A Lightweight Laser Welding Detection Method for Synchronized Sound and Photoelectric Signals Based on DC01 Cold-Rolled Steel Plate

During laser welding, the penetration state (not penetration, moderate penetration, excessive penetration) is a key indicator affecting the welding quality. Traditional photoelectric signal detection has insufficient sensitivity to deep penetration information, while the fusion of the synchronized sound and photoelectric signals can complement each other’s advantages. This study puts forward a lightweight detection method relying on synchronized sound and photoelectric signals. Firstly, the preprocessing of synchronized sound and photoelectric signals are achieved through L2 norm, logarithmic normalization, and grayscale image conversion. Then, a lightweight model is formed by combining depth-wise separable convolution (DSC) and convolutional neural network (CNN) to efficiently recognize the fusion state. The experimental results indicate that the proposed model achieves high accuracy while maintaining only 50% of the original model in terms of computational complexity and parameter scale, providing an economical method to monitor laser welding in real time.