PERFORMANCE Getting the most from a UniSim Design simulation

1

• PERFORMANCE Getting the most from a UniSim
  Design simulation
• Devon Clack

2
Problem Statement

• An interest in gasification and simulation led to
  the development of a Gasifier simulation model.
• The simulation has to be able to reliably,
  accurately and dynamically predict the product
  gas composition at specified conditions of
  temperature and pressure.

3
Background

• Gasification
• Is a process whereby carbonaceous materials are
  converted to a synthesis gas (or syngas) that is
  primarily composed of carbon monoxide, carbon
  dioxide, methane and hydrogen.
• The carbonaceous material reacts with a specified
  quantity of oxygen and steam. Pure oxygen or air
  may be used.
• Coal, petroleum or biomass can be gasified
• Gasification is carried out in a gasifier.
• Gasifiers can be classified according to the
  method of contact between the gas and solid
  phases.

4
Background

• Gasifiers
• Legacy Gasifiers are counter current, moving-bed
  coal gasification units. The coal bed moves
  downward by gravity, counter-current to a gas
  stream flowing upward due to a pressure
  differential.
• The process is of a semi-batch nature. Coal
  batches are loaded every 11 minutes, however,
  products are withdrawn continuously.
• Steam and oxygen are fed continuously at the
  bottom of the gasifier to provide the reactants
  for the combustion and gasification reactions.

5
Model and Simulation Development

• The model incorporated into the UniSim Design
  simulation was the product of extensive research
  conducted on many aspects of the gasification
  process.
• The most important assumptions incorporated into
  the model are
• The primary reactions believed to occur
• C(s) H2O(g) ? CO(g) H2(g) (1)
• C(s) CO2(g) ? 2CO(g) (2)
• C(s) 2H2(g) ? CH4(g) (3)
• CO(g) H2O(g) ? CO2(g) H2(g) (4)
• C(s) O2(g) ? CO2(g) (5)

6
Model and Simulation Development

• Equilibrium is thought to be reached for many of
  the reactions that occur inside of a gasifier. As
  a result it will be assumed that all of the
  reactions in the gasification zone (reactions 1
  4) are at equilibrium.
• The composition and temperature of the effluent
  gas can be predicted without any detailed
  kinetics by making the following assumption
• 100 conversion of the oxygen in the feed through
  oxidation (reaction 5)
• These assumptions were incorporated into a
  dynamic simulation in UniSim that also accounts
  for the counter-current nature of the flow of
  reagents.

7
Final Simulation PFD
8
Data Mediation

• Generic legacy Gasifier plant data was used, with
  operating temperature and pressure a given. The
  data also included the following information at
  one minute intervals
• Coal feed rate
• Steam feed rate
• Air feed rate
• The corresponding syngas concentrations of CO,
  CO2 and CH4
• In order to validate the model and simulation,
  the feed data was used in a simulation and the
  dynamic syngas concentration predictions of the
  simulation compared to the generic plant data.

9
Data Mediation

• To automate data entry, Microsoft Excel was set
  up as a dynamic data mediator using Visual Basic
  scripting.
• The development of the automation script made it
  possible to do the following, all from the
  Microsoft Excel interface
• Export equilibrium parameters from Excel to a
  UniSim Design simulation.
• Initiate a dynamic simulation and run it for a
  particular length of time.
• Input feed flow data from Excel during a dynamic
  simulation.
• Retrieve simulation syngas compositions at each
  time step, and write the values to a spreadsheet.
• Calculate an error using the generic data and the
  data retrieved from the simulation.

10
Model Validation

• So as to validate the model incorporated into the
  developed simulation, a dynamic simulation was
  run with the available generic Gasifier data set.
• A performance parameter had to be chosen to gauge
  the accuracy of the simulation.
• The primary components of interest in the syngas
  are
• Carbon monoxide
• Carbon dioxide
• Methane
• The performance parameter used was the sum of the
  squared errors (SSE) of the relevant syngas
  components.

11
Model Validation

• The SSE was defined as
• SSE (ECO)2 (ECO2)2 (ECH4)2
• Where the component errors (E) are defined with
  respect to the component molar fractions (Xi) in
  the syngas as follows
• Ei Xi-simulation Xi-plant
• Using literature values for the reversible
  reactions in the model, a dynamic simulation run
  over 50 minutes produced results with an SSE of
  0.159.
• These results needed to be improved if the
  simulation was to have a practical use in
  industry.

