Rolling Bearing Fault Diagnosis Based On Rough Set Theory And Artificial Neural Network

As the modern mechanical equipment increasingly become huge-size, preciseness and automatization, effective fault diagnosis has great value for the reliable and safe operation of the mechanical system. Rolling bearing is one of the most ordinary parts in mechanical machine, it has the advantages of small friction, easy assembling and so on. and its running status is directly related to the performance of machine operation. Therefore, applying condition monitoring and fault diagnosis for the rolling bearing is extremely essential, especially for the avoiding of economic losses and happening of major accidents.Vibration analysis is the most commonly used and effective way of rolling bearing fault diagnosis. To have the knowledge of vibration mechanism, confirm the method for the measure of vibration signal, simulate rolling bearing fault from the inner ring to outer ring, then construct the fault diagnosis experimental system. After gathering and processing the vibration signal, analysis and extract the eigenvectors in time domain, frequency domain, time and frequency domain, which can reflect the operational status of rolling bearing.EMD provide a finer decomposition by its characteristic time scales, it has high SNR and adaptability, especially good for the analysis and processing of nonlinear and nonstationary properties. As the advantage of multi-resolution, wavelet analysis is widely used in noise reduction and compress of signal. Window function is used in wavelet analysis to decompose frequency bands locally. From the point of wavelet energy, we can analysis the wavelet layer which we are interested in. Wavelet packet analysis is the development and extension of wavelet analysis, it performs the decomposition of low and high frequencies simultaneously, improves the time-frequency resolution, and has more practical values.Rough set is a new mathematics theory dealing with uncertain, inaccurate and incomplete information. It's widely used in knowledge discovery, decision analysis and artificial intelligence. In rolling bearing fault diagnosis, while the diagnosis accuracy is not changed, RS can reduce the dimensions, of feature parameters, keep the core attributes, to reduce the impact of calculation and uncertainty, lower the complexity and scale of fault diagnosis system.Artificial neural network is a powerful information processing system, which simulate the structure and cognition function of human brain. It has high self-adaptability, parallelism, adaptive learning and generalization. A trained neural network can judge the malfunction part by fault symptom, and applied in rolling bearing fault diagnosis and pattern recognition. This thesis mainly introduces three solutions for rolling bearing fault diagnosis. Firstly, taking the normalized feature vector as input eigenvectors of BP network, which leads to a rolling bearing fault diagnosis result; Secondly, construct a RS classifier, classify the rolling bearing operation status by self-learning; Thirdly, using RS as the pre-processor of feature vector, then discrete data, reduce attributes and get the final decision rules, lastly taking the optimized characteristic parameter as the input of BP network. Following the3experiments of the fault diagnosis, the RS theory is combined with ANN, and a more satisfactory fault diagnosis system with better diagnostic accuracy and diagnostic efficiency is established.This thesis focuses on the extraction of feature vector, making flexible use of RS to pro-processing data, eliminate redundancy and prevent information explosion, while fully using ANN advantages, such as strong fault-tolerance and generalization ability, to get a better result of rolling bearing fault diagnosis. At the same time, RS theory as a brand new technique for feature dimension reduction has a great development and wide application in intelligent fault diagnosis.