Research On The Feature Extraction And Performance Degradation Evaluation For Rotating Machinery

With improvement of technology and development of mechanical equipment towards higher speed,heavier load and higher stability,mechanical structure is getting to be complicated and the working environment is getting increasingly bad.For health condition of components will be worse as time goes by in the service process,that study on machinery fault diagnosis technology make it great significant for guaranteeing safe operation of machinery and improving productivity.Rotation machinery is one of the most important component of mechanical transmission system,such as bearings and gears,its health condition effects overall reliability of mechanical system.Fault features extraction and performance degradation assessment of bearings and gears is the key research for equipment fault diagnosis.This paper introduce the frequency components of fault signal for inner-outer ring of bearings and local gear on the base of analyzing fault mechanism.Time-frequency domain features of fault vibration signal are introduced further by simulation.A novel second generation wavelet transform(SGWT),called fully adaptive second-generation wavelet transform(FA-SGWT)was presented for fault features extraction,which optimizes simultaneously the orders and the coefficients of lifting operators by using the proposed variable chromosome length co-evolutionary algorithm.Meanwhile,a new index named Crest Factor of Envelope Spectrum is formulated as fitness evaluation function,considering both impact intensity and their cyclicity.The FA-SGWT is verified by the comparison between SGWT,spectral kurtosis approach and FA-SGWT and field date of the sound signals from wheel bearings of running wagons.For probability similarity at present has well entered into an early saturation stage,in other words,monitoring target has achieved upper limit while not fails yet,this paper only study on failure assessment model based on reconstruction and boundary.A gear fault degree assessment method basing on autoregressive(AR)model and auto-associative neural network(AANN)was presented in this paper for reconstruction failure assessment model.The core of method is to extract features of vibration signals by AR model coefficient as the input of AANN.Then,the assessment of gear fault degree can be accomplished by AANN and evaluated fault degree by proposed index--D-value of root mean square.The method can directly identify gear fault degree by analyzing data of gear vibration corresponds to different gear fault degree.On this basis,using experimental data during gear full life to verify the effectiveness of this method for further verification,the results shows that it can discover incipient fault in time which could not be observed by naked eyes and directly reflects the deepening of gear fault degree while gears’ performance is degenerating.A rolling bearing fault degree assessment method basing on ensemble empirical mode decomposition(EEMD)and kernel mahalanobis distance was presented for boundary failure assessment model.This method extracts feature information through the standard deviation of IMF obtained by EEMD decomposition,all the standard deviations of a sample consists of a feature vector.Combine the feature vectors extracted by normal samples and the feature vector extracted by test sample into a feature set,evaluating the bearing performance degradation degree by calculating the kernel mahalanobis distance between the feature set and the feature vector extracted by test sample.Fault degree assessment method basing on EEMD and kernel mahalanobis distance is verified by gears’ full life tests and data analyses.The method can exactly reflects the deepening of gear fault degree and keeps track of the development trend of faults.