Reliability Analysis Censored Type I data

[vnigam]

I have a data set where the test was concluded after a fixed period of time. I have 5% of the parts that failed 95% still running. In what catogory can I assign this data. I have individual failure time for all the parts that failed but for the ones still operating, I have the time at which we concluded the test. What I am trying to say is that I do not have an upper bound on the time, if I decide to group the data, for the suspened or still running parts.

Can you please advise?

Vandana

[David]

Hi Vandana,

I will assume that you are wanting to conduct a life data data. In this case, the data would be classified as times-to-failure with right censored (suspensions) data. Within the Data Type Expert of Weibull++ 6, simply select the Times to Failure and with Right Censored Data (Suspensions) options. Click OK and the data sheet should have two columns (not including any Subset ID columns): State F or S and Time to F or S. For the units that were still working at the end of the test, they would all be suspended at the test time.

I hope this helps.

David

[vnigam]

Thanks David.

If I assign the test end time as the time on the parts still operating, wouldn't my estimate over estimate the reliability?

Please advise.

Vandana

[David]

How would that over-estimate the reliability? That is the information you have at this time. You have some failures, but you also know that X units were still working after some time T. If you had continued the test they would have failed at some point, but you do not know when that would have been. So the units are right-censored at the test time. Now, if some units were removed from the test before the end time then these units would be suspended at the time they were removed. If the suspended units are not indicative of the population for some reason then that is another issue.

[vnigam]

Thanks David.

Please help me understand this:

Suppose we continued the test for an additional time period and the parts still operating at the previous time period still continued to run. Now I decide to terminate the test and consider the data as a right censored data set and do my analysis. What would we expect the difference between the two analyses?

The failure times are the same the suspended time for the operating parts are one additional time period than the previous data set.

What your your thoughts?

I really appreciate your assistance.

Vandana

[David]

If you are using rank regression then the parameters would not change even with additional time on the suspensions. This occurs because the suspensions are occurring at the end and this does not affect the rank of failures. Information on the ranking can be found at http://www.weibull.com/LifeDataWeb/analysis_parameter_methods.htm (http://www.weibull.com/LifeDataWeb/analysis_parameter_methods.htm) under the Shortfalls of this Rank Adjustment Method section. Because of this I would tend to use MLE. If you use maximum likelihood estimation (MLE), then the estimated parameters will change. How much they change will depend on the sample size and the suspension time. You can easily check this yourself by simply adding time to the suspended items. Assuming no additional failures, increasing the suspension time will tend to flatten out the line on the probability plot when using MLE. In other words, the shape parameter will tend to decrease. But now you are dealing with what-if scenarios. Yes, it is possible that you may come to a different conclusion if you had different/more information depending on what you are interested. Keep in mind that failures give you the most amount of information. There is only so much information which can be obtained depending on the data given (you work with the information you are given). Additional information on whether to use rank regression or MLE can be found at http://weibull.com/hotwire/issue1/hottopics1.htm (http://weibull.com/hotwire/issue1/hottopics1.htm).

I hope this helps.

[vnigam]

Thank you very much.

Regards

Vandana:)