Machine learning-driven design of support structures and process parameters in additive manufacturing

Support structure design is a critical aspect of additive manufacturing (AM), particularly for parts with overhanging features, as it influences both part quality and material efficiency. However, the combined effects of support geometry and process parameters on key qualities such as residual stress (RS) and geometric accuracy remain insufficiently explored. This study presents a hybrid framework integrating experiments, finite element method (FEM) simulations, and machine learning (ML) to simultaneously analyse process parameters and support geometries in laser powder bed fusion (LPBF). Results show that increasing support contact area and reducing support spacing enhance structural stability and heat dissipation, thereby lowering RS and residual displacement (Disp). Additionally, higher laser power and lower scan speed further reduce RS and Disp by improving melting quality in unsupported regions. Several ML models – including K-nearest neighbours, support vector regression, decision trees, AdaBoost, and random forest – were trained to predict RS and Disp, with random forest achieving the highest accuracy (R² = 0.947 for RS, R² = 0.965 for Disp). Shapley additive explanations (SHAP) provided interpretable insights into the influence of each factor. This integrated approach offers a comprehensive strategy for optimising LPBF process parameters and support structures, contributing to improved part quality and manufacturing efficiency.