Identification And Classification of Rail Damage Based On Ultrasonic Echo Signals

Rail ultrasonic flaw detection is an important method for railway operation and maintenance. The accuracy of damage detection directly affects the safety of railway transportation systems. Rail damage detection, identification and classification are the key to ultrasonic flaw detection systems. At present, most of the researches focus on the local damage of rails, such as rail head nuclear damage, rail waist cracks, or through holes at the bottom of the rail. The classification and identification of typical damages of the entire rail have not been systematically studied. In this paper six features are extracted by principal component analysis(PCA), they are the ratio of the energy of the signal processed by wavelet decomposition, the area surrounded by the 100-150 sampling points and the horizontal axis, the maximum amplitude and its position of the 100-150 sampling points, and the maximum amplitude in the 240-300 sampling points, and the position in the 300-400 sampling points, which are used to present the characteristics of the rail damage. Least square support vector machine (LSSVM) is introduced to create the classification analysis models, the eigenvalues extracted by PCA are the input of the models and any rail damage is the expectation output. In order to improve the accuracy of the classification models, particle swarm optimization (PSO) algorithm is introduced to optimize the parameters of the model, seven classification analysis models of the rail damage are built. Experiments results show that the typical damage in the standard test rail can be accurately identified and classified, the classification recognition accuracy is about 92%. This method can recognize the typical rail damage in the national standard. It has definite practical application value in the field of rail inspection.