Optimizing FDM process parameters: predictive insights through taguchi, regression, and neural networks

Abstract

Fused deposition modelling (FDM) is a popular additive manufacturing process used for rapid prototyping and the production of complex geometries. Despite its popularity, FDM's susceptibility to variations in numerous process parameters can significantly impact the quality, design, functionality, and mechanical properties of 3D printed parts. This study explores thirteen FDM process parameters and their influence on the mechanical properties of polylactic acid (PLA) polymer, encompassing surface roughness, warpage, tensile and bending strength, elongation at break, deformation, and microhardness. The optimum parameters were identified alongside key contributors by applying the Taguchi method, signal-to-noise ratios, and analysis of variances (ANOVA). Notably, specific FDM parameters significantly affect the surface profile, with layer thickness contributing 32.65% and fan speed contributing 8.59% to the observed variations. Similarly, warping values show notable influence from nozzle temperature (29.53%), wall thickness (16.74%), layer thickness (16.56%), and retraction distance (12.80%). Tensile strength is primarily determined by wall thickness (31.83%), followed by infill percentage (26.73%) and infill pattern (16.18%). Elongation at break predominantly correlates with wall thickness (44.82%), with a supplementary contribution from nozzle temperature (10.90%). Microhardness lacks a dominant parameter. Bending strength variations primarily arise from layer thickness (38%), wall thickness (37.6%), and infill percentage (9.17%). Deformation tendencies are influenced by layer thickness (19.20%), print speed (11.37%), wall thickness, and fan speed (10.9% each). The optimized dataset of FDM process parameters was then employed in two prediction models: multiple-regression and artificial neural network (ANN). Evaluation based on the correlation coefficient (R2) and root mean squared error (RMSE) indicates that the ANN model outperforms the multiple-regression approach. The results indicate that precise control of FDM parameters, coupled with ANN predictions, facilitates the fabrication of 3D-printed parts with the desired mechanical characteristics.