Research On Automatic Recognition Algorithm For Wheel Set Defects Of Ultrasonic Testing Images

Under the condition of high-speed and heavy-load operation, the fatigue cracks are easy to take place on wheel rim and spokes. Good quality of wheel-sets is the guarantee of the rail safety. As the rims and rails are in direct contact, the rims are under huge friction and rolling. So the rims are the key parts of safety inspection. The ultrasonic testing technology is mature. It can detect Surface defects and internal defects of wheel rim and spokes. The applications in the train wheel inspection overcome the limitations of the visual detection. The ultrasonic testing speed is fast and has high detection efficiency. It satisfies the requirement of the safety of railway transportation and the needs of the development of modern high-speed railway.In this thesis, we use wheel detection system to detect wheel set. This system combines the conventional ultrasonic detection technology with phased array ultrasonic flaw detection technology to detect the wheel-sets flaws. However, the defect localization is realized by the operators. It costs too much time and efforts. And the human errors exist. These situations reduce the detection efficiency, it is hard to satisfy the development of China’s high-speed trains.This topic aims to the PE images of the fixed wheel detection system. By discussing its localization methods and designing flaw localization algorithms to improve the detection speed and accuracy. In this thesis, the ultrasonic testing technology and systems are analysis and summarized. To combine with the structure and defect distribution characteristics of wheel-sets, the thesis focuses on the ultrasonic testing technology craft. By obtaining the B-scan images of PE method, first of all, according to the defect location method, we roughly determine the location of the defect, and then do feature analysis and extraction, a series of measures are put forward, such as image intensification, morphological image processing and centroid extraction. Through the experiment, the algorithm is feasible, it can achieve the automatic localization and reach the desired goal.