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: Andrew Rowland
Date posted: Wed Dec 4 12:26:23 US/Eastern 2002
Subject: Life Data Analysis
I am trying to perform life data analysis which I plan to roll-up to an RBD. I am doing this to provide examples in an attempt to convince management that these analyses are important for an organization concerned with reliability. Obviously, it will be more poignant if I use actual plant systems, but I want to ensure that my results are the best they can be. In light of this, I have a couple of questions. I thank you in advance for your input.
The system I have chosen to model is comprised of four identical “channels” of electronics. System level functions are satisfied by a 2-out-of-4 arrangement of the channels. The system has been in service for 25 years (late 70’s). Use of a computerized failure reporting system began in May 1991. Thus, the date of failures and time taken to repair each failure after May 1991 are easy to obtain. However, there is no failure data for the system prior to May 1991.
The only thing I know for certain is that each channel operated failure free from May 1991 until the first failure was reported. Intuitively, the first failure found in the computerized system for each channel would be a right censored data point. The time of suspension would be the date of failure minus May 1991. Is this the correct way to treat this situation when performing life data analysis? Would the model developed only be applicable to a channel during the 15-25 year portion of it’s life? That is, I couldn’t expect my model to describe the behavior of a channel that was “brand new”.
Subject: Maybe a RGT model?
Reply Posted by: Joe Dzekevich
Organization: Raytheon RAL
Date Posted: Fri Dec 6 14:50:49 US/Eastern 2002
Try analyzing all the data you have as a reliability growth model. I assume these things are repairable systems. The idea would be to see if you are seeing an increasing, decreasing or steady-state failure rate intensity. If it is steady-state, then just use the data you have over the time period as an MTBF and plug it into your model. If the system is aging, then use a Duane or AAMSA (Larry Crow) model and project a failure rate at any point in time of interest. If your items are not repairable, then try analyzing the data as a distribution, then time shift it to the right. Just thinking out loud. Joe