Performance Degradation Assessment And Maintenance Optimization Of Multi-state System Based On Monitoring Data

At present,sensor technology has been widely used in the industrial field,and the massive monitoring data collected by them has become an important data source for system condition assessment,residual life prediction and maintenance strategy optimization.In particular,with the continuous improvement of artificial intelligence theory,deep learning,as the latest research result in the field of artificial intelligence,can learn the multi-level representation of data through the specific deep network structure,that is,the deep abstract characteristics of data,so as to have more powerful data expression ability.Although the state or remaining life of the system can represent the change of the system over time,it can not directly guide the maintenance strategy.Therefore,combined with the state change of the system,the appropriate maintenance strategy is adopted and the optimization problem of the corresponding model is solved.Based on the above ideas,the main contents of this paper are as follows:(1)Degradation feature extraction based on monitoring data.The extraction of degradation eigenvalues is an important part of multi degradation state system modeling and state recognition.Whether the features are reasonable or not will directly affect the reliability of the model and the accuracy of the learning results.In this paper,principal component analysis,automatic coder model and variational automatic coder model are used to study the standardized monitoring data,and the effect of variational automatic coder model in the degradation feature extraction is compared.The variational automatic coder model can better retain the information of the original data and obtain low dimensional and effective degradation features.(2)Remaining life prediction and multi-degradation state division based on degradation features.According to the degradation features,this paper uses the sequential neural network based on time window to predict the remaining life of the system,and puts forward the models based on different circulation bodies.It also illustrates the degradation scenarios that the models with different circulation bodies adapt to and the accuracy of the proposed model for residual life prediction.At the same time,although the degradation characteristics can reflect the basic characteristics of system degradation,continuous degradation characteristics are difficult to directly show the health status of the system,and it is difficult to further develop scientific maintenance strategies based on the degradation characteristics.In this paper,clustering algorithm is used to identify the real number of clusters in the degenerate features,which is to map the continuous degenerate features to the discrete degenerate states.(3)Modeling and optimization of maintenance strategy based on multiple degradation states.On the basis of degenerate state,considering the hierarchical relationship between different states and degradation characteristics,a multi degenerate state system model is established by using hidden semi-Markov model.The complete degradation feature contains all the information from normal state to fault.Combined with the division result of multiple degradation States,the established model is trained and learned.After training,the model can learn the transition probability of different states and the feature distribution of different states,and then identify different degradation states.At the same time,the service age regression factor is used to describe the effect of incomplete maintenance,and to minimize the cost rate per unit time as the goal,the preventive maintenance optimization problem under non equal cycle is established,and the maintenance optimization model is solved by combining the enhanced elite retention genetic algorithm to obtain the optimal solutions of maintenance times and maintenance intervals,and formulate scientific and effective maintenance strategies.The artificial intelligence model represented by deep learning can extract the degradation characteristics from the monitoring data,so as to further realize the residual life prediction.At the same time,based on the degradation characteristics,the state stages can be divided to realize the identification of degradation States,and the multi state degradation model is further established.On the basis of multi state degradation model,the preventive maintenance strategy of non equal period is adopted,the effect of incomplete maintenance is described by service age regression factor,and the cost rate per unit time is minimized as the goal.Finally,the optimal solution of maintenance times and maintenance interval is obtained by combining the enhanced elite retention genetic algorithm.This paper provides the overall research ideas and technical implementation from monitoring data to maintenance strategy optimization.