Research On Application Of EMD And Entropy On Fault Diagnosis Of High Speed Rail Running Gear

High speed rail (HSR) has the advantages of large conveying capacity, fast speed, energy saving, etc, and it plays a very important role on the traffic, transportation, environment and economy. However, how to ensure the safe of HSR becomes a challenge task. High speed rail running gear is a important component which can affect the safe of HSR. Therefore, it is significant for HSR to identify faults of the train running gear. Signal analysis contributes to learn the running status of high speed rail running gear. In order to observe signals correctly and effectively, we always extract the valuable features of signal so that the important information hidden in signal becomes more obvious. Signal analysis is essential to the state recognition of high speed rail running gear, but the traditional signal analysis methods have some limitations when dealing with non-stationary signals. This thesis applies the EMD and entropy on the diagnosis of High speed rail running gear faults.The time-domain approximate entropy reflects the complexity of the vibration signal, and EMD decomposes the complex signal into a set of mode components. The time-domain approximate entropy theory and the power of mode components contain signal characteristics. So this thesis combines EMD and approximate entropy theory to extract the fault features of high speed rail running gear, and the BP neural network is applied as the classifier to recognize the status of HSR. The experimental results show that the proposed fault diagnosis method may identify high speed rail running gear faults effectively.In addition, a new feature extraction method for the fault diagnosis of HSR based on EMD together with fuzzy entropy is also presented. In detail, EMD decomposes the vibration signals which collected from the train in different running condition into a series of intrinsic mode functions. We compute the mean of fuzzy entropy of all intrinsic mode functions instead of the first few intrinsic modes of functions as features of signal. BP neural network, SVM and Bayes Network are employed as classifier to make the faults diagnosis of high speed rail running gear based on different features to observe effectiveness of our method. The experiments show that the extracted features may recognize fault patterns accurately and effectively.