Research On Optimization Of Predictive Maintenance Based On The Covariate Uncertainty Of Proportional Hazard Model

Predictive maintenance(PM)is a new maintenance strategy,which can improve the reliability of mechanical equipment,reduce maintenance cost and improve production efficiency.PM is used to analyze the equipment by using the condition monitoring data,so as to achieve the purpose of normal operation.However,due to the complexity of the internal structure and external operating factors of the system,it is difficult to measure the system parameters,which lead to the monitoring data can not reflect the internal degeneration of the system.Therefore,it is necessary to comprehensively consider the historical life data and the condition monitoring value to characterize the failure rate.The proportional hazard rate model(PHM),which takes monitoring state data as the covariate,can effectively solve this problem.In order to improve the prediction accuracy of degradation and the efficiency of maintenance decision-making,this paper carried out the optimization research of PM strategy based on the uncertainty of PHM covariate and took the failure rate evaluation as a health index.The main content and innovation are as follows.(1)The state monitoring data can not directly represent the degradation state,and the uncertainty of system parameters measurement is caused by the operating conditions.In this paper,a predictive maintenance model based on PHM is proposed,which takes into account the distribution parameters of uncertain covariates.Firstly,the Bayesian parameters updating prediction model is used to update the covariates parameters distribution.Secondly,the Monte Carlo simulation is used to simulate various covariate degradation scenarios to describe the uncertainty.The covariates considering multiple degradation evolution scenarios are fused with the historical life data and the failure rate is evaluated.Finally,the optimal maintenance interval is solved by the maintenance optimization decision based on life replacement strategy.(2)In view of the fact that it is difficult to model the degradation process,the neural network model based on mining the internal correlation of monitoring data can solve this problem,and the single model has the disadvantage of low prediction accuracy.Therefore,the predictive maintenance of covariate combination prediction model based on the PHM is put forward.Firstly,BP,RBF and GRNN neural network models are used to predict the degradation process respectively,and the prediction results of the sub-models are combined to reduce the uncertainty of covariate prediction by the combination prediction method.Then the prediction covariate and the historical life data are fused to evaluate the failure rate.Finally,the optimal maintenance interval is solved by the maintenance optimization decision based on life replacement strategy.Aiming at the limited prediction accuracy of the combination prediction model,a combination prediction model,based on the failure rate hierarchy is proposed to improve the maintenance efficiency.