Magnetic flux leakage (MFL) detection has been widely applied in defect identification for damaged oil and gas pipelines. However, current machine learning algorithms face challenges in classifying MFL defect data, such as low prediction accuracy and significant errors. To address these limitations, this study proposes a novel defect classification method for oil and gas pipelines based on a particle swarm optimisation-radial basis function (PSO-RBF) model. Firstly, a convolutional neural network (CNN) is employed to automatically extract features from MFL data, thereby improving the utilisation efficiency of three-dimensional MFL data. Secondly, a hybrid feature selection approach combining the Pearson correlation coefficient and random forest (RF) feature importance evaluation is implemented to refine the feature set, laying a solid foundation for enhanced classification accuracy. Subsequently, to improve the generalisation capability of the radial basis function (RBF) algorithm, the particle swarm optimisation (PSO) technique is utilised to optimise the RBF classification model. Finally, the proposed method is validated using MFL signals collected from an experimental platform, achieving a remarkable classification accuracy of 94.0%. Experimental results demonstrate that the proposed PSO-RBF model exhibits excellent performance in classifying various types of pipeline defect. A comprehensive comparison with other classification methods further confirms the superiority of the proposed approach.
Wei et al. (Thu,) studied this question.