Arc Welding Process Monitoring Using Neural Networks and Audio Signal Analysis

This paper investigates the potential of airborne sound analysis in the human hearing range for automatic defect classification in the arc welding process. We propose a novel sensor setup using microphones and perform several recording sessions under different process conditions. The proposed quality monitoring method using convolutional neural networks achieves 80.5% accuracy in detecting deviations in the arc welding process. This confirms the suitability of airborne analysis and leaves room for improvement in future work.