Convergence Hints (aspen)

Post on 13-Apr-2015

53 views 2 download

description

Convergence Hints (aspen)

Transcript of Convergence Hints (aspen)

Flowsheet Convergence in Aspen Plus

Strategies for Flowsheet Convergence

Some general guidelines are:

Start small. Make sure that individual blocks and elements of a flowsheet behave as expected, before slowly combining them into a larger simulation. Sensitivity analysis might help here.

Start with simple unit operation models. For example, converge the flowsheet with a DSTWU before switching to RadFrac.

Provide good initial guesses. Make sure the flowsheet starts converging from a reasonable point. If possible, select a tear stream that remains relatively constant.

Strategies for Flowsheet Convergence

Check physical properties. Make sure they are calculated correctly in the entire operating range of the simulation.

Check for correctness, variable accessing, spelling, and unit specifications.

Evaluate tear stream choice.

Analyze warnings and error messages.

Control Panel Messages • > Loop $OLVER02 Method: WEGSTEIN Iteration 198

• 5 vars not converged, Max Err/Tol 0.57095E+02

• Block: B9 Model: FSPLIT

• Block: B10 Model: MIXER

• Block: B1 Model: RSTOIC

• Block: B2 Model: FLASH2

• Convergence iterations:

• OL ML IL Err/Tol

• 1 1 10 45.720

• 2 1 10 20.797

• 3 1 5 13.609

• 4 1 5 7.0574

• 5 1 3 0.87879

Control Panel Messages

Each time the convergence block is executed in a recycle convergence loop, messages appear with the following format:

> Loop Block Method: WEGSTEIN Iteration 9

Converging tear streams: 34 vars not converged, Max Err/Tol 0.18603E+1

where:

Block = Convergence block ID

Max Err/Tol = Maximum error/tolerance for the unconverged variables

> = Symbol indicating nesting level of the convergence loop

> Outside loop, >> Loop nested one deep,

>>> Loop nested two deep, and so on

Control Panel Messages

Each time a convergence block for a design specification is executed in a convergence loop, messages appear with the following format:

>> Loop Block Method: SECANT Iteration 2

Converging specs: H2RATE

1 vars not converged, Max Err/Tol 0.36525E+03

Convergence is achieved when the value of Max Err/Tol becomes less than 1.0.

Debugging Convergence Problems

Specify blocks to be independent of flow rates (e.g. specify fraction instead of flow rate for Fsplit, or distillate/feed rate instead of distillate rate for RadFrac).

Check and confirm or alter calculation sequence.

Debugging Tear Convergence Problems In case of steady convergence progress without finally

achieving convergence the number of maximum iterations should be increased (either globally under Convergence / Conv Options or locally within a certain block).

It might be helpful to disconnect a recycle stream in order to obtain a good intial estimate and to investigate the sensitivity.

Try Broyden and Newton methods instead of Wegstein.

Reinitialize the simulation when appropriate.

Summary

1. Run the simulation using the default sequence generated by Aspen Plus ® .

2. Examine simulation results, look out for skipped and unconverged unit operation blocks. Check the Control Panel and results sheets for blocks that did not complete normally, had errors, or had unexpected results that might affect recycle convergence.

Summary

Problem Action

Incorrect block specifications. Correct them.

Feed conditions too far off. Provide better estimates for tear

streams and/or design variables.

Convergence specifications. Try different specifications,

different algorithm options, or

increase the number of iterations.

Algorithm options. Change options.

Not enough iterations. Increase number of iterations.

2 cont. Some common reasons for these problems are:

Summary

3. Check whether adjustments in the tolerance values or the algorithm parameters or a change of the algorithm itself is needed.

4. Check for unconverged design specifications.

5. Alter the calculation sequence, if necessary.

6. If the flowsheet is modified, rerun the simulation and go back to step 2.

The Golden Rules of flowsheet modeling

• Start as simple as possible

• Stay simple as long as possible

• Add necessary complexity one item at a time

• Test non-trivial additions separately

Acknowledgement

• Prof. Dr. Konrad Hungerbühler ETH Zürich, Institute for Chemical and Bioengineering.

• Prof. Jack Ponton

The School of Engineering, University of Edinburgh.