Railway Track Short-Wave Defect Identification Based on Axle Box Acceleration and Neural Network

Track welded joint impact and rail corrugation are common track short-wave defects in high-speed rail-way lines. Track short-wave defect inspection and analysis using vibration signals of vehicle dynamic response collected by axle box accelerators is relatively a proven technology, which is a great help of track health management. However, the real inspection data are pretty complicated. The accuracy of short-wave defect identification has much room for improvement. For deeper analysis of the characteristics of track short-wave defects and enhancing the quality of track health management, this paper pro-posed a new method for railway track short-wave defect identification from the perspective of real inspection data. Vehicle dynamic response vibration data are collected by a comprehensive inspection train and manually rechecked on site. By the on-site confirmation, typical samples of defects are integrated and divided into three categories: welded joint impacts, rail corrugations and non-defects. Signal processing approaches are applied to extract the features of those samples. The extracted features such as energy factor, crest factor, kurtosis, spectral entropy, etc. are input into a neural network for automatically learning and classifying. The results show that, by the combination of approaches of feature extraction and neural network, the impact defects and corrugations can be effectively identified from vehicle dynamic response vibration signals. Improving the quantity and quality of defect samples is the key point of identification accuracy enhancement. This method shows application prospects of axle box acceleration data mining and analysis of railway line health evolution.