Parameter design of PLA/Wood Fused Filament Fabrication using Taguchi optimization methodology

This study investigates the effects of four variables during fused lament fabrication of organic biocompatible composite material, PLA with coconut our, at the ultimate tensile strength and elasticity module of the printed parts. The parameter optimization uses Taguchi L18 design and regression models. The examined deposition variables are the layer thickness, the nozzle temperature, the raster deposition angle, and lament printing speed. The effects of the above variables on the strength of the parts are essential to enhance the mechanical response of the printed parts. The experimental outcomes are investigated using the ANOM and ANOVA analysis and modeled utilizing linear regression models. In addition, an independent experiment was repeated three times at optimum parameters' levels to evaluate the methodology, giving predictions errors less than 3%.


Introduction
Additive manufacturing (AM) is a sub-group of manufacturing processes that uses metallic, plastic, ceramic, and composite materials in solid-state (powders, laments, sheets, bars, etc.) or uids (liquid or droplets) that are deposited using plenty of different deposition mechanisms and techniques [1]. Fused lament fabrication (FFF), also known as Fused Deposition Modeling (FDM) or 3D printing (3DP), is the most popular among AM processes due to the simple deposition mechanism and the variety of available materials that exist in the market. Nowadays, more innovative materials and colours are extensively used in an eco-friendly way in applications, including the furniture industry, design, and fashion, among others [2].
The strength of the FFF parts is one of the main concerns of the end-users when the material deposition parameters are selected before the printing process starts [3]. The strength characteristics of FFF parts have been continuously explored for over two decades [4]. Part orientation inside the build space and the deposition parameters affect the interlaminar strength between the deposited material [5] and consequently the strength and the surface quality of the FFF parts [6]. The raster deposition angle (DA) also affects the tensile stress of polylactic acid (PLA) FFF parts [7]. It is observed that the PLA parts in X build orientation (zero DA) exhibit higher tensile stress than those in Y and 45° deposition directions. The solid print mode is used for all the experiments, which means that the FFF variables such as the nozzle temperature (NT), printing speed (PS), and layer thickness (LT) were kept constant. The LT and the orientation of the specimens on the X-Y platform also affect the tensile strength of the ABS-FFF parts [8].
Many other studies have been examined the effects of the LT, part build orientation, raster angle, raster width, and air gap on the strength characteristics of the FFF test specimen [9][10][11]. It is concluded that the strength of the FFF parts varies according to the quality of the interlayer bond formation. The thermal properties of the polymeric materials and the deposition rates need optimization to achieve a robust bond interface and better strength [12]. All the investigations conclude that different printing parameters are appropriate for each other material for optimizing the quality of FFF parts, such as strength and shape accuracy.
Concerning the organic biocompatible composites, PLA mixed with wood ours materials (PLA/W); research effort has been applied for investigating the strength properties according to material synthesis [13][14][15][16] and the FFF parameters [17][18][19]. ependable adhesion between PLA and wood ours is reported in Faludi et al [14]. In Chansoda et al. [13], the utilization of parawood powder derived from the furniture industry is tested as an in lled material in the PLA matrix. Zandi et al. [18] investigate the fatigue behavior of a PLA/W (Timber ll, 8% wood bers) processed through FFF process parameters, i.e., the layer height, the nozzle diameter, the in ll density, and the printing velocity. They used the Taguchi L27 orthogonal array and found that the layer height was the most critical parameter. In Kain et al. [20], two wood ber contents (15 and 25 %; Wood bers, ARBOCEL C100) mixed with PLA (Ingeo™ 3251D) are tested in different raster DA during FFF (0-90; step 15 degrees). It is proved that the 25% wood content gives better strength than 15% and that the in ll DA affects the strength of the FFF parts. The strength performance of PLA/W parts, changing the NT between 210 and 250 o C also investigated [21]. The increase of the NT from 210 to 230 o C improves the strength slightly. Above 230 o C, the strength is in uenced by the wood particle's degradation and is not suggested. Ayrilmis et. [19] alinvestigated the impact of LT on the water absorption and strength of FFF-PLA/W specimens. By increasing the LT, the water absorption and the cross-section porosity increase, and the durability decrease. More, it was found that the density of the printed PLA/W part increased as the PS decreased and that the UTS and exural properties of the FFF-PLA/W specimens were not altered signi cantly by the deposition rates [17].
The above researchers investigated the in uence of the FFF parameters on the strength of different PLA wood our contents. In conclusion, to the author's knowledge, it was not reported any research in literature where the in uence of the LT, NT, raster DA, and PS on 3D printed PLA/W is studied. Moreover, the proposed experimental area of the selected parameters is investigated for the rst time.
Therefore, this work is an experimental investigation of how these four FFF parameters, the LT, NT, raster DA, and PS, in uence the tensile strength and the elasticity module of the PLA mixed with coconut wood our FFF parts. The results were analyzed and presented using statistical tools such as main effect plots (MEP), analysis of variances (ANOVA), interaction charts, and linear regression models. So, the optimum parameters proposed to optimize the strength of the organic biocompatible composite FFF-PLA/W material.

