the cut-off parameters set for the CFRTP/Steel configuration.

**Figure 4.** Initial damaged region (IDR) due to the erosive effect of the abrasive particles (hydraulic pressure (P) 250 MPa, traverse speed (TS) 300 mm/min, and abrasive mass flow (AMF) 385 g/min) at: (**a**) 250× and (**b**) 500×.

On the other hand, when the traverse speed increases its value, this trend for a hydraulic pressure of 250 MPa is totally opposite due to the increase in the kinetic energy of the water jet. This produces a rougher RCR zone due to the curvature of the water jet due to the loss of kinetic energy [36]. Nevertheless, as hydraulic pressure increases, the surface quality is directly influenced by the traverse speed, especially for pressures of 420 MPa.

Thus, an increase in pressure and abrasive flow at reduced traverse speeds produces a stable water jet capable of homogeneously machining all composite material [37].

When the traverse speed is maximum, the increase in pressure has a varying effect depending on the amount of abrasive particles applied in the machining. With a minimum flow rate of 225 g/min, there is a significant increase in the machining capacity of the water jet, obtaining a more constant material removal and reducing the *Ra* values from 7 to almost 5 μm.

The increase of the abrasive particles improves the cutting capacity of the jet allowing a smoother region. This can be seen in the pressure of 250 MPa. However, a combination of high values of both hydraulic pressure and abrasive flow can be excessive resulting in a deterioration in surface quality. This may be due mainly to an excess of abrasive particles that intercollide, minimizing their erosive effect. In turn, this produces an increase in the IDR region increasing the abrasive particles adhered in the initial moments of machining and increasing the average roughness [15]. This in combination with the divergence between the inlet and outlet zones of the water jet due to a very high travel speed results in these variations when hydraulic pressure is increased.

Also, an increased pressure leads to a greater offset between the inlet area of the water jet and the outlet area. During this time, the kinetic energy of the water jet allowing the material to be machined is reduced. Thus, the impacts generated on the surface are more abrupt, generating irregularly shaped machined areas [26]. This could produce an area of higher roughness and poorer surface quality (Figure 5). These roughness profiles were made at three different levels as shown in Figure 2. The first profile at a distance equivalent to 25% of the thickness, the second profile at 50% of the thickness, and the third profile at 75% of the thickness of the material were obtained.

**Figure 5.** Roughness profiles for the combination of *P* of 420 MPa, *TS* of 300 mm/min, and *AMF* of 225 g/min for: (**a**) IDR, (**b**) smooth cutting region (SCR), (**c**) rough cutting region (RCR) region, and (**d**) overlapping roughness profiles.

This is corroborated by the results obtained for the parameter *Rz* (Figure 6). The trends obtained are very close to those obtained for Ra, which would justify the previously described trends. Thus, it can be seen that a combination of a pressure of 420 MPa, an abrasive flow of 385 g/min, and a traverse speed of 100 mm/min generate the minimum values of *Ra* and *Rz*.

**Figure 6.** Maximum profile height (*Rz*) values for CFRTP in CFRTP/Steel configuration.

In addition, the surface quality of the steel in terms of Ra is shown in Figure 7 and in terms of Rz in Figure 8. It can be seen that lower values are obtained in the metal alloy compared to the composite material. This is due to the composition of both materials. The anisotropy of the composite material causes each layer to behave differently when interacting with the water jet [30].

**Figure 7.** Steel surface quality results in terms of *Ra* as a function of the cut-off parameters set for the CFRTP/Steel configuration.

**Figure 8.** *Rz* values for steel in CFRTP/Steel configuration.

Furthermore, the different machinability of the reinforcement and matrix in combination with the dispersion of the water jet produces an effect known as hydrodistortion [21] (Figure 9). In addition, the main removal mechanism in the composite material was by microbending and fracture and in the matrix was by erosion. This results in a transversal removal of the matrix, leaving the reinforcement unprotected and generating a worsening of the surface quality (Figure 10).

Due to the monolithic composition of the steel, a more homogeneous machining has been obtained. This produces a smoother surface compared to composite material. However, as the second material is machined in the CFRTP/Steel configuration, a reverse situation is generated. A combination of a *TS* of 300 mm/min and an *AMF* of 340 g/min shows this effect. The divergence of the water jet between the inlet and outlet region is very high due to the destabilization of the water jet at this travel speed. This is enhanced by the difference in machinability between the two materials. In other words, the water jet is not capable of machining both materials consistently at the same time.

In addition, the reduced amount of abrasive particles minimizes the actual machinability of the water jet, resulting in a rougher surface [22].

**Figure 9.** Hydrodistortion defect at a pressure of 250 MPa, an abrasive mass flow of 225 g/min, and a travel speed of 50 mm/min.

**Figure 10.** Loss of thermoplastic matrix leaving the reinforcement unprotected at 1000× (*P* of 250 MPa, *TS* of 300 mm/min, and *AMF* of 385 g/min).

Due to the low machinability of steel, the increased abrasive mass flow has a positive effect on the surface quality. An increase in particles enhances the erosive capacity of the water jet allowing it to penetrate the steel more easily due to greater stabilization in the cut [38]. This can be seen in most tests where an increase in this parameter generates a reduction in both *Ra* and *Rz*.

The trend of each cutting parameter in the surface quality generated in terms of Ra for each material is shown in Figure 11. In terms of hydraulic pressure, steel is the most important material as it is more difficult to machine and is the second material in the structure. An increase in this parameter improves the penetration capacity of the water jet, facilitating shear impacts on the surface and obtaining a better surface quality. An increase in the amount of abrasive particles reduces the resistance of the material when machined.

Nevertheless, depending on the level of pressure and traverse speed, it can become a negative aspect. On the contrary, the traverse speed seems to be the most critical parameter for surface quality. The increase from 50 to 300 mm/min produces a 40% increase in both materials due to the destabilization of the water jet and the inability to machine both materials at the same time.

**Figure 11.** Cutting parameter trends in surface quality (*Ra*) in CFRTP/Steel configuration. (**a**) CFRTP Ra in function of *P*, (**b**) CFRTP Ra in function of *AMF*, (**c**) CFRTP Ra in function of *TS*, (**d**) Steel Ra in function of *P*, (**e**) Steel Ra in function of *AMF*, and (**f**) Steel Ra in function of *TS*.

#### *3.2. Steel*/*CFRTP*

The results obtained in the reverse stacking order for the composite material are shown in Figures 12 and 13.

Compared to the CFRTP/Steel configuration, the results obtained are slightly higher. This may be due to the positioning within the hybrid structure. When machining the metal alloy, a large part of the kinetic energy of the water jet is absorbed by this material, reducing the ability to penetrate the composite material [22,26]. This is especially outstanding when the pressure is minimal (250 MPa) and the amount of abrasive particles is insufficient. This produces an increase in the hydrodistortion effect between reinforcement and matrix resulting in a very rough surface where the reinforcement is not properly machined, generating high deviations (Figure 14).

**Figure 12.** CFRTP surface quality results in terms of *Ra* as a function of the cutting parameters set for the inverse Steel/CFRTP configuration.

**Figure 14.** Reinforcements that are not machined, resulting in a worse surface quality.

The formation of a turbulent jet in the interlayer and the consequent loss of power of the water jet results in an unstable flow that generates a very rough area. This, in combination with the low cohesion between the reinforcement and the thermoplastic matrix, generates a separation between both leaving the reinforcement unprotected and increasing the final roughness [39]. This can be seen in Figure 15.

Again, similar trends are observed in both *Ra* and *Rz*. However, in contrast to the CFRTP/Steel configuration, both hydraulic pressure and abrasive mass flow do not seem to have such a significant effect on surface quality. Only when the speed is maximum, an increase in these parameters generates a noticeable difference. In the CFRTP/Steel configuration, the influence of these parameters is more noticeable because the water jet dispersion is lower. In this sense, the increase in pressure minimizes the hydrodistortion defect in the composite material, minimizing the loss of kinetic energy prior to steel machining.

Thus, the machining capacity of the water jet is more constant and a variation in these parameters is more relevant. On the contrary, when the first material to be machined is steel, this energy loss is greater because it presents a greater difficulty to be machined, minimizing the effect of the cutting parameters in the composite material.

In contrast, the travel speed seems to have a more prominent effect in this configuration due to the dispersion of the water jet. When the jet starts machining at a very high speed and the first material (Steel) has a worse machinability, an excessive delay is generated between the machining of this and the second material (CFRTP) and an increase in the hydraulic pressure enhances the penetration capacity of the water jet improving the surface integrity in spite of obtaining very high *Ra* values [13].

**Figure 15.** Total loss of thermoplastic matrix leaving the reinforcement unprotected in 0◦ and 90◦ stacking orientation at 250× (*P* of 250 MPa, *TS* of 300 mm/min, and *AMF* of 385 g/min).

On the other hand, the results obtained in steel machining in terms of *Ra* (Figure 16) and *Rz* (Figure 17) are very close to those obtained in the reverse configuration. This would indicate that steel is the most decisive material in the machining of this structure. In terms of surface quality, the positioning of the steel does not affect the results obtained, but it does directly affect the final quality generated in the composite material.

**Figure 16.** Steel surface quality results in terms of *Ra* as a function of the cutting parameters set for the inverse Steel/CFRTP configuration.

The three cutting parameters generate an improvement in the quality obtained. When the traverse speed is between 50 and 100 mm/min, it seems that the most dominant parameter is the hydraulic pressure by reducing the Ra values from 4 to 3 μm and minimizing the deviations obtained, which would indicate that the surface is very homogeneous. This is corroborated by the *Rz* results, which would indicate that the surface is smooth with constant surface variations. On the other hand, as with the other results, the increase at a speed of 300 mm/min significantly worsens the surface quality due to the destabilization of the water jet, which leads to an increase in the lag defect in the RCR region.

With regard to the abrasive flow, its effect is more noticeable when the speed is higher than 300 mm/min due to the loss of kinetic energy. An increase in this parameter improves the tearing of the steel by shear forces. In addition, flows of 385 g/min offer a very reduced deviation in both Ra and *Rz*, which would indicate minimal variations in the surface obtained. Thus, for this level of abrasive mass flow and a pressure of 420 MPa, very close surface quality values are obtained for speeds of 100 and 300 mm/min, allowing an increase in productivity in the machining of hybrid structures.

**Figure 17.** *Rz* values for steel in the Steel/CFRTP configuration.

The trends for each cutting parameter are shown in Figure 18. A great influence is observed on the composite material by increasing the hydraulic pressure and traverse speed. Being the second material to be machined in this stack directly affects the results obtained. Steel has a higher resistance to be machined and makes it difficult to stabilize the water jet prior to the machining of the composite material. This, in combination with the turbulence that can be generated in the interlayer, affects the surface quality by varying these parameters.

On the contrary, different trends have been observed in the surface quality of the steel compared to the CFRTP/Steel configuration. Thus, the pressure does not produce a significant variation in the results and the abrasive flow seems to slightly reduce Ra values that improves the surface quality. However, the trend of the traverse speed is constant and similar in both materials and stacking configurations with an increase in the results due to the delay in the water jet and not using the same amount of particles per unit area.

The machining of materials of different machinability reflects that the surface quality can be very disparate and that the machined part does not fulfill its function. The ratio between the Ra values for the CFRTP/Steel configuration is shown in Figure 19 and for the Steel/CFRTP configuration in Figure 20.

**Figure 18.** Cutting parameter trends in surface quality (*Ra*) in the Steel/CFRTP configuration. (**a)** CFRTP Ra in function of *P*, (**b**) CFRTP Ra in function of *AMF*, (**c**) CFRTP Ra in function of *TS*, (**d**) Steel Ra in function of *P*, (**e**) Steel Ra in function of *AMF*, and (**f**) Steel Ra in function of *TS*.

**Figure 20.** Ratio values *Ra* CFRTP and *Ra* Steel for Steel/CFRTP configuration.

In general, the results obtained in other studies are corroborated, where the quality obtained in the composite material is superior to steel with ratios greater than 1. In turn, in both configurations, an increase in the abrasive flow parameter produces a greater dispersion between both materials.

This can be seen especially in the CFRTP/Steel configuration, due to the fact that a greater number of abrasive particles increase the detachment of the thermoplastic matrix, causing the reinforcement to be free and worsening the surface quality. On the contrary, this increase improves the penetration capacity of the water jet allowing a stable cut in the steel and reducing the *Ra* values compared to the composite material.

It should be noted that, although the traverse speed is a parameter that worsens the surface quality considerably, its tendency is very close in both materials, which generates close ratios.

In terms of cutting parameters, ratios close to 1 are obtained for a pressure of 420 MPa because the loss of kinetic energy of the water jet is not significant, especially in the Steel/CFRTP configuration. Thus, the stacking configuration that offers the closest values of surface quality in terms of Ra is Steel/CFRTP.

It should be noted that, in both stacking configurations, gaps between materials have not been observed (Figure 21).

**Figure 21.** Final quality of the bond between materials after machining at 250× (*P* of 250 MPa, *TS* of 300 mm/min, and *AMF* of 385 g/min): (**a**) CFRTP/Steel and (**b**) Steel/CFRTP.

In both cases, abrasive particles have remained adhered to the two materials and can affect the final surface quality. In the CFRTP/Steel configuration, remains of the thermoplastic matrix can be seen that have been pulled and adhered to the surface of the steel. On the contrary, in the inverse configuration, a cleaner surface can be seen in the interlayer.

#### *3.3. Statistical Analysis and Contour Diagrams*

The percentage contribution of each cutting parameter in the surface quality for each material and stacking configuration obtained by ANOVA analysis is shown in Figure 22.

It is confirmed that the traverse speed is the most determining parameter according to the results obtained. This is particularly evident in the Steel/CFRTP configuration. An increase in this parameter generates a greater destabilization in the water jet generating a rougher surface. In addition, due to the loss of kinetic energy, the RCR region increases, leading to the formation of grooves [25,40]. On the other hand, another key factor is the hydraulic pressure. An increase in this parameter improves the penetration and machining capacity of the water jet by facilitating the removal mechanism [41].

A balance between traverse speed and hydraulic pressure has been observed in the CFRTP/Steel configuration. A correct selection of these cutting parameters reduces the hydrodistortion defect in the layers of the composite material and minimizes the detachment of the thermoplastic matrix. This results in a less rough surface and improved surface quality.

**Figure 22.** Percentage contribution of cutting parameters on surface quality for: (**a**) CFRTP/Steel and (**b**) Steel/CFRTP.

In parallel, with the experimental results obtained, a series of predictive second-order polynomial models have been generated that relate surface quality in terms of Ra with cutting parameters for applications in the industrial sector.

The models obtained for the CFRTP/Steel configuration are shown in (1) and (2) with values of R2 of 67.48% and 86.98%, respectively, and the models for the Steel/CFRTP configuration in (3) and (4) with adjustments of 85.73% and 95.25%, respectively. It should be noted that, due to the anisotropy and the reduced thickness of the composite material, it generates a randomness in the surface quality that reduces the adjustment obtained.

$$\begin{aligned} \text{Ra} \ (\text{CFRTP}) &= \ 5.37 + 0.00211 \cdot P + 0.0050 \cdot AMF + 0.0350 \cdot TS - 0.000028 \cdot P\\ &\quad \cdot AMF - 0.000119 \cdot P \cdot TS - 0.000123 \cdot AMF \cdot TS \end{aligned} \tag{1}$$

$$\begin{aligned} Ra\ (Steel) &= \ 6.18 - 0.00568 \cdot P - 0.00615 \cdot AMF + 0.0116 \cdot TS + 0.000005 \cdot P\\ &\cdot AMF - 0.000011 \cdot P \cdot TS - 0.000045 \cdot AAMF \cdot TS \end{aligned} \tag{2}$$

$$\begin{aligned} Ra \, (\text{CFRTP}) &= -0.20 + 0.0123 \cdot P + 0.0128 \cdot AMF + 0.0299 \cdot TS - 0.000032 \cdot P\\ &- AMF - 0.000037 \cdot P \cdot TS - 0.000042 \cdot AMF \cdot TS \end{aligned} \tag{3}$$

$$\begin{aligned} \text{Ra} \ (Stel) &= \ 7.91 - 0.01441 \cdot P - 0.00765 \cdot AMF + 0.000025 \cdot P \cdot AMF + 0.000024\\ &\cdot P \cdot TS + 0.000003 \cdot AMF \cdot TS \end{aligned} \tag{4}$$

And the corresponding contour diagrams are shown in Figures 23 and 24, which relates the surface quality (*Ra*) to the cutting parameters for both stacking configurations. Values close to 4.5 μm in the composite material in the CFRTP/Steel configuration and close to 5 μm in the Steel/CFRTP structure are obtained by combining a pressure of 420 MPa, an AMF of 385 g/min, and a *TS* of 50 mm/min in the CFRTP/Steel configuration.

**Figure 23.** Contour diagrams for the CFRTP/Steel configuration: (**a**) Composite *TS* vs. *P*, (**b**) Composite *AMF* vs. *TS*, (**c**) Steel *TS* vs. *P*, and (**d**) Steel *AMF* vs. *TS*.

In this way, there is a direct relationship between the roughness generated and the ratio between the power of the water jet and the penetration depth *(* . *E*/*h*). Very high values of this parameter indicate surfaces with low roughness. Pahuja et al. [30] explain that the composite/metal configuration shows a high initial roughness and suffers a very fast decrease in the composite material and slower in the titanium due to the loss of kinetic energy of the water jet.

Furthermore, regardless of the material, the second material to be machined suffers an increase in *Ra* values compared to the reverse configuration. Thus, it is corroborated that no matter the composite material (thermoplastic or thermoset) and the metal alloy used, the roughness in a stacked configuration is mainly governed by the characteristics of the jet. This is enhanced when the pressure is minimal due to the reduction in machining capacity.

Conversely, lower values are obtained for the metal alloy with very similar results in both stacking configurations close to 3.5 μm. These values are achieved by combining a *P* between 320 and 420 MPa, an *AMF* of 385 g/min, and a *TS* of 50 mm/min.

**Figure 24.** Contour diagrams for the Steel/CFRTP configuration: (**a**) Composite *TS* vs. *P*, (**b**) Composite *AMF* vs. *TS*, (**c**) Steel *TS* vs. *P*, and (**d**) Steel *AMF* vs. *TS*.

#### **4. Conclusions**

Surface quality in machining processes is a key parameter in terms of functional performance. Abrasive water jet machining of hybrid structures of dissimilar materials generates a highly variable surface quality that depends directly on the correct selection of cutting parameters and stacking order.

Typical defectology in abrasive water jet machining of thermoset composite materials has been identified in thermoplastic composites. Small delamination and matrix loss have been detected leaving the reinforcement unprotected.

Stacking order is a key factor. Lower*Ra* and*Rz* values are obtained in the CFRTP/Steel configuration due to better conservation of the kinetic energy of the water jet. This allows for a better cutting capacity of the water jet, especially in the composite material by minimizing matrix loss and reducing fiber pull-out defectology. In contrast, in the Steel/CFRTP configuration, due to the difference in machinability, the steel absorbs much of the energy of the water jet reducing the ability to penetrate into the composite material and resulting in a rougher and more random surface.

With regard to cutting parameters, the traverse speed is the most critical factor. In both materials and stacking configurations, an increase in this parameter generates a notable growth in the *Ra* and *Rz* values due to the divergence of the water jet and the offset that is generated between the first material and the second during machining. Thus, smoother surfaces are obtained with a traverse speed close to 50 mm/min.

The lowest values of surface quality have been obtained by combining a traverse speed of 50 mm/min, a hydraulic pressure of 420 MPa, and an abrasive mass flow of 385 g/min, maximizing the machining capacity of the water jet.

Finally, a series of predictive mathematical models have been obtained with good fits that relate the surface quality in terms of *Ra* in both materials and stacking configurations to the cutting parameters and which may be of interest and application in current industry.

**Author Contributions:** F.B. and A.S. developed machining tests. F.B. developed data treatment. F.B. and A.S. analyzed the influence of the parameters involved. F.B. and J.S. wrote the manuscript, figures, and tables. M.B., B.S., and J.S. contributed to the experimental design and critical comments on the final manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been developed under support of a predoctoral industrial fellow financed by NANOTURES SL, mechanical engineering and industrial design department, and Vice-rectorate of Transference and Technological Innovation of the University of Cadiz.

**Acknowledgments:** The authors would like to thank the Laboratory of Corrosion and Protection TEP-231 (Labcyp) of the University of Cadiz for the support with scanning electron microscopy.

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

#### **Glossary of Terms**


#### **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* **E**ff**ects of Machining Parameters on the Quality in Machining of Aluminium Alloys Thin Plates**

#### **Irene Del Sol 1,\*, Asuncion Rivero <sup>2</sup> and Antonio J. Gamez <sup>1</sup>**


Received: 23 July 2019; Accepted: 22 August 2019; Published: 24 August 2019

**Abstract:** Nowadays, the industry looks for sustainable processes to ensure a more environmentally friendly production. For that reason, more and more aeronautical companies are replacing chemical milling in the manufacture of skin panels and thin plates components. This is a challenging operation that requires meeting tight dimensional tolerances and differs from a rigid body machining due to the low stiffness of the part. In order to fill the gap of literature research on this field, this work proposes an experimental study of the effect of the depth of cut, the feed rate and the cutting speed on the quality characteristics of the machined parts and on the cutting forces produced during the process. Whereas surface roughness values meet the specifications for all the machining conditions, an appropriate cutting parameters selection is likely to lead to a reduction of the final thickness deviation by up to 40% and the average cutting forces by up to a 20%, which consequently eases the clamping system and reduces machine consumption. Finally, an experimental model to control the process quality based on monitoring the machine power consumption is proposed.

**Keywords:** thin plates; thin-wall; machining; aluminium; cutting forces; roughness

#### **1. Introduction**

Aluminium fuselage skin panel machining is considered a challenging operation due to its dimensional and surface requirements. These parts are lightened by machining superficial pockets in order to increase the fuel efficiency of aircrafts by reducing their structure weight. These pockets have historically been machined using chemical milling operations, although green manufacturing approaches have been focused on the study of mechanical machining for this purpose [1]. In fact, different projects and research studies have invested hundreds of thousands of euros to remove chemical milling, designing specific clamping systems to ensure surface quality and dimensional requirements while maintaining clamping flexibility. These systems are focused on twin-machining heads [2,3], magnetic slaves [4] or flexible vacuum beds [5] that control the deflection of the part avoiding overcut during the operation.

Additionally, the conventional machining of low stiffness parts presents dynamic and static problems [6,7]. On the one hand, dynamic stability of machining strongly depends on system stiffness, its natural frequency response, and the selected cutting parameters. Vibrations—chatter and forced vibrations—can directly affect the final roughness of parts, increasing their value and forcing manufacturers to make reprocessing steps, therefore increasing the operational cost [8,9]. In order to avoid them, chatter influence is studied using stability lobe diagrams (SLD), a representation tool that commonly relates the stability areas of machining with the feed rate, the spindle speed, the depth of cut, or the tool position [10–14], and forced vibrations can be studied through dynamic models. In this case, the applied force is studied to reduce the dynamic deflection of the part. On the other hand, quasi-static deflection can take place when the elastoplastic behaviour of the workpiece, combined with a failure on the clamping, is not enough to counter the machining force effect, reducing the real depth of cut [15–17]. This fact was proved experimentally by Yan et al. [18], who optimized the depth of cut depending on the cutting force, reducing the part deflection and increasing the process efficiency. Similarly, Sonawane et al. [19] used a statistical approach to model workpiece deflection, addressing it to the machining parameters and cutting tool orientation. In contrast, Oliveira et al. [20] established that the most influencing factor on the real depth of cut was the milling strategy (up or down milling), while other studies [21,22] have focused on the analysis of the toolpath effect on the final quality of slim parts.

However, few studies have been focused on the analysis of the parameters effect into the quality characteristics of thin plates. The literature review has shown that most of the studies have focused on thin-wall machining rather than thin-plates or thin floors [7]. Few of them have focused on the analysis of the parameters effect into the final quality characteristics. For this reason, this paper focuses on the study of thin plates in order to simulate the machining of large skin panels to evaluate the effect of the machining parameters on the final thickness, surface roughness and cutting forces of the part.

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

#### *2.1. Machining Tests*

Two type of machining tests were performed. The first type was used to analyse the thin plates' behaviour through the machining of 50 <sup>×</sup> 50 mm<sup>2</sup> pockets on samples of 80 <sup>×</sup> 80 mm<sup>2</sup> and 2 mm thickness. Parts were screwed to a faced mill plate that was housed on a dynamometer. The samples were dry machined while keeping the axial distance constant, and the depth of cut (*ap*), the feed rate per tooth (*fz*) and the cutting speed (*Vc*) were variable. The values of each machining conditions are compiled in the levels listed in Table 1, in which the spindle speed (*S*) is also shown. The depth of cut was selected based on the geometrical requirements of industrial parts. Feed rate and cutting speed were selected based on the literature [14,23,24] and aerospace recommendations. The chosen strategy was down milling, following a toolpath from the centre of the workpiece up to the outside of it.


**Table 1.** Machining parameters and levels for skin sample tests.

The second test aimed to obtain the cutting force values following a mechanistic approach. In this approach, the specific force coefficients of the combination pair tool-material had to be experimentally defined by performing different slots on a rigid aluminium workpiece. Additionally, in order to avoid chatter and ensure the rigid behaviour of the samples, SLD were calculated using an impact hammer test. The maximum depth of cut for stable machining was calculated using the procedure described by Altintas and Budak [25]. Following this procedure, the frequency responses of the tool and the workpiece at four different steps of the cutting operation were obtained. This combination provided the SLD diagram at the four stages in order to analyse possible changes during the machining operation.

