Failure Pressure of Corroded-Pipeline Prediction—Artificial-Intelligence Model with FEM and Experimental Data

This research work aims to implement empirical models for predicting the failure pressure of a pipeline corroded with a single defect and internal pressure applications. For this purpose, five approaches were used: two artificial intelligence (AI) models with experimental data, a finite-element model (FEM), and two analytical models (ASME B31G modified and DNV-RP-F101). The two proposed artificial models were implemented and optimized in order to find their best performance. The first AI method is an artificial neural network with multilayer perceptron (ANN-MLP), and the second is a support vector machine radial basis function (SVM-RBF). The comparison of the results of these two models proves the good precision of the ANN-MLP, with the correlation coefficient R = 0.9906 and R = 0.9888, ahead of the SVM-RBF with R = 0.9290 and R = 0.9260, respectively, for the two phases of training and testing. In addition, the output results of ANN-MLP are analyzed by William’s diagram; it was noted that 97.82% (180/184) of points belong to the field of validity and applicability of the artificial optimal model. In addition, the sensitivity analysis demonstrates a linear correlation between depth defect and the failure pressure with 32.77%, followed by the length of defect and the inner diameter with 20.04% and 19.81%, respectively. According to the present regression analysis, the results obtained from the ANN-MLP and finite-element method (FEM) are more accurate compared with SVM-RBF and other analytical models. However, the predicted failure pressure values demonstrated strong agreement across all the five approaches.