12
Simulation Improvement

• Initially, literature values were used in the
  simulation for the equilibrium constants
  associated with the reversible reactions in the
  model.
• The equilibrium constants were determined
  experimentally, under strictly controlled
  conditions, without the presence of competing
  reactions.
• The parameters may not be accurate under the
  conditions that occur inside a gasification
  chamber.

13
Simulation Improvement

• In UniSim, equilibrium constant (K) expressions
  can be inserted in the form of a temperature (T)
  dependent, natural logarithm expression with two
  variable parameters (A and B)

14
Simulation Improvement

• With the aim of reducing the SSE of the results
  produced by the dynamic simulation of the generic
  Gasifiers, the parameters in the equilibrium
  expression for each of the reversible reactions
  were altered.
• Third party software capable of optimising the
  results by adjusting the equilibrium parameters
  had to be linked to UniSim to accomplish this.

15
Optimisation Method

• Matlab has the capability to communicate with
  Microsoft Excel using the ActiveX function.
• This makes it possible to
• Call the Microsoft Excel data mediation script
  from within Matlab.
• Input equilibrium parameters into the automation
  spreadsheet from Matlab, which are subsequently
  set from Excel into the UniSim simulation.
• Initiate a simulation through Excel which runs a
  dynamic simulation using time dependent data from
  the Excel spreadsheet.
• Retrieve an SSE from the Excel spreadsheet once a
  simulation is complete.

16
Optimisation Method

• With the capability of writing new parameters to
  a simulation from Matlab, and being able to
  retrieve a single measure of the simulations
  performance, the SSE, an optimisation routine
  could be employed.
• A non-linear, unconstrained optimisation routine,
  fminsearch, was used on a 50 minute dynamic
  simulation.
• The optimisation procedure is illustrated on the
  following slide and a depiction of the
  interaction between the three software packages
  during the simulation is depicted in the figure
  thereafter.

17
Optimisation Method
18
Solution Software Interaction
19
Results

• The optimisation was done on a 50 minute
  simulation, using generic feed data.
• The normalised feed rates of coal, steam and air
  are shown in the figure.

20
Results

• The optimisation routine was employed on the 50
  minute dynamic simulation to try improve the
  performance of the simulation.
• The reduction in the SSE as a function of the
  number of function evaluations is shown in the
  figure.
• One function evaluation in the optimisation
  procedure is one full 50 minute dynamic
  simulation, from which a single SSE is obtained.

21
Results

• After considerable computing time, and more than
  350 dynamic simulations, the SSE of the original
  simulation was reduced from 0.159 to 0.1068.
• The sum of the squared errors was reduced by 32.
• The dynamic concentration profiles predictions of
  the key components in the syngas are plotted with
  the corresponding plant data on the following 3
  figures.

22
Results CO Predictions
23
Results CO2 Predictions
24
Results CH4 Predictions
25
Results

• These are remarkable results considering
• The vastly simplified model incorporated into the
  dynamic simulation of a Generic gasifier.
• The quality of the data used to validate the
  model.
• The optimisation routine did not have enough
  computing time to find a local minimum in the SSE.

26
Conclusions

• A solution structure was developed to improve the
  performance of a dynamic simulation of a generic
  Gasifier. The solution structure incorporated
  three software platforms, namely UniSim Design,
  Microsoft Excel and Matlab.
• UniSim Design was used as the chemical simulation
  engine in the solution structure, Excel was used
  as a dynamic data mediator and performance
  calculator while Matlab was employed as an
  optimiser.
• By making use of a combined automation and
  optimisation solution method, UniSim Design
  simulations can be improved to produce more
  accurate results for a particular application.

27
Recommendations

• If this solution structure is to be employed
  again, it is recommended that the optimisation be
  run uninterrupted, giving it time to find a local
  minimum and thus ensuring the best simulation
  results.
• It is recommended that a simulation is developed
  that incorporates matching coal analyses and
  gasifier data before running the optimisation
  process.
• A constrained optimisation procedure should be
  used to alter simulation parameters to ensure the
  parameters are fundamentally correct. This should
  produce a simulation capable of accurate syngas
  concentration predictions.
• Investigate whether coal compositions or
  operating conditions have an influence on the
  optimised equilibrium parameters.