Prediction of Optimal Parameter Settings and Significant Parameter for Reduced Geometric Deviations Through Multi Criteria Decision Making and Machine Learning Algorithms

Abstract

Part dimensional inaccuracies serve as a barrier from adopting Additive Manufacturing (AM) processes in mass production. Fused Deposition Modeling (FDM) is a thermoplastic based low cost AM process which can create conceptual models, prototypes and end user industrial parts. The current study involves predicting the optimal parameter settings and significant parameter for reduced geometric deviations in printed part using Nylon filament reinforced with 20% carbon fiber. Five input factors such as build orientation, layer thickness, infill density, raster angle and infill pattern have been considered for preparing the experimental layout through taguchi�s mixed fractional factorial design. The changes in length, width and thickness of the printed part from CAD value have been evaluated individually through ANOVA and Signal to Noise Ratio method (Smaller the better). Layer thickness is significant only for variations in length, but build orientation affects both width and thickness dimensions. The geometric deviations are further analyzed using combined multi criteria decision making (MCDM) approaches such as Entropy-CoCoSo and PCA-TOPSIS. The optimal parameter settings obtained for reduced geometric deviations is found to be Flat orientation, 0.1mm layer thickness, 50% infill density, 0� raster angle and cubic infill pattern. Layer thickness is found to be highly significant parameter influencing the geometric deviations subsequently followed by build orientation from both the MCDM methods. The multi response performance index values obtained from Entropy-CoCoSo has been trained using classification algorithms such as decision tree, random forest and Naive Bayes. Naive Bayes algorithm outperformed other methods with highest classification accuracy of 99.4% in a training-testing split ratio of 75:25.