Automated multi-objective optimization for thin-walled plastic products using Taguchi, ANOVA, and hybrid ANN-MOGA

In this paper, an automated two-staged multi-objective optimization tool for plastic injection molding (PIM) is designed, implemented, and applied to industrial product. In the first stage, Taguchi method is utilized for design of experiments (DOE). Process parameters include mold temperature, melt temperature, flow rate, cooling time, and four parameters related to variable pressure profile. Quality characteristics of plastic product containing warpage, weld line, and clamp force are obtained by simulation experiments. ANOVA and S/N ratio are utilized for data analysis to identify the importance of factors, provide an initial optimal combination of process parameters, and obtain quantitative factor’s contribution percentage. The number of process parameters is compressed from 8 to 5 considering factors’ contribution percentage for reduction in time and computation cost. In the second stage, hybrid artificial neural network (ANN) and multi-objective genetic algorithm (MOGA) are utilized as further study to search for optimal solutions for synchronous reduction in three quality characteristics within compressed design space. Orthogonal array with Latin hypercube sampling (OA-LHS) is adopted for sampling with uniformity and stratification. ANN is utilized as a surrogate model to fit the surface between parameters and quality characteristics. MOGA is adopted to search the optimal solutions along the Pareto frontier. With the help of python, visual basic scripts, application programming interface (API) of Moldflow and DAKOTA, the two-stage optimization process is automated, which allows designers to study optimization for PIM, accurately and efficiently, without deep understanding of complicated optimization algorithms. The automated two-stage system is applied to an industrial plastic products and the result verifies that it is accurate and efficient in quality improvement.