The Early Fault Feature Extraction Of Train Bearings Under The Condition Of Variable Speed

Safety is the eternal theme of railway transportation. There are more than60railway trucks in china. Rolling bearing as one of the important mechanical parts easily has surface detachment fault, Which leads to accidents such as shaft overheating, shaft breaking or derailment. Therefore the implementation of early fault diagnosis is very important. Fault feature extraction is the primary part of the fault diagnosis technology. The current rotary bearing fault diagnosis based on steady characteristics of the stationary vibration system, such as frequency, amplitude and phase, it is not appropriate for the vibration signals under the condition of variable speed. In order to analyze the non-stationary signal of rotating bearing better, it is needed to explore the method of fault feature extraction and develop the new fault feature extraction technology. To achieve this goal, the article focuses on the vibration signal of rolling bearing under the condition of variable speed and studies the theory, implementation method and experiment research of the signal fault feature extraction.Speed estimation is one of key technology of the study on the vibration signal of the wheel-bearing under the condition of variable speed. For the variable speed estimation of wheel-bearings in strong background noise, a novel method with the short-time Fourier transform and BP curve fit (STFT-BPCF) is proposed. In the method, it calculates the time-frequency spectrum with STFT technique. Then the instantaneous frequency is estimated by peak detection. Taking the instantaneous frequencies as the input vectors, a BP neural network is trained to fit the discrete instantaneous frequencies. The effectiveness of proposed method is demonstrated by simulation. Experimental results show that proposed method provides better performance on variable speed estimation for wheel-bearings.For the adjacent order separation of rolling wheel-bearings under variable speed, a novel method with the fast ICA and peak detection (FastICA-PD) is proposed. In the method, it separates the mixed signal With fast ICA. According to the energy distribution characteristics of separated signal, the independent order signal is extracted by peak detection to improve the performance of adjacent order separation. The effectiveness of proposed method is demonstrated by simulation. Experimental results show that proposed method provides better performance on adjacent order separation for wheel-bearings.