A Data-Driven Based Research On Degradation Prediction And Reliability Assessment Method Of Corroded Subsea Oil Pipeline

Corrosion is the main reason for structural degradation of subsea oil pipelines,which may cause severe economic losses and environmental pollution.A data-driven method for degradation prediction and reliability assessment of corroded subsea oil pipeline is proposed in this paper.Based on the monitoring data,a corrosion rate prediction model is established.The residual strength prediction model is established by finite element simulation and machine learning technology.Using Monte Carlo method to calculate the failure probability and establish a reliability assessment model.It provides theoretical evidence and scientific guidance for the safety and integrity management of subsea oil pipelines,ensuring the operation safety of the in-service subsea oil pipelines.The research of this paper is as follows:1.Research on Corrosion Rate Prediction of Subsea Oil PipelineIntegrating kernel principal component analysis(KPCA)and Bayesian regularization artificial neural network(BRANN),a data-driven model for corrosion rate prediction of subsea oil pipeline is established.The model can eliminate the redundant information from the original monitoring data and improve the generalization by regularization constraints.KPCA is applied to reduce the dimension of the collected corrosion factors,and the obtained principal components are input into BRANN to establish a corrosion rate prediction model.The results indicate that the proposed model has high prediction accuracy and goodness of fit(MSE=0.46%;R~2=0.9901),and its output is an essential part of the risk management of corroded subsea oil pipelines.2.Research on Residual Strength Prediction of Corroded Subsea Oil PipelineThe residual strength can be directly used as evidence to assess the operation risk status of corroded subsea oil pipeline.The prediction method based on machine learning technology has become a remarkable alternative to the traditional empirical model.A data-driven model for residual strength prediction of corroded subsea oil pipeline is proposed.Considering the interaction between corrosion defects,the residual strength dataset is obtained by calculating the established finite element model of corroded pipeline.A manifold learning algorithm,i.e.,Local Linear Embedding(LLE),is used to extract data features.The residual strength prediction model of corroded subsea oil pipeline is established by BRANN.Compared with the traditional algorithm model,the proposed model has certain advantages in error and regression performance,with MSE reduced by at least 74.58%and R~2 increased by at least 0.15%,which can support the early warning of corrosion failure risk for subsea oil pipelines.3.Research on Reliability Assessment of Corroded Subsea Oil PipelineThe reliability assessment model based on traditional intensive computing is difficult to meet the engineering requirements of efficient assessment.A data-driven reliability assessment model for corroded subsea oil pipelines is proposed.According to the corrosion failure mode,the limit state function is constructed.The identified uncertain parameters are discretized according to the statistical characteristics.The failure probability dataset is obtained by Monte Carlo method,and the reliability assessment model is established.A set of corroded pipeline data is input into the assessment model,and the results indicate that the reliability of the corroded subsea oil pipeline is 0.59(moderate)and the maintenance-free life is 14.7 years,which provides a digital solution for the reliability assessment of corroded subsea oil pipeline.