Research On Rolling Bearing Fault Diagnosis Based On Wavelet Analysis And Neutral Networks

Rolling bearing is one of the most significant elements in rotating machines yet with the characteristics of high damage rate and large lifetime randomness caused by process and environmental factors. The malfunction of rotating machine is various whereas about one third is resulted from rolling bearing fault. Therefore, to handle the working status and the fault formation as well as development of rolling bearing is currently important topic in the area of mechanical fault diagnosis.In this paper, the bearing signal acquisition method is improved through analyzing the vibration principles, fault causes and signal features of the bearing architecture. The ZigBee technique is adopted to reduce the complexity of the fault diagnosis system, hence to improve the efficiency. Mechanism analysis and intelligence diagnosis are the current two ways to detect and diagnose the rolling bearing fault from vibrate signals. The common methods for mechanism analysis are stochastic resonance and wavelet analysis, and for intelligence diagnosis are neural networks and support vector machine. However, these methods above cannot conceal their disadvantages in practical application which will largely influent the effectiveness of bearing fault detection and diagnosis. Hence, it is considered of great significance to make a further research and exploration on the problems and disadvantages of current methods, mainly focused on the mechanism analysis and intelligent diagnosis of rolling bearing fault and stood on the developing theories and methods.(1) Aimed to solve the problem of the less flexibility and high fault rate in the traditional signal acquisition system, a wireless acquisition network of rolling bearing vibration signal has been designed based on the analysis of the vibration principles, fault causes and signal features. The acquisition network adopted802.15.4and ZigBee as standard protocol. With the transmission rate of250kbps and a mode of wireless deployment, it reduces the complexity of a system and its fault rate, supporting the followed diagnosing method of bearing fault in principle as well.(2) This paper proposes a stochastic resonance method based on inheritance immune partial swarm optimization concentrated on the problem that happens where useful messages are often flooded by the noises. This method not only realize the successful extraction of weak signals in large noises background, but also solve the confinement that the basic stochastic cannot work in processing large parameters signals but only small ones. Also a PSO based inheritance immune method is put forward after further experiment which is applied in the key parameter optimization of stochastic resonance. At last, a self-adapted stochastic resonance method based on inheritance immune partial swarm optimization is raised and corresponding analysis and test are conducted with practical bearing fault data.(3) A diagnosis method of rolling bearing fault based on the second generation wavelet transform (SGWT) is proposed used to solve the difficulties in constructing ideal wavelet basis functions in the practical application of wavelet theories. This method use SGWT to decompose the vibration signals of rolling bearing faults to different scales, and then extract the resonance frequency band. Then, the Hilbert transform is used to demodulate the signals, and the frequency analysis of the signals demodulated has been done to obtain the wavelet spectra from which the fault characteristic information of rolling bearing are obtained. Practice and analysis on bearing data manifest that this method extract the feature frequencies of damaged bearings on all ranks respectively and realize the quantitative diagnosis of rolling bearing fault successfully.(4) As for the self-limitation in neutral networks, an intelligent diagnosis method of rolling bearing based on SGWT and neural network is under experimenting. Staring from the quality improvement of neutral network import signals, this paper proposes an intelligent diagnosis model of rolling bearing based on SGWT together with neural network which takes the advantage of SGWT and feature assessment both. When put into experiment analysis and engineering practice, this model shows that,for one thing,more fault information can be revealed from the combined features extracted from SGWT decomposed signals, for another, aimed to properly classify the options on sensitive features based on various diagnosing object health status, feature assessment can largely increase the accuracy in BP (Back Propagation) neural network classification, which manifests the effectiveness of the intelligent diagnosis model set in this paper.(5) As for the problem that rolling bearing fault is of typical small sampled feature, a parameter optimization support vector machine based rolling bearing intelligent diagnosis method is proposed. The influence on the efficiency of algorithm caused by the difficult in selecting proper model parameters in basic support vector machine can be removed by the proposal of genetic-immune particle swarm optimization (PSO) algorithm and support vector machine based intelligent diagnosis model, which is set on the analysis of parameter influences, model of parameter optimization and adoption of optimized genetic-immune particle swarm optimization (PSO) algorithm. Furthermore, the forecasting model is used to diagnose bearing fault. The results show that diagnosis model of SVM optimized by genetic-immune PSO algorithm can achieve automatic optimization of parameters, increase diagnosis accuracy than the conventional cross-validation algorithm, and is more fitting to classify the faulty samples scattered greatly.Based on the varied shortcomings of current rolling bearing fault diagnosing algorithm, this paper proposes a series of optimization algorithm. Verified by experiment, the algorithms all get a good result. So the research in this paper provides an orientation for Rotating machinery fault diagnosing.