Novel Intelligent Deep Learning Model for Predicting the Burst Pressure of Pipelines with Outer Axial Surface Cracks

This paper proposes a novel pipeline burst pressure prediction model, transformer–bidirectional long short-term memory (BiLSTM)–Gaussian noise (GN) model (TBGN), to address the complex nonlinear coupling of elastic-plastic deformation and fracture damage in pipelines with axial cracks on the external surface. The study collected 180 full-scale burst test data sets for cracked pipes to build a comprehensive failure database. Based on the transformer-BiLSTM algorithm and the failure experimental database, an intelligent predictive model for pipeline burst pressure was developed. This model is integrated with the GN model to compute the optimal safety factor through a multidimensional performance evaluation system, balancing the trade-off between predictive risk and excessive conservatism. The resulting framework is termed the TBGN model. The performance of the TBGN model is thoroughly evaluated in comparison with the GN model and several established failure assessment models. Results indicate that the TBGN model outperforms its counterparts in various metrics. Its prediction accuracy (PA) follows a normal distribution, with a concentrated spread. In addition, its key indicators, such as the coefficient of variation (0.113) and correlation coefficient (0.012), reveal minimal dispersion in the prediction results, with no significant correlation between PA and the experimental burst pressure. The PA values predominantly fall within the 0.6 to 1 range, ensuring a robust safety margin without excessive conservatism. Furthermore, the mean PA deviates from 1 by only 0.179, confirming the model’s high accuracy. Overall, the TBGN model achieves a total performance score of 0.928, which is 110.39% higher than the classical models. The TBGN model is capable of effectively assessing the coupled failure resulting from the interaction between fracture and yield failure, thereby providing an accurate and reliable prediction of the burst pressure for pipelines exhibiting axial cracks on their external surface.