Real-Time Image Processing for Defect Detection in Laser Powder Bed Fusion Using Machine Learning

Laser powder bed fusion (LPBF) is a key additive manufacturing technique used in producing high-precision metal components. However, quality assurance in LPBF is challenged by defects resulting from poor powder spread-ability. This paper presents a machine learning-based system for real-time defect detection in laser powder bed fusion (LPBF) additive manufacturing. The study evaluates three deep learning architectures i.e YOLOv8, Mask R-CNN, and U-Net, for identifying and classifying powder bed defects, including re-coater hopping, streaking, incomplete spreading, and debris. The system achieves a mean Average Precision (mAP) of 71.73% with Mask R-CNN, demonstrating its potential to replace manual inspection in resource-constrained settings like Zimbabwe. A web-based decision support system provides actionable recommendations, reducing material waste and improving manufacturing efficiency.