Efficient prediction method of triple failure pressure for corroded pipelines under complex loads based on a backpropagation neural network. (March 2023)

• Record Type: Journal Article
• Authors: Zhang, Tieyao; Shuai, Jian; Shuai, Yi; Hua, Luoyi; Xu, Kui; Xie, Dong; Mei, Yuan
• Is Part Of: Reliability engineering & system safety. Volume 231(2023)
• Publication Date: 2023-03
• Subjects: Corroded oil/gas pipeline -- Finite element modelling -- Triple failure pressure -- Back propagation neural network -- Failure assessment diagram; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2022.108990
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: The triple failure pressure evaluation system for corroded pipelines is proposed. FE models are developed for predicting triple failure pressure of corroded pipes. The influence of external loads on failure pressure of corroded pipes is clarified. A BPNN model is established to efficiently predict the triple failure pressure. The multiple failure assessment diagram method for corroded pipelines is proposed. Abstract: With the complexity of geological conditions and corrosive environments, the evaluation of failure pressure for defective pipelines under external loads has gradually become an important part of pipeline integrity management and reliability assessment. In this study, validated, 3D, nonlinear finite element (FE) models are established. Considering the safety margin in the evaluation process of corroded pipelines, the evaluation system of triple failure pressure is proposed. Subsequently, many simulations have been carried out with validated FE models. The obtained dataset is selected for training the backpropagation neural network (BPNN) model. After hyperparameter analysis and comparison, the triple failure pressure prediction BPNN model, which is based on a 9-dimensional input layer, a 7-dimensional 180-node hidden layer and a 3-dimensional output layer, is established. Through double verification, it is considered that the established BPNN model has high accuracy in predicting the ultimate burst pressure of corroded pipelines. While rapidly predicting the ultimate burst pressure, the model can synchronously output the flow failure pressure and yield failure pressure. With the established deep learning model, the multiple failure assessment curves of defective pipelines are automatically generated within seconds, which provides a convenient evaluation reference for the defect problems encountered in practical engineering.