We present a novel approach for real-time defect detection and classification in laser welding processes based on the use of uncooled PbSe image sensors working in the MWIR range. The spatial evolution of the melt pool was recorded and analyzed during several welding procedures. A machine learning approach was developed to classify welding defects. Principal components analysis (PCA) is used for dimensionality reduction of the melt pool data. This enhances classification results and enables on-line classification rates close to 1 kHz with non-optimized code prototyped in Python. These results point to the feasibility of real-time defect detection.

Date Published: 1 July 2015
PDF: 5 pages
Proc. SPIE 9657, Industrial Laser Applications Symposium (ILAS 2015), 96570B (1 July 2015); doi: 10.1117/12.2177890

Published in SPIE Proceedings Vol. 9657:
Industrial Laser Applications Symposium (ILAS 2015)
Mike Green; Cath Rose, Editor(s)