Predicting the Burst Pressure of High-Strength Carbon Steel Pipe with Gouge Flaws Using Artificial Neural Network

Abstract

Predicting the failure pressure of pipelines is of paramount importance in design and integrity management in order for pipes to operate safely, efficiently, and cost-effectively in terms of repair costs. Given the increasing use of pipelines as high-strength materials, an accurate assessment of defective pipelines is of major importance. This study used data mining to investigate the burst pressure of pipelines containing gouge flaws. The required database was collected using nonlinear finite-element analysis. An artificial neural network method was adopted to predict the burst pressure in a gouged pipeline. The methods used in the artificial neural network are the multilayer perceptron (MLP) and support vector regression (SVR) by spline and Gaussian kernels. Finally, these methods were verified by a full-scale burst test, and the results were compared with those of other methods. The results indicated that the SVR Gaussian kernel had an accurate correlation with the results of the full-scale burst test data. However, the MLP results were less accurate than those of the Gaussian kernel. Moreover, the SVR model using the Gaussian kernel, as compared to other previous models, had the highest accuracy in predicting the burst pressure of high-strength pipelines with gouge defects.