Laser powder bed fusion provides high precision and the ability to fabricate complex structures, but the interaction among laser energy, powder behavior, and melt pool dynamics can lead to porosity, balling, and cracking, which affects part quality. This paper reviews recent advances in multimodal in-situ monitoring, including thermal, optical, and acoustic sensing, and assesses their performance in capturing melt pool behavior, temperature evolution, spatter activity, and acoustic features. It also discusses how deep learning improves multi-source data fusion, feature extraction, and defect prediction. The advantages of these technologies rely on multimodal in-situ sensing as well as post-processing measurements of defects such as porosity and cracks, both of which jointly provide high-quality annotated data for training deep learning models. Progress in closed-loop control is also examined, particularly the relationships between key process parameters and feedback signals such as melt pool size and temperature, which support real-time defect mitigation. Major challenges are identified, including low sensor standardization, difficulties in modeling multi-physical interactions, limited interpretability of deep learning models, and strong dependence on large datasets. These issues are particularly evident in the reliance on high-cost LPBF experiments and long post-processing cycles for labeling single data points, as well as the need to reconstruct datasets for parts with different geometric features due to geometry-dependent process signatures. Future research should focus on low-latency multimodal sensing, data-efficient and interpretable deep learning, and industrially adaptable closed-loop control to support large-scale and low-defect LPBF production.
Wang et al. (Sat,) studied this question.