*Article* **System Performance and Process Capability in Additive Manufacturing: Quality Control for Polymer Jetting**

#### **Razvan Udroiu \* and Ion Cristian Braga**

Department of Manufacturing Engineering, Transilvania University of Brasov, 29 Eroilor Boulevard, 500036 Brasov, Romania; braga.ion.cristian@unitbv.ro

**\*** Correspondence: udroiu.r@unitbv.ro; Tel.: +40-268-421-318

Received: 19 May 2020; Accepted: 3 June 2020; Published: 4 June 2020

**Abstract:** Polymer-based additive manufacturing (AM) gathers a great deal of interest with regard to standardization and implementation in mass production. A new methodology for the system and process capabilities analysis in additive manufacturing, using statistical quality tools for production management, is proposed. A large sample of small specimens of circular shape was manufactured of photopolymer resins using polymer jetting (PolyJet) technology. Two critical geometrical features of the specimen were investigated. The variability of the measurement system was determined by Gage repeatability and reproducibility (Gage R&R) methodology. Machine and process capabilities were performed in relation to the defined tolerance limits and the results were analyzed based on the requirements from the statistical process control. The results showed that the EDEN 350 system capability and PolyJet process capability enables obtaining capability indices over 1.67 within the capable tolerance interval of 0.22 mm. Furthermore, PolyJet technology depositing thin layers of resins droplets of 0.016 mm allows for manufacturing in a short time of a high volume of parts for mass production with a tolerance matching the ISO 286 IT9 grade for radial dimension and IT10 grade for linear dimensions on the Z-axis, respectively. Using microscopy analysis some results were explained and validated from the capability study.

**Keywords:** additive manufacturing; material jetting; polymer; machine capability; process capability; statistical process control; quality; variability; tolerance grade

#### **1. Introduction**

The applications of additive manufacturing (AM) to industry have developed from rapid prototyping (RP) and rapid tooling to rapid manufacturing (RM). Additive manufacturing will revolutionize future manufacturing as a key technology in the implementation of the new industrial revolution, Industry 4.0 [1].

Nowadays, the AM processes defined by ISO/ASTM 52900-15 [2] standard are starting to find applications in industry. An industrial additive manufacturing system [3] should have six main components: design, pre-processing, manufacture, post-processing, quality control, and maintenance. The performance of AM systems is an important task to be estimated for the production of parts in an industrial process. There are many AM processes [2] and technologies associated with them, as follows:


Polymers have become very popular as materials for AM, being used in most of the AM processes and targeting a variety of applications [4]. The performance of all the AM systems that are connected to the mentioned AM processes should be analyzed in order to determine their capability to produce parts for the industry. The artifacts or test pieces are primarily used to quantitatively assess the geometric performance of AM systems [5]. Additionally, the AM product characterization needs other tests such as feedstock materials characterization, mechanical tests [6,7], and surface texture characterization [8–10]. The test artifacts are intended to reveal the strengths and weaknesses of different additive manufacturing techniques. Furthermore, they allow the comparison of the performances of different AM systems and the same AM system over time [11]. According to [5], three main characteristics, accuracy, resolution, and surface texture, of the AM systems can be estimated based on some standardized artifacts. Thus, seven artifact geometries have been proposed as follows: linear and circular artifact artifacts to test the accuracy, pins, holes, ribs, and slots artifacts to test the resolution, and the surface texture artifact to test the texture of the surfaces.

Current geometric dimensioning and tolerancing (GD&T) standards have been developed based on the capabilities of traditional manufacturing processes as subtractive manufacturing and formative manufacturing methodologies [12]. New GD&T standards need to be implemented for the different AM processes that use a large variety of materials (plastics, metals, composites, ceramics etc.). The effect of process parameters on the mechanical and geometric performances of polylactic acid (PLA) based composite materials was investigated in [13–15] and the results show the great potential of 3D printed composites in different applications. The physical and chemical properties of polymers relevant to dimensional accuracy require different evaluation and quantification of geometrical tolerances in comparison to metal materials. The tolerance standards applicable for metal parts, therefore, cannot be adopted for plastic structures or can only be applied to a very limited extent.

In the production process, the variations and fluctuations in the manufacturing accuracy are influenced by many factors such as machines, workpiece, methods, people, and environment, etc. The inherent fluctuations have less impact on product quality [16]. The abnormal variations have a large impact on product quality [17]. The most known methods used to control and reduce the manufacturing process variation are the statistical process control, measurement system analysis, six sigma method, and Taguchi's design of experiments [18].

Statistical process control (SPC) uses statistical methods in quality control to monitor, maintain, and improve the capability of manufacturing processes to assure product conformance [16,19]. Akande et al. [20] analyzed quality characteristics of strength, bending stiffness, density, and dimensional accuracy of parts built by the SLS process using SPC control charts. They concluded that SPC ensures consistency in product quality for long term production.

Any quality control process needs to quantify, first, the machine capability (short-term study or machine performance) in one continuous production run and manufacturing process capability (long-term study) in series production [19,21]. Measurement process capability provides the evidence for conformity or nonconformity with specification according to ISO 14253:2017 [22].

Experimental and theoretical studies have been developed in order to characterize the performance of AM processes and have particularly focused on quality control in additive manufacturing. Additionally, the standards focused on AM systems are under development. The use of AM processes in mass production depends on the part quality. Some issues are the inconsistency of AM repeatability and reproducibility that have not been solved yet for all the AM processes. Singh et al. [18] analyzed the repeatability of acrylonitrile butadiene styrene (ABS) replicas built by the FDM process and chemical vapor smoothing, but the repeatability variation and the appraiser variation were not calculated. Baturynska [23], using statistical analysis, attempted to improve the dimensional accuracy of the parts built by polymer powder bed fusion. She developed linear regression models to predict the value of

the thickness, width, and length of rectangular specimens and to compensate for the shrinkage effect. The material jetting process allows time between the jetting of each layer of material to relieve internal stresses [24]. George et al. [24] reviewed the accuracy and reproducibility of 3D printed medical models from polymers, using material extrusion (FDM), powder bed fusion (SLS), binder jetting, and material jetting. They concluded that regular testing of the accuracy of AM systems and preventive maintenance are necessary steps for quality assurance. Preißler et al. [25] investigated a process capability for a fused filament fabrication (FFF) process using PLA material, based on a customized pyramid object manufactured in 25 samples. The results for a 30 mm dimension and tolerance of ±0.2 mm through the quality control chart shows that the process was not in the statistical control.

Singh [26] investigated the process capability of the linear dimensions of a prismatic component built by PolyJet technology from an EDEN 260 machine. The results of this study suggested that that process lies in the ±4.5 sigma limit with regard to the dimensional accuracy of the chosen specimen. However, the variability of the measurement system was not performed and the number of 16 parts used to determine the process capability was too low, according to the capability standards [27]. Kitsakis et al. [28] investigated the IT (International Tolerance) grades for the dimensions of eight samples printed with deposition layers of 30 microns on the Objet Eden 250 3D printer, and they assigned the IT11 grade for it. The variability of the measurement system used in this study was not accomplished. Yap et al. [29] investigated the design capability and manufacturing accuracy of the PolyJet 3D printing process on an Objet500 Connex3 PolyJet printer using artifacts with customized features and concluded that the accuracy of the parts printed in glossy mode was better than that in matte finishing, but the minimum clearance gap for parts was obtained in a matte finish. Minetola et al. [30] evaluated the dimensional accuracy of three AM systems for polymeric materials using the ISO IT grades of an artifact from the GrabCAD library, building two replicas. They concluded that a smaller layer thickness provided higher dimensional accuracy of the part dimensions.

From the literature survey, the results are as follows:


Having benefits in terms of cost reduction and shorten of the time-to-market in products, the implementation of polymer-based AM technologies within production depends on the process capability and control.

The main aim of this article was to define a methodology for statistically analyzing the AM system performances and AM process control. A case study regarding the EDEN 350 AM system and polymer jetting process was conducted to validate the proposed basic methodology.

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

#### *2.1. New Methodology for Statistical Quality Tools in AM Production*

The main objectives of the new methodology in AM are to define statistical quality tools based on standards for the assessment of the variability of the measurement system, the additive manufacturing repeatability, AM system capability or AM system performance, and AM process capability. This methodology includes experiments, statistical analysis, and results interpretation. SPC tools are used to provide the mean of identifying possible changes in the process [16].

The new methodology in AM consists of a preparatory step, followed by six main steps, as shown in Figure 1.

**Figure 1.** Flowchart of the proposed methodology/procedure named Quality Tools in AM (QT-AM).

The preparatory step defines the AM process specification as follows: STL (Standard Triangulation Language) or AMF (Additive manufacturing file) file conversion and its accuracy, feedstock material properties, artifact type, build orientation and position, the sample size of specimens, and the manufacturing and the post-processing plan. According to the ISO/ASTM 52901:2017 standard [31], the part definition made by AM, for a purchase purpose, should include the following characteristics: part geometry, tolerances, surface texture, feedstock material, build orientation, acceptable imperfections or deviations, and process control information (e.g., repeatability). The main characteristics of part geometry can be defined as a digital file containing the 3D model and a part engineering drawing.

