Train Bearing Fault Diagnosis Based On Time-Frequency Analysis And BP Network

Rolling bearing is an essential part of a train.The running state of the rolling bearing has a direct impact on the safety performance of a train.Rolling bearing faults may cause closure of railway lines,or even safety accidents leading to casualties.This paper chooses rolling bearing as the object of study,extracting its vibration signals in four different running states,normality,inner race faults,outer race faults,and rolling element faults,thus analyzes and extracts feature vector through wavelet analysis and Empirical Mode Decomposition(EMD)and completes a bearing fault diagnosis system based on neural network.Finally,the effect of rolling bearing fault diagnosis is verified by measured signals,which provides an effective way for the early fault diagnosis of rolling bearing.In general,this paper mainly studies the following subjects.(1)Fault mechanism and signal feature of rolling bearing.This paper subdivides the research directions by analyzing the structural composition of rolling bearing and characteristic frequency of vibration signals.(2)Noise reduction of original signals.Signals extracted by sensors contain environmental noises,which may cause interference to the judgment of faults.This paper elaborates on noise reduction by wavelets,which denoises original signals and raises signal-to-noise ratio.(3)Feature extraction of vibration signals.Decomposing signals with EMD and wavelets produces IMF components as well as components in different frequency bands.Feature vectors in signals are determined to be extracted based on RMS by analyzing those statistical indicators in time domain and frequency domain.(4)Research on the construction of BP network mode.EMD-BP network mode and wavelet-BP network mode are built by time-frequency analysis of EMD and wavelet,using self-adaptive learning rate algorithm to improve standard BP algorithm.Comparing and analyzing the performance of different modes,the result shows that the feature vector which is decomposed and extracted by wavelets can effectively accomplish the intelligent diagnosis of rolling bearing faults in the BP network improved by self-adaptive learning rate algorithm.