ALBERT

Notes: Reliability of products has long been considered as an important quality characteristic.Traditional methods of product reliability assessment are based on lifetime data. With products beingmuch more reliable and the growing need for developing new products within shorter period and atlower cost, we can hardly get enough lifetime data in many cases. Performance degradation data canalso be used for reliability assessment. Recently, they are found to be useful in some cases where theyare easier to collect. But when performance degradation data are also limited and some lifetime dataare available, it is preferred to utilize both information sources. This paper deals with the problem ofreliability assessment combining both performance degradation data and lifetime data. It is assumedthat two samples from the same product are tested differently. Degradation data are collected fromone sample and lifetime data from the other. First, the performance degradation model is established,using either statistical methods or methods from physics of failure. Then lifetime of the product,which is defined as the first time when the performance crosses the known failure threshold, iscalculated. The MLE method is used for parameter estimation where the maximum likelihoodfunction is multiplication of the one from degradation data and the one from lifetime data. Toillustrate the proposed method, an example of the metallized film capacitor, which is used in inertialconfinement fusion (ICF) facility, is given. We model the performance degradation data of metallizedfilm capacitor with Wiener process with drift. The failure of the capacitor is defined as the first timewhen its capacitance drops below a threshold. The lifetime distribution is deduced and the parametersare estimated from the joint maximum likelihood function. A comparison is conducted between theassessment results of degradation data only and those of combination of degradation data and lifetimedata. In conclusion we propose that both degradation information and lifetime information should beused when neither of them is sufficient enough for reliability assessment. Some directions for futurework are also discussed