On the application of in-situ monitoring systems and machine learning algorithms for developing quality assurance platforms in laser powder bed fusion: A review

Abstract

Laser powder bed fusion (LPBF) is one class of metal additive manufacturing (AM) used to fabricate high-quality complex-shape components. This technology has significantly progressed over the last several years allowing the fabrication of high-value components for a broad range of applications, normally unmatched by other metal AM processes. However, the full adoption of LPBF to serial production is still challenging due to several barriers such as repeatability and reliability of final product quality. The main obstacle could be the high sensitivity of LPBF to environmental and process disturbances. Additionally, LPBF is governed by many process parameters. These factors profoundly affect the process, causing defects formation. To achieve high quality parts, trial and errors are conventionally carried out to obtain optimum parameters that result in good quality for a specific application. However, in recent years attention to the development of quality assurance platforms in LPBF has been the cornerstone of research and development. To this end, researchers have proceeded with three steps: 1) Gaining knowledge from the process by installing in-situ sensing equipment and collecting information from the process. 2) Understanding how the print parameters affect the process, analyzing in-situ datasets and developing defect detection algorithms, and 3) Developing real-time closed-loop control systems using the detection algorithms of Step 2 to automatically adjust the undesired phenomena in the process by changing the print parameters. Although valuable studies were published for the two first steps, the development of real-time controllers has remained challenging. Thus, this study aims to critically review the two first steps to provide insights for researchers into moving toward the development of the control system. In this study, in-situ sensing devices implemented in LPBF are categorized, explained in detail, and mapped to the literature. Then, a comprehensive review is conducted on the latest machine learning (ML) algorithms applied to the in-situ data of LPBF, such as supervised learning, unsupervised learning, and reinforcement learning. Additionally, a comprehensive discussion is provided on in-situ sensors and ML methods applied to LPBF. Lastly, this article specifies trends and future research outlook on this topic.