Inference of Bivariate Degradation Model Based on Copula

• Main Authors: Hsin-YehYang, 楊欣曄
• Other Authors: Shuen-Lin Jeng
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/34392468213252246383

碩士 === 國立成功大學 === 統計學系 === 103 === In this study, the main goal is estimating failure probability of future products with degradation data and constructing prediction intervals. According to these prediction intervals, decision maker may utilize the bounds of prediction intervals to decide how many spare parts are needed or when to maintain the products. If they can precisely predict the failure probability of products, they may be able to keep an appropriate stock level or maintain products at appropriate time. This may reduce the inventory cost or maintenance budget. We use cumulative failure rate to monitor the performance of components of the products. The possible prediction error is evaluated by an interval. Degradation data can be described by degradation paths or stochastic processes. Here we consider the random effect degradation path model. When coefficients of a path are more than two and correlated, a general procedure is assuming a multivariate normal distribution of the coefficients. However this assumption may be violated in real case. The contribution of our work is that we combine Copula model to construct multivariate distribution function and estimate the failure probability. The prediction intervals are constructed through bootstrap approach. We apply our proposed methodologies to roughness degradation data of highway pavement.