In traditional manufacturing, the specific requirements (dimensions, tolerances, surface finish, material, etc.) of the 3D model and drawings are set based on standards according to product material. Thus, ISO 286 is usually used for parts made of metal [32] and DIN 16742 for plastic parts [33]. In AM, the general tolerances for linear dimensions are specified according to the general standard ISO 2768-1 [34], based on the ISO/ASTM 52901:2017 recommendation. The surface texture or surface finish of the part should be specified by a maximum value.

Feedstock material properties need to conform to the suppliers' specifications. Artifact manufacturing should be undertaken according to a manufacturing plan (layer thickness, build strategy, process temperature). The CAD model of the artifact is converted to a STL file format. The conversion parameters used within different CAD software as well as any maximum deviation (chord height and angular tolerance) should be chosen correlated to the 3D printing layer thickness. Where supports cannot be avoided, a supporting strategy should be documented. It includes the support geometry, support material, the removal technique, and the specific post-processing treatments. The support material can be made from the same material as the artifact (model) material or can be different. The application of support structures or support material should be minimized on the critical features of the part.

The amount of variability induced in measurements by the measurement system itself should be determined before any capability study is performed (Figure 1). The measurements were performed using a grade "A" measurement method according to the ASTM 52902-19 standard [5]. Therefore, for simple and inexpensive measurements commonly available in a shop floor, a digital caliper was used.

In the second phase of the methodology, the critical capability assumptions are analyzed, as shown in Figure 1. AM machine capability or AM system performance has the main purpose of the checking of existing 3D printers, objective arguments in case of 3D printer defects, and findings for target specifications when purchasing a new 3D printer. 3D printer/AM system capability and 3D printing/AM process capability studies are determined within the third and fourth steps. Capability is the ability of a system, or process, to realize a product that will fulfill the requirements for that product. Capability conditions under which the process is evaluated include the following, according to the ISO 22514-1:2014 standard [35]:


Capability analysis should be carried out for a new or changed production process and then over time to control the process according to the standard ISO/TS 16949 [36]. Capability analysis is summarized in indices that show the system's ability to meet its requirements. Machine and process capabilities provide results on how well a machine and a process performs in relation to defined tolerance limits. These two branches differ because they are determined in different conditions, but principally similar indices are calculated. The target capability indices commonly used in the automotive industry are greater than 1.67, which corresponds to a safety or critical parameter for a new process [16]. The quality condition is excellent if the capability indices are between 1.67 and 2 [16,37].

A quality inspection through a microscopy study is performed in the fifth step of the methodology. Optical micrographs were performed using a Zeiss O-Inspect (Carl Zeiss Industrielle Messtechnik, Oberkochen, Germany) multi-sensor measuring machines.

#### *2.2. Process Specifications. Materials, Artifact, and Manufacturing Method*

In this work, a part used in the pre-production of plastic parts has been selected as the benchmark. The part presents similar geometric basic features as the circular artifact shown in Figure 2. The circular artifact consists of a circular upper surface and a steep lower surface ending with a sharp edge (Figure 2). Two critical dimensions in terms of assembly and functionality of the artifact, the height H = 12 mm, and the diameter D = 14.5 mm, have been selected for the machine and process capability study. A fine tolerance class of ±0.1 mm was selected, taking into account the ranges of nominal lengths between 6–30 mm according to the ISO 2768-1 standard [34].

**Figure 2.** (**a**) Views and section view of the artifact. (**b**) The part used in the pre-production.

SolidWorks version 2013 software (Dassault Systèmes, Massachusetts, MA, USA) was used to design the 3D model and to generate the STL file. The 3D model of the part was converted into a STL file, which is the input file format of the Objet EDEN 350 PolyJet machine (Stratasys, Rehovot, Israel) [38]. The STL file conversion tolerances were set to a deviation of 0.01 mm and an angular tolerance of 4 degrees.

Feedstock materials used in this study were Objet VeroBlue RGD840 resin used as the model material and FullCure 705 as the support material [39]. The composition of the Objet VeroBlue RGD840 resin consists of an acrylic monomer, urethane acrylate oligomer, epoxy acrylate, and photo-initiator. FullCure 705 resin is made of an acrylic monomer, polyethylene glycol 400, propane-1, 2-diol, glycerol, and photo-initiator. The main properties of the Objet VeroBlue RGD840 material are shown in Table 1 [39]. Characteristics may vary if different orientations of specimens and test conditions are applied [6,7].


**Table 1.** Objet VeroBlue RGD840 properties [39].

The orientation of the specimens on the build tray affects how quickly, efficiently, and qualitatively they will be manufactured by the AM system [40]. Additionally, within the PolyJet process, the orientation of parts has an influence on the quantity and where the support material is used. The circular specimens were printed in a standing up position on the build platform, as shown in Figure 3. It is advantageous to print a circular model that has holes standing up on the build platform, so support material does not fill the holes [38]. Additionally, if a circular model is lying down on the build platform and printed in glossy printing mode, then the surface quality is affected by some errors [41]. The experimental roughness (Ra) values for the PolyJet material jetting process are specified according to the finish type as follows: for matte finish in the range of 0.5–15 μm, and for the glossy finish in the range of 0.5–4 μm [42]. The dimensional accuracy and the quality of the surface of a circular artifact built in standing up position are not significantly influenced by the orientation and positioning on the build platform.

**Figure 3.** (**a**) Layout of the EDEN 350 build platform illustrating the 50 parts patterned in an array. (**b**) Detail of the printed specimens on the build platform.

An Objet EDEN 350 PolyJet system was used tomanufacture the specimens. Based on drop-on-demand (DOD) inkjet technology [43], the PolyJet system deposits layers of resin droplets of 0.016 mm thick. It levels each deposited resin layer and hardens it using ultraviolet (UV) light. During the process, the print heads and the photopolymer resins are heated at around 72 ◦C. The print heads were vacuumed at 6.2 atm. The experiments were performed under a controlled laboratory temperature of 20 ◦C and relative humidity of 30%. PolyJet 3D printers only use a solid infill pattern on parts. A different infill type can be added in the design stage of the CAD model, but the part's interior will likely be filled with support material in the printing process. A solid infill pattern was used for all of the samples.

The build platform preparation (Figure 3), STL model slicing, and G-code generation were performed using the Objet Studio client/server software (Objet Geometries, Rehovot, Israel). The specimens were 3D printed in a glossy finish style. Only the bottom surfaces of the specimen were affected by the support material. The support material was removed with a pressure water jet from the bottom surface of the 3D printed specimens.

The density of the printed material was determined using the Archimedes density method [44,45] by calculating the volume of five specimens, in addition to determining the mass of the parts using a precision scale. The results showed a mean density measured of the printed material of 1.15 g/cm3.

One batch of 50 parts was 3D printed for the AM system capability study, and three batches each containing 50 parts for the AM process capability study. A batch of 50 artifacts was manufactured in 1 h and 40 min, using 78 g of model material and 54 g of support material.

#### *2.3. The Variability of the Measurement System*

Within both manufacturing processes and quality systems, there is variation. All measurement data had some degree of variance or errors. A robust statistical process control (SPC) process requires accurate data to have the greatest impact on product quality. The percentage of variance due to the measurement system has to be determined. The measuring system can be affected by various sources of variation, called factors [46]: measuring instruments, operators, measuring method, specifications (the engineering tolerance), and parts or specimens.

The variability of the measurement system was determined by Gage repeatability and reproducibility methodology. Repeatability is due to measuring instrument variation and reproducibility is due to operator (appraiser) variation. Gage R&R study was performed using the analysis of variance (ANOVA) method [47]. The ANOVA Gage R&R method estimates:


The measurement system used in the analysis included:


Using Minitab 19 software (Minitab, Ltd., Coventry, United Kingdom) [48], a worksheet for Gage R&R analysis was created and the order of the measurements for each operator was imposed. The total sample size was 60 measurements.

#### *2.4. System and Process Capability for PolyJet Technology*

The Gauge R&R should be proven before the capability analysis. Two critical assumptions need to be considered when performing the machine and process capability analyses with continuous data, namely, the process is in statistical control, and a normal distribution of the process is required. A process is considered stable if its output is within the predictable limits. In order to assess whether or not a process is in statistical control, it uses control charts [16,19].

