Physics-informed artificial intelligence for reliable failure pressure prediction of corroded energy pipelines

The accurate prediction of failure pressure in corroded energy pipelines is critical for the safe and reliable operation of global oil and gas transportation systems, which are essential components of modern energy infrastructure. While data-driven machine learning models have shown high predictive accuracy, they often lack physical consistency and generalization capability due to the absence of embedded physical mechanisms. To address this issue, this study develops a physics-informed neural network (PINN) that integrates data-driven learning with mechanical principles for failure pressure prediction of corroded energy pipelines. Interpretable feature analysis is first used to identify the influence direction of input features on failure pressure. Based on these insights, physical monotonicity constraints are formulated and incorporated into the loss function, enabling the model to learn from both data and established physical relationships. Model hyperparameters and the physical-loss weight are optimized using an automated optimization framework. The proposed model achieves the best test-set performance among the evaluated models, with a coefficient of determination (R 2 ) of 0.9655, mean squared error (MSE) of 2.2988. Compared with purely data-driven models, the proposed model shows stronger generalization and a more concentrated error distribution around zero. Bootstrap-based uncertainty analysis further confirms stable predictions, with most measured values falling within the 95% confidence intervals. These results demonstrate that physics-informed artificial intelligence can improve both prediction accuracy and physical reliability, providing a practical tool for intelligent integrity assessment of energy pipeline systems.