Each machining test was performed in a 5-axis NC centre ZV 25U600 EXTREME (Ibarmia Innovatek S.L.U., Azkoitia, Spain). The material used on the sample parts was aluminium 2024-T3, and the tool was a torus end mill KENDU 4400.60 (Kendu, Segura, Spain) with a 10 mm diameter, 30◦ helix angle, 18◦ rake angle, 16◦ clearance angle for the secondary edge and 9◦ angle for the primary edge.

Forces and accelerations on the workpiece were monitored using a dynamometer Kistler 9257B (Kistler Group, Winterthur, Switzerland) and an accelerometer Kistler 8728A500 (Kistler Group, Winterthur, Switzerland), connected to National Instruments acquisition boards NI 9215(National Instruments, Austin, TX, USA) and NI 9234 (National Instruments, Austin, TX, USA), respectively. The power consumed by the spindle speed and the whole machine was also recorded using a Fanuc Servoguide system (Fanuc Corporation, Oshino-mura, Japan). The test configuration and monitoring system are shown in Figure 1.

**Figure 1.** Scheme of the monitoring system.

#### *2.2. Force Mechanistic Model*

The expected forces could be calculated through a mechanistic approach [26,27] Tangential (*t*), axial (*a*) and radial (*r*) forces could be considered as a function of the cutting coefficient (*Kqc*) and the friction coefficient - *Kqe* .

$$\partial F\_{\eta}(\varphi, z) = \mathbb{K}\_{\eta \epsilon} \partial S + \mathbb{K}\_{\eta \epsilon} f\_z \sin \eta(\varphi\_i, z) \,\partial z, \; q = \langle t, r, a \rangle \tag{1}$$

where ∂*S* is the differential chip edge length and ϕ is the applied rotation angle which depends on the instant depth of cut (*z*), the number of teeth engaged (*j*), the total number of teeth (*N*), and the helix angle (β).

$$
\varphi(\varphi\_{i\nu}z) = \varphi\_i - (j-1)\frac{2\pi}{N} - \beta \tag{2}
$$

The cutting forces were converted to Cartesian coordinates using Equation (3), with κ being the angle referred to the torus part of the mill.

$$\frac{\partial F\_{\mathbf{x},y,z}}{\partial z} = \begin{bmatrix} -\cos\varphi & -\sin\varphi\sin\kappa & -\sin\varphi\cos\kappa \\ \sin\varphi & -\cos\varphi\sin\kappa & -\cos\varphi\cos\kappa \\ 0 & \cos\kappa & -\sin\kappa \end{bmatrix} \begin{bmatrix} \partial F\_t \\ \partial F\_r \\ \partial F\_d \end{bmatrix} \tag{3}$$

The cutting and friction coefficients were obtained by solving the equation using the force values obtained in the slot test performed in a rigid part. The coefficients were considered constant for all the *fz*, but the effect of the *Vc* was taken into account. The test conditions are shown in Table 2. The results were used to predict the SLD.


**Table 2.** Machining parameters and levels for cutting coefficient calculation tests.

#### *2.3. Quality Evaluation*

The quality of the parts was established through the final thickness distribution and roughness. Typical tolerance values in the industry were really tight, about ±0.1 mm for final thickness and under 1.6 μm for the roughness average (*Ra*).

Final thicknesses were measured using a single coordinate measurement machine with an electronic comparator set. Nine points of the sample were evaluated, as shown in Figure 2a. *Ra* was measured using a Mahr Perthometer Concept PGK120 roughness measure station (Mahr technology, Göttingen, Germany) on five different areas of the part (Figure 2b). The areas for the roughness measures were selected in order to study the whole machining process and to cover both machine *x* and *y* axes. Roughness measurements were taken following the standard ISO 4288:1996 [28].

**Figure 2.** Measure procedure. (**a**) points selected for the thickness analysis and (**b**) areas studied for the roughness analysis.

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

#### *3.1. Final Thickness Error*

The final thickness error can be defined as the difference between the experimental thickness and the expected one; this parameter measures the real part dimension. In addition to the static and dynamic phenomena that occur during low rigidity parts machining [7], aspects such as the machine positioning error and the thermal expansion of the spindle have influence over this parameter. Other possible thermal errors can be discarded due to the short length of the machining test, which allowed us to underestimate the effect on the accuracy of the machine due to temperature changes in its surroundings.

The analysis of the results showed that the feed rate had a negligible effect on the average final thickness error, while the cutting speed seems to have had a significant effect on this parameter. The higher the cutting speed, the greater the geometric error was, and the piece became thinner (Figure 3). Nevertheless, the thermal expansion of the spindle increased with the revolutions and therefore with the cutting speed [29]. This fact explains the part thinning, rejecting the direct effect of the cutting speed. Though the thermal expansion of the spindle was identified as an influencing parameter for the increase of the final average thickness error of the part, this error was easy to compensate when studying the elongation curves of the spindle and considering them in the CAM design [30] or implementing error compensation rules in the machine control [5].

**Figure 3.** Average final thickness error depending on the feed per teeth (*fz*) and the cutting speed (*Vc*).

Once the average error—not directly related to the cutting parameters—can be compensated, the target is to get a homogeneous thickness distribution. Higher values of feeds per teeth combined with higher values of spindle speed led to a reduction up to 40% of the standard deviation of the thickness errors measured in a test sample. This decrease was due to a better behaviour of the process in terms of dynamics. If the part was machined at higher rotation speed where cutting forces excited higher frequency vibration modes and created lower vibration amplitudes (Figure 4), therefore leading to a more homogeneous thickness distribution

**Figure 4.** Fast Fourier Transform of the part acceleration signal for test at *fz* = 0.1 and *ap* = 1.0 mm, under two different cutting speeds.

#### *3.2. Roughness*

Roughness results were not significantly affected by any of the studied parameters in average deviation, with all of them being inside the most restrictive tolerance values (1.6 μm) required in chemical milling process (Figure 5a).

**Figure 5.** (**a**) Average of the roughness values obtained for different feed rates (*f* = *N*·*fz*·*S*) as a function of the depth of cut. The inlet shows the partial effect of the feed rate per teeth and the cutting speed. (**b**) Average roughness values of each performed test against the depth of cut error considering the force module.

Measured forces and depths of cut errors did not have any impact on this quality characteristic (Figure 5b). However, lower cutting force values were related to roughness values under 0.3 μm. In fact, lower forces caused less tool deflection and vibrations of smaller amplitudes, leading to more stable machining processes that allowed us to produce more homogenous surfaces [31]. This revealed that the surface quality can be kept constant for any depth of cut under similar machining conditions. For this reason, the selection of parameters that decrease the forces should be considered.

#### *3.3. Forces and Power Models*

According to the SLD (Figure 6), thin plates almost behaved like a rigid part. This fact ensures that the tests were performed under stable machining conditions, proving that any variation occurring on the final thickness of the part was not produced by chatter issues.

**Figure 6.** (**a**) Stability Lobe Diagrams (SLD) variation depending on the machining stage. Red line, before the machining; green line, in the first instants of the machining; blue line, intermediate stage; and yellow line, after the whole machining operation was performed. (**b**) Scheme of the areas of material removed in each stage with corresponding colours.

However, the forces obtained on the tests performed on a rigid body did not completely agree with those measured on the flexible parts (Figure 7). The depth of cut was linearly related to *Fx* and *Fy*, but for medium and high cutting speed values, the *Fz* initiated a constant trend at cutting depths of 0.8 and 0.1 mm, respectively. Even though the machining operation was not in a high speed machining regimen for aluminium alloys, there was a decrease of the force values expected for the higher depths of cut. This fact had been previously observed by López de Lacalle et al. [32], where, due to the reduction of stiffness, the cutting forces decreased, obtaining behaviours closer to high speed machining at lower cutting speed.

**Figure 7.** Average forces for the *x*, *y* and *z* axes under rigid (solid blue line) and flexible (dashed orange line) consideration.

Aiming to provide a suitable model to relate the machining parameters to the process performance, the forces have been also studied by following a statistical approach. In this research, the force

module and the material removal rate (*MRR*), which can be calculated as is shown in Equation (4), were correlated.

$$MRR = f\_{\overline{z}} \cdot \mathbb{S} \cdot a\_{\overline{p}} \cdot \mathbb{N} \cdot a\_{\overline{r}} \tag{4}$$

A potential regression was chosen for the model in order to ensure an *R*<sup>2</sup> close to 0.95. The model equation is shown in Equation (5), where *e* is the Euler number, *N* is the number of teeth and *ar* is the radial depth of cut.

$$F = MRR^{0.63}e^{-0.44 - 7.78 \cdot \text{S} \cdot 10^{-5}} \tag{5}$$

This model relates the force module (*F*) to the machining parameters through the *MRR*, and, as such, the concept of process productivity is introduced. This approach makes it easier to select higher efficient parameters. Both experimental and predicted data are shown in Figure 8. The reduction of forces at high spindle speeds for the same *MRR* is remarkable. This fact can be explained by a working regime close to high speed machining, as was referred to in previous paragraphs. The increase of the process temperature reduced the effort needed to cut the material [33]. Additionally, the heat generated in the process was quickly evacuated, which could have reduced induced residual stresses [34] and tool wear [33].

**Figure 8.** Average force module predicted (dashed lines) and experimental data against the material removal rate (*MRR*) as a function of the cutting speed (*Vc*).

The force predicted by the proposed model, as a function of the cutting parameters, allowed the machining operation to be monitored using the electrical power registered by the ServoGuide System. Machining cutting power involved all the cutting parameters together, thus giving a closer idea of the overall interactions in the cutting process [35]. This monitoring option could be used as an input for adaptive control systems, in which the instant depth of cut can be controlled and modified online, as an alternative to others online depth of cut control based on ultrasound measurements [36]. This can reduce the number of overcuts and defective parts produced. In this case, the empirical model followed Equation (4), where *e* is the Euler number and *D* the tool diameter.

$$Power = \left(\frac{F \cdot \mathcal{S} \cdot D \cdot \pi}{60}\right)^{3.5} a\_p^{-2.2} e^{-16.7} \tag{6}$$

This equation was empirically obtained following an ANOVA approach, using the data represented on Figure 9. Combined relations between the different variables were neglected during the analysis. The *R*<sup>2</sup> for the final statistical model was 0.953.

**Figure 9.** Model correlation between the mechanical power, calculated based on the recorded forces, and the electrical power obtained on the ServoGuide system as a function of the *ap*.

#### **4. Conclusions**

The machining of aluminium skin panels is used as a sustainable alternative for chemical milling process in the aerospace industry; therefore, it requires tight quality tolerances. This work presents a study of how cutting parameters influence the final thickness, surface roughness and cutting forces of thin plate aluminium parts in order to pursue two main objectives: To ensure the final quality of the part and to find an easy way to monitor the process.

The influence of the cutting speed on the final thickness error map of the machined thin plates has been proven. Higher values of cutting speeds tended to reduce the standard deviation of thickness error values measured in the test samples. The higher the cutting speed, the lower the cutting force module and the higher its excitation frequency, leading to an increase of process stability and a reduction of the results variability. Consequently, an improvement by up to a 40% of the implicit process tolerances has been achieved using a cutting speed of 566 m/min. This fact suggests that these parts could present even more homogenous results in terms of final thickness if higher cutting speeds were used. Roughness values are always under the more restrictive requirements for chemical milling. Lower values of cutting forces under stable machining conditions could ensure roughness values under 0.3 μm.

Furthermore, forces are affected by the low rigidity of the part that obtain lower average values for the *z* axis than those expected, based on rigid body experiments. A statistical analysis of the tests also revealed high cutting speed parameters as the more efficient ones based on *MRR,* providing a force model that includes all cutting parameter effects.

Finally, this model was used to relate mechanical power and electrical power consumption, allowing us to control online the depth of cut. This model is proposed as an alternative method to implement adaptive control techniques in other to avoid overcuts on aeronautical panels, reducing defective parts at the final stages of the process chain when their value is very high.

**Author Contributions:** Conceptualization, I.D.S., A.R. and A.J.G.; methodology, I.D.S. and A.R.; formal analysis, I.D.S. and A.J.G.; investigation, I.D.S. and A.R.; resources, A.R.; writing—original draft preparation, I.D.S.; writing–review and editing, A.R. and A.J.G.

**Funding:** This research was funded by University of Cadiz, grant number University training plan UCA/REC01VI/2016.

**Acknowledgments:** The authors acknowledge the support given by the Fraunhofer Joint Laboratory of Excellence on Advanced Production Technology (Fh–J\_LEAPT Naples).

**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* **Influence of Elastomer Layers in the Quality of Aluminum Parts on Finishing Operations**

#### **Antonio Rubio-Mateos 1,\*, Asuncion Rivero 1, Eneko Ukar <sup>2</sup> and Aitzol Lamikiz <sup>2</sup>**


Received: 31 January 2020; Accepted: 20 February 2020; Published: 22 February 2020

**Abstract:** In finishing processes, the quality of aluminum parts is mostly influenced by static and dynamic phenomena. Different solutions have been studied toward a stable milling process attainment. However, the improvements obtained with the tuning of process parameters are limited by the system stiffness and external dampers devices interfere with the machining process. To deal with this challenge, this work analyzes the suitability of elastomer layers as passive damping elements directly located under the part to be machined. Thus, exploiting the sealing properties of nitrile butadiene rubber (NBR), a suitable flexible vacuum fixture is developed, enabling a proper implementation in the manufacturing process. Two different compounds are characterized under axial compression and under finishing operations. The compression tests present the effect of the feed rate and the strain accumulative effect in the fixture compressive behavior. Despite the higher strain variability of the softer rubber, different milling process parameters, such as the tool feed rate, can lead to a similar compressive behavior of the fixture regardless the elastomer hardness. On the other hand, the characterization of these flexible fixtures is completed over AA2024 floor milling of rigid parts and compared with the use of a rigid part clamping. These results show that, as the cutting speed and the feed rate increases, due to the strain evolution of the rubber, the part quality obtained tend to equalize between the flexible and the rigid clamping of the workpiece. Due to the versatility of the NBR for clamping different part geometries without new fixture redesigns, this leads to a competitive advantage of these flexible solutions against the classic rigid vacuum fixtures. Finally, a model to predict the grooving forces with a bull-nose end mill regardless of the stiffness of the part support is proposed and validated for the working range.

**Keywords:** vibrations; part quality; flexible vacuum fixture; AA2024 floor milling

#### **1. Introduction**

Monolithic aluminum components are widely employed in the aeronautical sector due to their good strength-to-weight ratio [1]. The final quality of these parts is normally obtained or improved in the finishing operation and it is influenced by static and dynamic phenomena [2].

On the one hand, from the static point of view, cutting forces and part clamping produce elastic deformation that can lead to deteriorating the final dimension and the surface of the workpiece [3]. On the other hand, vibrations increase the roughness of the parts. These dynamic instabilities become frequent in the milling operation and are produced by the lack of dynamic stiffness in one or more components of the system [4]. The most characteristic vibrations appeared in the milling operation are the self-excited vibrations or chatter [5,6]. However, even in the absence of chatter, it almost always exists forced vibrations derived from the periodic excitation of the intermittent cutting engagement of the milling cutter on the workpiece [7].

Thus, in order to improve the part quality different approach has been studied. First, the tuning of the cutting parameters can lead to static and dynamic improvements. Thus, a process forces reduction leads to a decrease of the elastic deformation of the workpiece that it is reflected in the part accuracy. For instance, Perez et al. [8] obtained a machining forces reduction and an improvement of the compressive residual stresses with the increase of the cutting speed. On the other hand, in terms of vibrations, different surveys have been developed for the improvement of the system dynamic stability and the obtained surface quality. Different stability models have been developed for the milling operations of compliant systems [9,10] and the effect of the cutting parameters on the process damping have been analyzed [11]. However, these solutions are limited for the inherent stiffness of the system.

The key element for the increase of the system stiffness is the workpiece clamping. Thus, different part-fixture systems have been employed to guarantee a suitable part positioning and fixing [12]. Due·to the lack of dynamic stability of some of these solutions, different damping features have been implemented in the system in order to improve the machining process. Thus, different active features based on the use of eddy currents [13,14], pivot mechanism [15,16] and piezoelectric dampers [17] have been studied. These solutions are cost efficient and their implementation is limited to certain applications.

The use of passive damping elements is increasing for milling operations as they are more cost efficient compared to the active developments. The passive damping systems are based in the implementation of different elements or fluids with outstanding damping properties to stabilize dynamically the system. Thus, by employing electrorheological [18] or magnetorheological [19] fluids, the vibration amplitude of the cutting processes varies and the part quality is improved. Moreover, in order to increase the narrow vibration band of these passive dampers, Yang et al. [20] developed a tunable passive devices. However, the industrial implementation of these passive solutions are challenging as they interfere with the clamping of the workpiece and the machining process.

In the present study, the use of an elastomer layer employed as a passive damping element is proposed and characterized. Elastomers, particularly rubber materials, are ideal materials for vibrations isolation as they are low in cost with high internal friction [21]. Moreover, the industrial implementation of these compounds for machining applications is feasible as they are employed as passive control of vibration [22–24]. In fact, the damping properties of these elastomers have been analyzed for low frequency [25] and high frequency applications [26], including under certain machining operations. For instance, Kolloru et al. [27] employed neoprene layers combined with torsion springs to reduce up to eight times the vibration in the milling process of circular thin-wall components. On the other hand, Liu et al. [28] implemented a viscoelastic material in the toolholder to increase by 99% its damping ratio. Nevertheless, there is no study of the direct application of rubber materials as the clamping element of workpieces in milling operations.

In this case, in order to combine a fixture and a passive damping system the use of a nitrile butadiene rubber (NBR) layer is proposed. This sort of elastomers is one of the most employed seal component in the oil and gas industry [29], and the proposed development benefitted from these outstanding sealing properties [30,31] to transform a flexible layer into a suitable vacuum table. Hence, these solutions enable milling in aggressive environments, with capacity to clamp different geometries. Moreover, as the passive damping element act as a fixture, its industrial implementation is feasible as the interference with the rest of the machining system is reduced.

In order to characterize the behavior of these elastic polymers under the machining processes loads, compression tests and milling tests have been performed. Thus, these flexible solutions have been characterized in terms of chatter and forced vibrations performed by the milling tool.

Regardless the milling strategy, the most aggressive machining zone is the entrance of each pocketing where the tool machines with an axial pitch equal to its diameter. Thus, the analysis is focused in the grooving application with depth of cuts defined by finishing operations. The suitability survey is performed in terms of part quality. First, the machined depth is measured to quantify the groove thickness error. Then, the floor roughness is measured and analyzed. Finally, the dynamic behavior of each system is characterized, and a universal force model is developed for the grooving operation in finishing applications.

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

Two different NBR layers were selected for the analysis as these passive elements are defined by different vibration bands. The mechanical properties of both vulcanized rubber materials are shown on Table 1. Besides the hardness and the density, the compounds ingredients are given, where the carbon black is a form of elemental carbon that is used to increase the resistance of rubber and also to improve the tensile strength [32].


**Table 1.** Materials mechanical properties.

The mechanical behavior of rubber depends on the amplitude, feed rate and frequency of the applied load, combined with the temperature of the material [33]. In the case of milling operations, the amplitude and feed rate of the applied forces are completely defined by the machining conditions.

Similarly, the load frequencies suffered by the part are generated by the milling tool rotation and by the workpiece fundamental modes. Finally, the temperature of the material is influenced by the heat generated on the cutting zone and the room temperature.

Based on the load application strategies employed on this survey, some simplifications were considered. For instance, the decrease in stiffness during the first few cyclic loads, the so-called Mullins effect [34], was neglected. Therefore, different loads prior to each test were performed over each elastomer layer. The characteristics of these loads were defined in terms of the test to perform. Thus, for compression tests, a compression load was performed prior to each test. Likewise, prior to each milling test, a previous groove was performed to reduce the Mullins effect on the rubber and to level the upper side of each slot.

Finally, due to the reduced compression loads during the machining operation and the wide part area in contact with the elastomer layer, the expected strain amplitudes are minimal. Therefore, it is not considered a rubber heat up due to material damping derived from large harmonic loads [33]. Hence, due to the part thickness located between the cutting zone and the rubber layer, the temperature of the rubber was considered as the room value.

Tests were performed in a standard 5-axis numeric control (NC) center. The selected geometry for the elastomer layers was a 300 <sup>×</sup> 300 mm2. The mean value of the thickness for both cases was 14.2 mm with a tolerance of ±5%. In order to guarantee a uniform contact and clamping conditions between the part and the elastic material, a slot grid was machined in each rubber layer (Figure 1a). Thus, the vacuum clamping force was distributed along the contact area by means of the channels. Then, the air was removed through a unique orifice and the part could be safely clamped during the machining operation, as shown in Figure 1b.

**Figure 1.** Adapted rubber: (**a**) vacuum channels distribution. (**b**) Rubber layer implementation as a vacuum fixture on milling tests.

Both, the compression of the rubber and the part profile before and after each milling tests were monitored with a GT1000 type linear variable differential transformer (LVDT) gauging transducer (RDP Group, Wolverhampton, UK). The forces were registered with Kistler 9257B measurement equipment (Kistler Ibérica S.L., Granollers, Spain). In each test, following another similar set-up [35], the part and the elastic element were attached to the force sensor with a synthetic rubber adhesive. This double-sided filmic tape TESA 64620 (Tesa Tape S.A., Argentona, Spain) guaranteed a homogeneous clamping due to the compressive nature of the axial loads in compression and milling tests.

#### *2.1. Compression Tests*

For most of engineering rubbers, material damping is caused by two different mechanism, resulting in amplitude and rate dependent behavior [36]. Thus, the objective of these compression tests is characterizing the effect of the feed rate on the strain and comparing the influence of the strain cycles on each rubber. In general, compression tests on rubber materials are performed with circular samples [37]. However, in order to include the effect of the slots in the material deformation, the tests are implemented in the same elastic layer employed as vacuum fixture, see Figure 2.

**Figure 2.** Compression test procedure: (**a**) set-up scheme. (**b**) Load application zone.

The axial loads were applied by the machine head by means of a cylindrical punch and its position was monitored with a LVDT. The feed rates conditions were selected based on the most extreme cases tested in the milling tests. Three repetitions were performed for each condition and, in order to evaluate the strain accumulative effect, a 15 min relaxation period was guaranteed between the successive tests.

#### *2.2. Milling Tests*

For the milling survey, the same adapted rubber layers were employed. The part samples to be machined were 20 mm thick AA2024-T3 rigid blocks. These parts were 240 <sup>×</sup> 240 mm2 wide and were located in the center of the elastic support. Hence, any rubber edge effect could be neglected, and its local thickness tolerance was diminished from ±5% to ±3%. In order to reduce the vacuum leaks a <sup>290</sup> <sup>×</sup> <sup>290</sup> <sup>×</sup> 0.7 mm<sup>3</sup> sacrificial porous layer was included between the elastic element and the specimen to be machined. Hence, the vacuum leaks depended on the part area and it was not influenced by the part contour. Thus, different part geometries could be clamped without any fixture redesign.

The air from the channels was removed through the hole with a standard Venturi guaranteeing a proper vacuum union between the rubber and the aluminum part sample for all the working range, see Figure 3. Then, the rubber was held to the dynamometric table with the double-sided filmic tape. In the tests with no rubber, the part sample was stuck directly to the Kistler by the same token.

**Figure 3.** Milling test set-up with and without a rubber layer.

Groove milling was the selected machining operation. These slots were dry machined side to side, in two steps. First, a previous 0.2 mm groove was performed in order to guarantee the same initial profile between tests. Then, the test with each condition was milled. The separation between each groove was 10 mm.

The selected tool was a two flutes bull-nose end-mill Kendu 4400 (Kendu, Segura, Spain), with a diameter of 10 mm and a 2.5 mm edge radius. Table 2 shows the tests conditions. Therefore, the effect of the depth of cut (*ap*), feed per tooth (*fz*), spindle speed (*n*) or cutting speed (*vc*) and the tool feed rate (*f*) on both elastic polymers could be studied and compared with the use of a rigid clamping. Two different depth of cuts were selected based on the values employed in finishing operations in the aeronautic field [38]. On the other hand, three different tool rotation values were studied in order to reduce vibrations generated out of the tool-part system while two feed per tooth values were selected for maintaining a suitable milling process of aluminum parts [9,38]. Hence, based on the feed per tooth and spindle speed configurations, six different milling conditions were analyzed for each depth of cut.



For each milling condition, three repetitions were performed in different random positions relative to the center of the part. The analyzed zone was restricted to each groove middle area. Then, the real machined thickness was evaluated by measuring the part profile beforehand and afterward each milling operation. This measure was performed with the previously presented LVDT attached to the machine tool head with an adaptor. Besides, the roughness of the floor of each slot was measured in four different zones equally separated by 20 mm. Thus, the *Ra* value was evaluated with a Mitutoyo Surftest SV-2000 (Mitutoyo, Kawasaki, Japan) roughness measure station. As a reference, the typical tolerance values in the aeronautical industry were about ±0.1 mm for final thickness and under 1.6 μm for the *Ra* [39].

