Computer vision assisted high-speed train hunting stability monitoring using an artificial neural network approach

The phenomenon of hunting motion in the process of high-speed train operation will greatly reduce the ride comfort, and may even cause serious safety accidents such as train derailment. According to the image characteristics of high-speed train wheel-rail contact, a train hunting motion monitoring method based on You Only Look Once (YOLO) neural network and machine vision is proposed. The method can detect the high-speed train hunting motion state and parameters, and send an alarm signal when the relative wheel-rail displacement exceeds the threshold. The wheel-rail contact image during high-speed operation is obtained by installing a camera system at the bottom of the high-speed train bogie. A detection model of wheel-rail contact parameters is then established and the key position of wheel-rail contact is identified. Combined with the image processing algorithms including the Canny edge detection and Hough linear transform methods, the relative wheel-rail displacement during the process of high-speed train hunting motion is calculated. The accuracy and reliability of the model is validated by field measured data. The results show that the proposed method can accurately detect the high-speed train hunting motion under normal illumination, overexposure and dark conditions. The image processing speed is up to 60 frames per second, which can meet the needs of real-time detection in the high-speed railway application environment.