Optimization of experimental design techniques for modeling volumetric shrinkage in injection molding experiment using artificial intelligence

Abstract

The study examined two types of design of experiments (DoE) methods for injection molding of a molded part. It evaluated them using an artificial neural network (ANN) and a support vector machine (SVM) via cross-validation and holdout validation. The innovative goal is to identify the most efficient and successful ways for modeling varied DoE. The influence of four processing parameters on the volumetric shrinkage of a thin polystyrene plate sample is simulated using factorial design and orthogonal Taguchi arrays design. As measured by root mean square error (RMSE), the prediction performance revealed that DoE with eight experimental points as in \({2}^{4-1}\) for fractional factorial design and L8 for orthogonal Taguchi design is particularly efficient for this modeling simulation problem. Both design methods are beneficial and efficient because orthogonal Taguchi arrays play an essential role when the accuracy of fractional factorial designs is insufficient.