COMPUTER AIDED DESIGN AND OPTIMIZATION OF POLYMERIZATION PROCESSES

1
COMPUTER AIDED DESIGN AND OPTIMIZATION OF
POLYMERIZATION PROCESSES
Professor Costas Kiparissides
Department of Chemical Engineering Chemical
Process Engineering Research Institute, Aristotle
University, Thessaloniki
Glasgow, April 3, 2001
2
OUTLINE

• Introduction
• Computer Aided Design of High Pressure Low
  Density PolyEthylene Tubular Reactors
• Computer Aided Design of Industrial PVC Batch
  Suspension Polymerization Reactors
• Computer Aided Design of Industrial Emulsion
  Polymerization Reactors
• Computer Aided Design of Gas Phase Olefin
  Catalytic FBRs

3
CADOC of Polymerizations Reactors

• In the Laboratory of Polymer Reaction Engineering
  at AUT/CPERI, computer-aided design (CAD),
  computer-aided process optimization and control
  (CADOC) are very actively promoted aiming at
  reactor design improvements, productivity
  increases, quality improvement, safer reactor
  operation, energy conservation, minimization of
  manufacturing waste and environmental impact,
  etc.
• The philosophy adopted by LPRE is toward the
  development of custom-made CADOC software for
  specific polymerization processes.
• The main problems to be addressed in the
  development of a CADOC software package are
  

• Deeper understanding of polymerization reactions
  (via mechanistic / data-based modelling and
  experimentation).
• Ability to measure and characterize a whole range
  of polymer quality variables using a variety of
  hardware and software sensors.
• Development of nonlinear model-based optimization
  and control policies with emphasis on achieving
  superior performance and constraint handling.

4
Key Ingredients of the Developed Software Packages

• State-of-the-art model and physical properties
  evaluation.
• Fusion of the modeling (academia) and operating
  (industry) expertise in order to
• consolidate the control and design objectives
• identify the key reactor measurements and model
  parameters
• structure the parameter estimation-control-optimiz
  ation problems
• Meticulous in their operations, transparent in
  their use.

5
Features of the User-Friendly Interface

• Menu driven - Problem Specific
• Input Output
• Automatically Created Formatted
• Input - Output File
• Database Aided Input
• Graphics Output

Input Files

Editing Input Files
Modified
Input Files

• Reactor Geometries and
• Configurations,
• Initiator Properties and
• Mixtures

FORTRAN Programs
FORTRAN
Output
'
Processing Programs Output

• Profiles of Temperature,
• Pressure,
• Heat Transf. Coefficient,
• Weight - Number Av.M.W.,
• Density, Melt Index,
• LCB, SCB, etc

Reactor Temperature
Formatted Printout
Graphs
Comparative Graphs
6
Functions of the Sofware Packages
7
COMPUTER AIDED DESIGN OF HIGH PRESSURE LOW
DENSITY POLYETHYLENE TUBULAR REACTORS
8
LDPE Tubular Reactor Configuration

• A tubular LDPE reactor consists of a
  spiral-wrapped metallic pipe with a large
  length-to-diameter ratio. The total length of the
  reactor ranges from 500 to 1500 m while its
  internal diameter does not exceed 60 mm.
• The heat of reaction is partially removed through
  the reactor wall by water which flows through the
  reactor jacket.
• The reactor is divided in to a number of sections
  in relation with process heat requirements and
  typically has multiple feedstreams and initiator
  mixture injection streams.

9
LDPE Modelling (Reactor Geometry Input)

• Detailed input of reactor geometry (reactor
  sections, tubes and cross section information)

• Input of reactor cooling systems information
  (information about coolant tanks, coolant
  flowrate and type of flow for each section)

10
LDPE Modelling (Input of Model Parameters)

• Input of estimated model parameters (reactor
  section fouling factors, values initiator
  efficiency)

• Input of kinetic parameters (Arrhenius factor,
  activation energy and activation volume for all
  elementary reactions of free radical
  polymerization)

11
LDPE Reactor Modelling (Graphics Output)

• User-selected display of any two curves (e.g.
  number average molecular weight and
  polydispersity index of the final product)

• Superposition of experimental data (e.g.
  temperature data)

12
LDPE Reactor Modelling (Graphics Output)

• Comparison charts of any property between
  different reactor runs (e.g. overall heat
  transfer coefficient)

• User-specified composite graphs (e.g. overall
  heat transfer, inside heat transfer and fouling
  heat transfer resistance)

13
COMPUTER AIDED DESIGN OF INDUSTRIAL PVC BATCH
SUSPENSION POLYMERIZATION REACTORS
14
The Suspension Polymerization Prosess

• Monomer dispersed as droplets (e.g. 20-250 µm).
• Initiator dissolved in organic phase (monomer).
• Dispersion maintained by agitation and addition
  of stabilizers (polymers, CMC, PVA,
  electrolytes).
• Particle size distribution difficult to control.
• Best for production of PS, PMMA, PVC, PVAc.
• Final PSD in the range of 50-500 µm.

