Mixing Weibulls from Different Data Sources

[Michael Rand]

Hi everyone

I have a product with multiple failure modes (A and B, say).

Let's say I have two subgroups of the products:
Subgroup 1 has adequate data for both A and B
Subgroup 2 has adequate data for A and very little data for B.

Assuming that Subgroup 1 and 2 are operated under different conditions (key), is it ok for me to create failure mode B from Subgroup 1 and use it with failure mode A from Subgroup 2 to come up with Subgroup 2's expected composite failure distribution? Wouldn't I be generating a composite for Subgroup 2 that might not represent an internally consistent set of probabilities of failure?

[Tarik]

If you want the ‘general’ reliability of your product (regardless of operating conditions), I would mix ALL data (subgroup 1 and 2) and come up with a general profile for your product’s life.

However, if you need to distinguish between these 2 subgroups for whatever reason (for example, you have 2 types of customers, or probably the product will either operate under Condition-1 or under Condition-2 (no mixing)) then I would separate the analysis by subgroups. Your product is probably experiencing, naturally, less Mode-B failure under Condition-2 (or do you suspect that you might have a problem with your data collection?), i.e, because of the nature of Condition-2, the failure mechanism that produce Mode-B failures don’t happen as often as in Condition-1, so it wouldn’t be appropriate to bring more Mode-B failures (from subgroup 1 data) into the picture.