Reliability predictions - continued reliance on a misleading approach

Abstract

Reliability prediction methodologies, especially those centered on Military Handbook (MIL-HDBK) 217 and its progeny are highly controversial in their application. The use of reliability predictions in the design and operation of military applications have been in existence since the 1950's. Various textbooks, articles, and workshops have provided insight on the pros and cons of these prediction methodologies. Recent research shows that these methods have produced highly inaccurate results when compared to actual test data for a number of military programs. These inaccuracies promote poor programmatic and design decisions, and often lead to reliability problems later in development. Major reasons for handbook prediction inaccuracies include but are not limited to: 1) The handbook database cannot keep pace with the rapid advances in the electronic industry. 2) Only a small portion of the overall system failure rate is addressed 3) Prediction methodologies rely soley on simple heuristics rather than considering sound engineering design principles. Rather than rely on inaccurate handbook methodologies, a reliability assessment methodology is recommended. The reliability assessment methodology includes utilizing reliability data from comparable systems, historical test data, and leveraging subject-matter-expert input. System developers then apply fault-tree analysis (or similar analyses) to identify weaknesses in the system design. The elements of the fault tree are assessed against well-defined criteria to determine where additional testing and design for reliability efforts are needed. This assessment methodology becomes a tool for reliability engineers, and ultimately program managers, to manage the risk of their reliability program early in the design phase when information is limted.

Reliability Engineering: Value, Waste, and Costs

“…These activities can never add value, and are examples of pure waste. The most prominent activity in this category is reliability prediction based on any 'parts count' or 'part stress' method such as Mil-Hdbk-217 (and all derivatives thereof) [Jais et al, 2013]. It is simply not possible to determine product reliability by merely adding part failure rates obtained from any published document.…”

Design for Reliability of Power Electronics for Grid-Connected Photovoltaic Systems

“…Previous research in lifetime prediction of power electronics (converters) has been mainly focused on statistic analysis. During this period, a handbook -MIL-HDBK-217 F [38] has been widely adopted to predict the lifetime of electronics equipment [32], [37], [39]- [43]. Basically, the lifetime predication can be achieved in a statistic way, where the reliability models (typically, constant failure rates) of subsystems are defined.…”

“…However, the models with constant failure rates in this handbook are out of date, which have not been updated since 1995, leading to the revocation of this handbook. Another reason for the termination is that the predicted lifetime or reliability has no direct or high guiding value for planning and design of the entire PV systems [39]. That is, it is difficult to use the predicted reliability data to design new products or systems with higher reliability.…”

Reliability and Mission Profiles of Photovoltaic Systems: A FIDES Approach

“…Further, MH217 does not explicitly consider the impact of temperature cycling on reliability [16]. In spite of its shortcomings, MH217 has been widely used to predict the reliability of power converters, albeit with the aforementioned limitations [17][18][19].…”