Multi-objective optimization of injection molding process parameters based on BO-RFR and NSGAⅡ methods

Abstract

Abstract Injection molding of thin-walled plastic parts with minimum deformation in warpage and volume shrinkage is crucial for part quality. Simulation combined Latin hypercube sampling approach was used to research the effects of different process parameters on deformation. Then, random forest regression (RFR) is used to construct the mathematical relationship between process parameters and defects, such as warpage and volume shrinkage. The gaussian process is used as probabilistic surrogate model, while the probability of improvement is used as acquisition function to construct a Bayesian optimization for RFR’s hyperparameters, and the performance of random search is compared. In addition, the gradient boosting regression (GBR) and support vector regression (SVR) were also adopted to establish the prediction models, respectively. Comparing all the above prediction models, it can be found that the Bayesian optimized random forest regression (BO-RFR) has the highest accuracy. The Non-dominated Sorting Genetic Algorithm-II (NSGA-II) is interfaced with the predictive models to find the optimum design parameters for the purpose of effectively predicting and controlling warpage and volume shrinkage. The results show that warpage is reduced by 66.03% while volume shrinkage is 46.20% after optimizing. The final finite element simulation and physical tests indicate that this proposed method can effectively achieve the multi-objective optimization of injection molding.