Customizing RADFRAC convergence

Advanced Distillation with Aspen Plus

Customizing RadFrac Convergence

Customizing RadFrac Convergence Advanced Distillation with Aspen Plus

© 2002 AspenTech. All Rights Reserved.

Lesson Objectives • Understand how to customize RadFrac convergence algorithms



9–1

Aspen Technology, Inc.



RadFrac Convergence Form


Basic Convergence Input From the Basic sheet, you can alter the following convergence parameters: • Algorithm (only for Custom Convergence). • Maximum iterations. • Error tolerance. • Initialization method (only for Custom Convergence). • Damping level. • Liquid-liquid phase splitting method. • Solids handling.



9–2




Convergence Methods • To customize convergence methods, first choose Custom on the RadFrac Specifications form. • This will allow you full freedom to select both the convergence method and the initialization strategy. • For details on convergence methods, refer to lesson 6.


Custom Convergence Specifying Custom Convergence on the Setup Configuration sheet allows the user to select convergence strategy:



9–3




Algorithms • Four algorithms can be selected on the Algorithm field: – Standard – Sum-Rates – Nonideal – Newton

• Parameters for the specific algorithms are on the Convergence Algorithm sheet.


Maximum Iterations • Most problems converge within the standard default 25 iterations. • Complex columns may need more. – Large numbers of components – Reactions – Chemistry – Liquid-liquid equilibrium

• In general, increase Maximum iterations if Err/Tol is decreasing but convergence is not achieved in 25 iterations.



9–4




Error Tolerance • The tolerance specified here is the outside loop convergence tolerance. Expect errors of this magnitude in: – Dew and bubble points – Mass balance (component and total) – Enthalpy balance

• The default value is 10-4. Do not increase this value. • Decrease (tighten) when necessary.


Tightening Error Tolerance Tighten tolerance in these situations: • Following trace components. • LLE where one liquid phase is small. • Any time greater accuracy is desired. • RadFrac blocks inside recycle loops. – Default tear stream tolerance is 10-5 – Errors in RadFrac can impede tear convergence – Make RadFrac tolerance 1 or 2 orders of magnitude tighter than tear stream tolerance (otherwise noise from RadFrac results may adversely affect higher level iterations)



9–5




Flowsheet with Recycle Recycle

RadFrac convergence tolerance should be tighter than tear stream tolerance. © 2002 AspenTech. All Rights Reserved.

Initialization Option Use this strategy For this situation …………………………………………………………………… Crude For wide boiling systems with multidraw columns …………………………………………………………………… Chemical For narrow boiling chemical systems …………………………………………………………………… Azeotropic For azeotropic distillation columns …………………………………………………………………… Cryogenic For cryogenic applications (for example, air separations) ……………………………………………………………………



9–6




Damping Level • RadFrac offers four damping levels: – None (default): – Mild: – Medium: – Severe:

No damping Mild level of damping Medium level of damping Maximum damping

• Use damping to stabilize convergence if excessive oscillation occurs during convergence (when Err/Tol oscillates when near solution). • Damping slows convergence – with Severe damping extra iterations are almost always needed.


Liquid-Liquid Phase Split Algorithm Gibbs (default) – Uses Gibbs Energy minimization – Stable, reliable – Requires that LLE model be thermodynamically consistent Eq-Solve – Equates component fugacities – May find false roots – Fast Hybrid (new for Aspen Plus 10) – Uses Eq-solve to find all roots and Gibbs to select one – Should be faster than Gibbs; requires consistency



9–7




Solids Handling Overall (default) – All solids are removed from feed streams and put into bottom

streams – Makes column heat balance easier to solve – Small errors in duties and bottoms temperature – Flow specifications (Bottoms flow, e.g.) do not include solids

Stage – Solids move down the column, are present on all stages below

feeds – Gives correct heat balances and temperatures – Flow specifications include solids


Fixing Convergence Problems Err/Tol

Steps

Decreases but slowly

Increase the number of iterations

Oscillates

Increase Damping Level

Diverges

1) Provide Initial Estimates 2) Check specifications for problems 3) Simplify the specifications 4) Try different algorithms 5) Tune algorithm parameters



9–8




Convergence Algorithm Sheet • This sheet allows input for details of three-phase calculations and for algorithm -specific parameters. • It has four Subsections: 1. Sum-rates parameters 2. Newton parameters 3. Non-ideal parameters 4. Three-phase parameters


Sum-Rates Parameters Jacobian Method controls when and how the Jacobian (matrix of partial derivatives) is calculated. – Initial - Calculates the Jacobian initially, updates it with the

Broyden method once inside loop has converged – Rmsol - keeps recalculating Jacobian until outside loop error falls below the Update threshold. Below the Update threshold Broyden is used. – The Update threshold default is 0.01 – The Rmsol method is slower (more Jacobian calculations are done) but more stable. Lowering the threshold increases this effect.



