A New Method for Wellhead Device Defect Identification with Ultrasonic Signals

In this paper, we use ultrasonic detection to study the feature extraction and pattern recognition of defects in wellhead devices and to classify the depth of defects in wellhead devices. Firstly, we use COMSOL finite element software to simulate the ultrasonic defect detection process, and set up defects of different depths in the inner wall of the inspected object, through which we can observe the propagation process of ultrasonic waves in the pipe and get the simulation echo signal of defects with different depths. Build an experimental platform, set up artificial defects of different depths in the inner wall of the pipe, and collect each defect waveform. For the denoising of the defect echoes, this paper uses the empirical mode decomposition (EMD) denoising method, wavelet packet decomposition for energy entropy feature extraction, and finally the extracted feature vectors are fed into the random forest classifier optimized by the sparrow algorithm for pattern recognition, and compared with the unoptimized random forest classifier, the results show that for the ultrasonic simulation signals with different defect depths, the results show that the recognition accuracy of SSA-RF classifier can be as high as 92.5% for ultrasonic simulation signals with different defect depths; for ultrasonic real data, the recognition accuracy of SSA-RF can reach 89.5833% for man-made defects with different depths.