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: Per Jonsson
Date posted: Fri Apr 22 5:21:28 US/Eastern 2005
Subject: prediction - books, software or what..
What is the best way of predicting failure rate of a component? I know there are some different kinds of software but are those sufficient or do I need more tools?
When it comes to software there are also an economical matter to consider before an investment. How much do one have to pay for a good tool by the way?
Thanks for any help on the matter!
This forum is a great thing.
Subject: Credible reliability prediction
Reply Posted by: Larry George
Organization: Problem Solving Tools
Date Posted: Sun Apr 24 13:26:33 US/Eastern 2005
The word "best" in conjunction with failure rate prediction creates a possible oxymoron. At RAMS 2000, the speaker before me, Lou Gullo, asked how many made MTBF predictions, ~65 out of 100 raised their hands. He then asked how many believed them? Not a single hand was raised.
Please forgive this biased survey of alternatives:
1. Benchmark MTBF (1/failure rate) prediction a la MIL-HDBK-217 or Telcordia (ne Bellcore) may be useful for relative comparisons. In this case, best may be the quickest and cheapest, http://www.fieldreliability.com/MH217F1.htm, so that you can make the failure rate prediction and move on to productive, useful work. "Benchmarking is doing as badly as or worse than best in class," anonymous curmudgeon.
2. Predictions are necessary for totally new products only. For old products, use observed failure rates. Generation of field failure rate observations have convinced me that, although designers and producers say products and processes have changed, failure rates don't change, much. That is because other important factors don't change, factors like environments and customers. This led me to recommend scaling old part failure rates to predict new-product failure rates. Biostatisticians use this method extensively. Clinical trials depend on it. See http://rac.alionscience.com/pdf/4Q2003.pdf.
3. Failure rates depend on age, so a single-number failure rate prediction is suspect and not particularly useful. Why not predict age-specific failure rate functions, using scaled, observed field failure rates from parts already in use? I call this "credible reliability prediction," because it seems more credible than alternatives 1 and 2 and lends itself to updating as field data accumulates on the new product. The formal procedure for this is known in the insurance business as "credibility."