Hybrid CatBoost for steel pipelines collapse prediction under bending and external pressure

The increased transportation of oil, gas and derivatives via offshore steel pipelines has led to a growing demand for accurate predictions of their collapse behavior under combined bending and external pressure over the entire period from installation to operation. In this study, a finite element method using ABAQUS was employed to numerically estimate the collapse pressure of steel pipes subjected to these combined loads. To address the complexity of the problem, a machine learning-based approach was implemented, consisting of three stages: feature design, collapse analysis, and the establishment of a machine learning model. A hybrid machine learning model that integrates CatBoost with Bayesian optimization (namely BO-CB) for accurate prediction of collapse pressure was developed using a dataset of over two thousand data points and five key features. To evaluate the performance of the established BO-CB model, five foundational ensemble machine learning models were utilized for comparison. Superior performance in terms of accuracy was exhibited by the hybrid BO-CB model when compared with the other models. Furthermore, the effect of input features on the pressure at collapse was evaluated using the Shapley Additive Explanations method. Lastly, a user-friendly graphical interface was developed, allowing for the seamless application of the BO-CB model.