Short-term performance studies are typically performed on machines where parts are produced consecutively under repeatability conditions and the sample size produced is at least 50 workpieces to be manufactured in one shift [27]. 3D printer capability or AM system capability is used to assess the quality and performance of a single AM machine. The AM system capability was evaluated within the following conditions:


The quality of the production processes is measured by establishing some characteristics and monitoring the long-term capability of its 3D printing process capability or AM process capability is a long-term study on a stable process that indicates the performance quality of the 3D printing process. The AM process capability can be evaluated within the following conditions:


D and H dimensions of the parts were measured using the Mitutoyo 500-196-30 digital scale caliper (Mitutoyo Corporation, Kawasaki, Japan). System and process capability is determined by calculating the capability coefficients described in Equation (1). The lower specification limit (LSL) and upper specification limit (USL) are the targets set for the process. The potential machine and process capability indices (Cm, Cp) represent the number of times the process spread fits into the tolerance interval. A high potential capability index does not guarantee that the process is close to the target value, which is why the position of the process spread in relation to the tolerance interval is determined by calculating the critical capability machine/process index (Cmk/Cpk).

$$\begin{cases} \mathbf{C}\_{i} = \frac{lISL - LSL}{\mathbf{x}\_{j} \cdot \mathbf{s} \rho\_{i} \mathbf{s} \mathbf{s} \mathbf{s} \mathbf{s} \mathbf{s}} - \mathbf{x}\_{i, 0.135 \mathbf{s} \mathbf{s}}}{\begin{pmatrix} lSL - \mathbf{x}\_{\mathbf{s} \mathbf{0}} \mathbf{x} \\ \mathbf{x}\_{i \mathbf{q} \mathbf{s} \mathbf{s} \mathbf{0} \mathbf{s}} - \mathbf{x}\_{i \mathbf{q} \mathbf{s}} \mathbf{x} \end{pmatrix}}, \text{ i.e.} \begin{cases} \mathbf{c} = \left\{ m - machine, p - process \right\} \\ \mathbf{C}\_{\mathbf{t} \mathbf{y} \mathbf{c} \mathbf{t}} = 1.67(1.33) \end{cases} \tag{1}$$

The location and dispersion were calculated using the M11,6 method according to the ISO 22514-2: 2017 standard [21]. Subscripts 1 and 6 refer to equations for calculating the estimator for the location and dispersion, respectively. This means that the arithmetic mean of the values is used for the location being assumed, and externally tested the normal distribution, and the distance between the edges 0.135% and 99.865% for dispersion. The reference interval of the product characteristic is bounded by the 99.865% distribution quantile, and the 0.135% distribution quantile. The length of the interval is X99,865%−X0,135% [35]. X50% represents the 50% distribution quantile.

The results of the short-term and long-term capabilities were analyzed based on the requirements from the SPC Reference Manual, from Automotive Industry Action Group (AIAG) [19]. Destra software [49] was used to perform the capability study. The capability can be evaluated graphically by drawing capability histograms and capability plots. The requirements for indices Cm, Cmk, Cp, and Cpk demand a minimum value of 1.67 for all of them.

#### *2.5. Capable Tolerance Specification for PolyJet Technology*

Tolerance specification (tolerance, lower and upper limits) for the dimensions of the 3D printed circular part was chosen based on the general tolerances standards [34] and plastics molded parts tolerances [33]. A tolerance of ±0.1 mm was selected. Based on this specification, the AM system and process capability for PolyJet technology was calculated. The capability indices were compared with a capability target index of 1.67.

Rather than estimating the process capability for a particular tolerance, a capable tolerance and its limit deviations were calculated based on a target capability index. The target capability index was set to 1.67. The index of the process K was calculated using Equation (2) and describes the level by which the process is off target value and represents an appropriate measure of process centering [37,50]. The lower (LSLT) and upper (USLT) specification limits of the capable tolerance were calculated based on Equation (3). Upper limit deviation (ULD) and lower limit deviation (LLD) from nominal size were then determined based on Equation (3).

$$K = \frac{(ULL + LSL) - 2x\_{\text{man}}}{USL - LSL} \tag{2}$$

The process mean is positioned between the midpoint of the specifications and one of the required limits if 0 < |K| < 1. |K| > 1 indicates that the process mean is situated outside the required limits.

$$\begin{aligned} \text{if } K > 0 \text{ then } \begin{cases} \begin{aligned} &LSL\_{T} = X\_{50\%} - \text{C}\_{\text{PK}}(X\_{50\%} - X\_{0.135\%}) \\ &LLD = T\_{\text{m}} - LSL\_{T} \\ &LISL\_{T} = LSL + \text{C}\_{\text{P}}(X\_{998.665\%} - X\_{0.135\%}) \\ &LLD = USL\_{T} - T\_{\text{m}} \\ &ILSL\_{T} = X\_{50\%} + \text{C}\_{\text{PK}}(X\_{998.665\%} - X\_{50\%}) \\ &LLD = ILSL\_{T} - T\_{\text{m}} \\ &LSL\_{T} = ILSL - \text{C}\_{\text{P}}(X\_{998.665\%} - X\_{0.135\%}) \\ &LLD = T\_{\text{m}} - LSL\_{T} \end{cases} \end{aligned} \end{aligned} \tag{3}$$

The capable lower limit deviation and capable upper limit deviation were determined based on the relations LLDC < LLDT and ULDC > ULDT. The capable tolerance is calculated as follows: Tc = ULDC – LLDC. A confirmatory analysis of AM process capability was performed using the determined capable tolerance.

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

#### *3.1. The Variability of the Measurement System*

The variance components (VarComp) compare the variation from each source of measurement error to the total variation. In these results, the %Contribution column (Table 2) shows that the variation from Part-To-Part for H and D dimension was 99.46%/99.12%, which is much larger than the total Gage R&R, which was 0.54%/0.88%. Thus, the largest part of the variation was due to the differences between parts. This means that the measurement system can reliably distinguish between parts.


**Table 2.** Variance components for the characteristic diameter and height.

<sup>1</sup> Diameter, <sup>2</sup> Height.

The measurement system variation compared to the total variation is shown in Tables 3 and 4. The total Gage R&R equaled 7.33%/9.37% of the study variation for the H and D dimensions. In order to evaluate the capability of the measurement system to evaluate parts versus specification, the values %Tolerance are used, these values being calculated for each characteristic as the ratio between the study variation for each source and the process tolerance.


**Table 3.** Gage evaluation for diameter (D).

**Table 4.** Gage evaluation for height (H).


The repeatability variation and the reproducibility variation, which shows the equipment variation (EV) and the appraiser variation (AV), respectively, were lower than 10%. Based on the requirements specified in the MSA 4 [46], the measurement system can be accepted. The number of distinct categories was greater than five (Tables 3 and 4), resulting in an acceptable measurement system [46].

The variability results of the measurement system are graphically provided in Figures 4 and 5. In the Components of Variation graph, the %Contribution from Part-To-Part is larger than that of the total Gage R&R. Thus, much of the variation is due to differences between parts. The R Chart by Operator shows that Operators measured parts consistently. In the Xbar Chart by Operator, most of the points were outside the control limits. Thus, much of the variation is due to differences between parts.

**Figure 4.** Gage R&R (ANOVA) report for height (H): (**a**) Components of Variation graph; (**b**) R Chart by Operator; (**c**) Xbar Chart by Operators; (**d**) By Parts graph; (**e**) By Operators graph; (**f**) Parts \* Operators interaction.

**Figure 5.** Gage R&R (ANOVA) report for diameter (D): (**a**) Components of Variation graph; (**b**) R Chart by Operator; (**c**) Xbar Chart by Operators; (**d**) By Parts graph; (**e**) By Operators graph; (**f**) Parts \* Operators interaction.

The By Operator graphs (Figures 4e and 5e) show that the differences between operators were smaller than the differences between parts. In the Parts \* Operators Interaction graphs (Figures 4f and 5f), the lines were approximately parallel and the p-value for the Parts \* Operators interaction was 0.779/0.195 for the H and D dimensions. This indicates that no significant interaction between each Parts and Operators exists.

The Gage R&R result shows that for the height as well as for the diameter, a variation due to the measurement system was much lower than the part-to-part variation, as a result, the next studies could be based on measurements.

#### *3.2. System Performance of Objet EDEN 350 PolyJet*

First, both critical assumptions for performing the machine capability (system performance) analyses were graphically checked. The control charts from Figures 6 and 7 show the manufacturing process information for all 50 measurements of the D and H measured dimensions. The distributions were stable over the period of study, as shown in Figures 6 and 7.

**Figure 6.** Control chart for the short-term capability study of the height dimension (H).

**Figure 7.** Control chart for the short-term capability study of the diameter dimension (D).

The dimensional values lay within the LSL and USL, indicating that the process is in statistical control for both dimensions. A normal distribution was detected based on the Anderson–Darling normality test. Figure 8 shows the histograms of the individuals and the distribution models.

**Figure 8.** Histogram of the individuals and the distribution model for short term capability study: (**a**) height (H); (**b**) diameter (D).

The measurements were located near the upper specification limit (USL) of the diameter (D) and near the lower specification limits (LSL) of the height (H), respectively. This graphics show the shape of the subgroup frequencies.

The numerical results of the machine capability analysis are shown in Tables 5 and 6 for both dimensions of the circular specimen, where Tm is the tolerance center, T is the tolerance of the characteristic, n is the sample size, *xmin* the minimum value of the characteristic, *xmax* is the maximum value of the characteristic, *x*mean is the median of all values, StDev is the standard deviation of all individuals, X0.135% is the 0.135% distribution quantile, X50% is the 50% distribution quantile, and X99.865% is the 99.865% distribution quantile. The potential and the critical capability index both showed three values (Figure 9) that specify the two-sided 95% confidence interval for the respective capability index: lower confidence limit, estimator, and upper confidence limit.

