Abstract

The increasing demand for highly reliable products has been posing big challenges on reliability assessment. A major challenge for life tests of these products is how to quickly and efficiently extract failure information to assess the remaining useful life of the devices. A degradation model which measures the physical degradation path as a function of time can provide a direct connection between the product failure time and the inherent degradation mechanism, and hence improves accuracy and credibility of the predicted reliability. Most existing work in the literature focuses on modeling and analysis of degradation data with a single characteristic. In some degradation tests, multiple characteristics of a degradation process are measured to understand different aspects of the reliability performance. The literature on modeling degradation data with multiple characteristics is scarce. We propose a methodology capable of helping to fill this gap in the literature of degradation models for data with covariates. The proposed methodology is a general path dynamic model for degradation data with covariates that allows the degradation rate to be writen as a function of two components. The first component represents the particularities of each unit and has a regression structure that accommodates the covariates. The other component represents the random effects of the common environment and evolve over time. In addition, the inspection times are included in the model from generic functions, allowing for different practical representations to be accommodated. The relation of the model parameters and the failure time is found and methods for estimating the remaining useful life for units under test and a future one are discussed. We applied the methodology to the scar width and train wheel degradation data. Results show that the proposed methodology is competitive in predicting failure times and estimating the remaining useful life. Joint work with Rosangela Loschi (UFMG) and Thiago Santos (UFMG)