Materials And Methods
Page 4/17

Preparation of experiment
According to ASTM D638 standards (see Fig. 1a), a dogbone was designed having a thickness of 4mm and translated in STL format at the SolidWorks CAD program. Then, the Craftbot Plus 3D printer (Fig. 1b) fabricated the specimens. Its platform is made of aluminum, thus giving the capability for printing specimens using different types of material. The maximum volume of the vat was 250 X 200 X 200 mm, and the printing speed was up to 200 mm/s. The commercially available material NEEMA3D™ WOODPLUS consisted of 30% wood bers of coconut and additives, and 70% pure PLA polymer was utilized. The lament diameter was 1.75mm. The speci c gravity was 1.2g/cc (ASTM D1505), the melting point between 140-150 o , and the nozzle's diameter was 0.4mm. The minimum produced tensile strength and modulus is 70 MPa 1900 MPa, respectively (ASTM D882).
The static tensile testing is performed on a strain-controlled Instron 3382 Universal Testing Machine with a load capacity of 100 kN, as per ISO, equipped with an especially gripping xture as shown in Fig. 2. All specimens were manipulated at a crosshead speed rate of 1mm/min. At the same time, force and displacement data for each test coupon were recorded through a data acquisition system and stored through BlueHill software for further treatment.

Design of experiment
In the current study, the Taguchi L18 (2 1 x 3 7 ) experimental approach was employed [22][23][24]. This methodology adopts balanced experiments according to Taguchi's proposed Orthogonal Arrays (OA). Although the number of the executed experiments is a fraction of the full combinatorial design, the results are proved to nd the best parameter levels and construct predictive mathematical models [25,26]. Finally, three validation experiments against the best conditions are applied to verify the experimental design and evaluate the experiment's spread at the optimum conditions. The rst step in the above approach is selecting the variable parameters that affect the aspired attributes: the UTS and the elasticity module (E). This task is critical and has two main concerns: (i) to select the appropriate variable parameters and levels and (ii) all the experiments of the decided orthogonal array (OA) to be achievable (should do all experiments contained in the OA).
After the literature review in the introduction section and a 'trial-and-error' procedure, the selected variables and the constant parameters are tabulated in Table 1. According to the literature review, layer height and nozzle temperature (LT and NT) in uence the dimensional accuracy, surface roughness, and mechanical response of the FFF parts [3]. In addition, raster deposition angle (DA) is also an essential parameter [27]. It affects the strength of the FFF parts, as this parameter de nes the direction of the strands of the woven pattern. Finally, the printing speed determines the time between layers and affects the interlaminar quality between layers [12]. All other FFF process parameters were kept constant in this study to minimize the noise error. The Taguchi orthogonal array, known as L18, is adopted for the above variable parameters and levels. This OA consists of eighteen arrays and eight columns [28,29]. The eighteen arrays show the number of the experiments, 18 experiments, and the columns show the possible variables [23]. We have used only 4 out of 8 columns, i.e., four parameters (LT, NT, DA, and PS). According to the Taguchi approach, the empty columns contribute to calculating the error. So, it is not necessary to repeat each combination of the 18 experiments. Table 2

Results And Discussion
All the results from the experiments conducted in this work are presented in Table 2. The UTS is the higher tension that a material can sustain while being stressed before breaking. At the same time, the elasticity module (E) is the linear proportion between the stress and the strain in the elastic deformation zone of the stress-strain curve. Thus, both the UTS and E are used for characterizing the material's withstand properties. The results basic statistics of the process performance are presented in Table 2 (mean, min, max, spread). It is evident that the spread values of the mechanical response are considerable, 100% and 69% for the UTS and E. Thus, process optimization is vital for sustainable FFF printings [30].

