An owner-operator engineer compared the failure rates for two similar products from an exida certificate with those from a TÜV Italia certificate.  exida numbers were an order of magnitude higher.  Who is right?  

There are some relevant fundamental facts in the field of Reliability Engineering:

1. Failure rate prediction methods depend on the assumptions made including the definition of “failure.” Different assumptions will result in order(s) of magnitude difference in the resulting predictions.
2. Any competent Reliability Engineer knows that failure rates come from field failure data – the higher the quality and quantity, the more accurate the failure rate result.
3. The highest quality process industry field failure data in the world is produced by the OREDA consortium based on Norway and analyzed by a company named Sintef.

So, considering these facts, whose failure rate data, exida or TÜV Italia, is right?

Here is what we know at exida:

1. In this recent case we did not find any assumptions or methods described in the TÜV Italia report we were given.  So we cannot know for sure why this failure rate data is an order of magnitude lower. But it is highly likely that different assumptions/definitions were used.  I recall one case where a TÜV engineer stated that most failures of mechanical devices were “systematic” therefore did not count as a random failure per IEC 61508 definitions.  In this case if an assumption was made to include “only random failures per their definition” then of course the failure rate would be much lower but very unrealistic. Another report from TÜV indicated that only failures due to manufacturing defects counted as random - far from realistic.
2. exida failure rate predictions include all failures not just random manufacturing defects. This policy follows the OREDA/Sintef philosophy and is quite realistic.
3. exida failure rate data has been compared with OREDA, DOW, and other field failure data and found to be amazingly close and well within uncertainly bounds.  The TÜV Italia data was far below these bounds.
4. exida has initiated nearly a dozen meetings with the engineers at competitors (including TÜV Italia) to reconcile methods and assumptions.  We were not successful with TÜV Italia as their analyst refused to explain the methods used or the assumptions.
5. Perhaps most importantly we know that a higher failure rate will result in a safer design. Low failure rates may result in dangerous designs that do meet risk reduction goals.

Which failure data set would you pick if you and your family lived near your plant?