An easy method of image feature extraction for real-time welding defects detection

Sensing technology is the key for intelligent robotic welding. AI alloy pulsed Gas Tungsten Arc Welding (GTAW) has been increasingly applied in several industries from aerospace, automobile to ships for light weight manufacturing, wherein vision sensor has shown better performance among others. However, with the disturbance of arc light and surroundings, on-line image feature extraction is still a huge challenge in terms of improving the real-time performance, stability and robustness of robotic welding vision system. In this paper, we proposed an easy methodology to quickly extract several image features for the purpose of detecting the typical welding defects of AI alloy in pulsed GTAW. First, based on the idea of vision attention, the gray level statistics have been calculated for three image regions of interested(ROI) both from welding pool and back seam. Then, experience-driven based certain gray interval is chosen to extract its total number of pixel as the main monitoring parameters. Furthermore, the background noise is successfully removed by using the proposed pixel ratio algorithm as well as enhancing the ratio of signal to noise. The test results indicate that the proposed method has the ability of predicting and identifying welding defects of under penetration, surface oxidation, over penetration and burning through, which certainly improves the intelligent level of robotic welding. This paper also provides some guidance for vision-based monitoring of other similar manufacturing process.