On search of optimal FDM parameter settings for reduced dimensional error using entropy-PCA-TOPSIS and regression machine learning techniques

Abstract

Fused deposition modeling (FDM) is a fascinating additive manufacturing technique that has the potential to create prototypes and industrial-grade parts from thermoplastic filaments processed through the extrusion technique. Higher processing time, poor surface finish, and low dimensional accuracy are the major barriers to adopting this technique on a wider scale in industries. The current research has focused on identifying the optimal factor settings for minimizing the dimensional deviations in a simple cuboid model with a circular hole at the center. The study has considered six different FDM factors, such as layer thickness, extrusion temperature, print bed temperature, print speed, raster angle, and part orientation, to understand its impact on errors in dimensional attributes such as length, width, thickness, and hole diameter for nylon filament reinforced with aramid fibers at 8% composition. The measured values are further analyzed for international tolerance (IT) grades, length, and width, resulting in IT08 to IT14, thickness, and hole diameter tolerance grade values ranging from IT09 to IT14. Entropy and principal component analysis methods have been deployed for criteria weight evaluation, and different weight values have been assigned to the experimental outcomes. Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) based ranking of alternatives correlates 72.34% (Entropy vs Principal component analysis), with the worst ranked alternative as similar. Regression models developed have R-sq values of 98.15%, 86.27%, 97.63%, and 99.96% for accurately predicting length, width, thickness, and hole diameter changes, respectively. Adaboost algorithm outperformed in training and testing closeness coefficient values with superior performance at all the considered split ratios. The decision tree structure developed indicates layer thickness is significant over the change in length, part orientation has significance over the change in width and hole diameter, and raster angle is significant towards thickness, respectively. Field emission scanning electron microscope (FESEM) images ensure the presence of pores and partially solidified material in the printed samples.