Pipeline Flange Bolt Loosening Detection Technology Based on Stress Waves and Deep Learning

Flanged connections are a critical joining method in modern industrial production, making the detection of bolt loosening in flanges a vital step to ensure industrial safety. Current research on bolt loosening detection in flanges mainly focuses on flat-face flanges without gaskets, while studies on bolted pipe flanges containing gaskets are relatively limited. To achieve bolt loosening detection in such gasketed pipe flanges, this paper analyzes the influence of bolt loosening on wave propagation in the gasket based on the stress wave principle and finite element simulation, and employs the hammer impact method to realize the detection of bolt loosening degree in pipeline flanges. The optimal knock force and hammer head material for the bolt loosening detection experiments were determined experimentally. Through comparative experiments, the Support Vector Machine-Recursive Feature Elimination (SVM-RFE) model was identified as being more accurate and efficient in assessing the degree of bolt loosening. Furthermore, the model was optimized by incorporating feature enhancement and cost-sensitive learning, thereby providing a reliable methodological solution for the rapid identification of bolt loosening severity in pipeline flanges.