Effects of the variable parameters on UTS and E
The diagrams that show the effects of the variable parameters on an attribute are known as ANOM diagrams (analysis of means) or main effect plots (MEP). Such diagrams explain the in uence of each variable graphically in consideration of the attribute measure and are utilized for nding the optimum levels of the variables. For example, by employing these plots, the effects of the four process parameters (LT, NT, DA, and PS) on both the UTS and E can be extracted (see: Fig. 3). A way for identifying how one parameter interacts with another is by utilizing the interaction plots [29,31,32]. Fig. 4a and 4b show how LT, NT, DA and PS interact concerning the UTS and E accordingly. These two plots show a strong interaction between the dominant parameter DA and the others (PS, NT, and LT).
After the MEP and interaction plots, the ANOVA analysis is the qualitative tool to investigate the importance of each variable parameter on the quality attributes (here, the UTS and E). Analysis of variances (ANOVA) decomposes the errors of each variable on the total error when a mathematical model is tted on the results. The 'MEP' plots results showed that the DA is the dominant parameter. Therefore, it is decided to use linear regression models for the 'UTS' and 'E' analysis of variances (see : Tables 3 and 4, respectively).  (4) and P values smaller than 0.05, which means this conclusion is statistically signi cant. The best value for the raster DA is zero degrees, and the worst the 90 o . The interactions of raster DA with the other parameters are slightly synergistic (Fig. 4). The mechanical response (UTS and E) increases by decreasing the raster deposition angle. It means physically that the zero degrees aligned lament show the highest UTS and E values.
The other three parameters (NT, PS, and LT) are of lower importance for the utilized experimental space.
The F values are smaller than 2, and P values are higher than 0.05 (F<2 and P>0.05). It is noted that the F value for the LT concerning the UTS attribute is between 2 and 4 (2<FLT<4), which statistically means that it is more signi cant than the NT and PS. These statistical values show that the PS, NT, and LT affect the optimal thermal bonding condition in an anti-synergistic way (see interaction charts, Fig. 4).
Considering previous experimental work and knowing that the interlaminar bonding conditions are affected initially by the PS and NT, and secondly, by the LT [33], the following MEP plots (Fig. 5) are drawn using the subsets of 0.1 mm and 0.3 mm of the LT parameter of Table 2. It is concluded that the NT is a signi cant parameter for 0.1 mm LT and not signi cant for 0.3 mm LT, denoting that the interlaminar conditions are different in the case of 0.3 and 0.1 LT.
The NT has different trend lines for 0.1 and 0.3 LT. In the case of 0.1 LT, when the NT increases, the UTS and E increase (Figs 5a and 5b). On the other hand, for 0.3 LT, the NT is insigni cant for both UTS and E (Figs 5c and 5d). In the total experimental space (Fig. 3), the trend line of the PS is similar to that of the 0.1 LT (Figs 5a and 5b). The ANOVA analysis shows that the PS is an insigni cant parameter (F<2 and P>0.05; Tables 3  and 4).

Modeling and validation
Considering the ANOM analysis (Table 3 and 4) and the MEP plots (Figs 3 and 5) as well as the interactions charts (Fig. 4), linear regression models were developed for both the UTS and E for the 0.1 mm and 0.3 mm LT respectively (eq. 1-4).
Finally, three evaluation experiments are executed to validate the developed regression models ( Table 5). The specimens for the validation experiments were built with the optimized parameters according to the MEP in Fig. 3 (zero DA, 0.3 mm LT, 30 mm/s PS and 220 o C NT). All three experiments give accurate predictions with an accuracy lower than ±5%, which means that the models accurately predict both the UTS and the E attributes. The experimental results were analyzed using the main effect plots (MEP), interaction diagrams, and ANOVA analysis and conclude the following: The DA parameter dominates on both the UTS and E values, while when this parameter increases, both UTS and E attribute decrease. The DA in uences the UTS and E at about 80% and 90%, accordingly. Physically the zero-oriented lament shows higher UTS and E values like in ber composites. This conclusion follows the literature [7].
The optimal thermal bonding condition is affected by the PS, NT, and LT in an anti-synergistic way (see interaction charts, Fig. 4) but trivially.
The PS parameter affects the UTS and E slightly. When decreases, improve both the UTS and E for the 0.1 LT and decrease the UTS and E for the 0.3 LT.
The NT parameter affects both the UTS and E for the 0.3 LT insigni cantly. Concerning the 0.1 LT, when the NT increases, both the UTS and E increase slightly.
The LT is a signi cant parameter but is highly dependent on the interlaminar bonding condition (PS and NT) and should be studied individually.
Finally, the developed linear regression models proved adequate, having an accuracy better than 5% by three independent validation experiments.
As future work, the authors propose a multi-parameter multi-objective optimization of the PLA/wood parts, including more wood our types and wt. %.

Declarations
Authors declaration: The submitted work is original and has not have been published elsewhere in any form or language Funding Source: No funding was received for conducting this study.
Con ict of interest: The authors have no competing interests to declare that are relevant to the content of this article.
Availability of data and material (data transparency): All data generated or analyzed during this study are included in this published article.
Code availability: Not applicable.
Ethics approval: Not applicable.
Consent to participate: Not applicable.
Consent for publication: Not applicable.