Integrated Simulation and Machine Learning Framework for Residual Stress and Distortion Prediction in LPBF Additive Manufacturing

This study presents an integrated methodology for predicting residual stress and distortion in Laser Powder Bed Fusion (LPBF) additive manufacturing by combining a streamlined finite element analysis (FEA) framework, high-resolution residual stress measurements using the contour method, and machine learning (ML) techniques. A simplified two-parameter temperature field is used in the FEA model to reduce computational cost while maintaining predictive accuracy. Three ML models, multi-layer perceptron (MLP), gradient boosting (GB), and random forest (RF) regressors, are trained on simulated data and validated against experimental results, achieving prediction discrepancies between 52 MPa and 84 MPa. The framework also enables accurate distortion prediction and mitigation: applying inverse predicted distortions to the CAD model reduces final part distortion in a bridge sample from 0.94 mm to 0.06 mm, achieving a 94% improvement. This approach provides an effective tool for improving the accuracy, efficiency, and robustness of LPBF process simulations, with direct applications in part design and quality control.