How do we get reliability information from vendors?
Reading data sheets and hoping for the best is often all we have time to accomplish. For high risk (new technology, new supplier, new… something) many organizations will examine the reliability claims to some extent. Some may even conduct reliability testing in-house to validate vendor claims.
That still leaves a lot of components with some risk for component level failure that you should address. Datasheets and websites often do not provide enough information to make a reasonable assessment.
You know we need reliable parts from reliable vendors, yet how do you determine if the parts and vendor are reliable without expensive component level testing?
First Determine what you need
Naturally, you do not need the MTBF value. Enough said.
What you need to do before contacting a vendor is determine what reliability you need for your application. This includes the
- environment (local stresses and use information),
- probability of success
Use an apportionment model based on the system reliability goals.
The most common mistake I see is setting only a duration or a failure rate (a variation of probability) target. You and the vendor needs both bits of information.
You can and in most cases should have at least three couplets of probability and duration. Early life failures (99% survive the first month). Warranty period which is a business decision could be 97% survive two years, and the long-term or expected useful life, 90% survive 5 years.
Be clear on what you want.
Questions to ask your vendor
If you could ask a vendor for 95% reliability over 3 years in XYZ environment and use conditions performing XYZ function and receive a complete report of expected failure mechanisms along with time to failure distribution estimates, life would be great.
Unfortunately, it doesn’t happy very often.
So, you need to ask a few questions to help them provide the information you seek.
- Given our use what would be expected to fail?
You want to know if they know what will fail. Do they know how their product will perform in your application? If so, that is a great first step to understanding how long before the failure is likely to occur.
- For that failure mechanisms, when would we expect the first 1%, 10%, etc. to failure?
In other words, we want the life distribution for the expected failure distributions. They may have that data, maybe even a Weibull plot, help them realize that is exactly what you want. Be specific about failure mechanisms as that often determines how the component is tested and modeled.
- What evidence do you have to support the reliability values?
Notice we’re not asking for a parts count prediction, for fit rate, MTBF (naturally), rather we want to know if they know how it will fail, when it will fail and what evidence they have to support the claims.
We want to see the testing details and results. Get the raw data if possible and do your own analysis. If they do not have anything more than anecdotal evidence, it’s not a good sign. If they have field data, test data, and modeling that shows an understanding of the failure mechanism and how it related to the life distribution, then you have a good answer that you can use.
Use open questions, rather than restrictive questions. If you ask for MTBF, they that is probably all you will get. If you ask how and when will a component fail in my application, you may receive useful information.
When evaluating components for use in your design, consider the reliability too. It may take a little work to find meaningful information. Asking a few questions, such as above, may help your vendors provide the information you need to make informed decisions.