The Intelligent Methodology for Monitoring the Dynamic Welding Quality Using Visual and Audio Sensor

Aiming at the problem of online real-time monitoring of weld quality in traditional welding process, in this paper, a real-time welding quality prediction scheme based on multi-information fusion was proposed. Firstly, in view of the collected arc sound signals, a feature extraction method of short-time average energy and MEL frequency cepstrum (MFCC) is proposed to characterize the energy, timing characteristics and Merle frequency domain characteristics of sound; the image characteristics of weld are analyzed and the image processing method is designed. For straight welds, boundary method should be adopted for segmentation. Centerlines of weld seams can be extracted by means of median filtering, operator edge detection and least square straight line fitting; Based on ROI visual attention mechanism, the front image of the molten pool was extracted, and the edge features of the molten pool were extracted based on SDM method, so as to further obtain the features of the molten pool area, width, and semi-length. The results showed that the arc sound and visual information could support to each other. Finally, combine these two features could achieve online welding quality monitoring.