An Intelligent Optimization System for PIM Process

Abstract

This study proposes an intelligent optimization system based on the Taguchi method, back-propagation neural network (BPNN), multilayer perceptron (MLP) and modified PSO-GA to find optimal process parameters in plastic injection molding (PIM). Firstly, the Taguchi method is used to determine the initial combination of parameter settings by calculating the signal-to-noise (S/N) ratios from the experimental data. Significant factors are determined using analysis of variance (ANOVA). The S/N ratio predictors (BPNNS/N) and quality predictors (BPNNQ) are constructed using BPNN with the experimental data. In addition, a modified PSO-GA algorithm in conjunction with MLP is used to find initial weights of BPNN and to reduce the training time of BPNN. In the first stage optimization, the S/N ratio predictors are coupled with GA to reduce the variations of the manufacturing process. In the second stage optimization, The combination of S/N ratio predictors and quality predictors with modified PSO-GA is empoyed to search for the optimal parameters. Finally, three confirmation experiments are performed to assess the effectiveness of these approaches. The experimental results show that the proposed system can create the best performance, and optimal process parameter settings which not only enhance the stability in the whole injection molding process but also effectively improve the PIM product quality. Furthermore, experiences of the novel hybrid optimization system can be transferred into the intelligent PIM machines for the coming up internet of things (IoT) and big data environment.