Reliability of complex mechatronics system

[joshi]

Hello!

Reliability prediction based on MIL STD 217 if undertaken will give Reliability prediction of system and sub-systems as ' xi' hrs as MTTF. Development testing commences : the initial failures are taken care and configuration frozen for that design; no failures during ESS, acceptance testing and also during type approval/qualification testing and say ' y ' hrs are logged. If y > x; then the designers reliability estimatee can be considered to be accurate based on the design, components and process.

- The question is if the value of ' y' is based on testing of one or two systems then how sure are we of calculated MTTF ?

- For a complex mechatronics system would use of exponenetial distribution for ' R' estimation correct?

- The system consists of say 25 major critical sub-systems : mech, electrical, embedded systems, pneumatics etc : With limited test data how accurate would be reliability estimation and that too at what CL ??

[Arai.M]

1. Given that you are using MIL STD 217 you are assuming an exponential distribution. In that case, the relation between time logged, demonstation time, failures allowed (or observed), and confidence level is given by the Chi-Squared test design. So given failures = 0, Tdemo = X, Ta (accumulated test time) = Y you can solve for the confidence level at which you can make the statement that you have proved a MTTF of X. See http://www.weibull.com/LifeDataWeb/test_design.htm#chi_squared (http://www.weibull.com/LifeDataWeb/test_design.htm#chi_squared) for calculations example. Again, since you are assuming an exponential distribution, you can run Y hrs with 1 unit, Y/2 hrs for 2 units, etc.

2. That depends and generalizing would not be wise. Ideally you validate that assumption through data and common sense. When a system is made out of a very large number of components often that assumption is made just because gathering data at the component level becomes too complex, too expensive and/or basically not feasible.

3. Bounds around estimates at a given CL can be obtained based on the variability associated with each of the components and how these components are reliability wise connected in the system. If the data you start with is not sufficient and/or inaccurate, this will compound at the system level and translate into poor results. The methodology for obtaining bounds when analysing a system at the component level (or a component at the failure mode level) can be found in the following link http://www.weibull.com/LifeDataWeb/competing_failure_modes.htm#cb (http://www.weibull.com/LifeDataWeb/competing_failure_modes.htm#cb)