Monitoring and characterizing melt pool dynamics in additive manufacturing is essential for understanding complex thermal and material behaviors that govern layer formation. Melt pool characteristics, such as size, shape, and dynamics, directly influence microstructure evolution, mechanical properties, and dimensional accuracy of the printed part. Therefore, accurate and efficient methods for melt pool detection and tracking are vital for advancing process control and part quality. This work investigates the application of advanced image processing techniques based on Semantic Segmentation to extract melt pool regions from high-speed imaging data. A semi-supervised learning framework employing teacher-student architecture is proposed to reduce the need for extensive labelled datasets while preserving segmentation accuracy. Comparative analyses with conventional approaches are conducted under varying printing conditions to demonstrate the method’s effectiveness. Results show the efficiency of the proposed methods addressing accuracy and labor intensity. Finally, the study discusses the challenges and bottlenecks of the implemented methods and their usability in an industrial perspective.
Hrechuk et al. (Thu,) studied this question.