LS-SVMs-based reconstruction of 3-D defect profile from magnetic flux leakage signals

Abstract

Magnetic flux leakage techniques are used extensively to detect and characterise defects in natural gas and oil transmission pipelines. Based on the least squares support vector machines (LS-SVMs) technique, this paper presents a novel approach for the three-dimensional (3-D) defect profile reconstructed from magnetic flux leakage signals. The basic theory of LS-SVM for function estimates is given. The hyper-parameters of the LS-SVMs problem formulations are tuned using a 10-fold cross validation procedure and a grid search mechanism, and applying the pruning algorithm to impose sparseness on the LS-SVMs. The training data are composed of the measured and simulated data. A mapping from MFL signals to 3-D profiles of defects is established, the reconstruction of 3-D profiles of defects from magnetic flux leakage inspection signals is achieved and 3-D error of reconstruction results is analysed. The experimental results show that the LS-SVM has high precision, good generalisation ability and capability of tolerating noise.