1. Home
  2. ⧑ Industrial Control Group
  3. Solving static flow networks in Simulink: Usual approach?

Solving static flow networks in Simulink: Usual approach?

For anyone with experience with Simulink...

Bit of preamble here, question is at the end. I'm assisting a colleague who is following the instructions in an old report he's unearthed, to model a noncompressible flow network (liquids in pipes and valves) using Simulink. The method involves introducing 'artificial' compressibility at the network nodes, so that the pressure balance can be calculated using ODEs, ie.

PressureAtNode = integral (flow in - flow out).

The model is taking forever to run, presumably because the amount of compressible material is kept small so that nonexistent dynamics isn't added to the model, this means that the equations are very stiff, and the solver needs to work with extremely small time increments.

My own approach in this situation is simply to calculate the system pressures and flows statically, that's usually pretty easy, just a nonlinear version of ohms law. The resultant models are efficient, and run very fast.

Is there something in Simulink that makes solving the flow network using background integrators the preferred approach? Or would one normally do something a bit more (IMHO) sensible, and do a static solution?

Reply to
Bruce Varley
Loading thread data ...

I gave up on Simulink a long time ago, because the unwieldiness of your model grows faster with a graphic-based system than with a text-based system, and because it seems that no sooner do you get a model working than they "improve" things and break it.

But if I recall correctly, Simulink does not handle algebraic loops well. My guess is that the method of using artificial compressibility and ODE's is to make the thing work at all under Simulink.

When you're simulating ODE's with Simulink (or anything else, for that matter), the algorithm you choose has a pretty significant effect on both speed and accuracy. I don't know if there's a better way, but any time I doubt the algorithm, or am just having frustrating or oddball problems, I try different algorithms either until I'm happy or I've convinced myself that joy is not to be found.

Reply to
Tim Wescott

Yep, confirms what I suspected. I have a real problem with introducing dynamics into models just so they'll solve, it hazards fidelity. In the report I mentioned, the actual system is incompressible and static, but the model responses are all perfect exponentials, guess where they come from.

Reply to
Bruce Varley

If you're running a simulation just to find a steady state solution, you might try using the TRIM function.

If you're having problems getting an algebraic loop to converge, you can try setting intitial conditons with the IC block.

dave y.

Reply to
dave y.

PolyTech Forum website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.