Recycle+subflowsheet solving time

[Usseglio]

Dear All,

Is there a given set up in case of a model configuration having a recycle with multiple (5) sub-flowsheets ?
The matter is that I am facing quite low solving time, each time when I run the simulation model either if I change mass composition at inlet streams or a temperature in a heat exchanger.

Thanks and a prosperous 2015 for all members !!

Marcelo

[jasper]

Best wishes for 2015.

Solver speed improvements are in the making, but will only partially be applied to embedded flowsheets (due to the thermo in there coming from an external source it is hard to solve embedded flowsheets in other threads than the main thread).

If the embedded flowsheets themselves are coupled by a recycle in the outer flowsheet an extra speed penalty comes from the fact the embedded flowsheets are fully converged for each iteration in the outer flowsheets, that will surely slow things down. Is there a recycle in the outer flowsheet as well as in the inner flowsheets?

[Usseglio]

The recycle is in the outer flowsheet. The subflowsheets are Unit Ops (simple operations inside without recycle) that are located inside the outer recycle.
I tried to manually set up the sort method aiming to have positive impact in solving speed, but I did not get an apparent improvement.

[jasper]

Sort method? You mean stream and unit operation sorting? That does not affect solving at all (well - actually may, in the existing solver, affect cut stream arbitration; perhaps if COFE finds two streams that are exactly equally suitable as cut stream it will pick the first one in the sort order).

If there are no recycles in the embedded flowsheets, the speed of solution should not be affected by the embedding per se. What aspect of the solution appears slow? Does the evaluation of the embedded units take a long time? Does it take a long time to converge? Does the perturbation for the Jacobian (the start of each Newton iteration) take a long time?

[Usseglio]

Yes, sorry, I meant the unit operation sort method. I thought it might have an impact in the resolution along the recycle.

The evaluation of the embedded flowsheets does not take a long time, it is quite quick.
What it takes too long, is each iteration to be solved and to get convergence. Generally, when I change temperature and/or mass composition, the residual of first iteration is high, and it takes too long to begin a second iteration.
Normally, with each iteration, residual is 10 times lower, but it takes too long to converge.

[jasper]

I am working on a new solver that might help with this issue; would you be interested to test an alpha version of this in a few days? If so, please drop me a private mail.