**Table 5.** Machine capability analysis for the H dimension.




**Figure 9.** Machine capability analysis report for: (**a**) dimension H; (**b**) dimension D.

The requirements for indices Cm and Cmk were met for the D dimension (Figure 9b). Based on the measured parts, the critical capability index was lower than the target for the characteristic H. Therefore, the 3D printer capability was not proven (Figure 9a).

#### *3.3. Process Capability of PolyJet*

The control charts of the process capability for both dimensions of the diameter and height of the circular specimen are shown in Figures 10 and 11. The Xbar-S control charts for the subgroups with the sample size of five pieces were chosen to check if the process variation was in control. The mean data and standard deviation data showed that none of the points were outside the control limits (UCL, upper control limit, LCL, lower control limit), and the points displayed a random pattern. Thus, the process variation was in control.

**Figure 10.** The control charts for the long-term capability study of height dimension (H): (**a**) mean data; (**b**) standard deviation data.

**Figure 11.** The control charts for the long-term capability study of diameter dimension (D): (**a**) mean data; (**b**) standard deviation data.

The model distribution of the data for the dimensions H and D showed a normal distribution, as shown in Figure 12. The entire production process was stable and controllable.

**Figure 12.** Normal probability plot graph for the long-term capability study: (**a**) height dimension (H); (**b**) diameter dimension (D).

The location of the process distribution (Figure 13) was near the upper tolerance limits for the dimension of diameter (D) and near the lower tolerance limits for the dimension of height (H), respectively.

**Figure 13.** Histogram of individuals and the distribution model for the long-term capability study: (**a**) height dimension (H); (**b**) diameter dimension (D).

The numerical results of the process capability analysis are shown in Tables 7 and 8 for both dimensions of diameter and height. The standard deviation of height was slightly larger than that of diameter.

**Table 7.** Process capability analysis for the H dimension.




Based on the requirements, the target for Cpk is very often established at a minimum of 1.67. Some of the industry manufacturers accept even lower values of 1.33 for Cp and Cpk [51]. Even so, the result for the height characteristic in terms of Cpk was lower than 1.67 or 1.33. The requirements for indices Cpm and Cpk were met for the diameter dimension (Figure 14), but were not met for the height dimension.


**Figure 14.** Process capability analysis report: (**a**) height dimension (H); (**b**) diameter dimension (D).

#### *3.4. Capable Tolerance and Its Limits Deviation for PolyJet Process*

The calculation of the capability indices was based on the location and dispersion of the characteristic value with respect to the specified tolerance. xmean indicates the location of the process. It can be observed from the process capability graphics (Figure 13) that the xmean was lower than the nominal value for the H characteristic and higher for the D characteristic, respectively.

Capable tolerance and its limit deviations were calculated based on a target capability index of 1.67 for both dimensions of height and diameter. The index of process K was calculated, and the results showed the value of 0.62 for the height and −0.38 for the diameter, respectively. The capable lower limit deviation and capable upper limit deviation were determined for both dimensions.

The capable limit deviations of the circular specimen were found as ULD = max{ULDD, ULDH} and LLD = min{LLDD, LLDH}, where the subscripts H and D represent the characteristic height and diameter, respectively. The results show that the capable lower limit deviation and capable upper limit deviation of the circular artifact were LLD = −0.13 mm and ULD = +0.09 mm, respectively. The capable tolerance interval of the circular artifact was TC = 0.22 mm.

A confirmatory analysis of AM process capability was performed using the determined capable tolerance of the circular artifact. The process capability result was "too high" (Cpk>1.67), as shown in Figure 15. Thus, the requirements were met.

**Figure 15.** Process capability analysis report based on the capable tolerance of the circular artifact: (**a**) height dimension (H); (**b**) diameter dimension (D).

#### *3.5. Determination of Tolerance Grade (ISO IT grade)*

Tolerance grades indicate the degree of accuracy of manufacture. Since IT grades provide guidance on how precise a manufactured feature of a particular size should be, they can be used to compare different manufacturing processes [52]. The lower value of IT Grade implies a better dimensional accuracy. The IT Grade was calculated for 50 specimens of the circular artifact, based on the standard ISO 286 specifications [32]. The dimensional accuracy and IT Grade depend on the size of the feature. Two dimensions of the circular artifact, the height and the diameter were analyzed. These sizes were within the ISO basic size range of (10–18 mm).

$$n\_i = \frac{|D\_N - D\_{M\_i}|}{0.45 \Big(\sqrt{D\_{\min} D\_{\max}}\Big)^\frac{1}{\epsilon} + 0.001\sqrt{D\_{\min} D\_{\max}}}, \mathbf{i} = \{1, \ldots, 50\} \tag{4}$$

The relative magnitude of each IT (International Tolerance) Grade is calculated relative to the standard tolerance unit i. The standard tolerance unit is i = 1.083 μm for the ISO basic size range of Dmin = 10 mm to Dmax = 18 mm. The tolerance unit "n" was calculated using Equation (4), where 'D' is the geometric mean of the ISO basic size range; DN is the nominal dimension; and DM is the measured dimension. Table 9 shows the IT grades for the height (H), and the diameter (D) of the circular specimen. The tolerance grades were determined based on the tolerance unit n.


**Table 9.** International tolerance grades for the circular artifact.

The results show that the International Tolerance Grade of the height dimension was IT10 for 86% of specimens (Figure 16). A significant variation in the IT Grade percent was detected for the diameter dimension with a 58% IT10 distribution, as shown in Figure 16. The IT Grade, which represents the dimensional accuracy of the AM systems for each interval of ISO basic sizes can be determined using the same procedure.

**Figure 16.** IT grades for the characteristics of height and diameter of the circular artifact within the size range (10–18 mm).

#### *3.6. Quality Inspection through Microscopy Analysis*

A microscopy analysis study was performed to conduct a quality inspection of the critical features of the circular workpiece. The dimension H was measured between the upper and lower surface of the specimen, and dimension D on the upper surface of the workpiece, respectively. The quality of these surfaces should be investigated. There was no support material deposited on the upper surfaces of the model printed in glossy mode, only on the bottom surfaces, as shown in Figures 17 and 18.

**Figure 17.** The lower surface of the circular artifact manufactured by the Objet EDEN 350 PolyJet. (**a**) Lower surface detail affected by support material; (**b**) lower edge detail in a matte finish.

**Figure 18.** The upper surface of the circular artifact manufactured by the Objet EDEN 350 PolyJet. (**a**) Upper surface in glossy mode; (**b**) upper edge detail in a glossy finish.

The lower surface of the circular specimen was affected by the material support. Small pieces of support material were detected on the lower surface of the specimen, even if the specimen was cleaned with a pressure water jet after 3D printing (Figure 17).

The quality of the lower and upper edges of the circular artifact may influence the artifact height dimension. A good quality surface without material defects was detected on the upper edge of the specimen (Figure 18b). The edges of the upper surface printed in glossy mode were rounded, as shown in Figure 18b. Additionally, the sharp edge of the lower surface affected by the support material was rounded, as shown in Figure 17. This edge roundness can explain why the distribution of the height measurements was located near the lower tolerance limits and was lower than the nominal value.

For both the glossy and matte finishes, microscopic investigations on the lateral surface of the circular artifact were conducted on a perpendicular and parallel direction to the X-axis (Figures 19 and 20). It can be seen as a clean and smooth surface in the X-axis direction (Figure 19a) for a glossy finish. Rough areas were detected on the perpendicular direction to the X-axis (Figure 19b). The steep surface affected by the support material indicates a homogenous material that contained small inclusions of the FullCure 705 support material (Figure 20).

**Figure 19.** Microscopic views (1:1 <sup>×</sup> <sup>10</sup>−<sup>4</sup> m scale) of the lateral surface of the artifact in the glossy finish area located: (**a**) parallel to the X-axis (0◦); (**b**) parallel to the Y-axis (90◦).

**Figure 20.** Microscopic views (1:1 <sup>×</sup> 10−<sup>4</sup> m scale) on the lateral surface of the artifact, in the matte finish area affected by the support material located: (**a**) parallel to the X-axis (0◦); (**b**) parallel to the Y-axis (90◦).

#### **4. Conclusions**

This paper contributes to the characterization of the dimensional accuracy, repeatability, system performance, and process capability of polymer-based AM processes and systems. The methodology used for quality control in additive manufacturing allows the polymer-based AM processes to be implemented in production. Additionally, this methodology can be used as the AM's machine monitoring technique.

The following conclusions are drawn:


These were located near the upper tolerance limits for the dimension of diameter (D) and near the lower tolerance limits for the dimension of height (H), respectively.


Further research is required for the capability characterization of other AM machines and processes using different types of materials.

**Author Contributions:** Conceptualization, R.U.; Data curation, R.U.; Formal analysis, R.U. and I.C.B.; Investigation, R.U. and I.C.B.; Methodology, R.U.; Project administration, R.U.; Writing—original draft, R.U. and I.C.B.; Writing—review & editing, R.U. All authors have read and agreed to the published version of the manuscript.

