High-density polyethylene buried natural gas pipe butt fusion joint defects automatic detection via total focusing method and deep learning

This paper presents a defect detection method for butt fusion joints of high-density polyethylene (HDPE) pipes based on deep learning, which aims to achieve both high accuracy and rapid detection. Firstly, a directivity-corrected circular coherence factor (D-CCF) weighed algorithm based on total focusing method (TFM) has been proposed to inspect the HDPE pipe butt fusion joints. The D-CCF utilizes the circular coherence factor (CCF) to reduce noise interference during the calculation of defect phase and employs a directivity function to compensate for sound field intensity variations across different directions. This approach leads to a significant improvement in the average signal-to-noise ratio (SNR) improvement compared to the TFM method. This stdy propose an improved YOLOX algorithm that introduces a convolutional block attention module (CBAM) and replaces the original regression loss with the complete intersection over union (CIoU). The experiment is carried out to automate the detection of HDPE pipe butt fusion welding defects. The improved YOLOX algorithm achieved a mean average precision (mAP) of 98.75%on a dataset of 2324 images, demonstrating its effectiveness in detecting small defects. Additionally, a defects automatic detection software is developed based on the proposed YOLOX model to improve the efficiency and accuracy of HDPE pipe butt fusion detection.