#### *2.3. Force Mechanistic Model in the Tool Axis Direction with a Bull-Nose Mill*

In order to evaluate the stability for each set-up, stability lobe diagrams (SLD) were calculated in the specimen center with an impact hammer test. A uniaxial PCB accelerometer model 352C22 (PCB Piezotronics, Inc, Depew, NY, USA) with a measuring range from 1 kHz to 10 kHz and a sensitivity of 1.0 mV/(m/s2) was employed to register the tool and part vibration. The maximum acceptable *ap* in the stable regime is calculated with the model described by Altintas and Budak [10]. The cutting forces (tangential (*t*), radial (*r*) and axial (*a*)) over the cutting edge *i* could be considered as a function of the friction coefficients (*Kte*, *Kre* and *Kae*) and the shearing cutting coefficients (*Ktc*, *Krc* and *Kac*)

$$
\begin{Bmatrix} \partial F\_t \\ \partial F\_r \\ \partial F\_d \end{Bmatrix} = \begin{Bmatrix} K\_{tc} \\ K\_{tc} \\ K\_{ac} \end{Bmatrix} \times \partial S + \begin{Bmatrix} K\_{tc} \\ K\_{tc} \\ K\_{ac} \end{Bmatrix} \times f\_z \times \sin \phi\_l \times \partial z \tag{1}
$$

In this equation, ∂*S* is the length of the differential chip edge, φ*<sup>i</sup>* is the angular position of the cutting edge *i* measured from axis *Y*, perpendicular to the tool feed direction, and ∂*z* is the depth of cut.

Compared with other tool geometries, bull-nose end mills have a variable radius and helix angle along the tool axis. Likewise, the lead angle increases its value from 0◦ to 90◦ in the toroidal part, and then kept constant and equal to 90◦ all along the flank [40].

This geometrical variation combined with cutting speed and the depth of cut leads to variable cutting coefficients. This nonlinearity could be solved using a linear model to calculate the SLD [9]. However, Altintas [41], simplified a circular insert geometry taking an average edge angle of 45◦.

In this case, the model was oriented to the floor finishing application. For these cases machined depths were usually focused in a range between 0.2 mm and 1.2 mm, mainly in low stiffness parts. This means that the edge angle was located between 11◦ and 29◦. Thus, for this survey, the average edge angle was defined as 20◦.

The friction and shearing cutting coefficients were obtained by solving the equation by using the force values obtained in the milling tests. These coefficients were considered constant for all the milling conditions. The model results were employed to predict the SLD for each flexible fixture and compared with the use of a rigid clamping underneath the part sample. Moreover, with the axial forces obtained in these tests, a model was proposed regardless of the hardness of the support.

#### **3. Results**

#### *3.1. Rubbers Compressive Behaviour*

The differences in the composition of each tested rubber led to a completely different stress-strain behavior. In the Figure 4 it can be observed the strain variation of each rubber based on the stress and feed rate evolution. This evolution is presented with a fifth-degree interpolation in order to visualize the more linear behavior of rubber B compared with rubber A.

**Figure 4.** Accumulative stress and feed rate increase effects on both rubber materials: (**a**) rubber A and (**b**) rubber B.

Both elastomers increased their elastic modulus as the feed rate rose. The rate effect was usually attributed to the resistance in reorganizing the polymeric chains during the loading period. Since this reorganization cannot occur instantaneously, the loss of energy is rate dependent [33]. Moreover, the polymeric chains in the rubber A lacked time in the relaxing period to return to the original state and, thus, the elastic modulus decreased for the second and third trial. This effect was not noticeable in the rubber B. However, it can be observed that, as the feed rates increased, the behavior of both elastomers tended to match for stresses under 0.4 MPa, which was within the work range for the milling tests. Hence, regardless of the rubber composition, these flexible fixtures presented stress-strain behavior influenced by the load amplitude and feed rate transmitted by the tool and by the previous deformation implemented on the rubber.

#### *3.2. Thickness Error*

The thickness error is defined as the difference between the experimental and theoretical thickness. In addition to the static and dynamic phenomena that occurred when applying loads over the elastic layers, other effects such as the machine precision, repeatability and the thermal expansion of the spindle had an influence over the real machined thickness. For instance, the repeatability for the rubber A was within ±9 μm, for the rubber B was within ±19 μm and for the use of no rubber was within ±8 μm.

In order to analyze this parameter, an analysis of variance (ANOVA) was employed. Thus, the influence of the main machining parameters in the thickness error was evaluated. Therefore, first, the normal distribution of the data was checked by the Anderson-Darling (AD) test, and the variance homogeneity with the Bartlett's test. In both cases, the confidence interval of 95% (α = 0.05) was employed. As it can be observed in the Table 3, for all the tests their *p*-values were over α and, thus, were suitable for an ANOVA.


**Table 3.** Analysis of the suitability of the thickness error data.

On the other hand, a variance analysis was performed to determine the main parameters affecting the machined depth inaccuracy. In this case, the null hypothesis was that the factors or their combination have no influence over the thickness error. As it is detailed in the Table 4, from this survey it was obtained that, with a 95% confidence, the null hypothesis was proved to be true. The only exception was the effect of spindle speed in the case of the rubber A as its *p*-value was under α, see Table 4 values in bold.


**Table 4.** Analysis of variance of the thickness error data.

This result was coherent with the compression tests, as the rubber A was the most sensible to strain changes. Furthermore, as it can be observed in the Figure 5, there was a global decrease in the thickness mean error as the *vc* increased. In this case, the positives values meant that the system was compressed, and the depth of cut was lower than programmed. This effect, as expected, was more noticeable with the use of rubber as a support. In the other hand, if the thickness error had a negative value, it meant that the tool machined more depth than expected. This last effect was mainly caused by the thermal expansion of the spindle, as it increased combined with the revolutions [42]. This error can be compensated previous to the machining [43] or even with in-process tool position adjustments [44,45].

**Figure 5.** Thickness error evolution based on the *vc* variation.

Another noticeable effect was that, as the cutting speed increased, the thickness errors tended to equalize. This effect matched with the fact that, due to the cutting conditions tested, as the cutting speed rose, the feed rates increased accordingly. Then, as it is observed in the compression analysis, the elastic modulus of the rubbers rose as the feed rates increased, leading to a more rigid-like support. This is aligned with the industrial implementation of this system in milling operations of aluminum, as the productivity of these applications tend to the employment of these or even higher cutting speeds.

Finally, there was a thickness error component that was caused by the system vibration and that produced the difference in the variability of each system. This vibration was analyzed in terms of roughness in the next section.

#### *3.3. Roughness*

In order to analyze the effect of the machining parameters on the floor *Ra* on the slots, another ANOVA was performed. Likewise in the thickness error analysis, the data suitability was analyzed with an AD and a Bartlett's test. As shown in the Table 5, it was demonstrated that data met, with 95% of confidence, the requirements for a valid ANOVA.


**Table 5.** Analysis of the suitability of the roughness data.

With this data, a variance analysis was replicated based on the effect of the machining parameter on the roughness obtained on the groove floor. As it is can be observed in the Table 6 in bold, compared with the thickness error, more parameters and their combination affected the part vibration. This influence was more noticeable in the rubber A, as it happened with the thickness error, due to a higher sensibility to strain variations.


**Table 6.** Analysis of variance of the thickness error data**.**

Despite the dependence on the machining parameters of the roughness, there was no direct influence of a single parameter into the behavior of the three systems at a time. However, it was clear that the most influential parameters were the spindle speed and the feed rate. Thus, in the Figure 6, it can be observed the evolution of the roughness with the increase of the cutting speed and the feed per tooth. In this case the repeatability for the rubber A was within ±0.13 μm, for the rubber B was within ±0.17 μm and for the use of no rubber was within ±0.11 μm.

Results show that roughness obtained with rubber A tended to match the one obtained with the part robustly clamped to the machine as the cutting speed increased. This effect, as explained in the case of the thickness error, was caused by the increase of the elastic modulus. However, this increase in the cutting speed has to be balanced with the feed rate in order to maintain the feed per tooth.

**Figure 6.** Roughness evolution: (**a**) based on the cutting speed and (**b**) based on the feed per tooth.

On the other hand, rubber B tests suffered higher roughness and wider variability. As the stiffness of rubber B was above the rubber A's, the instability must be caused by the vacuum union between the part and the rubber. Thus, due to the higher hardness of the rubber B, the contact with the part did not perform proper clamping conditions as the rubber A.

This analysis indicates that cutting loads applied by the machining tool did not affect exclusively the rubber compression but the clamping suitability as well. Despite rubber A having more variable compression behavior, its lower hardness improved the fixture clamping capacity and the obtained part quality.

#### *3.4. Force Model*

The SLD performed over the three systems, as shown in the Figure 7, presented the identical behavior of them. The reason was the combination of a hammer shot at a high feed rate and a wide supporting area of the rigid part. Then, as observed in the compression tests, these fixtures based on elastomers, at high feed rates behaved as a rigid system in terms of chatter vibrations. Thus, these results proved that there was no chatter on the performed milling tests as the maximum depths of the cut were below these curves.

**Figure 7.** Stability Lobe Diagrams (SLD) variation: (**a**) complete and (**b**) zoomed on the studied zone.

The analysis of the force harmonics, see Figure 8, confirmed that the vibration was mainly influenced by the tool cutting loads. The case of rubber A and no rubber had similar behavior, with lower amplitudes and with the cutting per tooth as the main driver of the vibration. However, the

forced vibrations at different harmonics were higher in rubber B. Once again, this evidence confirmed that the union between rubber B and the part was not completely suitable.

**Figure 8.** Fast Fourier Transform of the *Fz* signal for the test at *fz* = 0.1 mm/tooth and *ap* = 0.8 mm, under different spindle speeds: (**a**) No rubber - 2000 rpm, (**b**) No rubber - 4000 rpm, (**c**) No rubber - 6000 rpm, (**d**) Rubber A - 2000 rpm, (**e**) Rubber A - 4000 rpm, (**f**) Rubber A - 6000 rpm, (**h**) Rubber B - 2000 rpm, (**i**) Rubber B - 4000 rpm and (**j**) Rubber B - 6000 rpm.

Then, the main cutting forces in the axial direction were studied following an empirical approach in order to provide a suitable model able to relate them with the machining parameters regardless the part support. The repeatability for all the tests was within ±2 N. The process parameters were grouped around the material removal rate (MRR), where *N* is the number of teeth, and *ar* is the radial depth of cut:

$$MRR = n \times f\_z \times a\_p \times N \times a\_r \tag{2}$$

The model is based on a potential regression, see Equation (3). The *R*<sup>2</sup> of this model is 0.984. This equation emphasized, once again, the strong influence of the cutting speed on the machining process

$$F\_z = 67.22 \times n^{-0.58} \times MRR^{0.49} \tag{3}$$

Figure 9 presents how the model fits with the analyzed data. As it can be observed, the main forces could be modeled regardless of the part support. López de Lacalle et al. [46] noticed that the cutting forces decreased due to the reduction of stiffness. However, by using a rubber underneath a high stiffness part sample, the system flexibility can be considered not compromised as the cutting forces are maintained.

**Figure 9.** Axial mean loads predicted (lines) and experimental data against the material removal rate (MRR) as a function of the cutting speed (*vc*).

Finally, this model demonstrated that the process axial forces in finishing of the aluminum parts did not depend on the material hardness or the accumulative strain state of the rubber. This facilitates the implementation of these flexible fixtures in the industry and provides a calculation tool for the improvement of the milling process productivity.

#### **4. Conclusions**

In this paper the effect of clamping high stiffness aluminum part samples over elastomer layers was analyzed. The machining application was groove milling, simulating finishing conditions in the aeronautic field. First, by a compression test the influence of stress amplitude, feed rate and cycles were examined. Thus, the rise of the elastic modulus as the strain rates increased and the strong dependence of the stress cycles were proved, especially for the soft rubber.

Then, the effect of cutting speed, tool feed and depth of the cut were analyzed in terms of the machined thickness error, roughness and axial forces. The results show that, as the cutting speed increases combined with the feed rate, the rubbers tended to behave like a rigid support, guaranteeing the thickness and roughness tolerances required in certain aeronautic applications. Moreover, as on these applications high speed machining operations were performed with higher cutting speeds and feed rates, the results of this solutions could improve compared to the actual rigid clamping solutions. In terms of hardness, the softer rubber tended to provide more stable machining conditions due to a better clamping capacity.

Finally, an axial force model was developed and validated regardless the support stiffness and accumulative strain. This could lead to facilitate the implementation and improve the productivity of this solution into certain industrial applications, including the milling of aeronautical aluminum parts.

**Author Contributions:** Conceptualization, A.R.-M., A.R. and A.L.; methodology, A.R.-M., E.U. and A.L.; software, A.R.-M.; validation A.R.-M. and E.U.; formal analysis, A.R.-M.; investigation A.R.-M.; resources, A.R. and A.L.; data curation, A.R.-M.; writing—original draft preparation, A.R.-M.; writing—review and editing, A.R.-M. and E.U.; visualization, A.R.-M.; supervision, A.R., E.U. and A.L.; project administration, A.R. and A.L; funding acquisition, A.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Basque Government (Eusko Jaurlaritza) under the ELKARTEK Program, SMAR3NAK project, grant number KK-2019/00051.

**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*

## **Feasibility Study of Hole Repair and Maintenance Operations by Dry Drilling of Magnesium Alloy UNS M11917 for Aeronautical Components**

#### **Fernando Berzosa 1, Beatriz de Agustina 1, Eva María Rubio 1,\* and J. Paulo Davim <sup>2</sup>**


Received: 26 May 2019; Accepted: 28 June 2019; Published: 30 June 2019

**Abstract:** Magnesium alloys are increasingly used due to the reduction of weight and pollutants that can be obtained, especially in the aeronautical, aerospace, and automotive sectors. In maintenance and repair tasks, it is common to carry out re-drilling processes, which must comply with the established quality requirements and be performed following the required safety and environmental standards. Currently, there is still a lack of knowledge of the machining of these alloys, especially with regards to drilling operations. The present article studies the influence of different cutting parameters on the surface quality obtained by drilling during repair and/or maintaining operations. For this propose, an experimental design was established that allows for the optimization of resources, using the average roughness (*Ra*) as the response variable, and it was analyzed through the analysis of variance (ANOVA). The results were within the margins of variation of the factors considered: the combination of factor levels that keep the *Ra* within the established margin, those that allow for the minimization of roughness, and those that allow for the reduction of machining time. In this sense, these operations were carried out in the most efficient way.

**Keywords:** magnesium alloy; UNS M11917; AZ91D; hole repair; surface roughness; dry drilling; re-drilling

#### **1. Introduction**

Recently, the need to reduce energy consumption as well as environmental pollution has been highlighted, especially in the transport sector, which includes industries such as the aeronautical, aerospace, and automotive industries. This need has led to a constant search for the reduction of the weight of components by using lighter materials, which allow for mass reduction and, therefore, lower consumption of fuel and polluting emissions. In this context, there has been increasing interest in extending the use of materials such as magnesium, which has excellent specific mechanical properties and whose full potential has not yet been reached, due in part to the insufficient knowledge about magnesium compared to other materials such steel and aluminum [1–9].

The main advantages of magnesium alloys are their low density, high availability, high recyclability, and good properties for foundry and machining, such as high specific strength and good weldability under a controlled atmosphere. Nevertheless, there are also a few disadvantages such as low creep resistance above 100 ◦C, low resistance to corrosion, hardness, and they are difficult to form at room temperature [10,11]. Magnesium's high chemical reactivity is another drawback that is closely related to problems during machining [12].

Magnesium alloys are mainly formed by casting, of which about 70% is processed by casting in permanent molds, producing near net shape parts. After that, machining operations are necessary in most cases [13,14]. Magnesium is considered to have excellent machinability. This is due to its low specific cutting strength, low tool wear, excellent surface quality, short and brittle chips due to its hexagonal crystal structure, and high thermal conductivity, which maintains low temperature increases even using dry machining, allowing for high cutting speeds and feed rates [15,16]. As a result, all common machining operations such as turning, milling, drilling, threading, reaming, or grinding can be performed with these alloys without major problems. The fundamental difference between magnesium and other structural materials is the ability to use higher feed rates and depths of cuts for magnesium, to give low roughness and closed tolerances [17].

The published literature shows that research on magnesium alloy machining focuses on cutting speed, feed rates, depth of cut, precision, and quality of the machining surfaces, also on the formation of adhesions, mainly build-up edge (BUE) and build-up layer (BUL) [18]. In turning and milling processes, researchers pay attention mainly to cutting forces, surface roughness, tool materials, tool wear, lubricant-cooling systems, temperature, chip morphology, and hardness. The cutting conditions for turning used in previous experimental works were as follows: cutting speed from 75 to 2400 m/min; feed rate from 0.05 to 0.65 mm/rev; and depth of cut from 0.2 to 5 mm. For milling, the parameter values were the same order of magnitude [1,10,19–22].

In the aeronautical industry, the drilling process is key due to the high number of joints by riveting, threaded joints, and mechanical fasteners made in the whole vehicle. In fact, the operation that consumes the most time during the assembly of a plane is the pre-assembly operation in the fuselage. An important cause of problems in the structural integrity of the fuselage is the growth of cracks in the drilled holes. For this reason, effective hole drilling is fundamentally important. In the case of commercial aircraft, the number of drilled holes can reach up to 3 million. Twist drills are used for most metals, using High Speed Steels (HSS) for aluminum and magnesium alloys [5].

In most studies on the drilling of magnesium alloys, the cutting speeds were around 50 m/min and the feed rates ranged between 0.1 and 0.7 mm/rev. In these studies, the influence of machining conditions on variables such as surface quality, force, torque, and tool wear, among others, was studied. The majority of studies used average roughness (*Ra*) to evaluate the surface quality of the obtained surfaces [3,23–27].

Weinert et al. [28] carried out a study on magnesium drilling using wide cutting parameters, reaching cutting speeds up to 1100 m/min and extending feed rates to 1.2 mm/rev. In that study they found that the surface quality, quantified by the maximum height of the profile, *Rz*, remained approximately constant by varying the cutting speed between 100 and 1100 m/min, while keeping the feed rate constant at 0.2 mm/rev; increasing the feed increased the roughness. In addition, the mechanical load on the tool did not vary significantly in the range of cutting speeds from 100 to 700 m/min, being determined by feed rate.

Other studies focused on machining parameters that are not high performance, using average roughness (*Ra*) as a variable to quantify the surface quality [3,23,26]. There are potential risks in the machining of magnesium alloys; on the one hand, there is the danger of ignition when the chips reach temperatures of 450 ◦C, and on the other hand, with the use of water-based lubrication there is danger of a reaction between water and magnesium, which forms a hydrogen atmosphere that is flammable [20,29]. Considering these reasons, it was decided to carry out the present study using dry machining.

As discussed above, there are still gaps in our knowledge about magnesium alloys. There are not many scientific works that discuss the problems during solid drilling of these alloys, and we found only one work specifically about re-drilling or core drilling operations in magnesium alloys: Rubio et al. [30] studied this process, but for hybrid Mg-Ti-Mg components. This type of machining is used in the repair process of damaged holes, which is common in the aeronautical sector where the holes are machined to a larger diameter to insert oversized rivets. These repairs must be carried out with great care to avoid damage to the machined parts [30]. These types of operations can be framed as low performance operations since productivity is a secondary objective.

In machining plants, drilling operations have traditionally been carried out in two steps: first drilling and then enlarging the diameter of the holes. These operations are executed with a solid base of knowledge of the materials and operations. However, in maintenance and/or repair operations, this is not always the case, especially considering that occasionally a smaller increase in the diameter of the hole is sought in order to not weaken the pieces. In this aspect, there is a certain lack of understanding and, therefore, such drilling operations can be improved to increase the safety and quality of the holes.

The aim of this work is to analyze the feasibility of carrying out repair and maintenance operations on pre-drilled parts used in the aeronautical industry. To do this, a pre-drilled test piece was used to simulate the repair of housings in covers that are joined by elements such as rivets. This joint type is widely used in aeronautics and can be the origin of fatigue cracks, which can lead to catastrophic failures in the pieces if they are not repaired in time.

This paper presents the analysis of the surface roughness, in terms of *Ra*, obtained by drilling holes to a slightly larger diameter in magnesium alloys UNS M11917 (AZ91D) at low cutting parameters. In this way, the behavior of these alloys in maintenance and/or repair operations was studied. The use of twist drills in these operations has certain advantages compared to reamers, which have less availability in the machining sections, generally have a higher price, and have a smaller variety in terms of the machined final diameters. The final aim is to establish if it is feasible to carry out such operations under environmentally sustainable conditions, maintaining the surface roughness requirements within a range of values established for the aeronautical industry, that is, from 0.8 to 1.6 μm [31,32].

To achieve this goal, the experimental design was established taking into consideration the three most important factors, feed rate, cutting speed, and type of tool, at two and three levels according to the recommendations of the manufacturers and prior knowledge of drilling operations. In addition, the small variations of the diameters to be drilled and the depth of the holes where the roughness measurements were to be taken were considered factors in the experimental design. Blocks were considered for quantification, and if obtained surface roughness was constant along the machined surface, a replicate was performed in order to quantify the error. The statistical method used to study the results obtained was the analysis of variance (ANOVA).

The tests were carried out in two stages by machining a pre-drill and then re-drilling, maintaining a constant depth of 0.125 mm. The final diameters were drilled to 7 and 7.5 mm, in the first and second stages, respectively. The last stage served firstly to corroborate the data obtained initially and secondly to check if small differences in the diameter affected to the variables studied.

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

The UNS M11917 magnesium alloy is produced by the die casting method and was supplied as an ingot with a length of approximately 500 mm and a section of 118 × 60 mm. A rectangular parallel-piped block was milled using low machining parameters so as not to raise the temperature of the piece, until reaching the measurements of 110 × 62 × 50 mm, maintaining surface roughness below 2 μm. This alloy has a chemical composition of mass of 90% Mg, 8.30–9.70% Al, 0.35–1% Zn, ≥0.13% Mn, ≤0.1% Si, ≤0.03% Cu, ≤0.005% Fe, and ≤0.002% Ni and presents a microstructure formed by an α-phase matrix and an intermetallic β-phase whose composition is Mg17Al12 and is located at the boundaries of the grains [33]. The main mechanical properties of this alloy are shown in Table 1.


The block of magnesium was positioned in the hydraulic jaw of the machining center, aligning it so that the upper face was parallel to the plane of the machine table. To do this, a 3D tester (Haimer GmbH, Igenhausen, Hollenbach, Germany) was used. To avoid bias, pre-drills and drills were carried out without moving the specimen of the clamping jaw. The clamping of drills was done by a collet ER25 suitable for the drill diameter.

Two types of tool were used for the performance of the tests. They were both manufactured by Phantom (Van Ommen B.V., Beekbergen, Gelderland, Netherlands) in HSS, and are called 11.130 (type A), and 11.160 (type B), Figure 1.

**Figure 1.** Twist drills for the tests; (**a**) Tool type A; (**b**) Tool type B.

These drills are suitable for drilling depths up to three times the diameter. They are manufactured according to DIN 1897 [34], with a straight shank and two flutes of 34 mm. They have a helix shape normal type N according to DIN 1836 [35], and both drill points are sharpened using split form C point in accordance with DIN 1412 [36].

Drilling tests were carried out using a computer numerical control (CNC) controlled vertical machining center Lagun L650 (Lagun Machinery S.L.L., Legutio, Álava, Spain) under dry conditions. The cutting parameters were selected based on solid drilling operations, taking into consideration the values given by the manufacturer for the group of non-ferrous materials and those used in the previous published studies. Keeping in mind that the present work was focused on repair and/or maintenance, we selected the following test values: cutting speed (*S*): 60 and 120 m/min; and feed rate (*f*): 0.2, 0.4, and 0.8 mm/rev; the cutting depth was kept constant at 0.125 mm. All blind holes were drilled to a depth of 20 mm from the top face of the specimen.

In the first stage, drilling tests were performed to enlarge holes from a diameter of 6.75 mm to a diameter of 7 mm, for all the combination of cutting conditions pointed out above, whereupon the machined surface roughness was measured. In a second stage, drilling tests were carried out under the same combination of cutting conditions, in this case from a diameter of 7.25 mm to a diameter 7.50 mm, in order to also evaluate the influence of the diameter of the drill on the surface roughness.

Between each drilling operation, the upper surface of the specimen and its surroundings were cleaned using a brush and pressurized air. Before carrying out the drilling, the periphery of the block was covered with paper, with the purpose of collecting samples of the fragile chips produced in the machining, as shown in the Figure 2. Once the process was finished, the hole and the used drill were marked with a number, and a photographic record of the obtained chips was taken using a Nikon Coolpix P510 digital camera (Nikon, Tokyo, Japan).

**Figure 2.** Detail of the method for collecting chips.

The arithmetical mean deviation of the assessed profile, *Ra*, was used as a response variable to quantify the surface roughness of the machined surface, which according to the standard ISO 4287:1997 [37] is defined as the "arithmetical mean of the absolute ordinate values *Z*(*x*) within a sampling length". The range, that a priori would be expected to be the value of the measured *Ra*, should be between 0.1 and 2 μm according to ISO 4288:1996 [38], which also established the sampling length (*lr*) at 0.8 mm and the evaluation length (*ln*) at 4 mm. Subsequently, after the measurements of the roughness were made, these assumptions were confirmed.

In each of the drilled holes, the roughness was measured on eight different lines. That is, measurements of the *Ra* were taken along four lines equi-angularly separated by 90◦ in two cylindrical zones at different distances from the upper surface of the specimen. The first one, named the top plane (TP), was at a distance of 5.5 mm from top face and the second one, named the bottom plane (BP), was at a distance of 15 mm from top face, as can been seen in Figure 3. The measurement length along each one of the eight lines was of 4 mm.

