Failure analysis and control of natural gas pipelines under excavation impact based on machine learning scheme

Abstract

Third-party excavation operations pose a serious threat to the safe operation of natural gas pipelines, and quantifying the failure conditions of pipelines can effectively identify the hazards of excavation operations. Considering the current lack of test data on the full-size critical conditions of pipelines under different failure modes, to make the research results have better field application, this study aims to develop a failure prediction model, which is adopted for predicting the failure modes of the pipeline under different excavation conditions in order to propose control strategies. In this work, finite element analysis is combined with machine learning algorithms. The finite element analysis process derives the critical loads for different failure modes of the pipe and establishes the failure condition data set. Correlation analysis and sensitivity analysis were employed to investigate the influence pattern of the features. The prediction performance of different machine learning combination algorithms was tested, and a hybrid data-driven prediction model was established and combined with the excavation equipment parameters to determine the risk level of the excavation equipment and the risk area of the operation. The results demonstrate that the critical load value for failure grows when the strength of the pipe increases. The four features of yield strength, strength limit, pipe diameter, and wall thickness exhibit the highest importance scores. The bucket tooth wedge angle only influences the magnitude of the puncture critical load, with correlation and sensitivity coefficients being 0.315 and 0.116, respectively. In the test of the combined algorithm, the RFECV-CSVR-NSGAIII algorithm built in this study has the highest generalization performance, the mean absolute error percentage (MAPE) is lower than 0.0476, the coefficient of determination R2 reaches over 0.9960, and the prediction model has excellent accuracy. The prediction model was subjected to case studies to obtain rapid identification of risk levels and operational risk areas of excavation equipment.