Reliability Testing - Weibull Analysis Help

marcusja2002

Involved In Discussions
Its been a long time since I've done any Reliability testing.

i know the formulas, I've just forgotten what each piece means.

I know I want with 95% confidence that my product will survive at least 132.5 hours before failure. How many hours do I need to run it in order to verify it will meet that?

I currently have 1 being tested, but I can throw more units at it to lower the number of hours needed to verify that.

Anyone with more statistical knowledge then I please help point me in the right direction.

Thanks in advance.
 

Miner

Forum Moderator
Leader
Admin
You will need to provide some additional information:
  • What distribution are you assuming? Weibull? What shape parameter? If you don't know the shape parameter, what is the failure mode? Is it wear out, random?
  • What reliability do you want to demonstrate at 132.5 hours? This is separate from the 95% confidence.
 

marcusja2002

Involved In Discussions
What do you mean by shape of the parameter?

There is a material that will change shape depending on temperature. when above a specific temperature is completes a circuit. I need it to be able to at least make it 132.5 hours of sensing that temperature and completing the circuit. It can fail randomly.

I guess I would like it to have zero failures up to 132.5 hours.
 

Miner

Forum Moderator
Leader
Admin
If you assume that your failures follow a Weibull distribution, the shape parameter (beta), defines the type of failures seen. B < 1 indicates infant mortality failures, B = 1 indicates random failures due to over stress, and B > 1 indicate failures due to wear out.

When the material fails, why does it fail? Does the material age such that it no longer switches at the correct temperature? Does it fatigue and crack?

You need to define your requirement similar to: 95% reliability at 132.5 hours while operating at ## degrees C and cycling # times per hour with 95% confidence.
 

Daniel Cruz

Registered
You need to assume something about how the failure times are distributed to answer your question. If we prove, for example, that only 1% of parts fail by 50 hours, how do we know what that means at 132.5 hours? We need to project reliability at one time to reliability at another, so we need to assume a life distribution, as Miner correctly stated above.

The basic concept here is that some failure modes get worse with age (to varying degrees), some actually get better with age (such as manufacturing defects), and by assuming a Weibull distribution and a slope of say 2, we're basically saying "I know it will get worse at this rate." That allows you to make the connection between, let's say, 99% reliability at 50 hours and 95% reliability at 132.5 hours (which would correspond to a Weibull slope of 1.7).

Here's a more thorough explanation of the different mathematical methodologies of reliability test design, with explanations as to why you need to assume the life distribution in this case.

Full disclosure: I work for the company that produces this wiki (ReliaWiki) and the software referenced in it (Weibull++).
 
Last edited:
Top Bottom