Research On Fault Diagnosis And Prediction Method Of Rotating Machinery

Rotor system and bearing are key components of rotating machinery, but their long-term work under high-speed and full load is likely to result in failure. Based on such vibration signal processing features as online, real-time fault diagnosis, and the setbacks of spectral analysis extracting nonlinear vibration signal, we researched wavelet packet analysis and sample entropy analysis method to extract fault features, and diagnose faults by gray relational analysis and BP neural network in this dissertation. The main contents are as follows:1. Establishing rotating machinery vibration experiment platform, signal acquisition analysis and diagnostic software system.Set up the vibration test stand, which simulates the unbalance, misalignment fault of rotor system. The signal acquisition circuit is designed base on DSP as the processor, and fault signals are transmitted to LabVIEW platform which hosts computer. The PC software completes signal display, storage, Data query, wavelet de-noising, feature parameter calculation, BP network fault diagnosis and vibration trend forecasting, which can be used for the actual vibration signal’s analysis and diagnosis.2. Studying de-noising method for improved wavelet threshold quantization function, and proposing wavelet packet sample entropy vibration signal feature extraction method.The traditional soft and hard threshold de-noising method has a constant bias and discontinuity shortcomings at the threshold, so an improved threshold denoising function is used, with its parameters adjustable. Experimental results show that this method of de-noising performance is better than the traditional method. We studied the wavelet packet and wavelet packet sample entropy features extract of the vibration signal. Combining wavelet packet analysis and complexity theory, we proposed wavelet packet sample entropy feature extraction method. Firstly, the vibration signal was decomposed by wavelet packet, then the reconstructed signals in the first three sub larger sample entropy energy generation are computed as the characteristic parameters. The actual diagnosis result indicates that this method works better than the direct calculation for diagnosis, and reduces the dimension of the feature parameters.3. Studying gray correlational analysis and BP neural network in rotating machinery fault diagnosis, and proposing wavelet packet sample entropy BP network fault diagnosis method.According to wavelet packet energy feature vectors, we studied the gray correlation analysis for normal, inner, outer and ring fault diagnosis. Under small sample the identify results is better. Combining wavelet packet analysis, we extracted the entropy features of sample bearing wavelet package. BP network is taken as classifier. The results show that this method performs well in identifying inner faults and rolling faults, and the identification percentage is6percent higher than traditional BP network diagnosis.4. Researching the GM (1,1) gray model for vibration trend prediction, Metabolic model’s prediction accuracy is higher than GM (1,1) model. The gray model was used for simulation prediction vibration failure, which included rising, random fluctuations and growth trends. The GM (1,1) modeling data is fixed, and new information will be added to the model prediction, the old information removed, and the metabolism prediction model modified online by model parameters was predicted. Simulation results show that the average relative error of15-step prediction for rising trend is1.33%, lower than5.93%of GM (1.1); average relative error of10-step prediction for comprehensive growth trend3.674%, lower than5.17%of the GM (1,1). Finally, we set a BP neural network prediction model for bearing failure prediction threshold feature information RMS. Results show that RMS can be used as bearing failure prediction feature and can predict bearing faults in advance with higher prediction accuracy.