Defects Recognition of High-Density Polyethylene Pipe Butt-Fusion Joint via STSVD Filtering Total Focusing Method and Machine Learning

Abstract

The butt-fusion joints of high-density polyethylene (HDPE) pipes represent a critical vulnerability, being prone to various defects that potentially lead to structural failures. Therefore, it is essential to conduct nondestructive testing (NDT) on HDPE pipe butt-fusion joints for defect detection. This paper proposes a spatiotemporal singular value decomposition filtering total focusing method (STSVD-TFM), combined with machine learning (ML) to utilize A-scan signals for defect detection. Initially, the signal data is filtered using the STSVD method. Subsequently, feature parameters, including time domain features and spectral features, are extracted from the filtered data, and significant features are selected based on the Relief-F algorithm. Finally, the filtered signal data is employed for TFM imaging, with defect types determined by the training results of the ML models applied to the feature parameters. Detection experiments are conducted on HDPE pipe butt-fusion joint specimens, which included through-hole and square groove defects. The results demonstrate that the proposed method effectively reduces the amplitude of static clutter in the near-field areas, enhances the signal-tonoise ratio (SNR) of the detection images, and achieves high accuracy autonomous recognition of defect types.