1
$\begingroup$

Like you will be fired/sued/shot if it happens.

I use 1 / 100 million / year (global catastrophes happen on Earth every 100 million years so it is likely no one will care about me)

But maybe there is standard/more scientific/less harsh approach?

$\endgroup$
3
  • 1
    $\begingroup$ They use 1 in 50 years or 1 in 100 years for flooding events and then get 3 in 10 years... $\endgroup$
    – Solar Mike
    Apr 16 at 10:30
  • 1
    $\begingroup$ It depends very much on the application, the cost of failure, and the expectations of the market. If you say something like "1 in 100 million pieces/ year", how can you even tell it apart from "1 in 100 thousand"? It takes a lot of effort to do such a thing. And for technological products there are so many unknowns, that the assumption of uniform or normal or even just consistent distribution of defects unrealistic at system level, making statistical tools hard to apply. $\endgroup$
    – Pete W
    Apr 16 at 16:17
  • 1
    $\begingroup$ Civil engineering standards like EN 1990 deal with this. The standards are not for free, but you can find some free reports and guides from Joint Research Centre like this one. $\endgroup$ Apr 17 at 17:46

1 Answer 1

1
$\begingroup$

First, there is not such thing as "will not happen" for processes/systems that do not violate natural laws. Statistics is based on a subset of past data and can be a good tool for estimating future occurrences but in no way predicts the future. Also important to realize that most processes/systems do not have the exact same input variables in the test/past scenario, as they do in the field/current scenario.

  1. Reliability Engineering is a whole field in itself (not my profession, so do additional research)
  2. Part failure rates as low as 1/E8 would be difficult to measure let alone attain. If a company were to thoroughly asses the quality of even 100million low cost/complexity items like wood screws they would likely go out of business before they got started. And how would they ensure that statistical rate continued without continuing the expensive inspection process.
  3. Most high reliability systems (like boiler management systems) accept parts with higher failure rates, but have designed layers of redundancy and alarms built into the system to gracefully handle what would otherwise be a critical failure. This is governments enforce building codes like NFPA to ensure good design practices even though they ultimately do not control the quality level of the individual products.
  4. There are ways to design out modes of failure. For example there is a concept in cyclical metal fatigue design called "infinite life" where when the stress is kept below the threshold no cracks will form or propagate; avoiding a fatigue failure.
  5. Practical Engineering has a good video on how the value of a statistical human life is used in the engineering design process.
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.