Abstract The metal additive manufacturing industry has experienced significant advancement and growth over the past decades. Powder bed fusion stands out as a promising technology for industries such as aerospace and biomedical due to its ability to produce complex geometries with short lead times. A critical drawback of the technology is the potential for defects to arise, restricting full adoption by these highly regulated industries. Achieving process control remains challenging due to many interacting process variables, coupled with stochastic deviations which can result in defects. While these defects may be detected using ex-situ techniques, this is usually a very costly step. Hence, a critical challenge is the need for in-situ detection of defects to improve quality assurance. Despite the variety of in-situ sensors and defect detection algorithms that are emerging in commercial machines, detection accuracy and efficacy remain problematic. These sensors generate large, complex datasets that traditional pattern detection methods struggle to interpret. This presents an excellent opportunity for machine learning, which can uncover complex patterns in large volumes of multi-modal data. This paper offers a critical review of the current state-of-the-art with respect to in-situ monitoring of the powder bed fusion process. It includes a general overview of the various defect types, as well as analysis of different sensing techniques. Special emphasis is placed on the types of machine learning models employed and the methods used to generate their training data. Key breakthroughs are highlighted, and actionable recommendations are provided for future research and areas of improvement.
Edwards et al. (2026) studied this question.