**Figure 3.** Cylindrical sections inside the holes where the measurements of roughness were taken for the evaluation of average roughness (*Ra*) (striped in red) and, in each one of them, the specific zones to take the measurements are defined by the four points separated to 90◦ and marked as 1, 2, 3, and 4.

*Ra* was measured using a contact roughness surface tester Zeiss Handysurf E-35A (Carl Zeiss AG, Oberkochen, Baden-Wurtemberg, Germany). This model has the possibility of exporting the measured data to a computer or displaying it directly on a display panel; it includes several types of parameters of roughness, among them the *Ra*. To carry out the measurements, the roughness meter and the test block were placed on a surface plate. The probe of the roughness tester is portable, so to perform the eight measurements in each hole, the probe was placed in a tool coupled to a height gauge, allowing for vertical positioning at the desired height, then rotation of the specimen four times to measure the equi-angularly located points, and then positioning of the probe tip at the two established depths. Figure 4 shows the surface tester, probe tip, and positioning system.

**Figure 4.** Zeiss Handysurf E-35A roughness tester, positioning system, and probe tip.

Based on all of this, the experimental design was determined, and its objective was to determine the influence of the factors considered in the response variable, that is, the surface roughness studied by the *Ra*. The experimental design selected was a full factorial design with three factors at two levels, one factor at three levels, and a block at two levels, which is the measurement depth from the upper face of the block, including the performance of a replica, which supposes a total of 96 experimental runs. The considered factors and their levels are included in Table 2.


Before carrying out the statistical analysis, the assumption of normality was checked; an Anderson–Darling and Kolmogorov-Smirnov tests data were carried out, which was not overcome, hence a Johnson transformation was carried out. Once the normal data was obtained, the next step was the statistical analysis in order to study the influential factors and interactions on the surface quality measured by *Ra*. This analysis was performed by the ANOVA.

#### **3. Results and Analysis**

#### *3.1. Results*

The experiment consisted of carrying out 96 experiments and, for each of them, measuring the roughness on four separate zones at 90◦ in two cylindrical sections located inside the holes as can be seen in the striped red zones in the Figure 3. To study the measured roughness, it was evaluated considering the average of the four values of *Ra* measured in each one of the two cylindrical sections located at the different distances from the top face of the specimen in each experimental run. Before carrying out the statistical analysis, the assumption of normality of the average *Ra* was checked. To do this, normality was assessed using the Anderson–Darling and Kolmogorov–Smirnov tests. The results obtained in both tests are shown in Table 3, and indicate the non-normality of the data and therefore the need to carry out its transformation prior to the statistical study.


**Table 3.** The *p*-values of tests for normality of *Ra*.

Johnson transformation was used to convert the original non-normal data into a standard normal distribution. The best adjustment of the transformation was obtained using the equality function collected in Equation (1), where the transformed roughness values are designated by *Ra<sup>t</sup>* , being the *Ra* of the average roughness initial values. The probability plot for the population before and after the transformation can be seen in Figure 5.

$$Ra^t = -1.05518 + 1.01828 \times \sinh^{-1} \left( \frac{Ra - 0.562553}{0.128499} \right) \tag{1}$$

**Figure 5.** Probability plots of original non-normal data and Johnson transformed data of the *Ra*.

Once the measured roughness data were normalized, it was possible to analyze them using the ANOVA statistical method. For this, the arithmetic mean of the four measured values of the *Ra* in each elementary run was made. The results obtained in the experimental design are those included in Table 4.


**Table 4.** Replicas of original and transformed data of *Ra* in μm, at different measurement depths.

*\*S* = cutting speed; *f* = feed rate; *T* = type of tool; *D* = diameter; *Ra<sup>t</sup>* = transformed *Ra*.

#### *3.2. Analysis and Discussion*

From the data obtained, it is clear that the obtained surface quality achieved improved results, independent of the parameters tested. All the roughness values were between 0.38 and 2.49 μm. Only five of the 96 *Ra* values were above the 1.6 μm that is set as the upper limit in the aeronautical sector. In consideration of the 16%-rule established in the ISO 4288:1996 standard [38], the surface is considered acceptable because only 5.2% exceeded the upper limit.

For the statistical analysis of the transformed *Ra* values, the Minitab 17 computer program was used. The model was reduced to include only the significant factors, for which a stepwise procedure was followed that eliminates or adds terms to the model using a significance level α = 0.05, starting with an empty model and then adding or removing a term for each step. Table 5 shows the influential factors and interactions, in other words, cutting speed, feed rate, type of tool, diameter, sum of squares, degrees of freedom (DF), mean square, F value, and *p*-value.

**Table 5.** Analysis ANOVA of the transformed *Ra.* DF = degrees of freedom.


The analysis showed that among the main factors, only the cutting speed, *S*, and the type of tool, *T*, were statistically significant for *Ra*. Regarding the interactions of the factors, there were only two

significant second-order interactions: the cutting speed with the diameter, *S\*D,* and the feed rate with the diameter, *f\*D*. The contribution of each effect to the variability is shown in Figure 6. The most important effect was the type of tool with a percentage of 45.25%, the second most important was the interaction of cutting speed with the diameter of the drill with 36.06%, the third was the interaction between feed rate and diameter with 10.18% and the last was the cutting speed with 6.98%.

**Figure 6.** Percentage of contribution to the variability of the ANOVA model for each effect.

In Figure 7, it can be seen that the use of a twist drill type B enhances the surface quality of the machined surfaces with respect to those gained from the use of type A. From the untransformed values obtained in the tests, tool A obtained a *Ra* of 0.97 μm, while type B was 0.67 μm, which represents a considerable improvement. An explanation of this behavior could be the appearance of radial stresses that would produce deformation of the drill and therefore affect the roughness, this would be caused by the small depth of cut of only 0.125 μm in conjunction with the use of a non-high rigidity drill and relatively high feed rates and cutting speed values. Regarding the cutting speed, it is the factor that has the least influence, producing a higher *Ra* with increasing speed. This result is concordant with those obtained by Weinert et al. for the case of solid drilling [13].

**Figure 7.** Effects on the transformed *Ra* of the significant factors.

The second effect in terms of importance is the interaction between the cutting speed and the diameter of the drill, *S\*D*, although the diameter itself is not a significant effect. Figure 8 shows that increasing the cutting speed in drills with a diameter of 7 mm causes a lower superficial roughness in contrast to drills with a 7.5 mm diameter.

**Figure 8.** Box and whiskers plot graph for interactions between cutting speed and diameter, *S\*D*.

The influence of the feed rate on the *Ra* was the opposite and this is explained by the third most important effect: the interaction between feed rate and the diameter of the drill, *f\*D*, as can be seen in the Figure 9. This can be seen most clearly for feed rates of 0.4 and 0.8 mm/rev. Use of 7 mm tools at higher feed rates caused lower *Ra* values; in contrast, for 7.5 mm tools, the lower *Ra* values were obtained with lower feed rates.

**Figure 9.** Interaction graph between feed rate and diameter, *f\*D*.

The results regarding these two interactions of the cutting speed and the feed rate with the diameter of the drill on magnesium alloys are remarkable because the difference between the diameters tested was so small. Further studies should be carried out to clarify this point, considering greater values in the drill diameters studied.

*Metals* **2019**, *9*, 740

A model of surface roughness was developed to predict the variability in the transformed data, *Ra<sup>t</sup>* , through Equation (2). This equation uses the significant factors identified by the ANOVA, in other words, cutting speed, type of tool, feed rate, and diameter of the drill, where *s*, *t*, *f*, and *d* represent their effects, μ is the term to adjust the mean, and ε is the error. The estimation parameters of the equation are included in Table 6.

$$Ra\_{ijkl}^t = \left. \mu + s\_i + t\_j + sd\_{i\mathbf{k}} + fd\_{\mathbf{lk}} + \left. \varepsilon\_{ij\mathbf{kl}} \right|\right.\tag{2}$$


**Table 6.** Estimation parameters of the predictive model.

The residuals of the model were obtained by the difference between measured and predicted values and were used for checking the model hypotheses. As can be seen in Figure 10, the residuals satisfy the normality and homoscedasticity hypothesis; also, no patterns were found in the model.

**Figure 10.** (**a**) Probability plot and (**b**) residuals versus predicted values for the model.

Once the validity of the model was verified, the inverse of the Johnson transformation used in Equation (1), of the fitted values according to Equation (2), was carried out to predict the surface quality by *Ra* in re-drilling operations of UNS M11917 magnesium alloys. For this, Equation (3) was used.

$$Ra\_{ijkl} = \ 0.562553 + 0.128499 \times \sinh\left(\frac{\mu + s\_i + t\_j + sd\_{ik} + fd\_{lk} + \varepsilon\_{ijkl} + 1.05518}{1.01828}\right) \tag{3}$$

From Equation (3) and considering both the significant factors identified in the ANOVA, as well as their levels, it was possible to calculate the predicted values of *Ra* for the different combinations. These values are shown in Table 7 along with the *Ra* obtained from the values measured in the tests, and the absolute error between the predicted and measured *Ra*.


**Table 7.** Predicted *Ra*, measured *Ra*, and absolute error for levels and effects combinations.

The surface roughness values obtained by the model predicted values between 0.52 and 1.23 μm. The absolute error between the measured and predicted values was less than 0.28 μm in all cases except one, in which the error reached 0.88 μm. The minimum *Ra* value was obtained with the combination of a cutting speed of 60 m/min, type of tool B, feed rate of 0.4 mm/rev, and by using a drill with a 7.5 mm diameter.

#### **4. Technological Discussion**

For the framework in which the present study was set up, that of enlarging holes with a low depth of cut by drilling in magnesium alloy UNS M11917, it is important to highlight some technological aspects with implications to the practical application of these operations, mainly in the aerospace sector.

Considering all the tests carried out and according the 16%-rule established in the ISO 4288:1996 standard [38], it can be affirmed that within the margins of the levels and factors tested, the surface quality would be within the quality requirements established in this sector (0.8 μm < *Ra* < 1.6 μm) [31,32]. Of all the 96 measured *Ra* values, only five of them were above the upper limit, and those five were drilled using the type A drill. From this it follows, in addition to be the main significant factor, the importance of the type of drill in the performance of these operations.

It is of great importance to carry out these maintenance and/or repair operations in the shortest time possible, meeting the quality standards required for the parts, so it is important to optimize the rate of material removal (RMM). According to Astakhov [39], this rate is directly proportional to the product of the feed rate by the cutting speed, *f\*S*. Therefore, a way of increasing productivity and improving process time is to select the highest feed rate and cutting speed values, that is, 0.8 mm/rev and 120 m/min, respectively. Using these operating parameters, the average *Ra* values of 0.6 μm for the type B tool and 0.9 μm for type A were obtained.

The type of tool is the most important factor in terms of its influence on the *Ra*. The Anderson–Darling test shows clearly that for the case of type B drills, the population follows a normal distribution. However, for the case of the type A drill, with a value of *p* < 0.005, the Anderson–Darling test confirms that it follows a non-normal distribution with a strong asymmetry and positive kurtosis, as seen in Figure 11 with its displacement to the right. An explanation of this phenomenon could be in the appearance of greater radial efforts that, due to the low rigidity of the drills, give rise to deformations that affect the surface quality. Subsequent studies must to be carried out to confirm this hypothesis.

**Figure 11.** Histogram of data with an overlaid normal curve of the original *Ra* for tool types A and B.

#### **5. Conclusions**

This experimental study on the small-scale re-drilling operations in magnesium alloy UNS M11917 within the maintenance and/or repair processes of pieces in the aerospace sector, confirms that it is possible to perform such operations in a way that satisfies the requirements of the surface quality and safety and, at the same time, under environmentally friendly conditions, that is, using dry machining or without the use of lubricant coolants. The most important factor to consider is the type of tool used, obtaining the best results with type B drills, which have a point angle of 135◦, compared to type A drills, which have a point angle of 118◦. However, further studies have to corroborate if, besides the point angle, other variables of the drill, such as the type of coating, affect the surface quality. Using that type of drill and choosing the highest values in the cutting parameters, that is, a feed rate of 0.8 mm/rev and a cutting speed of 120 m/min, a surface roughness was obtained of approximately half of the maximum limit considered within this sector, in other words, 0.8 μm.

The depth from the upper surface of the specimens did not present a statistical influence on the roughness, so it can be considered constant throughout the depth of the drilling holes, which in this study was 20 mm. Contrary to what was initially assumed, feed rate did not have an influence on surface roughness. This result is consistent with some other work on solid drilling operations in similar magnesium alloys, however, there are other works that show contrary data. Feed rate does have a second-order influence in its interaction with the diameter of the drill, but its influence value is small in relation to the other significant factors.

**Author Contributions:** Conceptualization, E.M.R., B.d.A., F.B., and J.P.D.; Methodology, E.M.R., B.d.A., F.B., and J.P.D.; Software, E.M.R., B.d.A., and F.B.; Validation, E.M.R., B.d.A., F.B., and J.P.D.; Formal analysis, E.M.R., B.d.A., and F.B.; Investigation, E.M.R., B.d.A., F.B., and J.P.D.; Resources, E.M.R., B.d.A., F.B., and J.P.D.; Data curation, E.M.R., B.d.A., and F.B.; Writing—original draft preparation, F.B.; Writing—review and editing, E.M.R., B.d.A., F.B., and J.P.D.; Visualization, E.M.R., B.d.A., and F.B.; Supervision, E.M.R., B.d.A., and J.P.D.; Project administration, E.M.R. and B.d.A.; Funding acquisition, E.M.R., B.d.A., and F.B.

**Funding:** This research was funded by the Ministry of Economy and Competitiveness and the Industrial Engineering School-UNED (DPI2014-58007-R; Ref: 2019-ICF03 and Ref: 2019-ICF05).

**Acknowledgments:** The authors thank the Research Group of the UNED "Industrial Production and Manufacturing Engineering (IPME)" for the support given during the development of this work and to the Grupo Antolín for donating the materials used in this work.

**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* **Parametric Analysis of Macro-Geometrical Deviations in Dry Turning of UNS A97075 (Al-Zn) Alloy**

#### **Sergio Martín Béjar, Francisco Javier Trujillo Vilches \*, Carolina Bermudo Gamboa and Lorenzo Sevilla Hurtado**

Civil, Material and Manufacturing Engineering Department, EII, University of Malaga, 29071 Malaga, Spain; smartinb@uma.es (S.M.B.); bgamboa@uma.es (C.B.G.); lsevilla@uma.es (L.S.H.) **\*** Correspondence: trujillov@uma.es; Tel.: +34-951-953-245

Received: 28 September 2019; Accepted: 23 October 2019; Published: 24 October 2019

**Abstract:** Macro-geometrical deviations play a very important role in the functionality and reliability of structural parts for aircraft. The use of environmentally friendly techniques, such as dry machining, may negatively affect these deviations. Despite its importance, there is a lack of research that analyzes them as a function of the cutting parameters in the case of aluminum alloys for aeronautical purpose. In this work, the cutting speed and feed influence on several macro-geometrical deviations (parallelism, straightness, circular run-out, roundness, concentricity, total circular run-out and cylindricity) in dry turning of UNS A97075 alloy was analyzed. The main novelty of this work lies in the use of high slenderness parts used in further fatigue tests. The results showed that feed seems to be the most influential parameter in most of the deviations studied. In addition, the parts with lower rigidity exhibited higher sensitivity to change with the cutting parameters. Finally, different parametric models were proposed to obtain the geometrical deviations as a function of the cutting parameters.

**Keywords:** UNS A97075; dry turning; surface integrity; straightness; parallelism; roundness; concentricity; circular run-out; total run-out; cylindricity

#### **1. Introduction**

One of the most appreciated quality requirements in machining processes is related to the surface integrity (SI) concept [1]. Field et al. [2,3] defined the SI as the inherent or enhanced condition of a surface produced in machining operations or other surface generation processes. Griffiths [4] defined it as the topological, mechanical, chemical and metallurgical worth of a manufactured surface and its relationship with its functional performance. A new definition of this concept is developed by Astakhov [5], who defined it as a set of properties (both, superficial and in-depth) of an engineering surface that affect its service behavior. These properties include geometrical, physical-chemical and biological parameters.

In more recent works, Gómez-Parra et al [6,7] provided an expanded view of the SI concept, defining the SI as a set of properties that the material surface exhibits, acquires or becomes modified during a forming process. These properties can be analyzed from three points of view connected to each other: micro-geometrical (surface roughness, micro and macro cracks, waviness, particle adhesion), macro-geometrical (cylindricity, concentricity, straightness) and physical-chemical properties (micro hardness, residual stress, stress corrosion, tensile strength, fatigue behavior). The authors highlight that these properties may not only improve the functional performance of the part, but also worsen it.

The macro-geometrical deviations or material mechanical properties inclusion within the SI definition presents great controversy. On one hand, the macro-geometrical deviations and the mechanical properties affect not only the part surface, but also the bulk. On the other hand, the mechanical properties are included within the surface physical-chemical properties. The surface

is considered as a whole, and not only from the micro-geometrical and physical-chemical approach at a point on the surface. In this way, the macro-geometrical deviations may affect negatively the mechanical behavior or increase the appearance of micro-geometrical or physical-chemical defects, and vice versa. Thereby, these three points of view are interconnected and may not be considered in isolation. Regardless of their inclusion or not within the SI definition, the macro-geometrical deviations play a fundamental role in mechanical properties [8,9], such as fatigue behavior [10]. In that way, fatigue test standards are very demanding regarding the tolerances of these geometrical deviations [11].

Fatigue behavior is one of the most important properties to take into account in the behavior of aircraft structural parts in service [12]. The quality requirements of these components are highly demanding because they are placed in critical areas. Consequently, geometrical tolerances (at macro and micro scale) are usually very narrow in order to make their assembly easier and to improve their functionality, reliability and longevity [13,14]. However, these high-quality demands result in higher costs. Hence, one of the most important challenges is to balance the manufacturing process performance of these components, from four different points of view: functional, economic, environmental and energetic [15–17].

Light alloys (mainly Al and Ti alloys) are widely used in the manufacturing of structural parts for aircrafts, individually or combined with composites (such as carbon fiber reinforced polymers, CFRP) to form fiber metal laminates structures. Specifically, aluminum alloys series 2000 (Al-Cu) and 7000 (Al-Zn) are used in the components under fatigue load in service, such as the pressurized cabins fuselage, ribs, spars and wings upper/lower skins [18–20]. Machining (mainly milling, drilling and turning) is frequently used to manufacture these structures [21,22]. The current trend in machining of these alloys moves towards reducing or eliminating the use of cutting fluids (dry machining), due to environmental and occupational health reasons [23–25]. The machining process based on the minimum quantity of cooling lubrication (MQCL) or the minimum quantity of lubrication (MQL) are good alternatives [26]. However, the performance balance of dry machining is currently a challenge. On one hand, the CFRP/Al structures show a bad behavior under wet machining conditions and the mixture of aluminum chip, CFRP and cutting fluids is complex and expensive to recycle [27]. On the other hand, dry machining results in very aggressive cutting conditions, which gives rise to a fast temperature increase in the cutting area and fast tool wear [28]. This fact may negatively affect the surface integrity of the machined parts. Hence, the environmental component of the process performance is improved, whereas the economical and functional components may be reduced.

Within this context, the cutting parameters (cutting speed, feed and cutting depth) play a very important role [29]. A large amount of research can be found in the literature analyzing the cutting parameters influence on the SI micro-geometrical aspects of dry machined wrought aluminum alloys [28,30–33] taking in to account the influence of the tool wear, chip geometry or axial machining length. Usually, the mean average roughness (*Ra*) is the selected parameter to evaluate these deviations. Most of the research agree that feed is the most influential cutting parameter on *Ra*, showing *Ra* a general trend to increase with feed [34–36]. Some of these works develop parametric models that allow predicting the *Ra* evolution as a function of the cutting parameters. Usually, these models adopt a potential form [35–38] due to the simplicity of the model, where the exponent of each variable represent the influence in the general term, and to the good fit that the models usually show. In addition, some authors show a relationship between *Ra* and the cutting forces [8,39]. Therefore, an online monitoring of micro-geometrical deviations was carried out to analyze this relationship.

Nevertheless, there is a lack of research focused on the analysis of the cutting parameters influence on macro-geometrical deviations for these alloys, despite their importance and influence on the functional behavior of these parts [9]. Clares et al. [7] proposed an experimental methodology to evaluate the SI in dry turning of aerospace alloys from the three aforementioned points of view (geometrical, at micro and macro scale and physical-chemical features). Sánchez-sola et al. [40] studied the cutting speed (*v*c = 40–170 m/min) and feed (*f* = 0.05–0.30 mm/r) influence on straightness, parallelism and roundness of the UNS A92024 alloy. The cylindrical bars (200 mm length, 80–120 mm

diameter) were dry turned. The worst results were obtained when the highest *f* and lowest *v*c values were used. A general trend to increase the straightness and parallelism with *f* was found for low *v*c. An opposite trend was found for high *v*c, with a strong dispersion in the results for the highest *f* values. Regarding the roundness, its value tends to increase with *f*, regardless *v*c. No clear trend was found with *v*c. On the one hand, Sánchez-sola et al. explain that straightness and parallelism are measured along the whole part and, therefore, they are more influenced by the roughness profile, chip generation, built-up edge (BUE) detaching and, as a result, by the cutting parameters. On the other hand, roundness is measured from the transversal section. Hence, it is less influenced by these parameters. In addition, exponential parametric models were obtained. These models showed a good fit for high *f* and low *v*c.

Trujillo et al. [9] analyzed the cutting parameters influence (*v*<sup>c</sup> = 40–200 m/min; *f* = 0.05–0.20 mm/r) on roundness, circular run-out, straightness and parallelism of dry turned UNS 97075 alloy cylindrical bars (150 mm length, 30–60 mm diameter). The cutting depth (*a*p) remained constant (1 mm). These geometrical deviations showed low sensitivity to change with the cutting parameters, unlike what was observed for micro-geometrical deviations, strongly influenced by *f*. However, *v*<sup>c</sup> exhibited higher influence on straightness and parallelism, whereas *f* showed higher influence on roundness and circular run-out. These results can be explained in a similar way that were exposed in [40]. The exponential parametric models were also developed for each macro-geometrical deviation. Parallelism and straightness models exhibited a good fit for low *v*c. However, the roundness and circular run-out showed a good fit for all *v*c tested. Finally, Trujillo et al. highlighted that these models should be tested under different conditions (cutting parameters range, specimens' geometry) to check their generality.

It is necessary to point out that these previous studies have been carried out on low slenderness parts. Nevertheless, the structural parts for the aircraft usually show high slenderness rates [41]. Therefore, extended research on slender parts should be developed in order to analyze the cutting parameters' influence on the macro-geometrical deviations and their influence on mechanical properties, such as fatigue behavior. Given the aforementioned, the main novelty of the study, this work focusses on the analysis of the cutting speed and feed influence on several geometrical deviations (straightness, parallelism, circular run-out, roundness, concentricity and cylindricity) of dry turned UNS A97075 alloy. For this purpose, the specimens with high slenderness were used. These specimens were designed to be used in further fatigue tests. Finally, the different experimental parametric models were developed. These models allow predicting some geometrical deviations as a function of the cutting parameters within the studied range.

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

Several dry turning tests were carried out on UNS A97075-T6 alloy specimens in order to evaluate the cutting parameters' influence on different macro-geometrical deviations. This material is widely used in the manufacturing of aeronautical structural parts that work under compressive and fatigue loads [42]. Arc atomic emission spectroscopy (AES) was used to obtain the tested alloy composition (% mass). The results are shown in Table 1.

**Table 1.** Tested alloy composition (% mass).


The final specimens' geometry obtained from cylindrical bars (*D* = 20 mm) was selected according to the rotating bar bending fatigue test standard, ISO 1143:2010 [11]. Among the different geometries proposed in this standard, the cylindrical smooth geometry was selected. On the one hand, this geometry is less rigid that those used in previous research [9,40], being this one of the main novelties of this work. On the other hand, these specimens are being used in current works regarding the cutting parameters influence on fatigue behavior. The specimens' shape and geometrical dimensions are shown in Figure 1. In addition, it is necessary to highlight that the standard quality requirements regarding the cylindricity and concentricity are strongly demanding. Therefore, the specimen slenderness was taken into account and calculated as the relation between the lengths (*L* = 167 mm) and the lower diameter (*d* = 7.6 mm), being this value, 22.37.

**Figure 1.** Specimens geometrical dimensions (mm).

The machining tests were conducted in a Computer Numerical Control (CNC) turning center. The different combinations of cutting speed (*v*c) and feed (*f*) were used (Table 2). The cutting depth (*a*p) remained constant. Every test was performed under dry conditions, in order to use environmentally friendly techniques. It must be pointed that although low cutting speeds are not recommended for machining aluminum alloys, these alloys are often hybridized with other materials in which these low cutting speeds are required, such as fiber metal laminates, FML (CFRP +Al + Ti). In addition, this fact allows the comparison with previous studies on the geometrical deviations performed in the same cutting parameters range [9,40]. Every test was performed under dry conditions in order to use environmentally friendly techniques.

