Abstract: This article describes a method for predicting the values of track’s geometric characteristics during the period between two consecutive measurements. As is known, track condition analysis is based on periodic measurements of geometric parameters. Changes in track condition manifest as an increase in the size of “large” defects and a decrease in “small” ones. This process is non-stationary, dependent on the physical properties of the track and axle load, and is not linked to the measurement schedule. Therefore, there may be situations where defects (irregularities) of critical or near-critical size emerge between measurements, which can lead to undesirable consequences. The issue of predicting the emergence of new defects, larger than those recorded in the most recent measurement, during the period between measurements, has not been explored. Naturally, such defects should be addressed without waiting for the next measurement cycle. It should be noted that in reliability theory, this process corresponds to the concept of an unexpected failure. Our approach addresses the problem of preventive track maintenance by providing information on defects whose sizes exceed the maximum sizes registered during the most recent measurement, prior to the next measurement cycle (1-2 months). The method also allows to identify sudden spontaneous deterioration of the track, which does not follow from a regular trend. The method is based on using the homogeneity (compactness) property of the results from successive measurements of the geometric characteristics of the track. Calculations are performed using the Irregularity Size Distribution Function (ISDF). For the analysis, we use exponential approximation of this function. Classification of the results and decision-making regarding the occurrence of a defect larger than the maximum size before the next measurement is made using the k-nearest neighbors’ method, which is commonly used in artificial intelligence tasks. According to the results of the experiment, for defects of the surface type, the probability of correctly predicting the spontaneous increase in the maximum defect size before the next measurement is 0.89, while the probability of a false positive is 0.11.
Abstract: This article describes a method for predicting the values of track’s geometric characteristics during the period between two consecutive measurements. As is known, track condition analysis is based on periodic measurements of geometric parameters. Changes in track condition manifest as an increase in the size of “large” defects and a decrease in “s...Show More