Rail corrugation is a typical rail cyclical disease which often occurs on heavy haul, urban transit, and high-speed railways. Rail corrugation has a significant impact on vehicle dynamic performance, especially on the axle box acceleration. It may cause the bolts of axle box to loosen or break the rail fastener, and even affect the operation of vehicle. Therefore, it is necessary to discover rail corrugation in time and to repair it by rail grinding, which is an important way to improve the safety of the vehicle. A numerical analysis method based on adaptive time-frequency feature extraction is proposed in this paper. First, acceleration sensors are installed on both the left and right side of the axle box. Then the vibration features of the axle box are extracted according to the line mileage segmentation based on the adaptive time-frequency feature extraction method proposed in this paper. Finally, the impact of different wavelength and different section length of rail corrugation is compared using field test data. The test results show that the method proposed in this paper can accurately extract the features of different wavelength and different section length of rail corrugation. Moreover, compared with traditional methods, this method is proven to be strongly adaptive and highly accurate.