**Table 2.** Cutting parameters.


The tool (Figure 2) was an uncoated WC-Co insert (ISO DCMT 11T308-14 IC20) and each test was carried out using a new tool to ensure identical initial conditions. The cutting angles setup can be observed in Table 3. The tool geometry and position respect the specimen allow the machining of the entire specimen profile with a single operation. Additionally, these geometrical cutting conditions reduce the deflection effect on slenderness parts due to the predominance of the axial cutting forces over the radial cutting forces. Due to the cutting tool geometry (a major cutting-edge angle of 66.5◦), the force axial component is dominant over the radial component. Therefore, the bending effect on sample was limited.

**Figure 2.** Specimen positioning and turning operation tool.


**Table 3.** Cutting angles setup.

Once the specimens were machined, different macro-geometrical deviations were controlled in the calibrated area (parallelism, straightness, circular run-out, roundness, concentricity, total run out and cylindricity). The experimental setup is shown in Figure 3. A millesimal dial gauge with a measuring span of 12.5 mm, scale division of 0.001 mm and a maximum permissible error (MPE) of 4 μm, was used to control these deviations. This device was placed on the tool carriage to avoid removing the specimen from the turning center and achieve a faster process. Previously, the setup rigidity and the run-out of the spindle were controlled in order to assess their contribution to the part run-out. This contribution was found to be negligible. In addition, some of the specimens were off-line measured in a geometrical deviation measurement machine (Figure 4) in order to validate the experimental setup. The differences found did not exceed 10%.

**Figure 3.** Geometrical control setup on the specimen calibrated area.

**Figure 4.** Form measuring system.

The circular run-out (CRO) was measured along six sections (S1 to S6, separated 4 mm from each other) in the calibrated area (Figure 5a). For each section, twelve measurements were performed at 30◦ each. The parallelism (PAR) was controlled along twelve generatrix (G1–G12), separated 30◦ from each other, Figure 5b.

**Figure 5.** (**a**) The roundness (RON), circular run-out (CRO) and concentricity (CON) measured sections; (**b**) straightness (STR) and parallelism (PAR) measured generatrix.

The circular run-out (CRO) was obtained as the difference between the maximum and the minimum profile radius (*R*max − *R*min), as in Figure 6a [42]. The roundness (RON) and concentricity (CON) were calculated from the CRO experimental results (Figure 6a). Among the different mathematical methods available, the least squares circles method [43] was applied in this work. To evaluate the least square circumference center, a nonlinear iterative mathematical model was considered, minimizing the function error (Equation (1)) and taking the seed as the rotation center.

$$\text{SSE}(a, b) = \sum\_{i=1}^{n} \left( R - \sqrt{(\mathbf{x}\_i - a)^2 - (y\_i - b)^2} \right)^2 \tag{1}$$

where different variables corresponding with:


**Figure 6.** Macrogeometrical deviations: (**a**) RON, CON and CRO, (**b**) total circular run-out (TCRO), (**c**) STR and PAR and (**d**) cylindricity (CYL).

Once the center of the least square circumference has been calculated, *RON* is obtained as the difference between the radiuses of two concentric circumferences (*J*max − *J*min) which delimit the area containing all points of the profile, as in Figure 6a. The CON is the distance between the rotation center (0, 0) and the least squares circumference center previously obtained (a, b) (Equation (2)).

$$\text{CON} = \sqrt{a^2 + b^2} \tag{2}$$

The parallelism (PAR) was obtained as the distance between two parallel lines (*X*max − *X*min), which delimit the area containing all the profile points and are parallel to the work piece axis (Figure 6c). The straightness (STR) was calculated as the distance between two parallel lines (*D*max − *D*min) which delimit the area containing all the profile points and are parallel to the least squares regression line (Figure 6c).

Taking into account the total calibrated area volume, the total circular run-out (TCRO) was calculated as the difference between the maximum and minimum of every radio sections measured (*T*max − *T*min) in the calibrated area (Figure 6b). Finally, the cylindricity (CYL) was obtained as the difference between two co-axial cylinders, such that their radial difference is at a minimum (*P*max − *P*min), as in Figure 6d.

The theorical PAR respect CYL was taken in to account. Considering the tool tip radius (*r* = 0.8 mm), the depth of cut (*a*p = 1 mm) and the worst feed rate condition (*f* = 0.20 mm/r), the scallop calculated is 6.27 μm. The standard 1143:2010 allows a maximum of 20 μm. Therefore, the scallop value for each feed rate implemented in this work is considered within the standard values.

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

#### *3.1. Parallelism and Straightness*

Figure 7 shows STR and PAR experimental results as a function of *v*c and *f*. In Figure 7a, the PAR shows a general trend to slightly increase at low *f* (0.05–0.10 mm/r), regardless *v*c. From *f* = 0.10 mm/r to 0.20 mm/r, PAR tends to decrease, whereas it remains more or less constant for *v*c = 60 and 80 m/min. Notwithstanding this, a higher dispersion can be observed. From *f* = 0.15 to 0.20 mm/r, PAR remains constant for *v*<sup>c</sup> = 40 and 60 m/min. Nevertheless, PAR tends to increase for *v*<sup>c</sup> = 80 m/min for this *f* range. In fact, the worst results (and the highest dispersion) are obtained when the highest *f* and *v*<sup>c</sup> are combined.

**Figure 7.** (**a**) Parallelism and (**b**) straightness deviation as a function of cutting speed (*v*c) and feed (*f*).

Figure 7b shows that STR presents a higher dependence on the cutting parameters studied than PAR. On one hand, the highest values are always obtained for *v*c = 80 m/min, regardless *f*. In addition, a general trend to increase STR with *f* is observed, mainly from *f* = 0.05 to 0.10 mm/r and from *f* = 0.15 to 0.20 mm/r. On the other hand, for the low range of *f* studied (0.05 to 0.10 mm/r), STR tends to increase with *f*, regardless *v*c. Nevertheless, in general terms, *v*c seems to be the most influent parameter. Moreover, its effect increased when it is combined with the highest *v*<sup>c</sup> (80 m/min). In addition, these results show higher deviations than other research with a less slender workpiece [9,40].

These results can be explained taking into account how PAR and STR are measured along the machining length. This fact results in a higher dependence of those geometrical deviations on the built-up edge (BUE) formation and detaching vibrations and/or deflections of the specimen. Figure 8 shows the tool rake face for *v*<sup>c</sup> = 40 m/min, for *f* 0.05 and 0.10 mm/r. These images show that the indirect adhesion wear phenomenon is higher for 0.10 mm/r. A similar trend was found for *v*<sup>c</sup> = 60 and 80 m/min. Therefore, STR is more sensitive to the BUE formation at low *f*. For higher *f* values (0.15–0.20 mm/r), STR becomes less sensitive to BUE (with similar intensity at high *f* values) and more sensitive to vibrations, more noticeable for *v*<sup>c</sup> = 80 m/min (Figure 9).

**Figure 8.** Stereoscopic optical microscopy (SOM) of the tool rake face (40×) after the tests performed for (**a**) *f* = 0.05 mm/r and (**b**) *f* = 0.10 mm/r, for *v*c = 40 m/min.

**Figure 9.** Stereoscopic optical microscopy (SOM) of the tool rake face (40×) after the tests performed for (**a**) *f* = 0.15 mm/r and (**b**) *f* = 0.20 mm/r, for *v*c = 80 m/min.

These results have revealed two important differences with previous research on similar alloys, in which more rigid specimens were used [9,40]. First, both PAR and STR have shown higher values. As a result, these macro-geometrical deviations tend to increase with the specimen slenderness. In addition, previous work did not reveal a clear influence of the cutting parameters on PAR and STR. However, only *v*<sup>c</sup> seemed to be slightly more influential at high values. However, in this work, PAR and STR behavior was more sensitive with *f*, and its effect was maximized at a high cutting speed.

#### *3.2. Circular Run-out, Roundness and Concentricity*

Figure 10 shows CRO experimental results as a function of the cutting parameters for each section. With the different sections (1–6) in mind, the specimen slenderness and the experimental setup rigidity play a special role. Further sections from the chuck (1–3) have exhibited higher CRO deviations, whereas closer sections (4–6) have shown lower values. This general trend is more noticeable for *v*<sup>c</sup> = 40 and 60 m/min, regardless of *f*, whereas it is less evident for *v*<sup>c</sup> = 80 m/min. Therefore, the specimen slenderness and rigidity are more influential at low *v*c. Nevertheless, vibrations become more relevant at high *v*c.

**Figure 10.** *Cont*.

**Figure 10.** Circular run-out deviations for (**a**) *v*c = 40 m/min, (**b**) *v*c = 60 m/min and (**c**) *v*c = 80 m/min.

In addition, Figure 11 plots the average CRO (taking all sections into account) as a function of *f* and *v*c. In general, *f* seems to be the most influential parameter. The circular run-out (CRO) exhibits a general trend to increase with *f*, regardless *v*c. This fact becomes more noticeable at high cutting speed values. The general trend is less clear with regard to *v*c. Between *f* = 0.05 and 0.10 mm/r, the highest CRO value is obtained for *v*<sup>c</sup> = 40 and 60 m/min, respectively. However, for *f* = 0.20 mm/r, the worst result is found for *v*<sup>c</sup> = 80 m/min. As a result, the combination of the highest *f* and *v*<sup>c</sup> values result in the highest CRO deviations. The high vibration levels, produced when the high values of cutting speed and feed are used, should explain this behavior.

**Figure 11.** Circular run-out mean values as a function of cutting speed (*v*c) and feed (*f*).

The RON and CON deviations can be calculated from CRO experimental results and Equations (1) and (2).

Figure 12 plots the RON experimental results in the function of the cutting parameters, for each section. A general trend to increase RON deviations in the function of *f* can be observed, regardless of *v*c. However, a strong increment in the function of *v*c can be noticed for *f* = 0.20 mm/r. Trivial variations of RON in the function of the section relativity position can be observed, except for *v*<sup>c</sup> = 80 m/min and *f* = 0.20 mm/r, where severe cutting conditions show an increase on the central sections (2–5). Taking into account that feed is the most influenced parameter in cutting forces, regardless of *v*c [31,44], this fact can be explained due to an increment of deflections, especially in a slender specimen, where the cutting forces are increased in the higher range of *f*. Notably, the CON evolution for the different sections measured can be considered as a function or the results shown for CRO (Figure 10) and RON (Figure 12).

**Figure 12.** Roundness deviations for (**a**) *v*c = 40 m/min, (**b**) *v*c = 60 m/min and (**c**) *v*c = 80 m/min.

Figure 13 plots the RON and CON mean values (considering all the sections) as a function of *f* and *vc*. A general trend to increase RON mean values with *f* is observed, regardless of *v*c, as in Figure 13a. This increment is more evident from *f* = 0.15 to 0.20 mm/r and softer for *v*c = 40 m/min. Regarding *v*c, the general trend is less clear. Only between *f* = 0.15 and 0.20 mm/r there is a trend to increase *RD* with *v*c. The worst results are obtained when the highest values of *f* (0.20 mm/r) and *v*<sup>c</sup> (60 and 80 m/min) are tested. Hence, the trend is very similar as that observed for CRO. Nevertheless, the cutting parameters influence is more evident on RON than on CRO. This fact is a consequence of deleting the concentricity effect, which shows a less clear trend with the cutting parameters, as in Figure 13b. The CON average values (Figure 13b) show a general trend to increase with *f*, but softer than RON. Regarding *v*c, the CON values are more scattered, showing different behavior at a low and high feed.

**Figure 13.** Mean deviations values in function of *v*c and *f* for (**a**) roughness and (**b**) concentricity.

Considering that CRO is related with RON and CON, it is necessary to point out that CON takes more importance in CRO for low *v*<sup>c</sup> and *f* values, whereas RON influence is more noticeable for higher values.

The found trend is slightly different from those obtained in previous research for UNS A97075 and UNS A92024 alloys [9,40]. CRO and RON sensitivity to change with the cutting parameters was very low (in specimens with higher rigidity). Nevertheless, this sensitivity increases when the specimen geometry is less rigid, especially with *f*. In addition, the deviations values are significantly higher for high slenderness specimens.

#### *3.3. Total Circular Run-out and Cylindricity*

Figure 14a shows the TCRO values as a function of *v*c and *f*. No clear trend can be observed as a function of *v*<sup>c</sup> or *f* for this deviation. From *f* = 0.05 to 0.10 mm/r, TCRO tends to increase for *v*<sup>c</sup> = 60 and 80 m/min, whereas it tends to decrease for *v*<sup>c</sup> = 40 m/min. From *f* = 0.10 to 0.20 mm/r, its value remains more or less constant, for *v*c = 40 and 60 m/min, regardless of *f*. However, TCRO only shows a general trend to increase with *f* for *v*<sup>c</sup> = 80 m/min. This fact is more noticeable for *f* = 0.20 mm/r, where the worst result is obtained. Therefore, TCRO exhibits less sensitivity to change with the cutting parameters than CRO, RON or STR, and similar to CON. The value of TCRO strongly depends on the maximum value of CRO in each section, but also on the angular position where that maximum value is obtained. Therefore, the maximum value for CRO in one section may occur in a different angular position than another one and, as a result, the effect of the cutting parameters may be less evident.

**Figure 14.** Mean deviations values as a function of *v*c and *f* for the (**a**) total circular run-out, and (**b**) cylindricity.

Finally, Figure 14b plots the CYL values as a function of *f* and *v*c. For this deviation, the influence of *f* is more evident. A general trend to increase CYL with *f* can be observed in a wide range of *v*<sup>c</sup> studied. This effect is more noticeable at *v*<sup>c</sup> = 80 m/min, mainly at higher *f* values (0.15–0.20 mm/r). The worst results appear when the highest *f* (0.20 mm/r) is combined with high *v*c (60 and 80 m/min). For *v*<sup>c</sup> = 60 m/min, this trend is fulfilled from 0.05 to 0.10 mm/r and 0.20 mm/r. Nevertheless, for *f* = 0.15 mm/r, a significant decrease is observed. Therefore, in a similar way than CRO, RON or STR, CYL exhibits a higher sensitivity to change with *f* than *v*c. This behavior was to be expected, taking into account that CYL combines the effect of the profile deviations along the specimen length, as well as along its section.

#### *3.4. Geometrical Tolerance in Rotating Bar Bending Specimens*

ISO 1143:2010 standard establishes the different geometrical tolerances for the rotating bar bending specimen, in order to minimize their effect in the fatigue behavior. In this sense, in the expected fatigue fracture zone, the standard indicates a CYL and CON maximum tolerance deviation. These requirements are 20 μm for CYL and 15 μm for CON, between sections S1 and S6 [11]. Nevertheless, the geometrical deviations in a manufacturing process are usually far from the standard requirements. Therefore, it may be interesting to compare these requirements with those obtained in the manufacturing process under different conditions.

On the one hand, the CON values were below the required standard limit, in general. Only higher values of *f* (0.15 and 0.20 mm/r), in combination with high *v*<sup>c</sup> (60 and 80 m/min) exceed that limit. Therefore, the influence of CON on fatigue behavior may not be neglected under these cutting conditions. On the other hand, only for low *f* and *v*<sup>c</sup> value combinations (*f* = 0.05 mm/r, *v*<sup>c</sup> = 40 and 60 m/min), the CYL results were below the standard requirements (Figure 12b). This is due to the stronger influence of the cutting parameters on this geometrical deviation. As a result, there is a wide range of cutting conditions where the CYL deviations should be considered in the fatigue behavior analysis.

#### *3.5. Parametric Models for Macro-Geometrical Deviations*

The experimental results suggest the possibility of obtaining parametric models that allow relating some of the analyzed geometrical deviations with the cutting parameters. These experimental models may be useful to predict these deviations before machining [9,40]. These models were developed for those deviations (GD) that have shown a greater dependence on the cutting parameters (STR, CRO and RON). Different models were tested. The best fit was obtained for a potential model, as shown in Equation (3).

$$\text{GD} = \mathbb{C} \cdot v\_{\text{c}}^{\text{x}} f^{y} \tag{3}$$

where *C*, *x* and *y* are constant. Table 4 shows the results obtained for these constants. It is necessary to point out that these models have shown, in general, a reasonable fit (coefficient of determination, *<sup>R</sup>*<sup>2</sup> <sup>≈</sup> 0.6–0.7).


Figure 15 plots the experimental data versus the model results, for STR, CRO and RON respectively. Regarding STR, both cutting parameters (*f* and *v*c) show a strong influence on the model. Nevertheless, the higher value of the *x* exponent indicates a higher influence of the cutting speed. This is in good agreement with the experimental results (previously discussed). Only for *f* = 0.20 mm/r and *v*<sup>c</sup> = 80 m/min, the model shows a worse fit (Figure 15a).

**Figure 15.** *Cont*.

**Figure 15.** Potential models as a function of *v*c and *f* (**a**) straightness, (**b**) circular run-out and (**c**) roundness.

This fact may be a consequence of the higher dispersion in the experimental data. This can be considered as normal, taking into account the higher vibration levels obtained for this cutting parameter combination, as previously discussed.

With regard to CRO, the *x* low value indicates a negligible influence of *v*c. This fact can be also observed in Figure 15b. Therefore, *f* is the most influential parameter. This is in good agreement with the experimental observations, as previously commented. However, this is the model that showed a worse fit. This may be explained due to the fact that CRO includes the effect of RON and CON. The effect of the cutting parameters on CON was not clear. As a result, the CRO experimental results exhibited higher dispersion and, therefore, the model for CRO showed a lower fit (Figure 15b).

Finally, the model for RON shows a similar value for *x* and *y*. Hence, the influence of both cutting parameters (*v*<sup>c</sup> and *f*) is similar, as previously commented. In addition, the model for RON shows a better fit than for CRO. However, as it happens for STR, the model shows a lower fit for the highest values of *v*c and *f*. As a result, it is useful only for the low range of *v*c and *f* studied.

Additionally, Figure 16 plots these parametric potential models in 3D, where each geometrical deviation is represented by a surface. As previously commented, Figure 16a,c (STR and RON, respectively) show a strong dependence of *v*c and *f*, whereas Figure 16b exhibits a very low influence with *v*c. In addition, an increment in these deviations can be observed when the cutting parameters are increased.

**Figure 16.** *Cont*.

**Figure 16.** Potential models for (**a**) STR = *g*(*vc*, *f*), (**b**) CRO = *h*(*vc*, *f*) and (**c**) RON = *j*(*vc*, *f*).

It is necessary to point out that similar models have been obtained in previous research, for specimens with low slenderness (UNS A97075 and UNS A92024 alloys) [9,40]. Those models presented an exponential form for these macro-geometrical deviations under similar cutting conditions. Notwithstanding this, they exhibited a lower sensitivity to change with the cutting parameters due to the higher rigidity of the specimens' geometry. Therefore, the potential models presented in this work are more suitable for specimens with higher slenderness. Finally, it is necessary to highlight that these models are valid within the range of the tested cutting parameters and under the cutting conditions exposed. The general validity of these models should be contrasted in further works, under different conditions. However, they are useful for a better understanding of the experimental results obtained.

#### **4. Conclusions**

In this work, the influence of the cutting speed and feed on several macro-geometrical deviations (parallelism, straightness, circular run-out, roundness, concentricity, total circular run-out and cylindricity) in dry turning of UNS A97075 (Al-Zn) alloy was analyzed. The analysis was performed using high slenderness parts in order to compare the experimental results with previous research carried out with more rigid parts.

Straightness has exhibited a general trend to increase with the feed, regardless the cutting speed. This trend was more evident at low feed values. The worst results were obtained when the highest cutting speed and feed rate were combined. These results should be explained taking into account that this deviation is measured along the specimen length. Therefore, it is influenced by the BUL and BUE formation and detaching, the process rigidity, the part slenderness and vibrations. On one hand, at low feed, the BUE formation and detaching becomes more relevant and the feed rate influence is more noticeable. On the other hand, at high cutting speed, vibrations become more relevant and the sensitivity to change with the cutting speed is higher.

Regarding parallelism, the influence was less evident and no clear trend was found as a function of the feed rate or cutting speed. Notwithstanding this, the obtained results for both deviations were higher than those obtained in previous works with more rigid parts under similar cutting conditions.

The circular run-out and roundness showed a general trend to increase with the feed rate. This dependence was more evident for the roundness. The cutting speed influence was lower in both cases. The feed rate effect worsened when it was combined with high cutting speeds. With regard to the evolution of those deviations along the specimen section, further sections from the chuck have shown higher deviations, whereas closer sections have exhibited lower values. This general trend was more noticeable at a low cutting speed, regardless of *f.* In addition, an increased *f* influence was observed in the sections more distant from the chuck. Hence, vibrations became more relevant at a high spindle rotational speed, whereas the specimen slenderness and rigidity were more influential at a low cutting speed.

Regarding concentricity, no clear trend was found. Therefore, its sensitivity to change with the cutting parameters was lower. Usually, this deviation is more influenced by the rigidity setup and spindle concentricity than by the cutting parameters.

Furthermore, no clear trend was observed as a function of the cutting parameters for the total circular run-out. Therefore, this deviation exhibited lower sensitivity to change with the cutting parameters. This fact may be explained taking into account that this deviation strongly depends on the maximum value of the circular run-out in each section, but also on the angular position for this value.

With regard to cylindricity, this deviation showed a general trend to increase with the feed rate in a wide range of cutting speeds studied. Its sensitivity to change with the feed was higher than with the cutting speed, in a similar way than the circular run-out, the roundness or the straightness. This is due to the fact that cylindricity combines the effect of the profile deviations along the specimen length, as well as along its section.

Therefore, the experimental results revealed higher sensitivity to change with the cutting parameters than the results obtained in previous research with lower slenderness parts, for most macro-geometrical deviations. In addition, the feed rate seems to be the most influential parameter whereas the cutting speed has shown less influence. On the other hand, the obtained deviations have been noticeably higher in parts with lower slenderness, compared with those obtained in previous research with more rigid parts of UNS A92024 and UNS A97075 alloys.

Additionally, the experimental results for CON and CYL have been compared with the quality requirements of the specimens used in the rotating bar bending fatigue test standard (ISO 1143:2010). The results revealed that there is a wide range of cutting conditions where both geometrical deviations should be considered for the fatigue behavior analysis.

Finally, a set of potential parametric models were proposed for STR (*vc, f*), CRO (*vc, f*) and RON (*v*c*, f*). These models exhibited a reasonable fit for STR and RON, whereas the fit for CRO was lower. These models may be useful to analyze the influence of cutting conditions (*vc, f*) in these deviations before machining. It is necessary to point out that these models are useful in the range of cutting conditions evaluated and can be considered as a first step to obtain more complex models. In addition, these models were compared with other models obtained from the tests with less slender specimens (20 times less slenderness). In spite of this, the results dispersion is of the same order.

It is necessary to point out that although the study was carried out on the parts with a geometry different from that used for the manufacture of aircraft structural parts, this work revealed the importance of this kind of analysis in further works. In addition, it can be considered as the starting point to analyze the influence of the geometrical deviations on mechanical properties, such as fatigue behavior.

**Author Contributions:** S.M.B. and F.J.T.V. conceived and designed the experiments; S.M.B. performed the experiments; S.M.B., F.J.T.V., C.B.G. and L.S.H. analyzed the data; S.M.B. and F.J.T.V. wrote the paper; L.S.H. and C.B.G. revised the paper.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors thank the University of Cádiz for its form measuring equipment and the University of Malaga-Andalucia Tech Campus of international Excellence for its economic contribution to this paper.

**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* **E**ff**ects of Tool Edge Geometry on Chip Segmentation and Exit Burr: A Finite Element Approach**

#### **Muhammad Asad**

Mechanical Engineering Department, Prince Mohammad Bin Fahd University, AL-Khobar 31952, Saudi Arabia; masad@pmu.edu.sa

Received: 18 October 2019; Accepted: 13 November 2019; Published: 18 November 2019

**Abstract:** The effects of different tool edge geometries (hone and chamfer (T-land)) on quantitative measurement of end (exit) burr and chip segmentation (frequency and degree) in machining of AA2024-T351 are presented in this work. The finite element (FE) approach is adopted to perform cutting simulations for various combinations of cutting speed, feed, and tool edge geometries. Results show an increasing trend in degree of chip segmentation and end burr as hone edge tool radius or chamfer tool geometry macro parameters concerning chamfer length and chamfer angle increase. Conversely, the least effects for chip segmentation frequency have been figured out. Statistical optimization techniques, such as response surface methodology, Taguchi's design of experiment, and analysis of variance (ANOVA), are applied to present predictive models, figure out optimum cutting parameters, and their significance and relative contributions to results of end burr and chip segmentation. Various numerical findings are successfully compared with experimental data. The ultimate goal is to help optimize tool edge design and select optimum cutting parameters for improved productivity.

**Keywords:** tool edge preparation; segmented chip; machining simulation; burr; optimization

#### **1. Introduction**

Aluminum alloys are widely used in the aerospace industry due to their excellent strength-to-weight ratios and thermal properties. Aluminum alloys are categorized as easy to machine materials and are ideal candidates to subject to dry high-speed machining. However, certain complex combinations of tool materials, tool cutting angles (mainly rake angles), tool edge geometry (hone edge and chamfer edge), chip breaker profiles, cutting process parameters, machine dynamics, among others, greatly influence high-speed cutting processes and may result in high cutting temperatures and intense localized deformations, as reported in numerous experimental and numerical studies performed on aluminum alloys, such as AA2024-T351, AA7010-T7451, and AA7050-T7451. The severe cutting conditions lead to highly segmented chip morphology (higher "chip segmentation frequency" and higher "degree of chip segmentation"), poor surface finish, compromised surface integrity, along with high residual stresses and early failure of tools [1–5].