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

**Acknowledgments:** We hereby acknowledge the research platform PLADETINO (Platform for Innovative Technological Development) from Transilvania University of Brasov (grant no. 13/2008), and CNCSIS 78 for partly providing the infrastructure used in this work.

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

#### **References**


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

## *Article* **3D Direct Printing of Silicone Meniscus Implant Using a Novel Heat-Cured Extrusion-Based Printer**

**Eric Luis 1, Houwen Matthew Pan 2, Swee Leong Sing 1, Ram Bajpai 3,4, Juha Song <sup>2</sup> and Wai Yee Yeong 1,\***


Received: 30 March 2020; Accepted: 25 April 2020; Published: 1 May 2020

**Abstract:** The first successful direct 3D printing, or additive manufacturing (AM), of heat-cured silicone meniscal implants, using biocompatible and bio-implantable silicone resins is reported. Silicone implants have conventionally been manufactured by indirect silicone casting and molding methods which are expensive and time-consuming. A novel custom-made heat-curing extrusion-based silicone 3D printer which is capable of directly 3D printing medical silicone implants is introduced. The rheological study of silicone resins and the optimization of critical process parameters are described in detail. The surface and cross-sectional morphologies of the printed silicone meniscus implant were also included. A time-lapsed simulation study of the heated silicone resin within the nozzle using computational fluid dynamics (CFD) was done and the results obtained closely resembled real time 3D printing. Solidworks one-convection model simulation, when compared to the on-off model, more closely correlated with the actual probed temperature. Finally, comparative mechanical study between 3D printed and heat-molded meniscus is conducted. The novel 3D printing process opens up the opportunities for rapid 3D printing of various customizable medical silicone implants and devices for patients and fills the current gap in the additive manufacturing industry.

**Keywords:** 3D printing; additive manufacturing; material extrusion; silicone; meniscus implant

#### **1. Introduction**

The meniscus, which is a pair fibrocartilaginous cushion within the knee joint, acts as a weight bearing cushion, lubricating device and knee stabilizer. It is mainly C-shaped and triangular shaped in the horizontal and cross-sectional sections, respectively. Its relatively centralized acellular extracellular matrix and limited peripheral vascular supply from the meniscocapsular plexus severely restricts the meniscus regenerative capability when damaged or torn. The compressive modulus of 0.03 MPa is weakest at its supero-medial portion of the posterior horns and these regions are the commonest sites of tear and injury [1].

Treatment options for meniscus tears include direct repair, replacement with either scaffolds or implants and partial or total meniscectomy (removal of the damaged meniscus tissues). Repair methods for meniscal tears with sutures, pins and fibrin glue were used with limited success due to poor vascular supply. Replacement with meniscus scaffolds made of polycaprolactone (PCL) or collagen have produced equivocal results owing to inferior compressive mechanical properties. Partial meniscus implants which include collagen meniscus implant (Menaflex) and Actfit were used with limited success and have not yet gained approval from FDA. Total meniscus implant NuSurface is the only total meniscus implant currently undergoing Sun Clinical Trial 2 [2,3].

Silicone is used in biomedical manufacturing for biliary stents [4], cochlear implants [5], peripheral nerve sheaths [6], breast [7–10], chin, nose [11], testicular and hand implants [12], interphalangeal joint replacement implants [13], knee prosthesis [14], cardiopulmonary bypass tubing [15] and silicone membranes [16]. Recently, microfluidics [17], lab-on-a-chip [18] and biosensors [19] have also been manufactured with silicone. Despite these, the manufacturing of meniscus implants using silicone was limited. Meanwhile, most information in literature are on the general properties of silicone resins but little on the rheological properties. Takahashi et al. studied the rheological behavior of several silicone resins in the un-crosslinked state and found that the temperature dependence of the viscoelastic behavior can be described by a Williams–Landel–Ferry (WLF) equation [20].

Most of the implants are produced by indirect molding technique, whereby the mold is first 3D printed or machined and the silicone is subsequently poured into the mold and left to cure. This indirect molding technique, both time consuming and costly, also posed technical challenges when customized implants of different geometries or hollow silicone parts need to be manufactured.

Industrial silicone 3D printing began in 2015. The current state-of-the art in silicone 3D printing includes (a) using UV light of 365 nm to cure UV-sensitive silicone, as they are being deposited layer-by-layer by an inkjet printer (ACEO, Wacker, Munich, Germany), (b) freeform reversible embedding technique using a 2-part A/B silicone where the part A-silicone catalyst-cross-linker is extruded into a bath of part B-silicone (Fripp Design, Sheffield, UK) [21], (c) using a progressive cavity pump to extrude moisture-cured one part Oxime silicone elastomer [22], (d) a hybrid ink-jetting and UV-extrusion techniques with a printing speed which is 20 times faster [23], (e) a combination stereolithographic with low-one-photon-polymerization (LOPP) [24] and (f) 3D printing of PDMS elastomer in a hydrophilic Carbapol bath support [25].

The first successful direct 3D printing of heat-cured medical grade silicone meniscus implants is reported in this study. To date, no direct 3D printing of medical silicone implants has been described. First, optimum curing temperatures and times are obtained from the rheological characterization of the silicone resins [26,27]. This is followed by parametric optimization of nozzle diameters, nozzle temperatures and bed temperatures, using a range of parametric values, to print out standard ASTM cylindrical blocks, T-bones and finally meniscus implants. Computational fluid dynamics (CFD) simulation is used to study the heat transfer within the silicone resin across the heated nozzle and Solidworks is used to study the heat gradient distribution across the silicone meniscus implant. The results were validated against experimental values using standard heat probes. The new direct 3D printing technology provides an avenue for rapid 3D printing of various customizable medical silicone implants and devices for patients and fills the current gap in the additive manufacturing (AM) industry.

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

#### *2.1. Silicone Material Specification*

The silicone material used in this study is a water-white translucent, two-part platinum-catalyzed silicone elastomer that cure upon exposure to room temperature or can be accelerated by heat (Smooth-On®, Smooth-On Inc, Macungie, PA, USA). Ecoflex30 and Ecoflex50 are chosen for its safety and biocompatible profiles. They are mixed 1A:1B by weight or volume and cured at room temperature with negligible shrinkage. Their low viscosities permit easy mixing and de-airing. The properties of Ecoflex30 and Ecoflex50 are shown in Table 1.


**Table 1.** Ecoflex sample technical data (from safety data sheet provided by Smooth-On Inc).

#### *2.2. Experimental Setup*

The system consists of six key components: (1) a motion control platform based on an open-source 3D printer (CoreXY, Guangzhou, China), (2) a syringe-pump extruder with accuracy of 0.05 μm to dispense the material, (3) a double barrel syringe connected to a static mixer using tubing (TIUB05C, SMC Corporation, Tokyo, Japan), (4) stainless nozzles with tip size between 20 and 21 gauges (inner diameter from 0.5 mm to 0.6 mm), (5) heating elements on the printer nozzles and platform with their controllers, as shown in Figure 1.

**Figure 1.** Experimental setup for heat cure extrusion-based additive manufacturing (AM): legends: (1) motion control platform, (2) extruder, (3) static mixer, (4) printhead, (5) heating elements.

Parts A and B of the Ecoflex ® resins were added into the double barrel syringe with mix ratio of 1:1 by volume prior to fabrication. The resins were subsequently passed through the static mixer before extrusion. 3D printing software (Repetier-Host, Hot-World GmbH & Co. KG, Willich, Germany) was used to control the 3D printer while Slic3r was used for tool path generation.

#### *2.3. Process Parameters and Experiment Design*

In this study, several key process parameters, such as the solution flow rate (*Q*), nozzle-to-substrate distance (*h*), were fixed to simplify the optimization process. Before the 3D printing of meniscus implants, several experiments were conducted to find a suitable range of the process parameters. The analysis of the 3D printed meniscus fabricated under varied temperature of the heating elements was done to find out the best heat cured meniscus.

For the first part, inner nozzle diameter (*d*) and the print speed (*v*) were analyzed through fabricating several cylinders and T-bones. Meanwhile, checks were conducted on the layer-layer deformation and whether there is void inside the specimens. Specimens with complicated structure were fabricated to verify whether these ranges of the process parameters are feasible for the meniscus fabrication. In the second part, several printing experiments were conducted to find out the optimized combination of the temperatures of the heated platform (*T*1) and the heating elements on the print head (*T*2). Lastly, comparison between the compressive properties of the 3D printed meniscus and heat molded meniscus was done.

#### *2.4. Rheology Test*

Rheology test was conducted to measure the gelation in silicone since it is instantaneous using a TA Texas Instruments DHR2 rheometer (New Castle, DE, USA). The testing geometry used is a 40 mm steel Peltier parallel plate which is fitted after instrument inertia calibration, bearing friction correction and geometry inertia calibration. The rheometer was operated in parallel-plate mode.

