SRC Forum - Message Replies
Forum: Reliability & Maintainability Questions and Answers
Topic: Reliability & Maintainability Questions and Answers
Topic Posted by: Reliability & Maintainability Forum
Organization: System Reliability Center
Date Posted: Mon Aug 31 12:47:36 US/Eastern 1998
Posted by: chandu
Date posted: Tue Mar 4 6:33:32 US/Eastern 2003
Subject: PREDICITON OF PARAMETER
I have some laboratory data (accelerated test)for Grade 1 item and Grade 2 item, i have estimated parameters of weibull distribution. For the Grade 1 item i have field data and i have estimated the parameters of weibull distribution.
Here i have a problem in correlating the field data with laboratory data for grade 1 item. How can i correlate this Grade 1 data?
By using this same same logic How can one predict the field data parameters for Grade 2 ?
Please help me in this case....
Subject: Weibull accelerated and field reliability
Reply Posted by: Larry George
Organization: Problem Solving Tools
Date Posted: Sun Mar 9 16:28:59 US/Eastern 2003
The Weibull shape parameters estimated from accelerated and field reliability estimates should approximate each other, UNLESS: 1. acceleration changes failure modes, 2. Weibull is a lousy fit, or 3. other interesting reasons. The Weibull shape parameter in P[Life > t] = exp[-(t/a)^b] is b. The reason is that the failure mode(s) should be the same, and b contains information about infant mortality or wearout, which are presumably related to failure modes. I have actually seen nearly equal Weibull shape parameters for accelerated and field data!
First, use a likelihood ratio test to test whether the grade 1 Weibull shape parameters differ. If shape parameters accelerated and field don't differ statistically significantly, then estimate the grade 2 Weibull parameters from the accelerated data. Assume the grade 2 Weibull parameter is the same whether accelerated or not. Assume the grade 2 scale parameters have the same ratio as the grade 1 scale parameters. Estimate the grade 2 scale parameter as
a(2; field) = a(2; acc)*[a(1; field)/a(1; acc)].
Send your data, and I will check whether Weibull fits tolerably and, if so, do the analysis.