PHYSICS-INFORMED MACHINE LEARNING FOR PREDICTIVE QUALITY CONTROL IN METAL ADDITIVE MANUFACTURING

Abstract

Metal additive manufacturing has transformed advanced production by enabling the fabrication of complex components with high design flexibility and reduced material waste. However, maintaining consistent quality remains challenging due to complex physical interactions among thermal behavior, material properties, process parameters, and defect formation mechanisms. This paper proposes a PhysicsInformed Machine Learning (PIML) framework for predictive quality control in metal additive manufacturing by integrating physical process knowledge with machine learning, Industrial Internet of Things (IIoT), Digital Twin technology, multi-sensor data fusion, and predictive analytics. The proposed methodology combines sensor-driven manufacturing data with physics-based constraints to improve defect prediction, process understanding, and adaptive quality control. Experimental evaluation demonstrates improvements in defect prediction accuracy, model reliability, process stability, parameter optimization, and manufacturing efficiency. The proposed framework provides a trustworthy and scalable solution for intelligent additive manufacturing by combining data-driven intelligence with physical process awareness for Industry 4.0-enabled production environments. Keywords— Physics-Informed Machine Learning, Additive Manufacturing, Predictive Quality Control, Digital Twin, Machine Learning, IIoT, Smart Manufacturing, Industry 4.0.