Challenges of Reliability Assessment Based on Degradation Data: An Example

Traditionally, the reliability of components is estimated by analyzing failure times of representative units. But failures are difficult to obtain for ultra-reliable components without submitting the test units to extreme test conditions. This, in turn, often leads to failure mechanisms which are different from the ones encountered under normal use conditions. A new technique is being developed to estimate reliability based on monitoring reliability-related characteristics over time and in extracting from these `degradation data` information on reliability without waiting for failure to occur. This approach offers many challenges to statisticians, physicists, and engineers, such as: modeling both the physical degradation phenomena and their sources of variability, understanding the measurement process and its variability to minimize the effect of measurement errors, designing good experiments, modeling the relationships among the monitored degradation phenomena and their impact on failure, and extrapolating results to use conditions. This paper illustrates some of these challenges with an example, describes the techniques used to meet these challenges, and concludes with the lessons learned.