Reliability from Warranty data - We do not know the date put in service

[jacko]

I recently had a concern brought to my attention on some analysis that I want to implement.
We receive a warranty report, and I wanted to use Weibull analysis to compare failure date to date manufactured. The problem that was posed to me is the fact that product goes to a warehouse, and we do not know the date put in service. How can we relate to a MTTF analysis? Will this give us false data? Can we actually know what the MTTF is if we do not know how long the unit is in operation?

Are there any thoughts on how I can utilize this data?

[Miner]

I cannot help you on handling this through Weibull analysis, but you may want to consider nonparametric methods described below.

Check out:
ASQ Reliability Review
Could Firestone Have Known?
and www.fieldreliability.com

If you have sufficient data available on infant mortality failures, you can create a distribution for the time between date of manufacture and date put into service. If needed, you can then back this out in the analysis. I have used the techniques described by Larry George in the referenced links and have been pleased with the results. Known issues with specific manufacturing periods stand out clearly in the analysis.

[Jim Wynne]

Originally Posted by jacko

I recently had a concern brought to my attention on some analysis that I want to implement.
We receive a warranty report, and I wanted to use Weibull analysis to compare failure date to date manufactured. The problem that was posed to me is the fact that product goes to a warehouse, and we do not know the date put in service. How can we relate to a MTTF analysis? Will this give us false data? Can we actually know what the MTTF is if we do not know how long the unit is in operation?

Are there any thoughts on how I can utilize this data?

Miner's good advice notwithstanding, you seem to be asking two different questions. Are you concerned with comparing date of failure to date of manufacture, as in your second sentence, or failure date with date-in-service, as in your third? If this relates to warranty claims, what you might need to be comparing is date of failure vs. the date when the warranty clock starts ticking. What are you trying to analyze?

[jacko]

sorry for being somewhat vague.
The concern is the fact that if I collect warranty data, we do not the actual time that the units were in operation, since we only have production date information and failure date information.
i.e.,
if we produce a unit that sits in a warehouse for 2 or 3 months, then is put into operation and fails after 6 mos. we would incorrectly assume that it failed after 8 or 9 months instead of 6 mos.
My question is whether this can be accounted for, or is this even a valid concern?

[Michael Walmsley]

Yes, you should account for it AND it is a valid concern.