Information Fusion and Resource Allocation for Accelerated Life Testing-Based Reliability Assessment

Abstract

Accelerated life testing (ALT) has been widely used to expedite the analysis of a product's failure time when used under normal conditions and calculate its reliability. For engineering systems with multiple components, ALT could be performed at different levels, namely component level and/or system level. Present methodologies in ALT allows for the assessment of system reliability by testing at the system level or single components without considering dependence. This research aims at fusing testing information collected from different levels in order to draw system reliability conclusions. It tackles the dependence between the component failure times of a system, which is caused by unobservable factors. Two novel frameworks are proposed to analyze the reliability of systems with multiple components using ALT. We model the dependency using a Copula function in conjunction with Weibull distributions in the first framework and using shared frailty models with extended hazard model in the second. We present the concept of information fusion which is a method to fuse both component-level ALT data with system-level ALT data to calculate system reliability. An optimization model that is constrained by the cost of testing and bounded ALT design parameters, namely stress levels and number of tests at each stress levels, is developed to find the optimal and cost-effective ALT plans. This research relies heavily on different statistical concepts and Bayesian inference approaches. A four-arm robots and electrical circuit broad of autonomous vehicles examples are used to demonstrate the proposed approaches.