Feature selection and classification algorithm for non-destructive detecting of high-speed rail defects based on vibration signals

Many rail accidents were caused by rail defects, therefore the detection of the rail defects is of vital importance. Using simulated and experimental measurements, the rail defect detection was carried out. The feature parameters were extracted both from time domain and time-frequency domain. Then the sequential backward selection method was applied to select the important feature parameters. After optimizing of the feature parameter set, support vector machine method was applied to recognize and classify the rail defects. It has been proved that the proposed algorithm of analyzing and processing the rail defect vibration signals is an effective and non-destructive detecting method of the rail defects.