For the experiments, the two components Parts A and B of each silicone resin Ecoflex30 and Ecoflex50 (cross-linker with catalyst and polymer) were manually mixed according to the manufacturer's datasheet. Characterization of the rheological behavior of the silicone in triplicates was done in both steady and oscillation modes at different isothermal temperatures (30 ◦C to 80 ◦C) according to DIN 53529 which is similar to ASTM D 5289. The selected temperatures were controlled by the Peltier device with a cooling pump at an accuracy of ±0.1 ◦C. For the dynamic oscillatory rheology investigation, the samples were exposed to increasing strain rate (0.00 to 3.00 s<sup>−</sup>1) at a constant frequency (1 Hz) to determine the linear viscoelastic range of the samples [28].

The curing times, complex viscosities and complex moduli of the silicone at various temperature are determined using the temperature ramp mode from 30 ◦C to 80 ◦C with ramp rate of 5 ◦C/min. A normal force control of 2 N, with tolerance of 1.75 N is applied on the torsion geometry with gap change limit maintained at 550 μm. The control strain is set at 0.025% and constant angular frequency was set at 3 rad/s. The data are collected and plots of G', G" and tan (delta) against temperature and time are performed [29].

The frequency sweep mode is used to determine the viscoelastic properties of the silicone resin in steady shear and dynamic-mechanical experiments in a frequency range between 0.01 and 100 rad/s. The heat-treated material was first ground and subsequently pressed to dense discs at 70 ◦C like the untreated material. Before being measured, all samples were dried in a desiccator at room temperature for 12 h.

#### *2.5. Surface Morphology of 3D Printed Silicone Meniscus Implant*

Nikon SMZ 1000 (Tokyo, Japan) was used for 6× magnification surface visualization of the overall layer-by-layer step morphology of the 3D printed implants and Eclipse LV 150 (Nikon Metrology NV, Leuven, Belgium), was used for 50× magnification cross-sectional visualization of the 3D printed implants.

#### *2.6. Statistical Analysis*

Descriptive statistics (mean ± standard deviation) for all quantitative variables are obtained. Normality of continuous variables were tested by Kolmogorov–Smirnov and Shapiro–Wilk tests. An exploratory analysis of the correlation between meniscus length and width with nozzle and print bed temperatures using Spearman rank correlation due to small sample sizes is done. The univariate linear regression is used to evaluate the effect of nozzle and print bed temperatures on the meniscus length and width. Statistical analyses were carried out using Stata version 14.2 (StataCorp, College Station, TX, USA).

#### *2.7. Computational Fluid Dynamics (CFD) Simulation Studies for Heated Printer Nozzle*

The nozzle design and schematic is shown in Supporting Information, Figure S1. Using Autodesk computational fluid dynamics (CFD) 2017 software (Autodesk Inc, San Rafael, CA, USA), the input parameters for materials used are provided in Supporting Information, Table S1.

After applying the material properties, it is necessary to provide the boundary conditions. The flow simulation works only in a closed region. Thus, there is a need to create a computational domain and a fluid sub domain [30]. Boundary conditions are set as follows: Surface Temperature—60 ◦C, Pressure—0 Pa Gage, Volume Flow Rate—50 mm3/min, Temperature—25 ◦C.

#### *2.8. Solidworks Simulation Studies for 3D Printed Silicone Meniscus Implant*

Solidworks 2018 (Dassault Systems Solidworks Corporation, Waltham, MA, USA) was used to study the temperature gradient distribution in the 3D printed silicone meniscus implants. Analysis using the stimulation module can reduce the number of product development cycles and time to market (TTM), optimize the meniscus design and reduce cost by testing the model on the computer rather than physical tests. From the simulation task pane, thermal study and the material properties are first entered: Elastic modulus (0.290075 <sup>×</sup> 106), Poisson's ratio (0.47), shear modulus (0.0029 <sup>×</sup> 106 Psi), mass density (143.58 lb/ft3), tensile strength (0.797708 ksi), compressive strength (4.35113 ksi), yield strength (0.758547 ksi), thermal expansion coefficient (540 <sup>×</sup> <sup>10</sup><sup>−</sup>6), thermal conductivity (4.77 BTU ft/h), specific heat (1.00602 BTU/lb.P). Subsequently, this is followed by automatic generation of mesh of finite elements for FEA by the program. To set the boundary conditions, the perimeter of each of the eleven layers sliced by the Slic3r program was used to estimate the amount of time taken to print each layer and the estimated perimeter in millimeters and printing time taken in seconds are expressed below: From the first layer to the final (eleventh) layer, the perimeters are 273, 274, 165.7, 151.8, 138.9, 125.8, 112.3, 97.9, 82.0, 64.0 and 43.0 mm requiring estimated printing times of 330, 330, 183, 167, 153, 138, 124, 108, 90, 70 and 47 seconds, respectively. The perimeters of each layer was used to estimate the total amount of time taken for each layer to be printed as shown in Supporting Information, Table S2. Experimental temperature values at every layer laid down are measured with heat probe model HT with K-type thermistor.

#### **3. Results**

#### *3.1. Rheology*

#### 3.1.1. Steady Shear Flow Study

The uninterrupted extrusion of silicone is crucial for the 3D printing of the meniscus implants. The extrusion is dependent on their viscous behavior under different shear rates. The rheological fingerprints of both Ecoflex30 and Ecoflex50 silicone samples at the temperature 50 ◦C are shown in Figure 2. Both Ecoflex30 and Ecoflex50 showed non-Newtonian behavior. The shear stress of Ecoflex30 is proportional to the shear rate at lower shear rates. Its viscosity was around 3 Pa·s, the so-called zero-shear rate viscosity η0. At shear rates of about 0.5 s<sup>−</sup>1, the Ecoflex30 viscosity starts to decrease significantly, until it starts to level off at higher shar rates of 3 s<sup>−</sup>1.

Similar behavior for PDMS solutions is reported [31]. Figure 3 shows a sharp decrease in the viscosity of both Ecoflex30 and Ecoflex50, from about 200 to 20 Pa·s and from 500 to 5 Pa·s, respectively, over the shear rate range 0.01 to 5 s−<sup>1</sup> At very high shear rates the viscosity may again become independent of shear rates, approaching the infinite-shear rate viscosity η∞. Polymer degradation becomes a serious problem before sufficiently high shear rates can be obtained which made η<sup>∞</sup> not usually measurable. The behavior of Ecoflex30 under steady shear flow in the range above 3.5 s−<sup>1</sup> is shear thinning or pseudoplastic behavior. The decreasing viscosity with increasing shear rates is utilized for the current extrusion nozzle design.

**Figure 2.** (**a**) Ecoflex30 and (**b**) Ecoflex50 complex viscosity and complex modulus versus shear stress.

**Figure 3.** (**a**) Ecoflex30 and (**b**) Ecoflex50 complex viscosity and shear stress versus strain rate.

#### 3.1.2. Transient Shear Stress Response

For rheologically complex materials, understanding the transient shear behavior is important. To examine the transient behavior of Ecoflex30 and Ecoflex50, their shear stress response against time were measured. Both Ecoflex30 and Ecoflex50 show Bingham pseudoplastic shear thinning behavior on the first 250 seconds, as shown in Figure 2.

Over the tested non-Newtonian range of shear rates 0.5 < *ç*ù < 10 s<sup>−</sup>1, shear thinning behavior was recorded for all temperatures. The shear rate has significant effect on the flow behavior of Ecoflex30. For example at 55 ◦C, the viscosity of Ecoflex30 decreases from 3000 Pa·s at 0.05 s−<sup>1</sup> to almost 2 Pa·s at 2.5 s−<sup>1</sup> (Figure 3a) and the viscosity of Ecoflex50 decreases from 300 Pa·s at 0.05 s−<sup>1</sup> to almost 2 Pa·s at 2.5 s−<sup>1</sup> (Figure 3b). This effect represents a depression of 3 orders and 2 orders of magnitude, for Ecoflex30 and Ecoflex50, respectively, over a shear rate range of less than 3 s<sup>−</sup>1.

#### 3.1.3. Dynamic Test

The storage modulus *G*' (elastic response) and the loss modulus *G*" (viscous response) are measured using a dynamic test where oscillating stresses or strains are applied to the test samples. The total resistance versus the applied strain gives the complex modulus *G*\*.

At the start of a dynamic test, the linear viscoelastic range is defined by increasing the stress to cover a wide range. The range where complex moduli *G*\* is constant with stress is the linear viscoelastic range which indicates that the internal bonds of the sample are still intact. The linear viscoelastic range was found to be around 35 to 50 Pa range for Ecoflex30 and 100 to 150 Pa for Ecoflex50.

A frequency sweep test was carried out at the stress value 1.5 Pa to study the viscoelastic behavior Ecoflex30 and Ecoflex50. Both samples demonstrated elastic and viscous responses. Figure 4 show the elastic G' and viscous modulus *G*" for Ecoflex30 and Ecoflex50 over the frequency range 1–100 rad/s. Ecoflex30 has both higher elastic and higher viscous modulus (Figure 4a), when compared to those of Ecoflex50 (Figure 4b). Eco30 samples also demonstrated a greater elastic response than viscous response over the entire range of frequencies.

