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: Geoff Hampden-Smith
Date posted: Fri Jun 24 5:58:23 US/Eastern 2005
Subject: Component Confidence levels
The target availability for a 3 component parallel system (valve shutdown system) is 99.95%. The three components have similar failure rates. However, the failures of the components are only revealed on testing.
The question is what confidence limits should be chosen for the individual components to have a 95% chance of shutdown on demand for the parallel system.
Any direction on performing the calculation would be appreciated.
Subject: Rel Conf Limits
Reply Posted by: Joe Dzekevich
Date Posted: Mon Jun 27 13:12:05 US/Eastern 2005
Confidence limits apply to measured data. In general, for any item, Rl = alpha ^ 1/n. So, if you had 50 units on test, Rl (lower CL) at 95% would look something like: Rl = (1-.95)^1/50 = (.05)^1/50 = .9418 for a zero allowed failure test.
But....is the system repairable while it is operating? Then R is higher. My only thought is this: confidence limits depend on measured data and thus sample size (n). Do you really mean: what R do I need for each component so that I have a system R of .95?
Subject: Parallel Systems
Reply Posted by: JLR
Date Posted: Wed Jun 29 13:49:58 US/Eastern 2005
First, there are simulation packages where such parallel system can be simulated, and the reliability required can be obtained via Monte Carlo. One such package is Reliasoft’s BlockSim; and there are several others.
Then, there are system reliability bounds, that are calculated from parallel redundant or series subsystem reliability, obtained from test data. The user could define the subsystem reliability and, from there, obtain the required system bounds with his pre-specified confidence. More information about this approach can be found in the RAC report Confidence Bounds for System Reliability, (Romeu, J. L RAC SOAR-4. 1985).
Finally, there are also Markov (stochastic) models, that allow us to provide the availability conditions that we need, including the number of redundant subsystems, the desired operational levels and system characteristics (e.g. full, degraded, failed states; repair capability during the failure of any subsystem, as long as the others are still working, etc.). Markov models are more complex, but can be derived as long as the failure rates are constant (they require the memory-less Exponential distribution). More information about this third approach can be found in the RAC START sheet, Understanding Availability (Romeu, J. L, http://rac.alionscience.com/pdf/AVAILSTAT.pdf ).
Subject: Component confidence levels
Reply Posted by: V. Narayan
Date Posted: Thu Jul 28 17:14:32 US/Eastern 2005
I am intrigued by your statement that the three elements of the system have similar failure rates. In my experience, the logic or control element is extremely reliable, followed by the detector. In general the reliability of the executive element is poor, and is always the hardest to test. Here I am talking about real tests, not functional tests.
I think you are looking for system availability numbers, not merely the reliability of components. In the end, what matters is whether the system as a whole works on demand. There are many reasons to question numbers that are not fully supported by test results.
Coincidentally, I used to work for Shell UK too, till I retired in 2002.