Auto body welding image recognition algorithm

With the machinery, the nuclear industry, aerospace, energy, transportation, petrochemical, construction, electronics industry, the rapid development of modern industry, welding technology in mechanical manufacturing production is rising in status.Therefore, welding quality and welding the degree of automation requirements are also increasing, welding automation and intelligent welding technology in the 21st century has become an important direction of development. Especially in automobile production, in order to reduce production costs and maximize profits, improve the automation of production has become an inevitable choice. How to improve the automation of production lines and production speed, but also to ensure product quality, it is important to establish standards for automated production lines. The welding automatic production line is the technical difficulties, especially in key parts of the weld is qualified, which for the great impact the quality of products. This paper is to solve the existing problem of lack of weld. This thesis research was "based on CCD (charge coupled device image sensor) of the weld visual inspection system," part of the image recognition algorithms, this detection system is the truck's cab for the development of weld inspection, aimed at improve production efficiency. Rapid and accurate detection of the weld is the weld inspection system, an important indicator. There are over six hundred cab spot, only a small part of the weld is completed by the welding robot welding, robot welding work because the reliability of this part of the weld does not require testing. The remaining 400 other solder joints, due to the reasons for the location of solder joints, it is difficult to weld with a welding robot, which can only be welded by hand, so that the phenomenon exists in welding may have a leak. Many scholars use the CCD had to identify the vehicle, the image recognition for a lot of research. Many scholars also use CCD to detect the weld, and have achieved certain results. Weld image for face recognition accuracy and real-time problems, this paper based on wavelet transform and the Karhunen-Loeve (KL) transformation of the new recognition algorithm. This study through the use of wavelet decomposition and KL transformation and other means to effectively improve the recognition system of the traditional shortcomings of the weld, greatly improving the accuracy of recognition systems and computing speed. This new method can effectively keep the characteristics of the original image information. Differences for the characteristics of image brightness, we must first enhance the image of the image preprocessing, the aim of reducing the image brightness differences. Then extracted using wavelet transform image of the relative stability of the weld the low-frequency sub-band map, the main features of the image information extraction. Then use KL transform to extract and identify the main characteristics of components. The results show that the recognition algorithm has basically reached the requirements of accuracy and real-time. 

 

Keywords: image recognition, CCD, wavelet transform, KL transform, welding