**Figure 4.** (**a**) Eco30 and (**b**) Eco50 storage modulus, loss modulus, complex viscosity vs angular frequency.

#### 3.1.4. Gelation Times

Ecoflex 30 has a gelation time of 805 s with a crossing modulus of 130 MPa when heated to 40 ◦C. This is shown in Figure 5. It has a gelation time of 185 s with a crossing modulus of 365 MPa at 50 ◦C. With the this increase in temperature, the gelation time was shortened by 10 min, the complex modulus was also lowered by from 28 mPa to 8 mPa. However, there was an increase in complex viscosity from 5 Pa·s to over 40 Pa·s.

**Figure 5.** Plot of curing times (s) of Ecoflex30 (orange line) and Ecoflex50 (blue line) versus temperatures (◦C).

At below 60 ◦C, Ecoflex50 demonstrated longer curing times when compared to Ecoflex30. Above 60 ◦C, both Ecoflex30 and Ecoflex50 have similar curing times. As the temperature increases, the complex viscosity decreases according to the temperature dependence of the viscous properties of

the silicone resins. However, at higher temperatures, the crosslinking rates dominate and increases the viscosity. The change in storage and loss modulus over time for Ecoflex30 and Ecoflex50 is shown in Supporting Information, Figure S2.

#### *3.2. Optimization of Print Speed and Nozzle Inner Diameter*

An optimized combination of flow rate and print speed is critical in determining printing outcomes. In this study, all flow rates were fixed at 1.2 ml/min. An initial simple three-layer T bone was designed and printed to determine the optimized print speed. The T-bone CAD drawings, shown as Figure 6a,b, were designed with height of 3 mm, overall width of 19 mm, overall length of 55 mm and width of narrow section of 6 mm.

**Figure 6.** Schematic illustration of T bone: (**a**) overview and (**b**) side view. (**c**) Top view images of the 3D printed T bone under varied print speed. Scale bar: 1 cm.

In this section, print speeds, varying from 10 mm/s to 30 mm/s with 5 mm/s intervals, were used to print the silicone. Other process parameters were fixed (*T*<sup>1</sup> = 40 ◦C, *T*<sup>2</sup> = 100 ◦C, *Q* = 1.2 ml/min and *d* = 0.90 mm) and 3 specimens were printed with each set of parameters. The top view images of selected 3D printed silicone T-bone fabricated under different print speeds were displayed in Figure 6c. With increasing print speeds, the silicone line diameter gradually decreases until it becomes spotty or discontinuous. At *v* = 30 mm/s, silicone droplets were deposited as the print speed was too high. On the other hand, a low print speed at v = 10 mm/s resulted in silicone overflow. The best printout was obtained with a print speed of 20 mm/s.

Subsequently, cylinders were printed to determine the optimal nozzle diameter for more complex printings. The CAD drawing of the cylinders, Figure 7a,b, has a diameter of 30 mm and height of 10 mm. In this section, seven different nozzle diameters, *d,* (0.41, 0.51, 0.60, 0.70, 0.90, 1.21 mm) were used to print the cylinders. Other process parameters were fixed at *T*<sup>1</sup> = 40 ◦C, *T*<sup>2</sup> = 100 ◦C, *Q* = 1.2 mL/min and *v* = 20 mm/s.

**Figure 7.** Schematic illustration of 3D printed cylinder (**a**) overview and (**b**) side view. (**c**) Cross-sectional images of the cylinder. Scale bar 1 cm.

Results of printed silicone cylinders are shown in Figure 7c. Using a nozzle diameter of *d* = 0.41 mm resulted in discontinuous print lines, droplets and under extrusion. Best print results, with print heights of 10 mm, were obtained with nozzle diameters of 0.51 mm and 0.6 mm, while keeping other process parameters constant. Silicone overflow due to over-extrusion were observed from *d* = 0.70 mm to *d* = 1.12 mm. Except for *d* = 0.51 mm and 0.60 mm, using all other nozzle diameters may result in under- or over-extrusion. Based on these experiments, a combination of a print speed of 20 mm/s and a nozzle inner diameter of either 0.51 mm or 0.60 mm were selected for subsequent printings. These specific process parameters will need to be modified accordingly for the printing of different specimens.

A more complicated polyhedron with four slopes was printed. The CAD dimensions of the polyhedron are shown in Figure 8a,b, which is approximately 50 mm × 40 mm × 20 mm and 10 mm deep. After slicing, this polyhedron has 20 layers and the slope starts from the 11th layer. Figure 8c shows the top view images of the polyhedron, which fabricated based on the previous process parameters (*T*<sup>1</sup> = 40 ◦C, *T*<sup>2</sup> = 100 ◦C, *Q* = 1.2 ml/min, *v* = 20 mm/s, *d* = 0.51 mm and 0.61 mm). Overall, the layers can be clearly observed on both polyhedrons and the dimensions of these two were close to the designed values. However, with d = 0.60 mm, slight overflow was observed due to over-extrusion and inadequate heating. This observation demonstrates that void-less silicone polyhedron with slopes can be reliably achieved.

**Figure 8.** Schematic illustration of a polyhedron in (**a**) overview and (**b**) side view. (**c**) Top view images of the polyhedron. Scale bar: 1 cm.

#### *3.3. Fabrication of Silicone Meniscus*

To determine the effects of temperature variation on the silicone meniscus printing, a series of experiments were conducted. The temperature of the heated platform was varied from 80 ◦C to 110 ◦C with intervals of 10 ◦C and the nozzle temperature was varied from 40 ◦C to 80 ◦C with intervals of 10 ◦C. As shown in Figure 9, several combinations of the temperatures were selected to fabricate the silicone meniscus.


**Figure 9.** Top view images of the 3D printed silicone meniscus under varied heating temperature (*T*<sup>1</sup> = nozzle temperature and *T*<sup>2</sup> = platform temperature) and inner nozzle diameter. Scale bar: 1 cm.

In general, a larger volume of silicone is extruded with a larger nozzle diameter. This extruded volume decreases with an increase in temperature and is caused by clogging of the nozzle or tubing. Clogging is particularly significant at nozzle temperatures of more than 70 ◦C and above. Nozzle temperatures of 50 ◦C and below resulted in inadequate heating, inadequate curing and overflow. With increasing height of meniscus being printed, heat conduction via the thickness of silicone meniscus is impaired and this may similarly result in silicone overflow and collapse of the printout. Therefore, a higher platform temperature of *T*<sup>2</sup> = 100 ◦C to 110 ◦C is required. From Figure 9, a combined nozzle temperature *T*<sup>1</sup> = 60 ◦C platform temperature *T*<sup>2</sup> = 110 ◦C and nozzle diameter d = 0.51 mm gave the best printout of meniscal implant.

#### *3.4. Association between Meniscus Length and Width with Nozzle and Print Bed Temperatures*

Mean meniscus length and width were calculated to be 4.08 ± 0.14 cm (range: 3.90–4.30 cm) and 2.08 ± 0.19 cm (range: 1.90–2.54 cm), respectively. Similarly, mean nozzle and print bed temperatures were calculated as 57 ± 12.52 ◦C (range: 40–80 ◦C) and 102 ± 10.33 ◦C (range: 80–110 ◦C), respectively. A strong negative correlation has been observed between meniscus length vs. nozzle temperature (rho = −0.93; *p* < 0.01); meniscus length vs. print bed temperature (rho = −0.82; *p* < 0.01); meniscus width vs. nozzle temperature (rho = −0.84; *p* < 0.01); and meniscus width vs. print bed temperature (rho = −0.83; *p* < 0.01). The linear relationship between the meniscus length and width to the nozzle and print bed temperatures is presented in Table 2 and Figures 10 and 11. The univariate regression analysis demonstrated that high amount of variability in the meniscus length and width can be explained by the nozzle and print bed temperatures independently.

**Table 2.** Univariate linear regression analysis to predict meniscus length and width using nozzle and print bed temperatures.


Adj: adjusted; S.E.: standard error.

**Figure 10.** Correlation between meniscus length and (**a**) nozzle temperature and (**b**) print bed temperature.

**Figure 11.** Correlation between meniscus width and (**a**) nozzle temperature and (**b**) print bed temperature.

#### *3.5. Surface and Cross-Sectional Morphology*

The surface morphology observed in Figure 12a,d shows relatively smooth surface with pits measuring less than 20 micrometers due to impurities and bubbling. The orderly layer-by-layer stepwise deposition is seen in Figure 12b,c, in both the posterior and anterior horns of the meniscus, respectively. Supporting Information, Figure S3 illustrate the interlayer silicone bonding with clear lamination lines.

**Figure 12.** Bright field images of (**a**) surface, (**b**) posterior horn and (**c**) anterior horn of 3D printed silicone meniscus implant. (**d**) Scanning electron microscopy (SEM) image of meniscus surface at ×200 magnification.