Furthermore, burr formation is another unlikely phenomenon associated with machining processes. Burr (the undesired and detrimental sharp material formed on workpiece edges) is formed during machining of metallic materials and composite/metal stacks in all sorts of machining processes, such as drilling, milling, turning, and broaching. However, ductile of machining materials generally results in pronounced burr lengths [6,7]. Deburring or burr removal is a necessary process before the component is ready for its functional life, providing the required surface quality and allowing integration into product assembly. Various mechanical, thermal, electrical, or chemical deburring processes employed in industry are costly, require technical expertise, and are quite time consuming [6,7]. These non-value-added post-machining deburring processes undermine the benefits of high-speed

machining of aluminum alloys. All of this necessitates the optimization of cutting parameters, tool materials, and angles and edge geometries to improve machined component quality, improve tool life, and eventually increase productivity. Worthy analytical, experimental, and numerical efforts have been carried out in this context to comprehend the chip formation process [8–12] and optimize cutting parameters to control surface quality and residual stresses [13–15]. Most recently, an integrated finite element and finite volume numerical model was presented by Hegab et al. [16] to analyze nano-additive-based minimum quantity lubrication (MQL) effects on machining forces, temperatures, and residual stresses. A considerable decrease in cutting temperatures and residual stress was reported using nano-additive-based MQL. This ultimately will help to increase tool life and improve surface integrity. Furthermore, physical comprehension of burr formation mechanisms and burr control through parametric optimization and tool and workpiece geometry optimization have also been widely discussed in literature [3,4,6,7,17–19].

The present work aims to examine the effect of tool edge geometry design (hone (round) edge and chamfer (T-land) edge), also called "tool edge preparation", on chip formation, chip segmentation frequency, degree of chip segmentation, and exit burr formation processes. Various combinations of two macro-level parameters of chamfer edge geometry, namely chamfer length (*l*β) and chamfer angle (γβ), and the macro geometry of the hone edge radius (*r*β) are investigated (Figure 1, Section 2.1). Micro-level cutting edge geometry segments such as "cutting edge segment on flank face" and "cutting edge segment on rake face", as discussed by Denkena et al. [19], are not considered, as feed values taken in the current study are higher than the equivalent edge radii (Table 1, Section 2.1). Additionally, the workpiece material in the vicinity of the stagnation point (around which micro cutting geometry is defined by Denkena et al. [19]) is extremely deformed during machining and is removed during simulation after attaining the defined damage criteria (described later in Section 2.2).

To simulate chip segmentation and exit burr formation processes for orthogonal cutting of AA2024-T351 finite element analyses using various combinations of tool edge geometry, cutting speed and feed tests were performed. Higher values for the tool edge chamfer length (*l*β), chamfer angle (γβ), and hone edge radius (*r*β) will certainly increase the negative rake angle in the vicinity of the stagnation point, and the increased workpiece area will experience high thermo-mechanical load. This will largely influence the primary shear zone, negative shear zone (responsible for exit burr formation), and material degradation, in turn reducing the augmentation of chip segmentation and leading to longer burr lengths. Chip segmentation and exit burr formation processes are the main focus of the present work due to their direct and indirect effects on machined surface quality and tool life. For example, chip segmentation frequency and degree of chip segmentation directly dictate residual stress patterns, intensity, and depth on machined surfaces [11,20]. The chip segmentation phenomenon also causes fluctuating cutting forces and harmful chatter vibration affecting machined surface and tool life [21–23], whereas burr not only influences machined surface quality but also influences the fatigue life of machined parts [4,6,7]. A phenomenal shift from "thermal softening" to "crack initiation and propagation" has also been highlighted [12,21,24], causing formation of segmented chips using varying tool edge geometries, cutting speeds, and feeds. This paper also provides more comprehensive information on burr formation ("negative burrs" at the exit end of workpiece), crack propagation at the front of the tool edge, formation of negative shear zones and pivot point locations, boot-type chip formation, and associated burr generation phenomena. The eventual aim of the presented work is to provide further insight into chip and burr formation in machining of AA2024-T351 and to optimize cutting parameters and tool edge design for improved productivity, employing a finite element (FE)-based design and analysis approach. Numerically computed results of chip morphology, cutting forces, and chip segmentation frequency are compared with the ones obtained previously by performing orthogonal cutting experimental investigations on AA2024-T351 under similar cutting conditions [11].

A full factorial Taguchi's design of experiment (DOE) technique is employed to determine optimum combinations of tool edge geometry, cutting speed, and cutting feed to curtail burr lengths, chip segmentation frequency, and degree of chip segmentation. Analysis of variance (ANOVA) is performed to determine the percentage influence of these factors on exit burr lengths, segmentation frequency, and degree of segmentation. Response surface methodology (RSM)-based quadratic predictive models are also proposed.

#### **2. Finite Element Based Orthogonal Cutting Model**

#### *2.1. Geometrical Model, Mesh, Constraints, and Hypothesis*

Figure 1 shows workpiece and tool geometrical models for orthogonal cutting cases, conceived in Abaqus explicit software (Abaqus, 6.16, Dassault Systemes, Johnston, RI, USA, 2016). For the present work, six different cutting edge geometries are considered: two hone edge (*r*<sup>β</sup> = 5 μm and 20 μm) and four chamfer edge (chamfer length (*l*β) = 0.1 mm, chamfer angle (γβ) = 15◦; *l*<sup>β</sup> = 0.1 mm, γβ = 25◦; *l*<sup>β</sup> = 0.2 mm, γβ = 15◦; *l*<sup>β</sup> = 0.2 mm, γβ = 25◦) technologies. In the current work, chip separation is based on ductile damage of a predefined sacrificial material layer approach [11], named the "chip separation zone" in Figure 1. The width of the "chip separation zone" is kept to the order of the tool hone edge radius (*r*β), as per experimental evidence [25]. For hone edge radii of 5 μm and 20 μm, the "chip separation zone" width is taken as 20 μm, while for chamfer edge geometries, the "chip separation zone" is taken as the "equivalent radius (*req*)" of chamfer edge geometries, as shown in Figure 1 and summarized in Table 1.

**Figure 1.** Orthogonal geometrical model and constraints.


**Table 1.** The "chip separation zone" width for various tool edge geometries.

In the FE model, the tool rake angle = 17.5◦ and the clearance angle = 7◦, and the profile of insert chip breaker geometry are obtained using scanning electron microscope (SEM: Zeiss SUPRA 55-VP FEGSEM, Oberkochen, Germany) and are similar to that of Sandvik's "uncoated carbide insert: CCGX 12 04 08-AL 93 H10 (Sandvik Coromant Sandviken, Sweden)" geometry used in experimental work [11]. The workpiece geometry is modeled initially in three parts: the "machined workpiece", "chip separation zone", and the chip (with specific feed, *f*). Later on, parts are assembled, as per Figure 1, with the Abaqus built-in tie constraint algorithm, which ensures that all parts behave as a single entity during simulation. The objective for generating distinct parts (the "machined workpiece", "chip separation zone", and chip) lies in the ease of defining different material behaviors and governing equations in different sections of the workpiece.

During machining, heat is generated due to plastic work and friction at the tool and workpiece interface; therefore, to perform coupled temperature–displacement simulations, both the tool and workpiece are meshed with four-node, bilinear, quadrilateral continuum, displacement and temperature, reduced integration elements (CPE4RT), using the plane strain hypothesis. In these elements, along with displacement, temperature is also a nodal variable. Selection of an optimum mesh density in metal machining simulation producing physical results is quite challenging because of the non-availability of a specifically defined criterion in the literature. However, as a general rule, the finer the mesh, the higher the cutting force due to the size effect phenomenon [2]. A mesh sensitivity analysis for various mesh densities (Figure 2) was performed for *f* = 0.4 mm/rev and *VC* = 100 m/min. The increase in cutting forces as a function of mesh density can be figured out. An asymptotic value of mesh size of approximately 25 μm was achieved. Any further decrease in mesh density will not change cutting forces considerably, however, it will attract a time penalty in numerical simulation. A mesh density in the order of 20 μm is chosen in the "chip separation zone", chip, and upper layer (~0.3 mm) of the machined workpiece. The workpiece is fully constrained, while the tool advances with defined cutting speed in the x-direction during simulation, as shown in Figure 1. Cutting simulations were performed with twenty-four various combinations of cutting speed (*VC*), feed (*f*), and tool edge geometries (Table 2).

**Figure 2.** Average cutting force (N) for various mesh densities (μm) for plain strain conditions.


**Table 2.** Levels of cutting parameters.

#### *2.2. Material Behavior, Chip Separation, Friction, and Thermal Models*

The workpiece material's behavior is defined by the Johnson–Cook thermo-elasto-visco-plastic constitutive model (Equation (1)). This law adequately defines material behavior in high-speed metal deformation applications. Chip formation and separation are based on the evolution of ductile fracture [5]. The Johnson–Cook shear damage model (Equation (2)) is used to simulate ductile damage. Initially, Equation (3) is used to calculate scalar damage initiation. Then, modeling of damage evolution is based on Equation (4), representing the linear evolution of scalar damage evolution parameter (*D*), and Equation (5), representing the exponential evolution of scalar damage evolution parameter (*D*). Equations (4) and (5) are used in chip separation and chip regions, respectively. In the latter equation, *Gf*, represents the fracture energy required to open the unit area of a crack, as per Hillerborg et al.'s fracture energy proposal [26], and is considered a material property. As per the approach, the material softening response after damage initiation is characterized by a stress–displacement response rather than a stress–strain response, and fracture energy is then given as Equation (6). In the present work, *Gf* is taken as an input material parameter calculated by Equation (7). Finally, Equation (8) is used to calculate the equivalent plastic displacement at failure.

$$\overline{\sigma}\_{\text{fC}} = \underbrace{(A + B\overline{\varepsilon}^{\text{H}})}\_{\text{Elasto-plastic term}} \underbrace{\left[1 + \text{Cl}\ln\left(\frac{\dot{\overline{\varepsilon}}}{\dot{\overline{\varepsilon}}\_{0}}\right)\right] \underbrace{\left[1 - \left(\frac{T - T\_{r}}{T\_{m} - T\_{r}}\right)^{m}\right]}\_{\text{Softening term}}}\_{\text{Softening term}}\tag{1}$$

$$\overline{\varepsilon}\_{0i} = \left[D\_1 + D\_2 \exp\left(D\_3 \frac{P}{\overline{\sigma}}\right)\right] \left[1 + D\_4 \ln\left(\frac{\dot{\overline{\varepsilon}}}{\overline{\varepsilon}\_0}\right)\right] \left[1 + D\_5 \left(\frac{T - T\_r}{T\_m - T\_r}\right)\right] \tag{2}$$

$$
\omega = \sum \frac{\Delta \overline{\varepsilon}}{\overline{\varepsilon}\_{0i}} \tag{3}
$$

$$D = \frac{\overline{u}}{\overline{u}\_f} \tag{4}$$

$$D = 1 - \exp\left(-\int\_0^{\overline{\mu}} \frac{\overline{\sigma}}{G\_f} d\,\overline{u}\right) \tag{5}$$

$$G\_f = \bigcap\_{0}^{n\_f} \sigma\_{\mathcal{Y}} d\,\overline{u} \tag{6}$$

$$\left(\text{G}\_{f}\right)\_{I,II} = \frac{1-\nu^{2}}{E} \left(\text{K}\_{\text{C}}^{2}\right)\_{I,II} \tag{7}$$

$$
\overline{u}\_f = \frac{2G\_f}{\sigma\_y} \tag{8}
$$

During the progression of material damage, as the damage evolution parameter (*D*) approaches a value of one, it is assumed that the element's stiffness is fully degraded and that it can be removed from the mesh. Hence, chip separation from the workpiece body is realized. The tool (tungsten carbide) is modeled as a purely elastic body in the present work. Tool and workpiece material properties and model equation parameters are shown in Tables 3 and 4, respectively.


**Table 3.** Physical properties of tool and workpiece materials [11].


During the machining process, heat is produced due to friction and plastic work. Conduction is the only mode of heat transfer considered in the present work, while the definition of contact conductance between the tool and workpiece ensures thermal conduction between them. Heat generation due to plastic work is modeled via Equation (9).

$$
\dot{q}\_p = \eta\_p \overline{\sigma} \dot{\overline{\varepsilon}}\tag{9}
$$

where . *qp* is the heat generation rate due to plastic deformation and η*<sup>p</sup>* is the plastic (inelastic) heat fraction, taken as equal to 0.9. The heat generation rate due to friction is calculated by employing Equation (10).

$$\dot{q}\_f = \rho \mathbf{C}\_\mathbb{P} \frac{\Delta T\_f}{\Delta t} = \eta\_f \mathbf{J} \,\text{\textpi\,\text{\textdegree}}\,\text{\textdegree}\,\tag{10}$$

An amount of heat *J* (from the fraction of dissipated energy η*<sup>f</sup>* caused by friction) remains in the chip (1 − *J*) and is conducted to the tool. The fraction of heat *J* is a function of conductivities and diffusivities of tool and workpiece materials [27]. These thermal properties are temperature-dependent (Table 3) and vary with tool and workpiece contact during highly dynamic cutting processes. All of this makes it quite challenging to consider an accurate value of *J* for tool–workpiece contact. Therefore, in the present work the Abaqus default value of *J* = 0.5 is taken. The steady state, two-dimensional form of the energy equation is given by Equation (11).

$$
\lambda \left( \frac{\partial^2 \mathbf{T}}{\partial \mathbf{x}^2} + \frac{\partial^2 \mathbf{T}}{\partial \mathbf{y}^2} \right) - \rho \mathbf{C}\_p \left( u\_x \frac{\partial \mathbf{T}}{\partial \mathbf{x}} + u\_y \frac{\partial \mathbf{T}}{\partial \mathbf{y}} \right) \dot{q}\_f + \dot{q}\_p = 0 \tag{11}
$$

Accurate and precise definition of friction characteristics between the tool and workpiece is important as well as challenging, since it depends on tool and workpiece material properties and geometries, cutting temperature, cutting speed, contact pressure, cutting forces, and contact length, among others [28,29]. Valuable research studies have been dedicated to this important aspect of metal machining to develop a more precise and realistic friction model under variable cutting conditions, owing to its importance in affecting the chip geometry, built-up edge formation, cutting temperature, tool wear, and surface integrity, among others. Application of these friction models in finite-element-based machining models can be taken into account when numerical models are based on the Eulerian formulation; nevertheless, it is still challenging when numerical models are based on the Lagrangian formulation. In the finite element cutting models based on the latter formulation, the workpiece mesh experiences high deformation in the vicinity of the tool–workpiece interaction. Simultaneously, when

damage and fracture energy approaches are used in constitutive models, the contact conditions become highly dynamic and complex. As the present work is based on the Lagrangian formulation, to avoid complexities in simulation, a basic Coulomb's fiction law has been adopted.

#### **3. Results and Analysis**

#### *3.1. Finite Element Analysis and Discussion*

Coupled temperature displacement cutting simulations for 24 combinations of feed, cutting speed, and tool edge geometries were performed, as per Table 2. Computational results concerning cutting forces, chip segmentation frequency, chip segmentation intensity, temperature distribution in the workpiece and tool, and end (exit) burr are calculated. Results of average cutting forces, chip morphology, and chip segmentation frequency (with tool edge equivalent radius, *req* = 20 μm) are compared with the related available results of the experimental work [11]. Numerical results of cutting forces are found to have good correlation with the related experimental ones, as shown in Table 5. The results of chip segmentation frequencies for levels 15 and 16 (*VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 20 μm and *VC* = 400 m/min, *f* = 0.4 mm/rev, *req* = 20 μm) adequately correspond to their experimental counterparts. However, chip segmentation frequencies for levels 3 and 4 (*VC* = 800 m/min, *f* = 0.3 mm/rev, *req* = 20 μm and *VC* = 400 mm/min, *f* = 0.3 mm/rev, *req* = 20 μm) do not correspond well. The latter is due to the fact that at lower cutting feeds, segmentation intensity decreases (i.e., more uniform chip or less intense segmented chip morphology results). A more refined mesh would be required to obtain more accurate "segmentation frequency" results at lower cutting feeds, which would attract a greater time penalty in numerical simulations. Numerical findings (as presented in Table 5 and Figure 3a) only at levels 3, 4, 15, and 16 are compared with available experimental data results [11]. This comparison is made to validate the numerical model, whereas the rest of the numerical simulations made with various combinations of speed, feed, and tool edge geometry (levels 1, 2, 5–14, and 17–24) are merely exploitation of the validated numerical model (with no experimental results found in the literature). Numerically simulated and experimentally acquired chip morphologies (level 15 only) are compared in Figure 3.


**Table 5.** Numerical and experimental [11] comparison of mean cutting forces (at constant cutting depth, *a*<sup>P</sup> = 4 mm) and chip segmentation frequencies.

3.1.1. Cutting Parameters and Tool Geometry Effects on Chip Segmentation Frequency and Segmentation Intensity

In almost all parametric combinations of cutting speed, feed, and tool edge radius, a slightly segmented to highly segmented chip morphology is reported. This shows the high plasticity properties of the alloy. Segmented chips (with high segmentation frequency and segmentation intensity) negatively affect machined surface integrity in terms of the quality of the surface profile, residual stress patterns, and the intensity of residual stresses. In the literature, these chips were also reported to produce periodic fluctuations in cutting forces and tool vibrations, which eventually effect tool life. The mechanism of formation of segmented chips is still not well understood, owing to the complex nature of the machining process, which is greatly influenced by the material properties and microstructure, tool geometries, cutting parameters, machine tool dynamics, and friction, among others [12,30]. However, there are mainly two theories explaining the phenomenon of chip segmentation in most of the ductile materials: (a) thermoplastic deformation and formation of adiabatic shear bands because of thermal softening; and (b) fracture, where cracks initiate and propagate in the primary shear zone [12]. In the present work, both phenomena have been witnessed.

**Figure 3.** Chip morphology for cutting speed *VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 20 μm (level 15): (**a**) numerically simulated; (**b**) experimentally generated.

At high cutting speed, frictional resistance causes an increase in cutting temperatures at the tool–workpiece interface, resulting in thermal softening (Figure 4). The thermal softening phenomenon dominates strain hardening, the material stiffness degrades (lower stresses in the vicinity of the tool edge; Figure 3a), and the material flows in the primary shear zones with ease, leading to generation of adiabatic shear bands. Apart from obvious results of higher cutting temperatures due to higher cutting speed, it can also be seen from Figure 4 that an increase in tool edge radius (especially tools with chamfer geometry) results in lower cutting temperatures. Similar trends have also been reported by Ozel [31] for cutting of AISI H-13 with cubic boron nitride (CBN) cutting inserts. This phenomenon is due to the size effect (i.e., more specific cutting energy is required as the tool radius increases in comparison to uncut chip thickness). A wider area now experiences plastic deformation, which requires more energy, and more heat is generated. However, the heat due to inelastic work is more easily dispersed over a large surface area with a larger equivalent edge radius, and consequently maximum temperatures are lower. At higher feed, higher temperatures are produced due to larger amount of plastic work (Figure 4). However, the rate of increase of temperature is not high enough (for feed variation studied in this work this ranges from 0.3 to 0.4 mm/rev) to cause any considerable thermal softening. Furthermore, at higher feed values, due to length effect, longer segments of chips are generated (i.e., frequency of segments will decrease). This shows that higher cutting speeds supplemented with a lower feed rate and lower tool edge radius promote formation of more adiabatic shear bands (high frequency of segmented chip morphology), mainly due to thermal softening. Segmentation frequency is greatly influenced by variation of cutting speed, while segmentation intensity or degree of chip segmentation, calculated by "(*h*max − *h*min)/*h*max", seems to be least effected by speed variation, as can be seen in Figure 5.

**Figure 4.** Maximum nodal temperature evolution for cutting speeds, feed, and tool edge radius variations.

**Figure 5.** Cutting speed effect on segmentation frequency and degree of chip segmentation: (**a**) Segmentation frequency for *f* = 0.3 mm/rev; (**b**) Segmentation frequency for *f* = 0.4 mm/rev; (**c**) Degree of chip segmentation for *f* = 0.3 mm/rev; (**d**) Degree of chip segmentation for *f* = 0.4 mm/rev.

Figure 6 shows that an increase in cutting edge radius rarely influences the segmentation frequency, which largely influences the degree or intensity of segmentation. Indeed, as the chamfer tool angle (γβ) increases, the effective rake angle in the vicinity of the stagnation point becomes more negative, and as the chamfer tool length (*l*β) or hone edge radius (*r*β) increase, the workpiece area experiences high thermo-mechanical load, leading to initiation and propagation of fracture in the primary shear zone. Furthermore, it can be noticed that chamfer tool length (*l*β) contributes more than chamfer tool angle (γβ) in intensifying the degree of segmentation and the equivalent edge radius (Table 1). On the other hand, as discussed previously and depicted in Figure 4, the increase in cutting edge radius results in

decreasing temperature; hence, thermal softening is not the dominant or responsible mechanism for chip segmentation at higher values of tool cutting edge radii.

**Figure 6.** Edge radius effect on segmentation frequency and degree of chip segmentation: (**a**) Segmentation frequency for *Vc* = 800 m/min, *f* = 0.3 mm/rev; (**b**) Segmentation frequency for *Vc* = 400 m/min, *f* = 0.3 mm/rev; (**c**) Segmentation frequency for *Vc* = 800 m/min, *f* = 0.4 mm/rev; (**d**) Segmentation frequency for *Vc* = 400 m/min, *f* = 0.4 mm/rev; (**e**) Degree of chip segmentation for *Vc* = 800 m/min, *f* = 0.3 mm/rev; (**f**) Degree of chip segmentation for *Vc* = 400 m/min, *f* = 0.3 mm/rev; (**g**) Degree of chip segmentation for *Vc* = 800 m/min, *f* = 0.4 mm/rev; (**h**) Degree of chip segmentation for *Vc* = 400 m/min, *f* = 0.4 mm/rev.

Figure 7 shows a highly segmented chip morphology (with higher degree of chip segmentation) generated for *VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 180 μm (level 23). In shear bands, the stiffness is fully degraded, with almost zero value for stresses. This shows the probability of fracture in the primary shear zone. Similar trends can also be seen in Figure 8 with variation of feed. The degree of chip segmentation is highly influenced by the change in feed, although by decreasing feed, segmentation frequency increases (due to length effect, longer segments of chips are generated), but this effect is not as pronounced as can be seen for the degree of segmentation.

**Figure 7.** Chip morphology for *VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 180 μm (level 23).

**Figure 8.** Feed rate effect on segmentation frequency and degree of chip segmentation: (**a**) Segmentation frequency for *Vc* = 800 m/min; (**b**) Segmentation frequency for *Vc* = 400 m/min; (**c**) Degree of chip segmentation for *Vc* = 800 m/min; (**d**) Degree of chip segmentation for *Vc* = 400 m/min.

Considering the above, it can be summarized that cutting speed greatly influences the chip segmentation frequency, while feed and tool edge radius largely effect the degree of chip segmentation. The thermal softening phenomenon plays a vital role in chip segmentation at higher cutting speeds, lower feed rates, and with smaller tool edge radius values (mainly in increasing segmentation frequency), while crack propagation in primary shear bands occurs at higher values of cutting edge radius and feed (largely influence segmentation degree). To predict optimal combinations of speed, feed, and tool edge radius to minimize the generation of segmented chip morphology (segmentation frequency and degree of chip segmentation), statistical analyses are performed in the next section.

#### 3.1.2. Cutting Parameters and Tool Geometry Effects on End (Exit) Burr Formation

During the course of chip formation, as the tool keeps on advancing in the cutting direction towards the end of the workpiece, a negative shear zone starts to grow from the workpiece free end (exit end) towards the primary shear zone (Figures 3 and 7). The formation of the negative shear zone is specifically due to the bending load experienced by the workpiece free end during tool advancement in the cutting direction. As the tool advances further, the bending load keeps on increasing, the material experiences higher stresses in this deformation zone, and a pivot point (high stressed point) appears on the exit edge of the workpiece (Figures 3 and 7). The location of the "pivot point" is measured from the machined surface along the *y*-axis. The distance of the "pivot point" has a direct relationship with burr lengths (produced at the exit end)—longer distances represent longer burr lengths. The pivot point distance highly depends on the cutting parameters, materials, and tool geometry. During the course of cutting, the negative shear zone expands further around the pivot point and reaches the tool edge. Higher stresses far ahead of the tool tip position (due to the negative shear zone) promote the material's ductile failure and initiation of cracks in the chip separation zone far ahead of the tool tip (Figure 9). The material deviates from the actual cutting phenomenon, the chip formation process ceases, the tool pushes away the boot-type chip (combination of chip and uncut material), and the end burr (workpiece's deformed exit edge) appears at the end of the workpiece. Figure 9 shows early and advanced failure of chip separation zone material with formation of cracks and generation of an end burr for *VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 180 μm (level 23).

**Figure 9.** The chip separation zone's material advance failure and formation of negative burr for *VC* = 800 m/min, *f* = 0.4 mm/rev, *req* = 180 μm (level 23).

