An Efficient Prediction Method for Critical Buckling Axial Compression Load of Corroded Pipelines

Abstract Corrosion and external loads are unavoidable problems in the operation of pipelines. It is of great significance to clarify the local buckling critical load of corroded pipelines under the influence of various parameters for pipeline integrity management and safety assessment. In this paper, a backpropagation neural network (BPNN) model with efficient prediction ability is established by combining finite element method (FEM) and machine learning (ML) method, which can realize rapid and accurate prediction of critical buckling axial compression load of X80 corroded pipelines with any 9 input parameters. Specifically, the accuracy of the 3D, nonlinear FE model was verified by experimental data, and then many buckling simulations of corroded pipelines with different input parameters were carried out. The data set with input-output relationship obtained by the FEM is used as the training data of the deep learning model. Through the hyperparameter debugging process, a 9-dimensional input layer, a 5-dimensional 80-node hidden layer and a 1-dimensional output layer BPNN model with deep learning and efficient prediction ability is established. Compared with the FEM, the prediction efficiency of this model has been significantly improved. Finally, the influence of corrosion defect geometry on the critical buckling load of corroded pipelines is studied by using the established efficient prediction model. It is clear that the defect depth and defect width have a great influence on the critical buckling load, while the larger defect length almost no longer affects the buckling load.