A Novel Method for Identification of Pipeline Stress Concentration Segment Based on ACSM and IMU In-Line Inspection Data

Abstract Pipeline stress concentration is a major factor leading to pipeline failure, often causing deformation and cracking, which further results in structural damage and secondary hazards such as oil and gas leakage. Therefore, accurately identifying stress concentration segments is crucial for ensuring the safe operation of pipelines. Currently, the widely used Inertial Measurement Unit (IMU) in-line inspection technology in engineering primarily focuses on bending strain analysis and struggles to identify stress concentration segments caused by longitudinal loads and local defects. The recent developed Alternating current stress measurement (ACSM) technology make it possible to derive the pipe axial stress directly, but there is few research focus to the in-depth data analysis. To address this, combined with IMU strain data, the features of in-line inspection data for different stress states are analyzed in depth, identifying the characteristic patterns of normal segments, axial stress concentration segments, bending stress concentration segments, and elbow segments. A stress concentration segment identification database is constructed, extracting 50 representative data features from each sample. Key features are selected using shapley additive explanations (SHAP) analysis, and Light Gradient Boosting Machine (LightGBM) and other machine learning models are employed to achieve segment classification and identification. The results indicate that the LightGBM model performed best in identifying stress concentration segments, achieving an accuracy of 96.8%, precision of 97.64%, recall of 79.17%, and F1 score of 84.01%. The AUC values for all four segment types exceeded 0.9, demonstrating the model’s ability to accurately classify normal segments, axial stress concentration segments, bending stress concentration segments, and elbow segments. By integrating ACSM and IMU in-line inspection data with the LightGBM machine learning model, accurate identification of stress concentration segments can be achieved. This research provides an efficient and feasible method for identifying stress concentration segments along entire oil and gas pipelines