SRC Forum - Message Replies
Forum: Reliability & Maintainability Questions and Answers
Topic: Reliability & Maintainability Questions and Answers
Topic Posted by: Reliability & Maintainability Forum
Organization: System Reliability Center
Date Posted: Mon Aug 31 12:47:36 US/Eastern 1998
Posted by: Mike D'Aquila
Date posted: Mon Dec 20 10:23:09 US/Eastern 1999
Subject: The Applicabilility of Reliability Prediciton Techniques
I, first of all, extend an apology to all RAMP professionals who conclude from readnig this question that I am trivializing the scope and depth of the [correct]answer. I do not mean to, however: While spending many years in the COTS and defense contractor environments, on numerous occasions I personally participated ( or took notice of ) in debates about - for instance - using some interaiton of *217* as a prediction technique for, say, a next generation COTS processing architecture vs. a Bellcore-based algorithm or even, perhaps, one that considered Weibull or physics-of-failure, or whatever hybrid was in vogue at the time. I even recall the Redstone Arsenal organizational debate that was founded on a belief that the USA should move away from 217 totally, as the default technique.
Yes, there were many valid arguments against using 217 as a technique in the COTS development arena - one being that the appendices never were able to keep up with the lightship speed of the day's ic-development race to revenue fame and glory....i.e. always "..one step behind.."...or whatever. Generally, too, the knowledged concluded that 217's felxibility with setting Quality Factors allowed for too much subjectiveness so as to cloud viability and consistency. And, in the end, certain RAMP-guys said that actual measurement proved that 217 was a pessimist.
So, with all that drivel comes my question: Are there any current "white paper" or "treatise" type discussions known to RAC where this the applicability of one technique vs another for COTS product development is at the focus ? Even though I have not looked into this for 18-24 months, I am confident that the synchronization between academia and commercial product development is again skewed......even more, perhaps. I read in one of the FAQs teh fact that some believe that 217 can be 2-3 times too pessimistic, or more, as compared to actual demonstration. PRISM, as I understand it, it to close this gap ??????
Thanks for your time........ Mike D'Aquila, 603.434.8231, "firstname.lastname@example.org"
Subject: Reliability Prediction
Reply Posted by: Seymour Morris
Organization: Reliability Analysis Center
Date Posted: Mon Dec 20 11:44:14 US/Eastern 1999
In the Winter Edition of the RAC Journal, which should be in the mail as I write, are two articles on reliability prediction. One is a general overview, which compares and contrasts various reliability prediction methodologies, including PRISM. The second is a detailed article focusing specifically on PRISM. However, neither of these articles differentiates between COTS and military developments. RAC considers the techniques to be applicable to either type of development.
You are correct in asserting that, in general, many using MIL-HDBK-217 do claim it to be pessimistic by a factor of 2-3. However, one company I am aware of that is required to provide a warranty for up to five years based partially on prediction results, claims that MIL-HDBK-217 is about right, or even a bit optimistic, in some cases. They are very conservative with their estimates since substantial money is involved. Thus, there is a need to be careful with unsubstantiated claims.
In general, PRISM models will predict lower failure rates than MIL-HDBK-217. This is in large part due to a "Growth Factor" which models industry trends in defect density. This in effect makes the PRISM models dynamic over time, based on the year of component manufacture, while MIL-HDBK-217 models only change with new Handbook releases. If "the clock is turned back" with the PRISM models, then they yield results similar to the MIL-HDBK-217 results for the timeframe when the underlying MIL-HDBK-217 data was collected.
Also, in the 1998 Annual Reliability and Maintainability Symposium Proceedings is a paper titled "From reliability-prediction to a reliability-budgetedddd" which addresses the reliability prediction issue. This paper discusses which advantages a prediction can offer to a manufacturer and a customer and concludes that instead of prediction one should talk about a reliability budget, which will guide the project manager to the part of the design where early analysis and tests should be made. It may be of interest to you.