Multi-objective optimization of water-assisted injection molding using machine learning and CFD simulations

This study proposes two integrated optimization frameworks for Water-Assisted Injection Molding (WAIM), which combine Computational Fluid Dynamics (CFD) modeling, automated post-processing, and machine learning. A curved tubular flow channel with a square cross-section, representing a typical configuration used in WAIM research, is modeled in Moldex3D software as a plastic product. The CFD model is validated against experimental data available in literature to ensure reliability. Key quality indicators, i.e. , the Hollow Core Ratio ( R HC ) and the Wall Thickness Deviation ( D WT ), are quantified automatically from the numerical flow field using a Python-based image processing pipeline. Surrogate models are then constructed using Artificial Neural Networks (ANNs) to approximate the behavior of computationally intensive CFD simulations. Subsequently, Sobol sensitivity analysis is performed on the trained ANN surrogates to quantify the relative influence of process parameters on each quality response. Multi-objective optimization is subsequently performed through the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and the Non-dominated Sorting Genetic Algorithm III (NSGA-III), separately. The optimized process parameters predict R HC and D WT values deviating from the corresponding CFD results by less than 2% and 5%, respectively. This dual-framework approach enhances both design efficiency and product performance in WAIM, offering a robust solution for intelligent manufacturing of plastic components. • CFD model validated with experiments ensures reliable WAIM simulations. • Automated image processing enables accurate core and wall thickness metrics. • Neural network-based surrogate model replaces costly CFD runs with fast and precise predictions. • NSGA-II and NSGA-III optimize hollow core ratio and wall thickness deviation. • Frameworks improve efficiency and product quality in WAIM manufacturing.