Electric Multiple Unit Mechanical Transmission Bearing Vibration Signal Analysis And Fault Diagnosis

In recent years, with the development of China’s opening up and the social economy, construction speed of high-speed railway is also accelerated. With the increasing railway mileage and traffic, the installation of EMU (Electric Multiple Unit) also increases rapidly, therefore the safety of the high-speed railway operation is becoming the most important issue of operators’concerning. Rolling bearing, as an important rotating part in EMU, is also one of the main sources of the device fault. Early fault of rolling bearings can be easily expanded in high-speed and heavy-loaded train, which could cause the train failure. To ensure the safe operation of EMU, the study of the rolling bearing fault diagnosis has practical significance. This thesis analyses the vibration signal from four working states (inner ring fault, outer ring fault, roller fault and normal bearing) of EMU rolling bearing in both time domain and frequency domain. Combining with the optimized Ensemble Empirical Mode Decomposition (EEMD) and Back Propagation (BP) neural network, bearing fault intelligent diagnosis and bearing working status recognition are achieved.Firstly, this thesis introduces the rolling bearing fault forms, natural frequencies and fault characteristic frequencies of its components’vibration signal. Then we analyse the changes and laws of five time domain features of vibration signal in four operating status; and analyse the signal in frequency domain by EEMD. Two optimization methods about the calculation efficiency of EEMD are proposed and its feasibility is demonstrated. Secondly, we use the optimized EEMD method to get8Intrinsic Mode Functions (IMF) and analyse their energy distribution. The energy of each IMF is extracted as the feature vector input of BP neural network. By using the nonlinear mapping function of BP neural network, a neural network mapping between its feature vectors and rolling bearing operating status is established. And then we train the neural network, test and verify its ability of status recognition. Finally, this thesis introduces the rolling bearing fault diagnosis system used in experiments. The hardware and software of the system, which are tested in experiments, are described respectively.