The merits and limitations of reliability predictions

A lot has been said and published about the limitations of reliability predictions based on the models first introduced by MIL-STD-217 (now MIL-HDBK-217) and commercialized by many. Some of the information in the literature denounces these methods as inaccurate and unreliable and promotes qualitative methods of ensuring reliability such as HALT/MASS, or quantitative methods such as Physics of failure, accelerated life models, etc. This paper presents the merits and limitations of reliability predictions as contrasted to reliability testing and assurance techniques from a product development standpoint. It also attempts to answer questions such as: are MIL-based reliability prediction methods useful? At what stages of the product development process it is useful? Which elements of the prediction can be practically used, and which should be discounted? How can the accuracy of reliability predictions be improved? Every method offers a certain benefit at a certain cost, is limited by a time element. No single answer exists in accurately predicting and demonstrating reliability. Balancing cost, benefit and time, the essential elements of a new product reliability & quality assurance program, provide a framework for selecting the methods. Specific, theoretical and practical examples are used to demonstrate the concepts and illustrate the methods that have been successfully used with encouraging results. In addition, useful interpretations of reliability predictions are presented, since it appears many popular misconceptions exist in the electronics industry.