Sequential modular vs. Simultaneous (equation oriented)

[rpseng]

Dear all,

What will be the solution tecnology to be used by OpSim?

Currently there are basicaly two main approaches:

1. Sequential modular (SM): each equipment class is not just a model
but it has a built-in solution procedure. Then the simulator has to
find an order to solve the equipments and pass the output solution of
the first to the second and so on. If there are recycle streams you
need to break the loop and iterate until convergence is achieved.
Examples: HySys, Aspen Plus and Sim42 (I'm not sure if Sim42 is SM or a
hybrid methodology).

2. Simultaneous or equation oriented (EO): each equipment class is just
a model (a set of equations for a set of variables). Then the simulator
just assemble a big model and solve it by a Newton's like method (for
steady state case) or with some DAE solver (dynamic case). EO is by far
more suited for dynamic simulation. Examples: EMSO, gPROMS, Aspen
Dynamics, ASCEND, Speed-Up.

A good article comparing the technologies is by Marquardt (1996):
Trends in computer-aided process modeling.
For Portuguese speakers I could send my master dissertation about the
design of an EO tool.

Each approach has advantages and draw-backs, what do you think?

With the best wishes.

[Matt Henley]

Our current discussions are based around using the methodology
presented in Sim42's solver. This has already been written and we can
use it as a go-by (its in python.. not Object Pascal). For me
personally, I am looking for something that looks and behaves like
hysys. I have never used a non sequential modular system. The systems
I have encountered are Hysim, Prosim, TSweet, Promax, Pro2 and a
couple of smaller more specialized systems.

[Kiran Pashikanti]

>From my examination of sim42's code, it appears that sim42 uses
sequential modular (SM) approach to solve flowsheets. Each indvidiual
unit operation has its own Solve() method whose results need to be
converged. It also appears that sim42 propagates results forwards and
backwards (leading to quicker solutions perhaps). I'm currently trying
to pair down the sim42 code a simple reference implementation that
makes it clear what the simulator is actually doing when solving a
flowsheet.

I think rewriting sim42 with an equation orientied (EO) approach is
elegant but ill-advised. The EO approach generates flowsheets with
thousands of equations and errors in the sheet are hard to find.
Additionally, the EO approach will require writing (or porting) a lot
of numerical code which does not seem to exist for the implementation
language chosen for opsim.

If opsim is designed carefully, a transition to a EO approach may not
be too difficult to make.

Good luck,

Kiran

[Rafil Elyas]

Sim 42 is the same sequential modular solver as HYSIM and HYSYS. However,
unlike the other SM solvers, it's stream based, most of the unit ops (other
than distillation columns and reactors) pass state information from inlet to
outlet stream and vice versa (bidirectionally) allowing you to calculate
backwords and forwards, minimizing the need for adjusts/controllers.

The EO solver is useful when you get a large number of recycles in your
system. And get faster solutions if you're using your model to supply
functions for an optimizer. Important if you're doing real time
optimization. That's one of the reasons Simsci (now Invensys) replaced their
SM ROM with ROM-EO.

I find SM solution speeds are adequate design and plant troubleshooting
models. SM solutions also make it easier for third parties to add their own
objects easily.

[Rafael de Pelegrini Soares]

Dear Rafil,

I agree, SM is very well suited for plant design (excluding dynamic analysis),
troubleshooting and maybe steady-state optimization.

SM is not adequate for dynamic simulation, data reconciliation, real-time
simulation/optimization.

But I do not agree when you tell that SM make it easier for third parties to add
their own objects. In EO you dont need to write a 'solve' routine and then
program-compile-link (you just write a set of equations).

Anyway I started this thread just to make clear the final objective of the
project:
- A steady-state SM simulator;
- Models are programmed in the same programming language of the system deriving
from a base unit operation class;
- Thermodynamics is part of the project (maybe an interface for already existin
g packages could be a good option to get something working sooner);

Please correct me if I get it wrong.

With the best wishes.

--
::M.Sc Eng. Rafael de Pelegrini Soares
::raf...@enq.ufrgs.br - www.rps.eng.br

Citando Rafil Elyas <rafil...@east101.com>:

[Craig Morris]

For the record, Sim42's solver is modular, but it is not sequential -
hence the non-sequential modular tag we invented way back in the Hysim
days. Because I, not surprisingly, am fond of the forwards-backwards
stuff, Sim42 does share this non-sequential characteristic with Hysim
and Hysys, but it is completely wrong to say it has the same solver. It
was written entirely from scratch without reference to the earlier
solvers and is structurally quite different.

Craig

--
Craig Morris - cr...@redtree.com
Fernie, BC, Canada

[Rafil Elyas]

Heh, I should have phrased it, "I perceive it behaving like the HYSIM
solver".