Methodological Approach to Optimizing Process Parameters, Monitoring, and Predicting Fatigue Life in Laser Powder Bed Fusion via Machine Learning

Author of the article:WANGXinlian, LI Jie, WAN Jie, YUAN Ruihao, LI Jinshan, WANG Jun

Author's Workplace：State Key Laboratory of Solidification Processing, Northwestern Polytechnical University, Xi'an 710072, China

Key Words： parameter optimization; defect monitoring; fatigue life prediction; data-driven; physics-driven

Abstract： Laser powder-bed fusion (LPBF) is recognized as a predominant method within additive manufacturing because of its high precision and shortened manufacturing cycle. However, the process still faces significant challenges regarding the repeatability of its manufacturing techniques, the interpretability of the production process, and the reliability of the formed components. The LPBF formation process involves a multitude of parameters, and the selection of different process parameters can lead to various types of micro/macrodefects within the components, thereby affecting their service performance. Therefore, clarifying the interconnections among process parameters, defects, and performance represents a current hot topic and a formidable challenge in laser powder bed fusion manufacturing. As an inevitable product of the evolution of big data and artificial intelligence, machine learning (ML) methods offer opportunities to address the complex nonlinear relationships between high-dimensional physical quantities effectively. In the realm of additive manufacturing, ML has garnered sustained interest for its applications in process parameter optimization, defect monitoring, and performance prediction. This article reviews common ML models, summarizes the input information for ML in the LPBF, and focuses on analysing the applications of data-driven and physics-driven ML models in various domains of the LPBF. Finally, it highlights the current limitations of ML and explores its development trends and technical prospects.