Research On Flaw Recognition Algorithm For CRH Mobile Wheel Sets Detected By Ultrasonic Phased Array

In recent years, CRH has been widely put into rail transport, while, there are new challenges for the railway traffic safety inspection department. Currently, the ultrasonic phased array inspection technology has been applied in the detection of high-speed train wheels, and achieved the preliminary results. In most cases, trained and experienced technical Inspector is needed when evaluating defects, and it takes a long time. in order to reduce the interference of human factors, and make the evaluation results more standardized and professional, automatic identification and evaluation methods are badly needed. The ultrasonic phased array inspection images of CRH wheels have their own characteristics, and due to the characteristics of ultrasonic testing and the detecting environment, sometimes the detected images are seriously disturbed by noise and clutter, additionally, there are echoes of other objects, such as the interface waves wave and web hole echoes, etc. So the recognition and evaluation of defects is difficult. To achieve automatic identification and recognition of flaw in ultrasonic phased array detection images of wheels, this article carried out the following two studies from shallow to deep:Firstly, this paper develops an automatic locating algorithm of train wheel flaws in ultrasonic B-scan images based on cross-correlation. Firstly, Preprocessing is acted on the images, including denoising with wavelet packet denoising method and adaptive threshold processing. Secondly, as the echo images of web hole have periodic feature in spatial domain, image cross-correlation method is used for extracting the period, and then the echo images of web hole is removed from the B-scan images. Finally, another threshold is needed to remove some small bright areas, and then important information of flaws (like centroid and size of flaws) is obtained to locate the flaws lastly. Experimental results show that this algorithm can effectively locate the wheel flaws in the B-scan images, and the average locating accuracy is higher than95%. It achieves the initial goal.Secondly, SVM (support vector machine) is adopted to automatically identify flaws in ultrasonic phased array detecting B-scan images. Both interface wave and web hole echoes are treated as identified object, then there are three types of identified targets. They are the interface wave, web hole echoes and defect echoes. The main processes include feature analysis, feature extraction and optimization, sample training, parameter optimization and identification. Flaw identification method vised SVM is more systematic and comprehensive, the recognition accuracy rate of SVM algorithm is more than90%, and achieves the desired objectives.In this paper, research of flaw location and recognition in phased array ultrasonic detecting images of CRH wheels is carried out, and research results can provide some information and reference for the development of automated detection and identification systems of CRH wheels.