Identification of Rail Crack Defects Based on Support Vector Machine and Artificial Neural Network

In order to ensure the safety of people's livelihood, it is necessary to grasp the damage degree of the rails in time. In this paper, the most common crack defects in rails are considered as an example. Utilizing the far-field advantage of guided wave detection, experiments and simulations are used to obtain detection signals for eight different depth cracks under multiple excitation/receiving modes. The characteristics of each detection signal are analyzed in time domain and frequency domain, and the 6 best damage feature values are extracted to depict the damage degree. Taking the feature vector as input, the support vector machine model and artificial neural network model for rail damage recognition were constructed respectively. The influence of the hyper-parameters and feature numbers of the two models on the accuracy of the defect recognition model was analyzed to optimize these models. The results is showed that the support vector machine model has superior performance and better recognition effect for rail crack damage under small sample conditions than the artificial neural network model.