9–9




Newton Parameters Newton’s method requires large amounts of intermediate storage space. Two parameters control the amount of space allocated: – Maximum size – Storage factor

These may need to be increased for large problems. If so, you will get an explicit error message when you first run the column.


Non-Ideal Method Parameters Gamma model controls what activity coefficient model is used to extrapolate activity coefficients in the inside loop – Margules – Combined (Wilson and Margules)

User can also specify limits on extrapolated activity coefficients at the midpoint and at infinite dilution – Increasing max gamma at infinite dilution can reduce iterations

needed for highly non-ideal (high activity coefficient) systems – Increasing max gamma at midpoint may reduce outside loop iterations but increase inside loops



9 – 10




Three-Phase Parameters Mole-fraction threshold for 2nd liquid key component

100 75 25

50

Temperature C

• Default is 0.5.

125

150

• Allows user to control whether a single liquid phase is called liquid1 or liquid2. Does not affect phase equilibrium calculations. Water-Butanol Liquid-Liquid Phase Diagram

0.5

0.55

0.6

0.65 0.7

0.75 0.8

0.85 0.9

0.95

1

Mole Fraction of Water


Additonal Three-Phase Parameters Gamma-model option for 2nd liquid phase: – Constant (computed from the outside loop) – Margules (default -- acts like non-ideal algorithm) Maximum liquid-liquid phase split iterations: – Controls LL iterations in inside loop – LLE done rigorously in outside loop – Default is 20 Maximum initialization passes: – Controls number of initialization iterations in which two liquid phases are assumed to exist on all stages – Default is 0. Using 1 or 2 may aid convergence when initialization does not predict the region of LLE well.



9 – 11




Convergence Advanced Sheet • A large number of unrelated options and parameters in alphabetical order. • Duplicates parameters on some other forms (Maxol). • Keeps all the old input language functionality. • Often uses input language terminology.

» This section covers the most important parameters in a (somewhat) more logical order. © 2002 AspenTech. All Rights Reserved.

Advanced Tab Parameters • Absorber (Yes or No) – Controls inside-loop convergence for Standard algorithm – Use ‘Yes’ for absorbers and strippers without condenser or

reboiler

• Fminfac – Lower limit of tray flow rate (vapor or liquid) as a fraction of

sum of all feed flows – Default is 10-5. Lower this only if there are expected to be some very small flows



9 – 12




Design Spec Parameters • Dsmeth – Controls the design spec method used with Newton algorithm – Simult (default) solves them with all the other equations

(usually quicker) – Nested solves them in a separate loop

(gives results even if spec is not satisfied)

• Rmsol0 – Specifies the RMS outside loop error below which design spec

iterations are performed – Default is 0.1. Decreasing this may help if column appears to diverge once middle loop iterations begin


Feed Flash Parameters • Flash-Maxit – Controls how many iterations are allowed for the feed flash – Default is 50 – Often 75 or more are needed for difficult flashes, especially • •

Liquid-liquid Equilibrium Electrolyte Chemistry

• Flash-tol – Controls feed flash tolerance (to override global flash-tol) – Rarely changed



9 – 13




Local Enthalpy Model Parameters • Hmodel1 – Mole, Mass, or Pseudomass weighting

• Hmodel2 – No-Temp: local enthalpy dependence only from ideal gas

(usually adequate) – Temp: local T dependence calculated initially (needed for absorbers because sensible heat effects are important) – Update: recomputes local Temp dependence every pass (rarely used)

Both are automatically chosen, but can be overridden © 2002 AspenTech. All Rights Reserved.

Local K-value Parameter • Kmodel choices: – X weights based on liquid mole fractions – Y weights based on vapor mole fractions – K based on (vapor mole fractions)/(k+1)

– This is chosen automatically based on type of distillation

calculation – Change to X or Y for systems with electrolytes



9 – 14




Inside Loop Parameters (1) • Maxil – Controls how many Inside Loop iterations are allowed per

outside loop – Increasing Maxil can help middle loop (design spec) calculations converge – Do not increase unless there are design specs – Do not increase above about 20 (if 20 is not enough, the problem lies elsewhere)


Inside Loop Parameters (2) Ilmeth – Controls how the inside loop is converged – Broyden (default) • Usually the fastest – Wegstein • May be necessary for very large columns (200+ stages) • Uses less storage than Broyden, but is slower • May require increasing Maxil – Newton (bigger and faster than Broyden; good with tray

efficiencies) – Schubert (default when using Sum-Rates)



9 – 15




Acceleration and Damping Controls • Acceleration (and damping) for the Bounded Wegstein method: – Qminbwol – Qmaxbwol – Qminbwil – Qmaxbwil

» In Aspen Plus 10, use of the Damping control on the Convergence Basic tab is preferred.