During machining of aluminum alloys, for various combinations of cutting parameters, both negative and positive burrs at the end of the workpiece have been reported in the literature [3]. Positive burrs (without considerable damage to workpiece edge) are normally generated at lower feed values, and vice versa [3]. In the present work, for AA2024-T351, with investigated combinations of cutting speed, feed, and tool edge geometry, only negative burrs (with edge breakout) were formed. It is found that machining performed with higher feeds along with larger tool edge radii produces highly stressed and more widened shear zones (both primary and negative), and the pivot point location is further away from the machined surface, generating longer burrs than for machining performed at lower feed rates and with smaller tool edge radii. Figures 10 and 11 quantify and produce a trend for exit burrs as a function of the feed and tool edge radius. On the other hand, speed variation was been found to have non-noticeable effects in changing exit burr lengths (Figure 12). The results, in general, are consistent with the findings of experimental burr formation studies performed on aluminum alloys [3,32]. Table 6 details numerically computed exit burr lengths for twenty-four various combinations (defined in Table 2) of cutting speed (*VC*), feed (*f*), and tool edge geometries.

**Figure 10.** Feed rate variation effects on exit burr lengths: (**a**) Burr length at *Vc* = 800 m/min; (**b**) Burr length at *Vc* = 400 m/min.

**Figure 11.** Edge radius variation effects on exit burr lengths: (**a**) Burr length for *Vc* = 800 m/min, *f* = 0.3 mm/rev; (**b**) Burr length for *Vc* = 400 m/min, *f* = 0.3 mm/rev; (**c**) Burr length for *Vc* = 800 m/min, *f* = 0.4 mm/rev; (**d**) Burr length for *Vc* = 400 m/min, *f* = 0.4 mm/rev.

**Figure 12.** Cutting speed variation effects on exit burr lengths: (**a**) Burr length for *f* = 0.3 mm/rev; (**b**) Burr length for *f* = 0.4 mm/rev.



#### *3.2. Statistical Analysis and Optimization*

In the preceding section, finite element method (FEM) approach was employed to predict the likelihood of chip segmentation features (segmentation frequency and degree of chip segmentation) and exit burr formation under various combinations of speed, feed, and tool edge radius. Various associated phenomena such as maximum nodal temperature, material stiffness degradation, early fracture of material in the tool's advancement direction, and location of the pivot point are also discussed. Interesting conclusions can be drawn for optimizing the machining of AA2024-T351 using tungsten carbide inserts. Nevertheless, further investigations are required to predict optimum combinations of speed, feed, and tool edge radius to minimize the generation of segmented chip morphology (segmentation frequency and degree of chip segmentation) and reduce burr formation. The relative significance of each cutting parameter on the latter phenomenon would also be interesting from a

production engineer's perspective. Predictive models of chip segmentation features (segmentation frequency and degree of chip segmentation) and exit burr lengths would be advantageous to minimize the cutting trials to optimize the cutting. In this framework, the present section exploits statistical analysis tools, such as Taguchi's design of experiment (DOE), analysis of variance (ANOVA), and response surface methodology (RSM).

#### 3.2.1. Statistical Analyses on Burr Optimization

To determine the optimum combination of cutting parameters (speed, feed, and tool edge radius) for minimum end burr lengths, Taguchi's DOE is employed. The quality criterion approach "the-smaller-the-better" is used for the data (exit burr lengths computed over twenty-four tests via finite element analysis (FEA) and Equation (12) is used to determine the signal-to-noise (S/N) ratio.

$$\frac{S}{N} = -10 \log \left( \sum \left( \frac{y\_i^2}{n} \right) \right) \tag{12}$$

In the relationship, "*yi*" represents the response value of the *i*th test and "*n*" is the number of test repetitions (taken as one). Feed and speed have two levels of variations (*f* = 0.3 and 0.4 mm/rev and *VC* = 800 and 400 m/min), while tool edge radius has six levels of variations (*req* = 5, 20, 80, 90, 166, and 180 μm). The parametric combination *VC* = 800 m/min, *f* = 0.3 mm/rev, *req* = 5 μm (Level 1, Table 2) represents the optimum combination for generation of minimum burr, as can be figured out by the plots of main effects of S/N ratio (Figure 13a) and data mean (Figure 13b). Table 7 results show that the edge radius is the most influential factor and speed is the least influential factor in burr formation. Results show a good match with the experimental findings of Niknam and Songmene [32].

**Figure 13.** Exit burr lengths as a function of cutting parameters: (**a**) variation of signal to noise (S/N) ratios and (**b**) data means.

**Table 7.** Response table for data means.


Next, to establish a relationship between exit burr lengths and machining parameters, a second order multiple regression model (Equation (13)) based on RSM is used. The developed regression model (Equation (14) using Minitab software (Minitab, 16.2, Minitab-LLC, State College, PA, USA, 2010). The predicted value for burr length (for optimal cutting parameters: *VC* = 800 m/min, *f* = 0.3

mm/rev, *req* = 5 μm) to generate minimum burr using Equation (14) matches the value acquired through finite element simulation (Table 6).

$$y = \beta\_0 + \sum\_{i=1}^{3} \beta\_i \mathbf{x}\_i + \sum\_{i=1}^{3} \beta\_{ij} \mathbf{x}\_i \mathbf{x}\_j + \sum\_{i=1}^{3} \beta\_{ii} \mathbf{x}\_i^2 + \varepsilon \tag{13}$$

*Burr length* = <sup>−</sup>0.0160032 <sup>−</sup> 1.41323*e*−6(*Speed*) + 0.290049 (*Feed*) <sup>−</sup> 0.000581417 (*Edge radius*) <sup>−</sup>1.34362*e*−5(*Speed* <sup>×</sup> *Feed*) + 6.58783*e*−8(*Speed* <sup>×</sup> *Edge radius*) <sup>−</sup> 0.00104982 (*Feed* <sup>×</sup> *Edge radius*) +2.17843*e*−7(*Edge radius* <sup>×</sup> *Edge radius*) (14)

In Equation (14), quadratic terms of speed and feed are been included as they are insignificant. Finally, to determine the significance of the regression model and relative contribution of each of the machining parameters, analysis of variance (ANOVA) is performed. Terms used in ANOVA Table 8 are defined in Equations (15)–(18).

$$\text{Sum of Squares } (SS) = \frac{N}{nf} \sum\_{i=1}^{nf} (\overline{y}\_i - \overline{y})^2 \tag{15}$$

where *N* is the total number of tests, *nf* represents the level of each factor, *y* is the mean of the response, and *yi* is the mean of the response at each level of the respective factor.

$$\text{Mean square}(Variance) : MS\_i = \frac{SS}{DF\_i} \tag{16}$$

$$\text{Fisher\\_Coefficient}(\text{F-value}) = \frac{MS\_{\bar{i}}}{MS\_{Error}} \tag{17}$$

$$\text{Percent Contribution} : \text{PP(\%)} = \left(\frac{SS}{SS\_T}\right) \times 100\tag{18}$$


**Table 8.** Analysis of variance (ANOVA) results for exit (end) burr.

In the ANOVA table, significance or insignificance is attributed to each of the source factors based on the Fisher coefficient value (*F*-value). ANOVA for significance level = 5% (95% confidence level) was performed. The probability values (*P*-values) of the regression model, feed, and edge radius are <0.05. This shows the significance of the regression model and the factors that contribute the most: feed and edge radius. Speed, "quadratic terms", and "interactive terms" have the least effect on burr formation. Table 8 also shows that the edge radius has the highest contribution in producing burr at 74.77%, the feed contribution is 17.39%, while speed variation has the least effect in exit burr formation. This hierarchy of contribution also confirms the findings of Taguchi's DOE methodology (Table 7). It is interesting to note that ANOVA produced for "pivot point location" (considering it as target

function, Table 9) has similar trends in term of % contribution of machining parameters in producing burr (Table 8). This helps to conclude that a distant pivot point location (for larger edge radius and higher feed values) is a strong sign that longer burr will be produced.


**Table 9.** ANOVA results for pivot point location.

3.2.2. Statistical Analyses of Segmented Chip Morphology (Segmentation Frequency and Degree of Chip Segmentation)

Figure 14a,b presents the plots of the main effects of S/N ratios and data means on segmentation frequency, respectively. Analyses of plots show that segmentation frequency increases as speed increases, while higher feed and larger edge radii suppress the segmentation phenomenon, though their effect is negligible. The parametric combination *VC* = 400 m/min, *f* = 0.4 mm/rev, *req* = 180 μm (level 24, Table 2) represents the optimum combination for generating the least amount of segmentation frequency.

**Figure 14.** Segmentation frequency as a function of cutting parameters: (**a**) variation of signal-to-noise (S/N) ratios and (**b**) data means.

A second order multiple regression model based on RSM is presented in Equation (19) to define the relationship of segmentation frequency as a function of cutting parameters. In the model, quadratic terms of speed and feed are not included as they are insignificant.

*Segmentation Frequency* = −39.8568 + 0.11879 (*Speed*) + 105.777 (*Feed*) + 0.44496 (*Edge radius*) <sup>−</sup>0.120458 (*Speed* <sup>×</sup> *Feed*) <sup>−</sup> 8.47311*e*−5(*Speed* <sup>×</sup> *Edge radius*) <sup>−</sup> 1.00696 (*Feed* <sup>×</sup> *Edge radius*) <sup>−</sup>5.16853*e*−4(*Edge radius* <sup>×</sup> *Edge radius*) (19)

To outline the significance of the model and relative contribution of each of the cutting parameters on segmentation frequency, analysis of variance (ANOVA) is performed and results are summarized in Table 10. Results show that speed has the highest contribution in producing segmentation frequency at 76.63%, edge radius contributes 5.37%, while feed variation has the least effect in generating chips with high segmentation frequencies.


**Table 10.** ANOVA results for segmentation frequency.

As discussed in Section 3.1.1, machining performed at higher speeds generates higher cutting temperatures (Figure 3), leading to thermal softening and generation of adiabatic shear bands (segmented chips). In this context, ANOVA analysis is performed for "maximum nodal temperature" (considering it as target function, Table 11) to figure out the % contribution of machining parameters (speed, feed, and tool edge radius) in influencing the temperature rise. It can be seen (Table 11) that edge radius has the highest contribution to temperature variation; indeed, temperature decreases as the edge radius increases (Figure 4), whereas speed is the second highest contributor in effecting the temperature; temperature increases as speed increases (Figure 4). Feed has been found to have the least effect on maximum temperature variations. Further analyses of Tables 10 and 11 help to conclude that higher temperatures produced at higher cutting speeds promote thermal softening and generation of more frequent adiabatic shear bands (higher segmentation frequency), whereas higher feed and larger edge radii reduce segmentation frequencies, though their effects are minimal.

**Table 11.** ANOVA results for maximum nodal temperature.


Figure 15a,b presents the plots of the main effects of S/N ratios and data means on "degree of chip segmentation", respectively. Analyses of plots show that all cutting parameters promote segmentation degree, though speed's effect seems negligible. The parametric combination *VC* = 400 m/min, *f* = 0.3 mm/rev, *req* = 5 μm (level 2, Table 2) represents the optimum combination for generation of chips with the least degree of segmentation.

**Figure 15.** Degree of chip segmentation as a function of cutting parameters: variation of signal to noise (S/N) ratios (**a**) and (**b**) data means.

A second order multiple regression model based on RSM is presented in Equation (20) to define the relationship of the degree of chip segmentation as a function of cutting parameters. In the model, quadratic terms of speed and feed are been included as they are insignificant.

( ) () ( ) ()( ) ( ) − ( ) × + × − × − × = − − + + − − *Degree o f Segmentation* = −0.317506 − 5.7063 (*Speed*) + 1.1144 (*Feed*) + 0.22237096 (*Edge radius*) +0.000102409 (*Speed* <sup>×</sup> *Feed*) <sup>−</sup> 1.36297*e*−9(*Speed* <sup>×</sup> *Edge radius*) <sup>−</sup> 0.00342995 (*Feed* <sup>×</sup> *Edge radius*) <sup>−</sup>3.03896*e*−6(*Edge radius* <sup>×</sup> *Edge radius*) (20)

To outline the significance of the model and relative contribution of each of the cutting parameters on the degree of chip segmentation, analysis of variance (ANOVA) is performed and results are summarized in Table 12. Results show that speed has the least contribution to producing highly segmented chips (with high degree of chip segmentation), while feed (43.895%) and edge radii (36.46%) significantly affect the production of highly segmented chips. Finite element analyses provide explicit explanation in this context (Section 3.1.1). Larger material area experiences severe plastic deformation when cutting is performed at higher feed rates supplemented with larger tool radii. Material stiffness degrades, leading to crack initiation and propagation in primary shear bands, resulting in highly segmented chips.


**Table 12.** ANOVA results for degree of chip segmentation.

#### **4. Conclusions**

The paper provides a staggered comprehension-to-optimization approach for chip segmentation and end burr (exit burr) formation phenomena in machining of an aerospace-grade aluminum alloy AA2024-T351. These phenomena effect tool life, workpiece machined surface quality and integrity, and hence the overall productivity. Primarily, a finite-element-based cutting model has been established and used to simulate orthogonal machining and chip formation processes for multiple parametric combinations of cutting speed, feed, and tool edge geometry. Results concerning chip segmentation (segmentation frequency and degree of segmentation) and end burr are numerically computed and comprehensively analyzed. To validate the numerical machining model, cutting forces, chip segmentation frequency, and chip morphology results are adequately compared with their experimental counterparts. Then, statistical optimization techniques such as Taguchi's DOE and ANOVA are employed to identify optimum cutting parameters and their % influence in effecting chip segmentation and end burr formation processes. Lastly, RSM-based quadratic predictive models for the aforementioned phenomena are presented.

The results presented in the current work are equally interesting for designers and researchers, providing further insight into machining and related phenomena. From a production engineering perspective, they provide optimum cutting conditions to enhance productivity through optimum selection of tool geometry and cutting parameters. Important findings of the present work are listed below.


In future studies, a more realistic friction model along with the most accurate heat fraction coefficient, *J*, will be incorporated into the finite element model to present more realistic results of industrial interest. Furthermore, the study will be extended for other materials and processes, such as drilling.

**Funding:** This research received no external funding.

**Acknowledgments:** Technical support provided by Francois Girardin of Laboratoire Vibrations Acoustique, INSA de Lyon, France is highly appreciated.

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

#### **Abbreviations**



#### **References**


© 2019 by the author. 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* **Predicting Continuous Chip to Segmented Chip Transition in Orthogonal Cutting of C45E Steel through Damage Modeling**

#### **Ashwin Moris Devotta 1,\*, P. V. Sivaprasad 2, Tomas Beno <sup>3</sup> and Mahdi Eynian <sup>3</sup>**


Received: 9 March 2020; Accepted: 8 April 2020; Published: 17 April 2020

**Abstract:** Machining process modeling has been an active endeavor for more than a century and it has been reported to be able to predict industrially relevant process outcomes. Recent advances in the fundamental understanding of material behavior and material modeling aids in improving the sustainability of industrial machining process. In this work, the flow stress behavior of C45E steel is modeled by modifying the well-known Johnson-Cook model that incorporates the dynamic strain aging (DSA) influence. The modification is based on the Voyiadjis-Abed-Rusinek (VAR) material model approach. The modified JC model provides the possibility for the first time to include DSA influence in chip formation simulations. The transition from continuous to segmented chip for varying rake angle and feed at constant cutting velocity is predicted while using the ductile damage modeling approach with two different fracture initiation strain models (Autenrieth fracture initiation strain model and Karp fracture initiation strain model). The result shows that chip segmentation intensity and frequency is sensitive to fracture initiation strain models. The Autenrieth fracture initiation strain model can predict the transition from continuous to segmented chip qualitatively. The study shows the transition from continuous chip to segmented chip for varying feed rates and rake angles for the first time. The study highlights the need for material testing at strain, strain rate, and temperature prevalent in the machining process for the development of flow stress and fracture models.

**Keywords:** chip segmentation; damage modeling; dynamic strain aging

#### **1. Introduction**

The machining process remains one of the critical manufacturing processes in the 21st century and it has critical engineering applications [1]. Developments in the fields of plasticity and fracture mechanics are used to improve the understanding of the machining process [2]. This improved understanding has become even more necessary with the ever-increasing reliability need for engineered components [3].

Machining process modeling has been carried out while using different approaches, such as empirical, analytical and numerical methods [4]. The finite element (FE) method has been quite extensively used in the modeling of the machining process to model the chip formation process. FE modeling of the machining process provides the ability to incorporate a newer advanced understanding of material behavior. This improved understanding is usually obtained through other methods, such as material testing or more sophisticated modeling techniques. Within FE modeling of the machining process, workpiece material modeling requires the material response to large deformations (large plastic strains) at very high strain rates and very high temperatures.

The ferritic-pearlitic steel group is one of the essential engineering steels having a wide area of application [5]. Material deformation mechanisms, such as strain hardening, strain rate hardening, and thermal softening are vital and have been used in FE modeling of the machining process for more than two decades. Besides, there are deformation mechanisms that are material specific, such as dynamic strain aging (DSA). Previous studies carried out in understanding the behavior of ferritic-pearlitic steel have shown that DSA is a function of temperature and strain rate [6–9]. Earlier studies have been uncertain about the need for the incorporation of DSA in machining process modeling [10,11]. However, recent studies [9,12] have shown the need to incorporate DSA to improve the model's prediction accuracy. DSA has been shown to influence strain hardening behavior, thermal softening behavior, and strain rate hardening [13]. C45E steel falls within the ferritic-pearlitic dual-phase steel group. Flow stress experiments in C45E steel through compression testing at varying temperatures and strain rates have shown the presence of DSA [14]. A modified form of the Johnson–Cook (JC) model was developed using the regression modeling approach. In another recently reported work [8], a material model (Voyiadjis-Abed-Rusinek (VAR) model) was developed, where a phenomenological model is combined with a physics-based model to capture DSA in C45 steel and it is better suited for FE simulation implementation. Using the VAR approach, an attempt has been made to modify the JC model to incorporate DSA. Further, the modified JC (MJC) model is to be implemented to simulate chip formation in the machining process.

Predicting chip segmentation through FE simulations requires the modification of flow stress curves through strain-softening modifications [15] or damage modeling [16–20]. Direct strain-softening modifications of flow stress are based on the adiabatic shear theory [21]. They have been used primarily in chip segmentation prediction in the machining of Ti alloys. Damage modeling where the material failure due to the ductile shear failure has also been used in the chip segmentation of ductile material machining. Most of the studies have concentrated on improving chip segmentation prediction accuracy in terms of chip segmentation intensity [22]. Besides, from a machining process design perspective, the transition from continuous chip to segmented chip is necessary and it has been the focus of very few studies [22,23]. The continuous chip to segmented chip transition with increasing cutting velocity is attributed to the adiabatic shearing process [24]. Recently developed models [16] have explored this area where the influence of damage parameter on chip segmentation has been evaluated. The ability of finite element simulations to predict the continuous chip to segmented chip transition is scarce [23], and none exists for steel machining to the authors' knowledge.

The present work aims to modify the Johnson–Cook model for the incorporation of the DSA influence. The modified Johnson–Cook model, in combination with the damage model, needs to be implemented then within an FE framework for simulation of chip formation in machining. The fracture initiation strain model's influence within the ductile damage model approach is to be evaluated in the prediction of the continuous chip to segmented chip transition in the machining of C45E steel under orthogonal cutting conditions.

In this section, the development of the modified JC model for C45E steel based on the approach developed for the VAR model [8] is presented.

#### **2. Modified Johnson–Cook Model Development Incorporating DSA**

#### *2.1. Modified Johnson–Cook Model*

The JC model, which is one of the most used material models in the finite element simulation of the machining process, as described in Equation (1).The JC model is built in the multiplicative form in the form of strain hardening component, strain rate hardening component and thermal softening component [25]. The strain hardening component is defined by initial yield stress (*A*), strain hardening coefficient (*B*) and strain hardening exponent (*n*), respectively. The strain rate hardening and thermal softening terms are fitted using the parameters *C* and *m*, respectively.

*Metals* **2020**, *10*, 519

The fitting of the strain hardening component is traditionally carried out using standard quasistatic testing using tensile or compressive loading. In the previously reported work, *A* and *B* were modeled using flow stresses at varying temperatures and strain rates using regression modeling approach. The strain hardening exponent, *n*, was fitted as a first-order function of temperature. The strain hardening behavior of the JC model itself is written in the form, as shown in Equation (2) to reduce the FE implementation complexity.

$$\overline{\sigma} = (A + B\overline{\varepsilon}^n) \Big( 1 + C \ln \left( \frac{\dot{\varepsilon}}{\dot{\varepsilon}\_0} \right) \Big) \Big( 1 - \left( \frac{T - T\_o}{T\_m - T\_0} \right)^m \right) \tag{1}$$

$$A + B\varepsilon^n = A\left(1 + \frac{B}{A}\varepsilon^n\right) \tag{2}$$

$$A = A\_0 + A\_\Delta \left[ 1 + \tanh\left(\frac{T - T\_l}{\xi}\right) \right] \left[ 1 - \tanh\left(\frac{T - T\_h}{\xi}\right) \right] \tag{3}$$

$$A\_{\Lambda} = \Lambda \sigma \Big(1 + A\_1 \log \left(\frac{\dot{\varepsilon}}{\dot{\varepsilon}\_0}\right)\Big) \tag{4}$$

$$T\_l = T\_{ll} \Big( 1.15 + t \log \left( \frac{\dot{\varepsilon}}{\dot{\varepsilon}\_0} \right) \Big) \tag{5}$$

$$T\_h = T\_{bh} \Big( 1.15 + t \log \left( \frac{\varepsilon}{\dot{\varepsilon}\_0} \right) \Big) \tag{6}$$

The initial yield stress parameter A is modified similarly to the VAR model to incorporate the DSA influence, leading to yield stress increase with temperature increase at specific temperature ranges [8], as in Equations (3)–(6) and shown in Figure 1. Reported experimental investigations have shown that the dynamic strain aging temperature range is a function of strain rate [6,7,14]. The value of *A*<sup>0</sup> is obtained from the room temperature quasistatic compression test. The value of *A*<sup>Δ</sup> corresponds to the peak initial yield stress increase in the DSA regime. The constants *A*<sup>1</sup> and *t* are fitted with flow stress curves that are obtained at high strain rate conditions from El Magd et al. [6] and Hokka et al. [7]. The temperatures *Tl* and *Th* represent the start and end of DSA, the regime at the reference strain rate. *Tl* and *Th* are modeled as a function of strain rate instead of a constant as in the VAR model. This methodology is necessary to accommodate the observation of dynamic strain aging temperature range being dependent on strain rate [6,7]. The model assumes the DSA regime to extend to higher strain rates at which experimental results are unavailable. Nevertheless, the DSA regime for machining process modeling has been incorporated by modifying the temperature component of the Power-law model while using a regression equation by Childs et al. [26], with the limitation of the temperature range to be a constant. The constants, *B*/*A* and *n*, are fitted for three different temperature ranges to accommodate the influence of temperature on strain hardening.

**Figure 1.** Schematic of the Voyiadjis-Abed-Rusinek (VAR) model-based modification of the JC model's initial yield stress parameter with corresponding mathematical equations with A0 representing the constant, A, of the JC model.

#### *2.2. Fitting of Flow Stress Curves Using the Modified JC Model*

The ability of the newly developed flow stress curves from the previous section to predict DSA has been validated using the Gleeble test data from previous work [14]. The flow stress curves are obtained for three different strain rates of 1 s−1,5s−1, and 60 s−<sup>1</sup> with programmed temperatures between 200–700 ◦C. The temperature range at which DSA is exhibited is a function of the strain rate [6,7]. In the temperature-strain rate range where DSA is active, the initial yield stress increases with temperature and decreases with the strain rate. The presence of DSA has also been shown to lead to serrations in the flow stresses. The temperature at which DSA peaks has been identified to be around 650 ◦C at high strain rates present in machining conditions [12]. Table 1 presents the constants of the modified JC model. The parameter *B* and *n* in the modified Johnson-Cook model has been modified for three different strain hardening regimes. This approach is carried out to ease the implementation of the FE method. The constants are provided, as shown in Equation (7).

$$\begin{cases} B = 1.5A\_{\prime} & n = 0.5 & T < 400 \, ^\circ \text{C} \\ B = 1.2A\_{\prime} & n = 0.3 & 400 \, ^\circ \text{C} < T < 500 \, ^\circ \text{C} \\ B = 1.0A\_{\prime} & n = 0.2 & T > 500 \, ^\circ \text{C} \end{cases} \tag{7}$$

**Table 1.** Modified JC model parameters for normalized C45E steel.


The strain rate hardening and thermal softening parameters of the original JC model for C45E steel are used in this study (*C* = 0.0018 and *m* = 1) [11]. With the parameters obtained from the MJC model, the flow stress curves at a strain of 0.1 are predicted under different temperatures and three different strain rates are shown, as in Figure 2. The flow stress prediction extrapolated to higher strain rates prevalent in machining conditions is also plotted to visualize the DSA regime being a function of strain rate.

**Figure 2.** Initial yield stress (σ0.1) for varying temperature and strain rate with the modified Johnson–Cook (JC) model extrapolated to high strain rates to accommodate machining conditions (Experimental data is plotted with dashed lines and model-predicted data is plotted with continuous lines).

The modification of the initial yield stress parameter of the JC model can capture the increase in initial yield stress with temperature increase. At the reference strain rate of 1 s<sup>−</sup>1, the DSA regime is active in the temperature range of 200 ◦C to 500 ◦C. With a strain rate increase between 10<sup>3</sup> s−<sup>1</sup> to 106 s<sup>−</sup>1, the modeled DSA regime's initial yield stress moves to higher temperatures of 650 ◦C and it correlates with other published experimental results [6,7].

