Ultrasonic inspection and pattern recognition of weld defect based on manual ultrasonic scanning method

Аннотация

As is known to all, the data of conventional manual ultrasonic testing are hardly stored in real-time, and the diagnosis of defect is performed completely by the experienced operator, therefore, the automatic assessment of defect in qualitative and quantitative is unreliable. In this paper, a manual ultrasonic scanning system based on USB-camera was developed. The position of probe in the scanning path could be extracted through the image obtained by the camera. At the same time, echoes reflected from defect were stored, which could offer more information for defect identification. Experiments were carried out by this system. Several welds, containing defects of hole, slag and crack, were inspected, and the images of weld defects were described intuitively by the method of 3D-projection imaging technology. According to the large number of stored echo signals of each defect, signal features were extracted in time domain, frequency domain, time-frequency domain and morphological features were also obtained through the image processing of weld defects. Then these features were optimized by classification criteria based on Euclidean distance. Finally, a back propagation (BP) neural network was adopted and trained by the optimized features to classify the three kinds of flaws. The classification result is satisfying and it will be helpful for weld assessment. Compared to the simple signal features, the fusional features of signal features and morphological features could offer more information of weld defects, thus the recognition rate of weld defect was improved by using these fusional features