Study On Fault Diagnosis Of Gearbox Based On Noise Reduction By Particle Filter

Gearbox system is commonly used in the transmission of rotating machinery equipment, which performance fine or bad is directly related to quality of the entire device. It has great significance to monitor its state, detect faults timely and classify the faults, which also can prevent unnecessary loss. Generally, sampling signal is mingled with strong background noise, which some useful signal may be covered for the badly environment of gearbox working. So it needs to carry out noise reduction before failure analysis.Particle filtering is a new model-based state estimation technique. Studying the principle of particle filter in-depth then use it to the noise reduction of gear vibration acceleration signals. Provided that the signal model and noise statistics is know when use particle filter technology to denoise. In this paper, it's realized as follows: firstly, establish time series AR model of the vibration acceleration signal, then the coefficients of this AR model as the particle filter coefficients of the state equation; Use wavelet transform threshold de-noising method to extract noise signal when comprehend the principle of noise reduction by wavelet theory. The particle filter observation equation will be use the extracting noise which assumed additive.Based on the above theoretical analysis, the gearbox vibration acceleration signal sampled through experiment is analyzed and processed. The first step is denoised by particle filter; second step is classifying the fault mode by BP neural network. As an adaptive pattern recognition technology, neural network has been widely applied in the field of fault pattern recognition and the theoretical research is mature. In this paper, we use two sets of data to diagnose faults by BP neural network, of which one is denoised by particle filter, and the other is not. Extracting the energy spectrum scales as the BP neural network input vector. The diagnosis results show that the data denoised by particle filter is better than other after network training. It is also confirmed that the effect of noise reduction by particle filter is good.