Improved product reliability quantification methodology making use of physics of failure based prognostics

Abstract

Quantifying accurate reliability at (sub-)system level is not an easy task. Despite the availability of different tools allowing reliability estimation, e.g. reliability handbooks as MIL217-F, the accuracy of the obtained results is not guaranteed. For instance, the data used in these handbooks are outdated, referring to old technologies and assuming stresses that are not always realistic. Other methods exist which should allow a more accurate reliability estimation e.g. the physics of failure prognostics. However, for an industrial end user, following such an approach at (sub) system level is too expensive. Typical steps to obtain reliability data of one component following physics of failure prognostic approach would require (i) understanding a given failure mechanism and developing its corresponding physics of failure model, (ii) identifying stress accelerators of this failure mechanism, and (iii) planning and implementing an accelerated life test to collect failure data in order to validate the model. A typical accelerated life test would require failures of components collected during the test time (in the order of months) at different stress levels. Another approach to get more accurate reliability at (sub-)system level is collecting and analyzing field data. However, this would require a complete process within an organization, by tracking the products in the field and collecting failure information for many years.In order to overcome these limitations for companies, we propose a methodology allowing to obtain quick and accurate estimations of the (sub-) system reliability by combiningcomponent’s reliability information from different sources, e.g. using physics-of-failure models for some critical components where test data are historically available, and / or using reliability prediction handbook for proven in use components, and / or using field data if available.