Strain Demand Prediction Model of Buried Pipeline Subjected to Tectonic Fault Displacement Via Deep Learning Models

Abstract The geological hazard such as strike-slip fault is one of the primary threats to the integrity and the safety of in-service pipeline in China, which triggers ground movement leading to the large-scale deformation and serious tensile and compressive failure on pipeline which will cause serious economic and social disruptions, so it is significant to carry out the precise pipeline mechanical response for effective assessments of safety status prediction. Therefore, the accurate numerical simulation of pipeline crossing geological hazards area is established based on the finite element method software ABAQUS with the consideration of the elastoplastic property of high-grade X80 pipe steel. With the combination of ABAQUS and Python, the pipeline under different working condition such as pipeline geometric dimensioning, soil surface movement, pipeline inner pressure and soil type was formulated and the maximum strain of pipeline database for machine learning and deep learning models’ training and validation dataset was formed according to the database above. Base on this database, models including the Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost) and improved Residual Neural Network (ResNet) model were trained and validated. The results show that the finite element-deep learning conjoint analysis method has the significant advantages of high prediction accuracy. This research serves as a valuable reference and useful guide for the safety state analysis of pipeline under geological hazard.