With the ability of the MJC model's initial yield stress parameter to capture the DSA regime in Figure 2, Figure 3 shows that the MJC model can capture the strain hardening behavior observed from Gleeble tests with reasonable accuracy. In the temperature ranges below 400 ◦C, the material strain hardens with increasing strain. At 500 ◦C, dynamic recovery influences strain hardening, leading to a lowering slope, and are well captured by the MJC model. Beyond 500 ◦C, at 600 ◦C and 700 ◦C, the flow stress curves exhibit constant flow stress with increasing strain.

**Figure 3.** Flow stress curves predicted by the modified JC (MJC) model compared with the flow stress curves obtained using compression tests using Gleeble thermomechanical simulator at strain rates 1 s<sup>−</sup>1,5s−<sup>1</sup> and 60 s−<sup>1</sup> (Model: Green color; Experimental: Red color).

MJC model predicted <sup>σ</sup>0.1 at a . <sup>ε</sup> of 0.52 <sup>×</sup> 104 s−<sup>1</sup> is plotted in Figure 4 to evaluate the validity of extending the hypothesis of DSA presence at very high strain rates. The model can predict σ0.1 with reasonable accuracy. The flow stress at a strain of 0.1 is chosen to avoid the transient conditions during the early stages of loading in compression testing.

**Figure 4.** Experimental flow stress at 0.52 <sup>×</sup> 104 s−<sup>1</sup> from El Magd et al. [6] fitted using the MJC Model.

#### **3. Flow Stress Modification due to Damage**

JC and MJC models both assume no change in material behavior with increasing strain. However, in reality, the flow stress is altered because of fracture. Therefore, to model chip segmentation in ductile materials at a lower cutting velocity where the possibility of adiabatic shear is less incorporation of fracture is essential. Figure 5a provides the schematic modification of the flow stress curves due to fracture from the works of Childs et al. [16,26]. The flow stress before the fracture is defined by Equation (8) and is marked by 'BEFORE FRACTURE' in Figure 5a. During plastic deformation, at the microstructure level, nucleation and growth of defects, such as micro-voids & micro-cracks and their coalescence into macro-cracks takes place leading to material damage [27]. The damage due to plastic deformation in the workpiece is accumulated through the damage factor (*D*) and is defined, as shown in Equations (8)–(12) and Figure 5a.

**Figure 5.** (**a**) Flow stress model for chip segmentation prediction with (**b**) flow stress modification factor as a function of stress triaxiality and temperature from Childs et al. using (**c**) the fracture initiation strain as a function of stress triaxiality.

The damage parameter is defined as the ratio of accumulated plastic strain to fracture initiation strain (ε*fi*). The fracture initiation strain is presented in detail in the following section and it is the focus of the study. As the accumulated plastic strain equals fracture initiation strain, the damage parameter equals one and it is shown by "FRACTURE" in Figure 5a.

$$D = \int\_0^{\overline{\varepsilon}} \frac{d\overline{\varepsilon}}{\overline{\varepsilon}\_{fi}} \tag{8}$$

$$
\overline{\sigma}\_{D} = \begin{cases}
\overline{\sigma} & D < 1 \\
q(\eta\_{\nu}T, D\_{\nu}D\_{+}) \cdot \overline{\sigma} & 1 < D < D\_{+} \\
f(\eta\_{\nu}T) \cdot \overline{\sigma} & D\_{+} < D
\end{cases} \tag{9}
$$

$$q(\eta\_{\prime}T, D, D\_{+}) = \left(\frac{1 + f(\eta\_{\prime}T)}{2} - \frac{1 - f(\eta\_{\prime}T)}{2}\tanh\left(a\frac{2D - D\_{+} - 1}{D\_{+} - 1}\right)\right) f(\eta\_{\prime}T) \tag{10}$$

$$f(\eta, T) = \begin{cases} \tanh(-\sqrt{3}\mu\_i \eta) & T < T\_S \\ p(\eta, T) & T\_S < T < T\_E \\ 1 & T\_E < T \end{cases} \tag{11}$$

$$p(\eta, T) = \tanh\left(-\sqrt{3}\mu\_i \eta\right) + \left[1 - \tanh\left(-\sqrt{3}\mu\_i \eta\right)\right] \left(\frac{T - T\_S}{T\_E - T\_S}\right) \tag{12}$$

The modeling of the flow stress behavior after the fracture point is defined by "AFTER FRACTURE" in Figure 5a. The flow stress curve is modified based on the loading conditions, η, and temperature, *T*, as shown in Equation (8) using the flow stress modification factor, *f*(η, *T*). In an earlier development of Childs et al. [16], *f*(η, *T*) was assumed to be a constant. With further development [26], *f*(η, *T*) is defined as a function of stress triaxiality and temperature, as shown in Equation (8) and Figure 5c and presented in detail in the following sections.

#### *3.1. Fracture Initiation Strain*

The need for the fracture initiation strain and its use in the damage factor is presented in the previous section. The fracture initiation strain is modeled as a function of stress triaxiality (η), lode angle parameter - θ , temperature (*T*), and strain rate -. ε , as shown in Equation (13).

$$
\varepsilon\_{fi} = f(\eta, \overline{\theta}) \cdot \lg(\dot{\varepsilon}) \cdot h(T) \tag{13}
$$

The function, *f* η, θ , is used to define the loading conditions. The stress triaxiality (η) defines the varying loading conditions in two-dimensional (2D) and the lode angle parameter, θ, extends the model to loading in three-dimensional (3D) space. In the case of the 2D cutting process (orthogonal cutting process), which falls under the plane strain condition, the lode angle parameter is zero. The stress triaxiality parameter defined by, η, as <sup>σ</sup>*<sup>m</sup>* <sup>σ</sup> defines the varying loading conditions from pure compression to combinations of shear/compression and shear/tension to tension, as shown in Figure 6.

**Figure 6.** Fracture initiation strain predicted using Shear compression specimen under quasi-static loading, Shear specimens and compression testing as a function of stress triaxiality and fitted using an exponential function.

In order to obtain the fracture initiation strain under different conditions of η (−1 to 1), specimens of varying geometrical shapes are to be used [28]. Interestingly, the fracture initiation strain for variants of C45E steel has been studied for nearly 75 years [29]. With the advances in material testing, the variants of the C45E steel's fracture strain has been studied. Autenrieth et al. [30] obtained the fracture strain as a function of stress triaxiality using torsion and torsion-tension tests. These tests provide stress triaxiality from zero to positive stress triaxiality conditions. Recently, the same fracture strain as a function of stress triaxiality for conditions that range from negative to positive has been obtained using the Shear-compression disk specimen by Karp et al. [31]. In this study, the latter two fracture strain models identified as Autenrieth fracture strain data and Karp fracture strain data are used to evaluate their influence on chip segmentation prediction. The difference in the fracture

strain between the two models can be attributed to the difference in processing history and the stress triaxiality evolution before fracture.

To implement the fracture strain in the finite element model, the exponential function used in the Bai–Wierzbicki damage model [32] is used and it is shown in Equation (14).

$$
\epsilon\_{fi-X}(\eta) = D\chi\_1 e^{D\_{X2}\eta} \tag{14}
$$

The Autenrieth fracture strain data are fitted using the parameters (*DX*<sup>1</sup> = 0.2339 and *DX*<sup>2</sup> = −1.035) and the Karp fracture strain data are fitted using the parameters (*DX*<sup>1</sup> = 0.7 and *DX*<sup>2</sup> = −0.9).

The function *g* -. ε is used to model the influence of strain rate. *h*(*T*) is used to model the influence of temperature on the fracture initiation strain. In this study, *g* -. ε and *h*(*T*) are obtained from the JC damage fracture model, as shown in Equation (15) and Equation (16) and *D*<sup>3</sup> = 0.0018 and *D*<sup>4</sup> = 0.58 are obtained from the literature [33].

$$\log\left(\dot{\varepsilon}\right) = \left(1 + D\_3 \ln\left(\dot{\varepsilon}/\dot{\varepsilon}\_0\right)\right) \tag{15}$$

$$h(T) = \left(1 + D\_4 T^\*\right) \tag{16}$$

#### *3.2. Flow Stress Modification Factor*

As previously mentioned, the flow stress is modified, as shown in Figure 5a using the flow stress modification factor, *f*(η, *T*) shown in Figure 5c. The flow stress in the 'AFTER FRACTURE' region is modeled while using a steady-state flow stress modification factor defined by Equation (9) and a transient flow stress modification factor. The steady-state is defined by *D* > *D*+ and the transient state is defined by *D*+ > *D* > 1. In this study, *D*+ is defined to be 1.25, as suggested by Childs et al. [26].

In the steady-state, the flow stress modification is a function of stress triaxiality and it is defined by Equation (12) below a certain temperature, *TS*, as shown in Figure 5c. A negative stress triaxiality condition (e.g., η = −1) characterizing compression, the flow stress is modified only slightly due to incompressibility. A positive stress triaxiality condition, e.g., η = 1 characterizing tension, the flow stress is drastically reduced to zero, similar to necking in tensile testing. At temperatures above *TE,* the material is assumed to be healed and continue to flow plastically with the flow stress defined by σ, as shown in Equation (8). Between temperatures *TS* and *TE*, the flow stress curve is defined using the tanh function, as shown in Equation (11). The transient flow stress modification factor is defined as Equation (9) by multiplying the steady-state flow stress modification factor with a tanh function of *D* and *f*(η, *T*).

The values for μ*i*, *TS*, and *TE* used in this work are 1 ◦C, 600 ◦C, and 700 ◦C, respectively. The values for *TS* and *TE* based on the strain hardening transition behavior observed from the Gleeble tests.

#### **4. Experimental Investigation of Chip Formation in Orthogonal Turning**

The orthogonal turning process has been carried out in a normalized C45E steel tube with a diameter of 150 mm with a thickness of 3 mm, determining the uncut chip width. The material hardness is measured and an average hardness of 220 HV is recorded with an average grain size of 13.5 μm. The tubular workpiece reduces grain size variation influence. All of the experiments were carried out under dry cutting conditions. The cutting tool material is H13A grade carbide with TiCN coating. The cutting velocity is set at 150 m/min. The rake angle is varied as −5◦, 5◦, 10◦, and 20◦, and the feed is varied as 0.05 mm.rev<sup>−</sup>1, 0.1 mm.rev−1, 0.15 mm.rev−1, 0.25 mm.rev−1, 0.4 mm.rev−1, and 0.6 mm.rev−1. The cutting forces and feed forces are obtained through the Kistler dynamometer attached to the tool turret. The sample chips were collected for all cutting conditions and they have been used to quantify chip segmentation, as shown in Figure 7. The experimental investigation is described in detail in previous work [34]. The chips have to be plotted with varying magnification to capture the chip segmentation features at very low feed rate conditions (e.g., 0.05 mm.rev<sup>−</sup>1) and similarly at very high

feed rate conditions (e.g., 0.60 mm.rev−1). This approach with varying magnification is carried out when the chips from simulations are also presented in the following sections.

**Figure 7.** Chip morphology obtained through experimental investigation under varying rake angle and feed at a constant cutting speed of 150 m/min. with the boundary differentiating between continuous and segmented chip. Note: Image magnification not to scale.

#### **5. FE Modeling of Chip Formation in Orthogonal Turning Process**

The FE simulation of the cutting process was carried out using a commercial finite element software, Thirdwave Advantedge [35]. The software has been specifically built ground up for metal cutting simulations while using a dynamic explicit coupled Thermo-elastoplastic Lagrangian formulation [36].

The main requirement for metal cutting simulation is the ability for adaptive remeshing to account for the large plastic strains in the primary and secondary deformation zones. The simulations are carried out in 2D. The software provides the ability to customize the adaptive remeshing parameters. The adaptive remeshing parameters control the mesh transition in the tool vicinity (primary deformation zone) from coarse to fine and also the mesh transition from fine to coarse (chip mesh). The software through the mesh refinement and coarsening factors automatically control the conditions for adaptive remeshing. With predicting chip segmentation being the main aim of the study, mesh refinement factor of 6 (1-coarse → 8-fine) and mesh coarsening factor of 3 (1-fine → 8-coarse) is chosen.

A further increase in mesh refinement factors leads to computation time to increase by a factor of 3, making it practically unviable. Besides, the suggested minimum element size is set at 0.01 mm. The friction coefficient of 0.5 is used for all cutting conditions. The software provides the possibility to input the custom material model through a FORTRAN Subroutine, which and has been employed in the study to implement the modified JC model. Two different solution algorithms are supported by the software: Newton method and the secant method. At this stage, the secant method is used, as the method does not require the derivative implementation with the compromise of a slower convergence rate.

Simulations are carried out for all rake angles (−5◦, 5◦, 10◦ and 20◦) and feed from 0.05 mm.rev−<sup>1</sup> to 0.6 mm.rev<sup>−</sup>1. The relief angle and the cutting-edge radius are kept constant with the values 7◦ and 30 μm, respectively, both in the experimental conditions and simulations. Simulations are run with the JC model and MJC model combined with the damage modeling approach. Within the MJC model + damage model framework, the influence of two fracture initiation strain model's (Autenrieth ε*fi* model and Karp ε*fi* model) capability to evaluate the transition from continuous chip to segmented chip is studied.

#### **6. Results**

The FE simulations were carried out using two different material models (JC and MJC). With the MJC+ damage model, two different parameter sets of fracture initiation strain models' ability to predict the continuous chip to segmented chip transition are evaluated. The FE simulations were run till a steady state was achieved, and the chip morphology was obtained. The chip morphology was plotted in the chip chart form. Further on, the continuous chip–segmented chip boundary is identified in the chip chart. The cutting forces predicted by the material models are also presented.

#### *6.1. Validation of Material Model under Continuous Chip Formation Conditions*

This section presents the MJC model validation in the simulation of the chip formation process. Experimental results with continuous chips are used to avoid the damage model influence. The rake angle of 10◦ was used with feeds of 0.05 mm.rev−<sup>1</sup> to 0.6 mm.rev−<sup>1</sup> and Figure 8 presents the cutting forces.

**Figure 8.** Cutting force predicted by the JC model and modified the JC model for a constant cutting speed of 150 m/min and rake angle of 10◦.

The experimental result showing the increase in cutting force with feed is well established within the metal cutting with the influence of uncut chip thickness increase. The cutting forces predicted by the JC model and the MJC model have a similar trend. The MJC model improves the prediction for all of the cutting conditions quantitatively. This improvement in cutting force prediction as compared to the JC model is attributed to the material testing used and the ability of the MJC model to incorporate the material behavior with improved accuracy. The highest discrepancies in quantitative prediction for both JC and MJC models are at the maximum chip thickness conditions. The MJC model's cutting force prediction error can be attributed to the extrapolation that was carried out with the fitting of the MJC model at very high strain rates. Further improvement has to be carried out, among other things, by improving the friction models.

#### *6.2. Experimental and FE Modeling of Continuous Chip–Segmented Chip Transition for Varying Rake Angle and Feed*

The chips obtained from experimental investigation for varying rake angle and feed rates at a constant cutting speed of 150 m/min are plotted in the form of chip chart in to evaluate the ability of the different fracture initiation strain models in chip segmentation prediction. Figure 7 shows that, at constant cutting speed, continuous chips are produced for rake angles 10◦ and 20◦. With a rake angle of 5◦, the cutting process is in a transition zone. Continuous chips are produced for feed of 0.05 mm.rev−<sup>1</sup> and 0.10 mm.rev−<sup>1</sup> and segmented chips are produced for feed from 0.15 mm.rev−<sup>1</sup> to 0.6 mm.rev−1. For a rake angle of <sup>−</sup>5◦, the produced chips are segmented for all feeds. At a constant feed of 0.40 mm.rev<sup>−</sup>1, a change in the rake angle from 5◦ to 10◦ changes the chips from being continuous to segmented.

Similarly, at a constant rake angle of 5◦, the change in feed rate from 0.10 mm.rev−<sup>1</sup> to 0.15 mm.rev−<sup>1</sup> leads to continuous to segmented chip. This leads to the rake angle of 5◦ being a transition zone. The experimental results clearly show that chip segmentation/continuous chip is stable under certain cutting conditions (−5◦, 10◦, and 20◦) and in transition mode under certain conditions (5◦). It is also seen that, in the transition zone, the feed also plays a role in chip segmentation. This states that minor changes in the cutting zone to be highly sensitive to the machining output. This would lead to challenges in the process capability in a production environment.

#### *6.3. Prediction of Chip Segmentation Using MJC Model and Two Di*ff*erent Fracture Initiation Strain Models*

Initial simulations were run with the JC flow stress model [37] and the JC fracture model [38] reported for AISI 1045 steel to predict chip segmentation. The continuous chip to segmented chip transition predicted by the JC flow stress-JC fracture model, as shown in Figure 9, with the JC fracture parameters that were obtained from the literature [38]. The JC flow stress–JC fracture model is unable to predict the influence of the rake angle and, at the same time, predicts lower chip segmentation intensity when compared to experimental investigation. With the inability of the JC flow stress–JC fracture model established, the flow stress behavior is modified while using the damage model to improve chip segmentation prediction. The chip morphology predicted by the MJC model with the damage model using the two fracture initiation strain models is presented here.

**Figure 9.** Continuous chip to segmented chip transition prediction predicted by the JC flow stress model in combination with the JC Fracture model. Note: Image magnification not to scale.

#### *6.4. Autenrieth Fracture Initiation Strain Model Predicted Chip Segmentation Continuous Chip Transition*

Figure 10 shows the transition of the chip morphology that was predicted by the Autenrieth fracture initiation strain model and MJC material model. The continuous chip to segmented chip transition is predicted with reasonable accuracy. At a constant feed of 0.6 mm.rev−1, the model can predict the segmented chip for rake angles: −5◦ and 5◦ and the transition to continuous chip with a rake angle of 10◦. The model can predict the transition from continuous chip to segmented chip as the feed is increased from 0.05 mm.rev−<sup>1</sup> to 0.1 mm.rev<sup>−</sup>1. At a feed of 0.05 mm.rev−1, a continuous chip is predicted for both negative and positive rake angle. A qualitative comparison of chip segmentation between the experimental investigation and the Autenrieth fracture initiation strain model prediction shows that the level of chip segmentation intensity is lower when compared to the experimental results. This leads to a further quantitative evaluation in terms of chip segmentation intensity become counterproductive. Chip segmentation intensity has been predicted with improved accuracy in literature but only for a constant rake angle and varying cutting speeds [39]. The models in these studies have shown cutting speed's influence and not the rake angle's influence on chip segmentation.

**Figure 10.** Continuous chip to segmented chip transition prediction predicted by the MJC flow stress model in combination with the Autenrieth Fracture initiation strain model. Note: Image magnification not to scale.

#### *6.5. Fracture Strain Through Shear Compression Disk (SCD) Experiments Predicted Chip Segmentation Continuous Chip Transition*

Figure 11 shows the transition of the chip morphology predicted by the fracture initiation strain that was obtained by Karp et al. [31] while using shear compression disk specimen. The model predicts segmented chips for all rake angles for feeds above and including 0.1 mm.rev−1. For the feed of 0.05 mm.rev−1, the segmented chip for a negative rake angle is not predicted correctly compared to experimental results. Qualitative evaluation of the chip segmentation intensity when compared to the experimental investigation is relatively low for a negative rake angle. On the other hand, the chip segmentation frequency is higher qualitatively compared to the Autenrieth fracture initiation strain model predicted chip segmentation frequency.

**Figure 11.** Continuous chip to segmented chip transition prediction by the fracture initiation strain obtained Shear compression specimen. Note: Image magnification not to scale. Note: Image magnification not to scale.

#### **7. Discussion**

The FE simulation of the machining process's ability to predict chip segmentation is dependent on the accuracy in which the material behavior is modeled to a large extent, as shown by Fernandez-Zeliaia et al. [40]. The other well-known influencing factors are friction behavior and thermal behavior input. Within flow stress behavior, the initial yield stress, the strain hardening behavior, the fracture initiation, and the material behavior after fracture are essential inputs. For the material under study, C45E steel, strain hardening behavior, and thermal softening behavior are more influential when compared to strain rate hardening [41]. The MJC model's ability to accurately predict three different strain hardening behavior as a function of temperature, therefore, systematically improves the accuracy of FE modeling of chip formation. The material behavior at the strain, strain rate, and temperature occurring in the chip formation conditions is vital in accurately predicting the chip morphology. In this study, the material flow stress is initially validated while using the flow stress that was obtained from compression testing through Gleeble tests (Figure 3). Although the temperatures at which the material behavior is validated through Gleeble tests are comparable, the strains and strain rates are lower by orders of magnitude. The material model validation through cutting process simulation is carried out by avoiding the cutting conditions where chip segmentation is not observed. The MJC model's DSA incorporation, which is active in temperatures occurring at the primary and secondary deformation zones, is an important outcome of this work. The ability to incorporate DSA keeping all other things constant has been shown to improve the cutting forces prediction accuracy, as shown in Figure 8. The fracture initiation models with the JC model have not been able to predict the chip segmentation boundary. This shows the influence of the MJC model on the improved prediction of chip segmentation boundary.

The damage behavior modeling involves fracture initiation strain and the flow stress after the fracture initiation strain. The results of Figure 9, Figure 10, and Figure 11 show that the continuous chip to segmented chip transition and chip segmentation intensity is sensitive to the fracture initiation strain input. The Autenrieth fracture initiation strain model better predicts the transition between the continuous chip and segmented chip when compared to the Karp fracture initiation strain model. Simulations run with the JC material model combined with the fracture initiation strain models have shown poor results in terms of prediction of the chip segmentation (Figure 9). This shows that the ability of the fracture initiation strain models to predict chip segmentation is dependent on the components of the flow stress behavior, i.e., initial yield stress and strain hardening behavior. The strain hardening behavior and the resulting temperature increase influences the fracture initiation strain. The strain rate difference and the initial yield stress variation between the two experimental conditions (for the Autenrieth fracture initiation strain model and Karp fracture initiation strain model) have been reported as being possible causes for the significant variation in the fracture initiation strain. The significant fracture initiation strain variation correspondingly leads to significant variation in the continuous chip to segmented chip transition and the chip segmentation intensity prediction in the finite element models that were developed in this study.

The experimental investigation (Figure 7) shows that the chip segmentation at a constant cutting speed is influenced by tool geometry and chip thickness. In this study, the normal rake angle is the tool geometry parameter modified, and its influence on the chip formation is studied. With the normal rake angle being modified from negative to positive rake angle under certain uncut chip thickness (feed) conditions, the chip formation process is transformed from a continuous chip to a segmented chip, as observed with the segmented chip boundary.

With the normal rake angle modification, the mechanical loading of the workpiece ahead of the tool is modified with varying levels of shear and compression. The mechanical loading is not uniform along the shear plane from the cutting edge to the free surface in the primary deformation zone. The stress that was observed by the workpiece material close to the cutting-edge is highly compressive when compared to the free end of the primary deformation zone, where the stresses are a combination of shear and compression. The workpiece material that is close to the cutting-edge rounding is mainly

under the stress state defined by combined shear and compression defined by a large negative stress triaxiality. The workpiece material on the free end of the primary deformation zone has a relatively lower negative stress triaxiality; still, they are in the negative stress triaxiality conditions. Figure 6 shows that the fracture initiation strain in the negative stress triaxiality condition to be sensitive and varies based on testing conditions. This fracture initiation strain sensitivity is attributed to be the main reason for chip segmentation to be sensitive to minor changes in cutting conditions. At a constant chip thickness and cutting speed, the tool geometry influences this mechanical loading and it is observed to influence the onset of chip segmentation. Another critical factor that influences the material behavior is the material's history. From a machining perspective, the influence of the previous cut on the strain profile has been shown by Childs et al. [26]. The tool edge geometry controls the influence of the previous cut on the stress profile on the workpiece surface, tool wear, and tool vibration. The tool edge geometry could also influence the transition zone between the continuous chip to segmented chip. In this work, the influence of the previous cut is not taken into account and it could be a source of improvement in chip segmentation prediction in future work.

It is clearly understood that the uncut chip thickness, in addition to the rake angle, influences the transition from continuous to segmented chip at constant cutting speed. This is understood from the results, showing that, at a constant rake angle, the chip thickness influences chip segmentation. The chip thickness increase leads to the heat being concentrated in the primary shear zone. This concentration of heat leads to a segmented chip, as predicted in the presented FE simulations. On the other hand, chip thickness increase needs to be significant in this case at a rake angle of 5◦ up to 0.6 mm.rev−<sup>1</sup> to produce segmented chips.

#### **8. Conclusions**

In this study, a new modified JC model is developed based on the Voyiadjis–Abed–Rusinek [8] constitutive modeling approach. The modified JC model with the fracture modeling approach by Childs et al. [26] has been implemented for the first time in the FE framework for the chip formation simulation. The influence of the fracture initiation strain model on the continuous to segmented chip transition prediction has been successfully demonstrated. The following conclusions are made from the study.


**Author Contributions:** Conceptualization, A.M.D.; Formal analysis, A.M.D.; Supervision, P.V.S., T.B. and M.E.; Validation, A.M.D. and P.V.S.; Writing–original draft, A.M.D.; Writing–review & editing, P.V.S., T.B. and M.E. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Sandvik Coromant AB and the Knowledge Foundation through the Industrial Research School SiCoMaP, Dnr 20110263, 20140130.

**Acknowledgments:** The authors would like to thank T.H.C. Childs for his support and guidance with the finite element implementation of the damage model and for invaluable discussions.

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

#### **Nomenclature**



#### **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*
