Mean time between failure caused by natural death

JBS

New Member
#1
I'm working on extending mean time between failure caused by one specific defect that's a build up in nature. It's average 10 hrs. But there is data points that's higher that 10hrs but caused by accidental failure like maintenance failure or product change over. I don't take them into acct currently. That means when we run really well, I don't have any data points which is counter intuitive. I try to include all data points that characterize the run time before build up failure but dont know where to draw the line.
It's like if a 90 year old person died of car accident. In my mind, that's still a long time and should be contribute to the life time data. In this analogy, the build up is the aging process and maintenance failure etc. s accidental causes. And I'm trying to characterize the how long a population lives before death by natural cause. Any idea??
 
#2
My memory is a little hazy on this, but i believe something like Cox regression can deal with, which if i understand you correctly is essentially the same as right censorship. Check out survival analysis too. Hopefully that'll give you something to google before a grown up answers.
 

Miner

TS Contributor
#3
Is this a product reliability issue? If it is and I understand your explanation correctly, there are two ways to deal with this. One would be to use only the failures due to the failure mode of interest and treat all other failures as right censored data. This will provide you with reliability statistics for the failure mode of interest only. You cannot interpret the results for any other failure mode. The second approach is to include all failures, but to code them by failure mode. This is similar to a regression analysis using a dummy variable, with the failure mode being the dummy variable. This will provide reliability statistics by failure mode.

BTW, don't use MTBF or failure rate (1/MTBF). It is a meaningless statistic. The MTBF of a male human from the USA is around 850 years. Use either the reliability statistic or an Lq life instead.