Railway wheel defect identification using the signals reconstructed from impact load data

Wheel Impact Load Detectors are common devices that measure the rail response made by the wheel-rail contact to estimate the condition of the in-service wheels. The data collected by the multiple sensors can be fused to reconstruct a wheel-rail contact pattern over the circumferential coordinate that provides some description of the wheel condition. Moving of a defective wheel with different velocities and axle loads influence the wheel-rail interaction and the pattern reconstructed. As a result, there is a range of variation in the pattern reconstructed for each defect. Therefore, this paper aims to tackle this challenge and to classify the railway wheel defects using pattern recognition tools. Due to the lack of real data, ADAMS/Rail is used to model the wheel-rail contact and to simulate the data collected by the sensors. Then, based on the fusion algorithm, the condition state signals are reconstructed for different wheel defects with different velocities, and axle loads. Then a dataset based on these patterns is generated that is used for training, and testing the classifiers. In this paper, the magnitude of the signal is directly used as the features. The results of the classification show that the wheel defects including a minor defect can be correctly classified with zero error.