Questions to ask a Supplier
Especially if they list MTBF on their data sheets.
My first questions, which I generally keep to myself, is ‘MTBF, yeah, right. Do they know better or not?” This is generally not a good way to start a conversation with a vendor about the reliability information you need to make appropriate decisions.
MTBF is very common on data sheets. I’ve been told that is because customers ask for MTBF more often than any other reliability measure. Maybe we need to stop asking for MTBF values and ask for something meaningful instead.
Until the time when vendor use informative measure for reliability, or even if they do so already, we generally want to know something about the measure, the data source, or evidence to supper the listed value.
Here are a few questions to include in your next discussion with a vendor.
1. What evidence is there to support the reliability claims on the data sheet?
What is the body of evidence and supporting logic that justifies the reported reliability claims? How can we adopt this information for use with our application?
If the evidence is mostly engineering judgment or unsupported, we may have to discount the reported value due to uncertainty. If the body of evidence is overwhelming, we can find justification to use the reported values directly.
The discuss is about where did the reported number come from and are we interpreting it appropriately.
2. What is the expected failure mechanism(s) for this device?
This question focuses on the fundamental understanding of how the device fails. If the vendor understands how various stresses interact with their product and cause a failure to occur, they are more likely to have test supported models fully describing the time to failure distribution.
An organization that does not understand their product an it’s failure mechanisms also doesn’t know the design and assembly elements that impact reliability.
3. What are the assumptions in the model or testing that lead to the reported reliability?
A revealing example is the assumption of constant failure rate when the dominant failure mechanism is wear out. It indicates they do not understand reliability statistics well enough to describe the changing rate of failure over time.
Asking for the assumptions also may reveal the level of understanding of the failure mechanisms, the material or assembly process relationship to the failure mechanisms, and their ability to make improvements. George Box said that all models are wrong, some are useful. Most models rely on assumptions to permit the calculation of time to failure estimates.
4. What model(s) if any are in use to convert test data to a reliability prediction?
Like assumptions, the choice and use of a time to failure model impacts the quality of the estimate. For example, if the failure mechanism is accelerated with temperature and they are using the Arrhenius equation as part of the model, do they understand and have accurate estimates for the activation energy? Furthermore, does the Arrhenius model accurately describe the failure mechanism’s relationship with temperature?
What models are in use and why were they selected? What are the limits of the model and connections when other stresses are also applied (as in real use)?
Of course this is just the start. Based on the responses you may want to explore additional areas, such as post testing failure analysis results or sample size determination. The intent of the questions above and any additional questions is to understand the device, it’s risks, and estimated reliability performance for your application.