Process optimization of injection molding by integrating numerical simulation with surrogate modeling approaches

An integrated optimization system that can adaptively and intelligently determine the optimal process conditions for injection molding has been developed. Nonlinear statistical regression techniques and design of computer experiments are used to establish adaptive surrogate models that can substitute time-consuming numerical simulation and quickly provide predictions with adequate accuracy for system-level optimization. While the Gaussian process (GP) surrogate model is being refined, a multi-objective genetic algorithm (GA) is employed for the global optimal solutions in a concurrent fashion. The performance and capability of various surrogate modeling approaches - i.e., Gaussian process (GP), artificial neural network (ANN), and support vector regression (SVR)-are also investigated and compared in terms of accuracy, robustness, and efficiency. The examples presented in this paper show that the adaptive optimization procedure helps engineers determine optimal process conditions more efficiently and effectively.