Research On Prediction Method Of Product Life Based On Stochastic Process

With the continuous development of technology,the high integration and high complexity of modern equipment,the system's fault diagnosis,maintenance support and reliability issues have been highly valued by people,in order to save more manpower and resources and avoid unnecessary Property life safety,residual life prediction has become the core issue of management and maintenance and fault prediction of product equipment in recent years.Since the structure of many equipment systems is difficult to model directly,the data-based residual life prediction has gained wide attention and continuous progress in the past decade,and it is also the focus of current research.This paper mainly studies how to apply stochastic process theory to better describe the real degradation process of products,and then improve the accuracy of residual life prediction.In this paper,the residual life prediction methods in recent years are reviewed and analyzed.In view of the shortcomings of existing models,three time-continuous stochastic process theory-based models are proposed for residual life prediction.Firstly,a state space model based on Gamma process is used to characterize the degradation process.This is because the Gamma process has the necessary attributes for stable modeling such as stationary and independent increments,which can be used to describe the irreversible degradation process of product equipment.In order to determine the hidden layer parameters in the model,this paper uses the empirically maximized EM algorithm to estimate.For some problems in the solution process that are difficult to directly solve numerical problems,the particle filter algorithm is applied to solve the problem.Secondly,a two-parameter Wiener process model with random error is proposed.Then a Wiener process model with adaptive drift parameters is synthesized,which is combined with historical data and measurement error.Then the Kalman filter is applied to the real degradation state and corresponding.The state of the drift coefficient is estimated.For the unknown parameters in the model that are not easy to solve directly,the RTS smoothing algorithm and the empirically maximized EM algorithm are applied to estimate.Based on the research of the current Weibull distribution application,the Weibull distribution is applied to the product's residual life prediction,and the Weibull distribution-based degradation model is established.For the parameters that are difficult to solve in the model,the least squares method and the maximum likelihood are used.The estimation method numerically solves it.Finally,the above three stochastic process models are experimentally verified by the degraded data of some lasers and batteries.The experimental results show that the gamma process model inthis paper is the best to describe the degradation process of the laser among these three models,and the Wiener process is the most appropriate to the degradation of the battery.Weibull process model has a good effect on the case of known life data.