#### *3.6. Heated Nozzle Computational Fluid Dynamics (CFD) Simulation Studies*

The following simulation tests results illustrate the distribution of temperature, velocity and viscosity of the silicone resin along the nozzle, as shown in Figure 13a–c below, respectively. Figure 13a shows a temperature of 60 ◦C at the nozzle inlet, shown in red, where it is in direct contact with the heating block. The subsequent reduction in temperature along the nozzle tip to around 40 ◦C is shown in dark green. These simulation temperatures correlated quite accurately with the actual temperatures of thermocouple readings of aluminum heating block thermocouples and heated platforms. Figure 13b shows a relatively slow traveling speed of silicone resin at 0.5 mm/s at the nozzle inlet, shown in blue. Within the barrel of the nozzle, the velocity of the silicone increases centripetally to about 6 mm/s, shown in yellow and 8 mm/s in the central axis, as shown in orange. Figure 13c shows a relative constant viscosity throughout the nozzle from the inlet to the outlet at 3 Pa·s, shown in red.

**Figure 13.** (**a**) Temperature distribution along nozzle, (**b**) velocity and (**c**) viscosity distribution of silicone resin along nozzle.

#### *3.7. Solidworks Meniscus Implant Heat Gradient Simulation Studies*

One-convection simulation and on-off layer simulation methods were performed and the results of heat gradients of printed meniscus are shown in Figures 14 and 15, respectively. The simulation temperatures from one-convection model correlates more closely with the experimental results, inferring the adoption of this model for future 3D printing of implants. The comparison between thermal results using one-convection and on-off simulations is shown in Supporting Information, Figure S4.

**Figure 14.** (**a**) Vertical and (**b**) horizontal temperature distribution of the 3D printed silicone meniscus implant using one convection block simulation.

**Figure 15.** (**a**) Vertical and (**b**) horizontal temperature distribution of on-off layering simulation.

#### *3.8. Comparison of Compression Modulus of 3D Printed and Heat Molded Silicone Meniscus*

The student's independent *t*-test was used to compare the compressive modulus of the 3D printed silicone meniscus (0.838 +/− 0.070) MPa vs that of the heat-molded silicone meniscus (0.131 +/− 0.024) MPa. The 95% confidence interval of this difference range from −96,564.16 to 1665.17 MPa. There is no statistical difference between the two groups. The two-tail P is more than 0.05 (*p* = 0.058). The results demonstrate that the 3D printed silicone meniscus has similar compressive mechanical properties as that of the heat-molded silicone meniscus. Figure 16 shows the stress vs strain plot comparison of the 3D printed meniscus versus heat-molded silicone meniscus.

**Figure 16.** Stress vs strain plot of 3D Printed (blue line) vs heat molded silicone meniscus (red line).

#### **4. Discussions**

This study describes the first successful direct 3D printing of heat-cure silicone meniscal implant, using biocompatible and bio-implantable silicone resins. Previous successful works with silicone extrusion, using non-heat curing technology include catalyst extrusion onto a silicone bath (Fripps, Sheffield, UK], multi-materials silicone 3D printing using UV-cured silicone technology (ACEO, Wacker Cheime, Germany] and moisture-cured silicone extrusion actuator [18].

Previously, 3D printing has focused mainly on the printing of bio-models for medical education and preoperative planning and trainings, with special cases of 3D printed titanium calcaneal, spinal and dental implants. To print medical models, implants or devices, one can first use the FDA-approved MIMICS software (Materialise, Belgium] to convert the DICOM files of the patients' CT-MRI to the STL 3D printable format or an open-source Slice3r software for practice.

This new technology certainly opens up the gateway to rapidly 3D print various customizable medical silicone implants and devices for patients and fills the current gap in the AM industry, since the current AM technology have not involved direct 3D printing of medical silicone implants.

The current study has shown that by the precise control of flow rate, nozzle diameter, nozzle temperature and platform temperature, it is possible to accurately print a customized meniscal implant. The starting heat temperatures required for the resins are obtained from the calibration curves shown in the result section. These temperatures ensure adequate degree of gelation prior to extrusion and optimal degree of inter-laminar bonding with previously extruded layers. With the progression of printing, these temperatures require fine modulation with time to cater for different geometries and thicknesses of the end-products.

In addition, the univariate linear regression analysis showed that there is a close correlation between the accuracy and variability of both the lengths and widths of the 3D printed meniscus and both the nozzle and print bed temperatures. A higher optimal temperature not only reduces variability of both the printouts of lengths and widths of the meniscus but also improves the precision and accuracy of printouts.

The simulation results also reflect quite accurately the precise setting of experimental temperature of the heating block at 60 ◦C to produce a nozzle output of silicone resin at around 40 ◦C, the extrusion of which provides optimal consistency and lamination with the previous layers. In contrast to the body of the printed meniscus implant, the anterior and posterior horns of the 3D printed meniscus implant retained the highest temperatures of about 80 ◦C upon completion of printing, making these sites are not suitable for further manipulation or incorporation of micro-channels, drugs or cellular components.

To ascertain the functionality of 3D printed silicone meniscus, the compressive mechanical properties are compared to heat-molded silicone meniscus. The results confirmed that the 3D printed silicone meniscus has similar compressive mechanical properties as that of the heat-molded silicone meniscus. This shows that the new process developed has potential to replace the current heat molding process.

Although several biodegradable and biocompatible scaffolds are available on the market to reconstruct the segmental meniscus defects of previous parts, these scaffolds still need to consider the bulk material properties, structure design and functional requirements and the fabrication process also should be reproducible and reliable. 3D silicone direct printing based on extrusion technique are compared with common fabrication technologies used in meniscal tissue engineering and are tabulated in Table 3.


**Table 3.** Comparison of direct silicone print with 3D scaffold for meniscus.

#### **5. Conclusions**

A novel silicone 3D printer was successfully built for the direct 3D Printing of silicone meniscus implants which demonstrate mechanical properties as the conventionally heat-molded meniscus. The nozzle diameter, nozzle and bed temperatures were shown to be critical factors in determining the precision and accuracy of the lengths and widths of the meniscus. In this study, heated printer nozzle and meniscus implant designs were evaluated to determine the thermal distribution along the nozzle and across the meniscus implant by employing CFD and heat simulations using Solidworks.

Incorporating complex interior lattice or micro-channels and composite multi-material 3D silicone printing will be the future challenges in direct silicone 3D printing. Scaffold-based tissue regeneration based on AM technology is another promising approach to meniscus surgical treatment. In theory, the meniscal scaffold should provide appropriate biomechanical functions after implantation to shield cells from damaging compressive or tensile forces, maintain their shape integrity (without shrinkage, etc.), mechanical stability and strength on the defect area until enough host tissue was regenerated, produce mechanical stimuli to promote tissue regeneration.

Unique challenges present in silicone 3d printing are: (1) difficulty in the handling of silicone resins, (2) difficulty in printing multi-materials or different silicone resins, (3) finding suitable post-processing methods and 4) coming up with suitable standards in medical silicone 3D Printing.

Handling of silicone resins require meticulous even mixing to avoid trapping of air bubbles. Two-part resins are prone to disproportionate mixing and uneven curing. One-part resins are susceptible to moisture contamination and premature curing.

Different grades of silicone resins or different materials require different printing parameters and printing processes for optimal output. Consequently, modifications and additions to the 3D printers are necessary to achieve multi-grade silicones or multi-material printing.

Unlike post-processing methods in the 3D printing of other solid, liquid or powder substrates, silicone rubber products are highly susceptible to cuts, fissures, abrasions and lacerations while undergoing post-processing. Conventional processing methods for the above substrates cannot be applied to silicone printing. Therefore, this is a unique case where post-processing should be avoided as much as possible.

Currently no ASTM or similar standards was prescribed for medical silicone 3D printing. It is therefore imperative that one of the major areas of future works focus on the setting of standards in medical silicone 3D printing.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4360/12/5/1031/s1, Figure S1. (a) Model dimensions and (b) schematic of CFD design of heated nozzle. Figure S2. Changes in storage and loss modulus over time for (a and b) Eco30 and (c and d) Eco50 under heat curing of 40, 45, 55 and 65 ◦C. Figure S3. Bright field images of (a) topmost layer and (b) vertical cross-section of body of 3D printed silicone meniscus implant. Figure S4.. Comparison of thermal results using one-convection (above) and on-off (below) simulation. Table S1. Material Properties of Aluminum, Stainless Steel 304, Ecoflex Silicone. Table S2. Boundary conditions for printed silicone meniscus implant.) and on-off (below) simulation.

**Author Contributions:** Conceptualization, E.L. and W.Y.Y.; data curation, H.M.P.; formal analysis, R.B.; funding acquisition, W.Y.Y.; investigation, E.L.; methodology, R.B.; resources, S.L.S., J.S. and W.Y.Y.; supervision, S.L.S., J.S. and W.Y.Y.; writing—original draft, E.L.; writing—review & editing, H.M.P., S.L.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research is supported by the National Research Foundation, Prime Minister's Office, Singapore under its Medium-Sized Center Funding scheme and the NTU Start-Up Grant.

**Acknowledgments:** This research was supported by the National Research Foundation, Prime Minister's Office, Singapore under its Medium-Sized Center Funding Scheme and the NTU Start-Up Grant. The authors thank all administrative and research staff in Singapore Center for 3D Printing, Nanyang Technological University that have contributed to the successful publication of this work.

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

#### **References**


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