Multi-objective optimization of injection molding via interpretable ensemble learning and a multi-strategy improved MOCGO algorithm

Abstract

To address the nonlinear coupling and performance conflicts among warpage deformation, volumetric shrinkage, and clamping force in injection molding, a collaborative framework integrating multi-model learning and multi-objective optimization is proposed. High-coverage simulation samples are first generated using Central Composite Design (CCD) and Latin Hypercube Sampling (LHS). On this basis, an ensemble prediction model is constructed by combining Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM), with hyperparameters optimized via the Improved Runge-Kutta (IRUN) algorithm. Model interpretability is enhanced using SHapley Additive exPlanations (SHAP). In the optimization phase, multiple strategies – including good point set (GPS) initialization, Cauchy local perturbation, Lévy flight mutation, and adaptive PBI-based sorting – are incorporated into the Multi-objective Chaotic Game Optimization (MOCGO) algorithm to improve performance. Finally, the proposed method is validated through coupled simulations using Moldflow and Ansys. Warpage, shrinkage, and clamping force are reduced by 24.0 %, 6.3 % and 7.4 %, respectively, with improved stress distribution, demonstrating both effectiveness and engineering applicability.