Integrating Taguchi method and artificial neural network to explore machine learning of computer aided engineering

ABSTRACT Plastic injection molding has been a very important technology in industries; however, problems that arise during molding cannot be understood or predicted by general linear rules; therefore, one needs to rely on experienced professionals to assess problems. Although computer aided engineering (CAE) technology has flourished in recent years, it is limited due to long analysis time and is not suitable for on-site real-time judgments. The Back Propagation Neural Network (BPNN) has excellent predictive ability for nonlinear problems, and can immediately provide accurate results after multiple sets of data trainings. In this study, CAE analysis data is used to train the BPNN, and the Taguchi orthogonal method is used to optimize the hyperparameters in the neural networks to construct a neural network that can predict CAE analysis results. The results of this study show that the prediction of the maximum injection pressure and the maximum cooling time is pretty good. However, there is still a big gap related to warpage prediction. Therefore, this study adds more training data on warpage, and conducts secondary training for the neural network. The results show improved predictions.