A rail profile matching method based on dynamic time warping

Precise matching of rail profiles is an important prerequisite for studying changes in their geometric shapes. Due to random vibration is generated when the track inspection trolley travels, the vibration of the sensor is caused, the acquired data are distorted. In the measurement process, there is some difference in the vibration of each sensor because the absolute symmetry of the sensor cannot be guaranteed. The traditional feature point algorithm can only rigidly compensate for vibration-induced noise, and sometimes feature points cannot be extracted. Based on this, a dynamic time warping algorithm was proposed. From the preliminary matching data, the data of the standard rail profile and the lumbar profile of the measurement rail were selected as two time series, and the distance matrix was established. The recursive idea was used to calculate the cumulative cost matrix with twisting the time axis. The optimal regulatory path was obtained. The correspondence was found between each point in the two time series so as to achieve the exact match of the track profile. Experimental results show that compared with the feature point algorithm, the matching accuracy of the dynamic time warping algorithm is improved by nearly 7 times, the matching effect is good, and the stability is high. © 2019, Editorial Office of Journal of Vibration and Shock. All right reserved.