Research On Reliability Modeling Methods Based On Wiener Process

Reliability modeling and analysis based on performance degradation, which is a heated area of current reliability study, is a key technique to tackle reliability design, analysis, test and assessment for products characterized as small sample, highly reliable and long life. Due to the capablility of modeling the degradation processes of many typical products and its good analytical and computational properties, Wiener process is discussed by more and more researchers in reliability filed. However, most existing studies of reliability modeling based on Wiener process are applications with respect to some particular products. A systematic study of the theory and methods has not been found, which is an obstacle of its further application and development in reliability modeling. Hence, this thesis focuses on dealing with various problems encountered in reliability modeling and analysis based on Wiener process in such aspects as degradation model recognition, parameters estimation, reliability assessment and lifetime distribution prediction. To be more specific, the problems include reliability modeling and analysis of one-dimensional and multi-dimensional degradation failure, competing failure modeling and analysis of degradation failure and random failure, reliability assessment and lifetime distribution prediction. The main achievements are as follows.(1)Model recognition methods of Wiener process are discussed. The definition, properties and the lifetime distribution of Wiener process are presented. Then model recognition methods are given, including autocorrelation method, likelihood ratio test method and sequential method. The proposed methods are of practical value in degradation model recognition in engineering practice.(2)Reliability modeling methods of one dimensional Wiener process degradation failure are investigated. We first present the methods based on degradation data and on lifetime data. In case that both degradation and lifetime data are available, we give a modeling method combining them. The proposed method is validated through the reliability assessment of metalized film pulse capacitors. For destructive measurement products and the ones whose performance degradation can not be measured, we study the modeling based on marker data. Methods of parameter estimation and lifetime distribution prediction for either kind of products are given. The methods are demonstrated using simulated examples. These methods provide a valid means to the reliability modeling when degradation data are few.(3)Reliability modeling methods of multi-dimensional Wiener process degradation failure, to be more specific, the two-dimensional one, are studied. In case that the degradation of the two performance characteristics are independent, modeling methods based on degradation data, lifetime data and the combination of them are present. In case that the two degradation processes are correlated, modeling methods based on balanced and unbalanced degradation data are forwarded. The proposed methods provide a solution to solve the reliability modeling problem of multi-dimensional degradation failure.(4)Competing failure modeling of degradation failure and random failure is studied. First, we deal with the competing failure modeling as well as its parameter estimation under normal stress level where the degradation process is one-dimensional Wiener process and the random failure is Weibull distributed. We apply the method above in competing failure modeling for xenon lamps. In addition, the case under accelerated stress levels is discussed.(5)Lifetime prediction methods in fusion of population data and degradation data from the product itself are studied. Existing lifetime distribution prediction method uses only degradation data of its own, so the precision is hardly satisfying when its degradation data are few. We solve this problem through data fusion of historical lifetime data or degradation data from other products. We establish a Bayesian framework in which the prior distribution of the parameters are built from historical lifetime data or degradation data from other products. The parameters of the posterior distribution are updated recursively and the Bayesian estimates are worked out. The proposed methods are demonstrated in a case study of lifetime distribution prediction of metalized film pulse capacitors. The proposed methods provide a solution to the lifetime prediction in engineering practice.