IDAES simulation backend¶

cheme process simulate solves a flowsheet in DWSIM by default. A flowsheet within a bounded scope can instead be solved by IDAES (idaes-pse, on Pyomo, solved with IPOPT), an optional second backend (ADR 0039). Both backends take the same simulator model and write results through the same --fill and --check path, run record and provenance.

Choosing the backend¶

The backend is a field of process/simulation.yaml, not a command-line option, because it is an input of every simulated value (its fingerprint) and of the run record:

schema: cheme/simulation/v1
kind: simulation
simulator: idaes            # dwsim (the default) or idaes
property_package: IDAES ideal VLE
specs:
  H1: {outlet_temperature: {value: 600, provenance: assumed, source: ASSUMED}}
  R1: {mode: adiabatic}
  F1: {temperature: {value: 325, provenance: assumed, source: ASSUMED}}

Switching simulator changes the model's fingerprint, so values filled by the other backend become cheme.process-sim-stale until --fill runs again.

Installing IDAES¶

cheme-as-code never imports IDAES or Pyomo. The worker (src/chemeascode/integrations/idaes_worker.py) runs in a separate Python environment:

python -m venv idaes-env
idaes-env/bin/pip install idaes-pse
idaes-env/bin/idaes get-extensions      # IPOPT and the other solver binaries
export CHEME_IDAES_PYTHON=$PWD/idaes-env/bin/python
cheme process simulate --fill

CHEME_IDAES_COMMAND takes a full command instead (a container or a remote host), which must run a copy of the worker with the request on its stdin. Without either variable the run stops with cheme.process-sim-unavailable. The worker was written against idaes-pse 2.13 and Pyomo 6.10.

What the backend runs¶

The scope is what the toluene hydrodealkylation loop needs (examples/process-hda-idaes). Everything outside it is refused at translation, before any solver runs, with cheme.process-sim-backend-unsupported (error) naming the step:

Kind	IDAES model	Notes
heater, cooler	`Heater`	outlet temperature or duty; one outlet (no condensate outlet)
valve	`PressureChanger`, adiabatic
pump	`PressureChanger`, pump	efficiency
compressor, expander	`PressureChanger`, isentropic	`IDAES Peng-Robinson` only; no intercooled stages
mixer	`Mixer`	outlet at the lowest inlet pressure
splitter	`Separator`, total flow	fractions from the step's `split:`
flash	`Flash`	temperature or adiabatic; pressure or no pressure drop; one inlet
conversion reactor	`StoichiometricReactor`	one outlet; conversion reactions in the vapour or the liquid phase; adiabatic, isothermal or outlet temperature
recycle	tear streams	two sequential-decomposition passes initialize the units, then one simultaneous IPOPT solve closes the loop

Not supported (use simulator: dwsim): heat exchangers with two sides, shortcut and rigorous columns, component separators, equilibrium, Gibbs and kinetic (fixed-bed) reactors, compressors with intercoolers, saturation feeds (vapour fraction instead of temperature), interaction parameters, --save.

Property packages and what they claim¶

`property_package`	Method	Built from
`IDAES ideal VLE`	Raoult's law and ideal gas (methods.yml `IDEAL`)	IDAES's `BT_ideal` (benzene, toluene) and `HC_PR` (hydrogen, methane) component data
`IDAES Peng-Robinson`	Peng-Robinson, every kij 0 (methods.yml `PR`)	IDAES's `BT_PR` and `HC_PR` component data

Both use IDAES's generic (modular) property framework; the worker reads the component data from the IDAES installation at run time. Two deviations from that data and IDAES's defaults:

IDAES ideal VLE takes its vapour-liquid equilibrium in the complementarity form of Dabadghao et al., doi:10.1002/aic.18029, which IDAES ships for cubic equations of state only, without the cubic-root terms. IDAES's smooth VLE has no solution for a subcooled near-pure liquid (pure toluene, benzene at its trace), or one with a free vapour fraction: the bubble and dew points are closer than its smoothing. IDAES Peng-Robinson keeps smooth VLE and does not converge on such a stream (neither does IDAES's cubic complementarity form reliably in idaes-pse 2.13).
BT_ideal's Perry's liquid heat-capacity coefficients are read in kJ/kmol/K: IDAES labels them J/kmol/K, a thousand times below its own BTX_activity_coeff_VLE values for the same correlation, which leaves a liquid's enthalpy almost independent of temperature.

Compounds: benzene (71-43-2), toluene (108-88-3), hydrogen (1333-74-0) and methane (74-82-8), hydrogen and methane in the vapour only. Enthalpies include the formation enthalpies at 298.15 K (IDAES's example data, not verified by cheme-as-code), so a reactor needs no separate heat of reaction.

No IDAES package is benchmarked by cheme-as-code. Translation always reports cheme.process-sim-method-unbenchmarked (info); methods.yml claims no application range for the IDAES route, and simulated values are not predictive. A recorded method_selection is compared by method (IDEAL, PR) as for DWSIM. Agreement with a DWSIM run of the same flowsheet is a consistency check, not a validation.

Results and provenance¶

As with DWSIM, every simulated stream value gets provenance calculated, origin simulated, the model fingerprint and the source

cheme process simulate: IDAES <idaes-pse version>, <property package>

and process/simulation-run.json records the run. With IDAES its simulator entry also names the numerical solver (solver: ipopt <version>), iterations is the IPOPT iteration count of the final solve, and notes keeps an initialization that did not finish (the simultaneous solve may still converge; converged is what counts). A run that does not converge writes no stream value. The conversion and energy checks of a DWSIM run apply, except that a reactor's energy closure is reported as unsupported: the IDAES packages' enthalpies include the formation enthalpies and they report none separately, and a closure on them (no formation term added) is not yet verified by cheme-as-code. The plant closure's basis is then simulator. The record's energy.reference (and energy.formation, with a reactor) states IDAES's basis: formation_enthalpy_included: true, no formation term added.

Tests¶

tests/test_process_simulate_idaes.py checks the translation, the refusals and a fill and check from a recorded worker response without IDAES. With CHEME_IDAES_TESTS=1 and CHEME_IDAES_PYTHON set it also runs the HDA example in IDAES and checks the material balance, the declared conversion and the flash outlets.