Signal Analysis On Damaged Rail Tread Detection Based On Wavelet Packet And MFCC

Following rapid development of urban rail transit, rail transportation safety performance is highly valued. Now, the developing direction of the railway vehicle running as rapidly and safely as possible is the mainstream, therefore, the condition of the wheel becomes the main concern. The separation and detection of wheel tread surface damage signal is related to the train operation state, and passengers safety.At present, the wheel tread surface damage of the vibration signal processing method includes short-time Fourier transform and wavelet transform. However, two methods have advantages and disadvantages. This paper uses the wavelet packet transform of the signal analysis on the one hand. And on the other hand, MFCC feature extraction method is also tried in detecting the damage signal of the wheel tread surface, although it was only regarded as efficienct in speech signal processing in the past.. The main work of this paper includes:(1) Firstly, the Physical model of the wheel damage characteristics of the vibration signals is established through analyzing the vibration signal acquired by the vibration acceleration sensor Factors that affects the vibration signal are discussed.(2) Using wavelet packet transform and PNN method, MFCC and HMM method, the damage signal is analyzed. By changing the relevant parameters, the vibration signal detection is realized.(3) In the wavelet packet transform, it uses six different feature extraction methods and two different PNN model, based on four different feature extraction method of MFCC and 8 to 16 state of the HMM model, and a large number of experiments in MATLAB software and validation and optimization algorithm. Finally more than 70% of the results is detected by both the two different methods.(4), based on wavelet packet transform and PNN model and MFCC and HMM model, a GUI software package developed in MATLAB is applicable.