An automated long range ultrasonic rail flaw detection system based on the support vector machine algorithm

Abstract

This paper presents an automated long range ultrasonic (LRU) flaw detection system based on an effective pattern recognition approach. The method comprises three stages: 1) multiple feature extraction techniques were applied for better representation of the signals in time and frequency domain, 2) feature selection was employed next for feature ranking, according to discrimination power, and finally 3) the classification task was accomplished by means of a kernel-based support vector machine (SVM). For the training and validation of the algorithm, an extensive experimental investigation was carried out on the foot section of a 4.30m long rail (CEN 56). Different depths of transversal slots have been induced in the foot at 3m from the excitation point. The results show that the proposed method is able to effectively detect flaws. © 2012 WIT Press.