Research On Remaining Useful Life Prediction Of Rotating Machinery Components

With the continuous improvement of modern industrial level,machinery and equipment are increasingly developing in the direction of large-scale,precision and complexity.As the most common mechanical equipment,rotating machinery components are the basis for continuous industrial production.The performance status of key rotating machinery components such as bearings,gears,and main shafts directly or indirectly determines whether major mechanical equipment such as aero engines,hydro-generators and gas turbines can operate normally.Rotating machinery components that run for a long time under complex working conditions will inevitably experience performance degradation,which will cause the remaining useful life(RUL)to decrease continuously,and the possibility of failure will gradually increase.Once a fault occurs,it can cause huge economic losses,serious casualties and a severe social impact.Given the importance of rotating machinery components in various types of machinery and equipment,research on the prediction method of its RUL will help enterprises to transform from the traditional passive maintenance mode to the active maintenance mode so that enterprises can make targeted production plans and maintenance strategies before failures occur.And it is of great significance for reducing the risk of equipment use,reducing the cost of equipment maintenance,improving the economic benefits of the enterprise and promoting the overall development of social economy.The traditional RUL prediction method based on the probability and statistical model often lacks systematic consideration of the failure process during modeling.It also has high requirements for the number of failure samples,and it is difficult to predict the RUL in real time.Therefore,this paper will investigate the RUL prediction methods of rotating machinery components from the perspective of failure system and performance degradation.The main contents are as follows:(1)Based on the connection relationship between rotating machinery components and driving system,a RUL prediction method for rotating machinery components based on an improved Exponential-Weibull distribution is proposed.Based on the reliability theory,this method develops an improved Exponential-Weibull distribution model by introducing correlation coefficients and uses maximum likelihood estimation to determine model parameters.Aiming at the problem that the accuracy of the parameter estimation results is relatively low due to the small number of failure samples,this method uses an improved self-service method to expand the sample size,thereby reducing the parameter estimation errors and improving the accuracy of the final RUL prediction results.A practical application of RUL prediction of bearings is given in this paper,and the results show that the method can describe the statistical characteristics of bearing life more comprehensively and has higher prediction accuracy than the traditional Exponential-Weibull model.(2)In the process from the healthy state to the end of life for rotating machinery components,various random external factors such as vibration,temperature and humidity will affect the end of life to varying degrees.However,the RUL prediction method based on the probability statistical model is often difficult to fully characterize the coupling effects of various external factors on rotating machinery components and depends on engineering practice experience when modeling.Aiming at the above problems,a method for RUL prediction of rotating machinery components based on ensemble Stacked Auto-Encoder(SAE)and degradation trajectory similarity is proposed.This method learns the performance degradation characteristics of mechanical equipment from different perspectives by establishing multiple SAEs with different initial hyper-parameters.Considering the characteristics of mechanical equipment performance degradation,the Cri index and characteristic frequency are introduced to filter the features and build ensemble feature set.Self-Organizing Mapping(SOM)network and SAE are used to fuse degradation features to generate health index/indicator(HI)that characterizes the degradation degree.Finally,the RUL prediction method based on the degradation trajectory similarity is used to realize the dynamic RUL prediction for rotating machinery components.The RUL prediction results of the PRONOSTIA bearing data set show that this method not only realizes the adaptive extraction of performance degradation features,but also obtains the ideal prediction accuracy.(3)Aiming at the problem that different features have different representation capabilities for the degradation process,this paper proposes a RUL prediction method for rotating machinery components based on improved SAE and degradation trajectory similarity.This method quantifies the importance of each feature in the feature fusion model according to the Cri values of different features and trains the feature fusion model by adding penalty coefficient to the weight of the input layer neurons corresponding to the features of different importance.Finally,this method realizes the RUL prediction of rotating machinery components by combining the RUL prediction method based on the degradation trajectory similarity.The bearing RUL prediction results show that this method makes full use of the representation capabilities of high Cri features to reflect the performance degradation characteristics of mechanical equipment more accurately.And compared with the RUL prediction method proposed in(2),the prediction accuracy of this method has been significantly improved.