The machining error and the running wear of railway wheelset will lead to the deviation at the radius of the rolling circle, which will affect the running safety and comfort. However, during the running process of a railway vehicle, it is difficult to accurately identify the wheel diameter difference (WDD) with the existing detection methods. The main purpose of this study is to combine the advanced signal processing method adaptive chirp mode decomposition (ACMD), the feature extraction method fractal box dimension and the pattern recognition algorithm to effectively detect the WDD. The chirp modes from the lateral acceleration signal of axle box are obtained by ACMD. The first three chirp modes are selected and divided into several segments by the windows with the same time. Then, the chaos degree of signal in each window is measured by fractal box dimension algorithm, which is used as the input feature vector of the kernel extreme learning machine (KELM). A dynamic model of railway vehicle is established based on SIMPACK software, and the proposed detection method is analyzed and verified. The results show that the fractal box dimension can effectively mine the potential data features of chirp modes, and the proposed method has strong detection ability for standard wheel diameter, in-phase WDD, and anti-phase WDD.
 

Wheel diameter difference; Adaptive chirp mode decomposition; Fractal box dimension; Axle box; Pattern recognition