Physics-informed GANs for thermal simulation and anomaly detection in PBF-LB

Laser-based Powder Bed Fusion (PBF-LB) enables the fabrication of complex metal components; however, achieving consistent part quality often requires accurate thermal–structural simulations. Traditional numerical approaches, such as finite element methods, are computationally expensive and time-consuming, whereas data-driven machine learning methods typically require large, labelled datasets. In this work, we propose a Physics-Informed Generative Adversarial Network (PI-GAN) that incorporates governing physical laws to generate realistic thermal field data and improve prediction accuracy when only limited training samples are available. Furthermore, an anomaly detection pipeline is trained using the physically informed simulation data, enabling early identification of process deviations and potential defects. • Physics-Informed Thermal Generation: Integration of heat-transfer physics into a Generative Adversarial Network framework enables accurate synthesis of thermal fields from limited laser based powder bed fusion process data. • Simulation-Based Anomaly Detection: The generated physics-consistent thermal data is used to train an anomaly-detection pipeline, allowing detection of process deviations and potential defects. • Efficient Process Optimization: The approach reduces reliance on large experimental datasets while providing actionable insights on early defect identification.