A Robust Design Method for Low-Pressure Die Casting Process Based on Surrogate Models

The batch-to-batch variability in low-pressure die casting (LPDC), caused by inherent process parameter fluctuations, poses a significant challenge to consistent quality. However, traditional single-point optimization methods ignore parameter fluctuations. This study presents a robust design framework to overcome this limitation. First, an integrated simulation workflow was established by coupling ProCAST casting simulation with Abaqus finite element analysis to predict shrinkage pore volume and load-bearing capacity (LBC). Subsequently, a dataset was constructed from the integrated simulations, and then served to develop a surrogate model using the Extreme Gradient Boosting algorithm. Finally, robust process windows were derived via an inverse search employing the swarm intelligence algorithm. The framework is implemented with a case study on A356 aluminum-alloy wheel casting. The results showed that the surrogate model for defect regression and LBC prediction has high prediction accuracy. SHapley Additive exPlanations analysis identified the interfacial heat-transfer coefficient and half-mold temperature as the dominant factors. Three optimization algorithms, Bayesian-optimized Particle Swarm Optimization (BO-PSO), Bayesian-optimized Genetic Algorithm (BO-GA), and Logistic-Chaos Sparrow Search Algorithm (LCSSA) were evaluated. LCSSA consistently identifies robust process windows satisfying both defect-control and LBC requirements across all target levels, whereas BO-PSO exhibits premature convergence at higher targets and BO-GA yields dispersed solutions with insufficient robustness. The proposed framework provides a systematic methodology for robust process-window design in LPDC applications.