Rapid failure risk analysis of corroded gas pipelines using machine learning

Pipelines are critical to the urban development of modern cities, closely entwined with both production and residential activities. This study introduces a rapid and efficient methodology for assessing the failure risk levels of corroded gas pipelines. Initially, a comprehensive dataset is meticulously generated according to industry standards and engineering practices, with failure probabilities estimated using a physics-based probabilistic approach that employs gamma processes and linear growth models for defect characterization. Existing standards conservatively estimate the residual strength of corroded gas pipelines, resulting in overestimated failure probabilities. Consequently, this study introduces a validated failure pressure model capable of accurately predicting the strength of pipelines across various steel grades. The application of Monte Carlo simulations enables precise failure risk level assignments. This study explores six machine learning models and employs Bayesian optimization for hyperparameter tuning, resulting in enhanced model performance. The ANN model demonstrates superior performance in capturing complex nonlinear relationships between input parameters and failure risk levels. Model interpretability is enhanced through SHapley Additive exPlanations (SHAP), providing clear insights into the contribution of each feature to the model's predictions. Validation using real-world PHMSA data confirms the accuracy and practical applicability of the proposed methodology. This comprehensive framework advances pipeline integrity management, providing valuable insights for strategic planning of monitoring, inspection, maintenance, and replacement activities, ultimately enhancing the safety and reliability of gas pipeline networks.