Advancing Defect Detection in Laser Welding: A Machine Learning Approach Based on Spatter Feature Analysis.

Area of Science:

• Materials Science and Engineering
• Manufacturing Technology
• Artificial Intelligence in Engineering

Background:

• Full-penetration laser welding (FPLW) is crucial for pipeline manufacturing but faces challenges with in-process defect formation, hindering industrial scalability.
• Incomplete penetration is a significant defect that compromises weld integrity and requires effective monitoring solutions.

Purpose of the Study:

• To develop a sensor-driven framework for non-destructive monitoring and automated defect detection in FPLW.
• To leverage infrared (IR) thermography and deep learning for real-time identification of welding defects.

Main Methods:

• Utilized high-speed IR thermography to capture thermal signals during welding.
• Processed IR data to extract spatiotemporal features like spatter dynamics and weld zone temperature.
• Employed a hybrid CNN-transformer model for multi-label classification of defects such as incomplete penetration, sagging, shrinkage groove, and linear misalignment.

Main Results:

• The proposed framework achieved a mean Average Precision (mAP) of 0.85 in defect detection on 09G2S pipeline steel.
• Demonstrated near-real-time inference capabilities on a CPU.
• Validated that IR thermography-based spatter dynamics provide effective signatures for automated defect prediction.

Conclusions:

• IR thermography combined with deep learning offers a viable solution for automated, non-destructive monitoring of FPLW.
• The developed framework can serve as a foundation for closed-loop quality control systems in industrial laser welding applications.
• Spatiotemporal features derived from thermal imaging are key indicators for predicting and detecting welding defects.