Wegstein Parameters (1) Qminbwol: The lower bound on qOL (Wegstein acceleration parameter). The default is 0.

qkOL = 0

x kOL+1 = xˆ kOL

qkOL = 0.5

xkOL+1 =

(direct substitution)

1 ˆk ( xOL + xkOL ) 2

(simple midpoint damping)

Set Qminbwol to 0.5 if outside loop convergence behavior is very erratic or is divergence occurs.



9 – 16




Wegstein Parameters (2) • Qmaxbwol The upper bound on qOL (default is 0.5). NOTE: If you increase Qminbwol beyond 0.5, for consistency, you must increase Qmaxbwol. • Qminbwil The lower bound on q IL (default is 0). Set Qminbwil to 0.5 if outside loop convergence behavior is very erratic or if divergence occurs.


Wegstein Parameters (3) • Qmaxbwil The upper bound on q IL (default is 0.5). NOTE: If you increase Qminbwil beyond 0.5, for consistency, you must also increase Qmaxbwil.



9 – 17




Acceleration and Damping Controls Rmsol1 – Controls when Broyden method is used to accelerate the

convergence of selected variables – Below this Outside Loop RMS error, Broyden is used – Decrease this number if RadFrac converges to a point and then stalls or becomes erratic – If Rmsol1 is less than the overall convergence tolerance, Broyden is never used » In Aspen 10, use of the Damping control on the Convergence

Basic tab is preferred.


Rmsol1 • RadFrac converges the outside loop by a combination of bounded Wegstein andk Broyden methods: k+ 1 k k k – Wegstein

x OL = q OLi • x OLi + (1- qOL i )xˆ OL i

(x (x

OLi

, x OLi

OLi

, x OLi k +1

– Broyden

) )

k

k → qOL i

k -1

→ q kOLi

k J OL (x OL − x OL ) = xˆ OL − x OL k

k

k

k = iteration index x OLi = assumed value of the ith OL variable xˆ OLi = calculated value of the ith OL variable



9 – 18




RadFrac Convergence Recap • Recheck that the problem is well-posed (physically reasonable) and as simple as realistically possible. • If Err/ Tol is decreasing, give it more iterations. • Change Convergence algorithm • Improve Initialization by: – Choosing a better initialization method, or – Giving estimates

• Increase Damping level • Tune convergence parameters • Reinitialize!


Workshop 8A: Absorber Column GAS-O U T

WATER COLUMN

GAS-IN BOTMS

START WITH: WS8A-ABSORBER.BKP © 2002 AspenTech. All Rights Reserved.


9 – 19




Workshop 8B: Acetone/Water Column

DIST

FEED

COL-FEED

COLUMN

HEATER

BOTMS

START WITH: WS8B-ACETONEWATER.BKP © 2002 AspenTech. All Rights Reserved.

Workshop 8C: Sour Water Stripper OVHD

FEED COLUMN STEAM

BOTTOMS

START WITH: WS8C-SOURWAT.BKP © 2002 AspenTech. All Rights Reserved.


9 – 20




Workshop 8D: VCM Column

OVHD

FEED1

COLUMN

FEED2 BOTMS

START WITH: WS8D-VCMCOL.BKP © 2002 AspenTech. All Rights Reserved.

Workshop 8E: Ethanol Dehydration AZ-OVHD

FEED CH-PHS AZEO-COL

ETOH-PRO

START WITH: WS8E- ETOHDEHY.BKP © 2002 AspenTech. All Rights Reserved.


9 – 21




Workshop 8F: Propanol/Butanol/Water

COLUMN

OVHD

FEED

SIDE

BOTMS

START WITH: WS8F-PBW.BKP © 2002 AspenTech. All Rights Reserved.

Workshop 8G: Electrolyte Column GAS-PROD WATER

COLUMN

SOURGAS BOTTOMS

START WITH: WS8G-ELEC.BKP © 2002 AspenTech. All Rights Reserved.


9 – 22


Customizing RADFRAC convergence

Recommend Documents