Reliability Block Diagram (customer question)

Question

A customer who is in the process of evaluating JMP Pro asked the following question:

"Is it possible for RBD to back calculate a components required life distribution to meet an overall system level reliability goal?"

I do not think it is possible for RBD to do this, but it's a great idea and would likely be a useful feature to add.  But does anyone have a suggestion on a work-around that would allow the customer to get a similar result?  The only thing that comes to mind is to manually tweak the component's life distribution to get the desired system level reliability outcome, but was wondering if there is something more clever than that which could be done (ex. JSL with a looping variable, but not sure how to integrate a RBD into a JSL looping script).

Hope everyone continues staying safe,

Mark Zwald

peng_liu · Answer

Consider a system with reliability R_s = f(R_1, R_2, ..., R_k), where R_i's are component reliabilities. The system has infinite solutions given only R_s is known. One must add more conditions, so the system has a unique solution.

And the additional conditions determine what might be the next step, or whether it is even possible to have a solution.

The tweaking(calibration) approach implicitly assumes all components are fixed except one, so you can tweak that particular one. But it could be difficult, too, if the life distribution of that component has more than one parameter. Then it means infinite solutions again.

DonMcCormack · Answer

AFAIK, this isn't available in RBD. On top of that, RBD components aren't exposed via JSL. You might be able to read in the RBD file as text and substitute a range of values for that component. Since component distributions have no more than three parameters, you could use a DOE to cover the parameter space. Getting the reliability info out would require a bit of JSL, too. All in all, I think this would be quite a bit of work and it would be simpler to do it interactively.

martindemel · Answer

No real clue, but just brainstorming to the sky ;)

Each component's life distribution profile could be thought as of a function. This function could be used to extract FPCs. These FPCs could be used as factors for a model to predict overall systems reliability. Assuming this all is possible you would get a model and a profiler. Gen Reg can be to match the distribution. Setting all FPCs for the other than the one of interest component to fixed, then set the response specs to match a Target, and optimize for the remaining components FPCS. This will lead back to the shape of  the components life distribution (in best case).

As said, brainstorming to the sky, much work and not sure you will find a solution. Also not sure if this would be mathematically feasible without trying to look at the equations and assumptions for each step.

However, a very interesting idea!