15
Kinetics Reactor Modeling (GRAPHICS)

• User-selected display of any two curves (e.g. VCM
  Conversion and Reactor Pressure)

• Superposition of experimental data (e.g. VCM
  Conversion)

16
Particle Size Distribution (GRAPHICS)

• Diameter Density Distribution evolution with time.

• Volume Density Distribution evolution with time
  (log scale).

17
Morphology DLA Simulations
Evolution of PVC grain morphology up to 80
conversion for an actual industrial case.

• Critical conversion xc16.
• Final average porosity e11.
• Pore sizes range from 50nm to 5µm.

(Length scale is in µm).
18
COMPUTER AIDED DESIGN OF INDUSTRIAL EMULSION
POLYMERIZATION REACTORS
19
Modelling of Emulsion Polymerization
Reaction Kinetics
Thermodynamics of the emulsion mixture
Aqueous phase radical concentration
Diffusion-controlled termination and propagation
reactions
Homogeneous
Rate of particle growth Average number of
radicals per particle
Rate of radical entry
Rate of particle formation Nucleation mechanism
Micellar
Rate of radical exit
Coagulative
Radical termination
Numerical methods for solving population balances
Reactor Dynamics

• Conversion
• Particle number
• Average particle radius
• Molecular weight
• Copolymer composition

• PSD
• MWD
• CCD
• DBD

20
Kinetics Reactor Modeling (INPUT)
21
Kinetics Reactor Modeling (GRAPHICS)

• Superposition of experimental data (e.g. Diameter
  of Swollen Particles)

22
Particle Size Distribution (GRAPHICS)

• Predicted Particle Size Distribution at different
  conversions

23
COMPUTER AIDED DESIGN OF GAS PHASE OLEFIN
CATALYTIC FBRs
24
Olefin Polymerization FBR from reaction
kinetics to FBR Modeling
25
Modeling of the Borstar FBR
Flash
Separator
26
Dynamic Measurements from the Borstar FBR
27
Steady-state simulation of the Borstar FBR

• Steady-State predictions of the developed model
  were compared with the industrial data The
  kinetic rate constants the active site fraction
  of the catalyst and the bed voidage were
  considered as tuning parameters.

Deviation of model predictions from industrial
data
Variable Error Reactor Temperature
0.28 Production Rate 0.99 Residence
Time 4.6

• The model predictions are in close agreement with
  the industrial data for two months of continuous
  operation of the reactor.

Dynamic simulation of the Borstar FBR

• Dynamic model simulations were performed and
  compared with the industrial data for 2 months of
  plant operation. The tuning parameters are kept
  constant with time. Dynamic simulations were
  performed using a numerical integration routine
  under FORTRAN as well as the gPROMS simulator.

28
Dynamic simulation of the Borstar FBR
29
Prediction of the Borstar FBR PSD

• The steady-state population balance model was
  used for the prediction of the particle size
  distribution of the Borstar FBR over a wide
  variety of steady-states of different time
  instances of plant operation. The model
  predictions proved to be quite accurate
  (deviations in the order of magnitude of sampling
  and measurement errors). The reactor operating
  conditions corresponding to four representative
  PSD measurements (denoted as SS1-4) are as
  follows

30
Prediction of the Borstar FBR PSD

• Steady state 1

• Steady state 2

31
Prediction of the Borstar FBR PSD

• Steady state 4

• Steady state 3

32
Optimal Grade Transition in an FBR using
gPROMS-gOPT Simulation-Optimization Package

• An integral quadratic objective function is used
  as a performance index for this grade transition,
  penalizing the deviations of polymers MI and
  Density from their desired values.

• Hydrogen (FH2) and Comonomer (Fmon2) feed rates
  are the only two manipulated variables required
  to achieve the desired polymer properties and
  they are used as optimization variables to carry
  out the grade transition.

• The manipulated variable profile is discretized.
  The manipulated variables are expressed as
  piecewise constant functions while the time
  domain is divided into a number of discrete
  intervals. The optimal control problem is solved
  sequentially as an NLP. Therefore the optimizer
  must choose a set of parameters that construct
  the manipulated variable trajectories, in order
  to minimize the performance index shown above.

• The effect of the discretization of the time
  domain on the grade transition is studied and the
  optimal policies are compared for different cases
  where the time domain is divided into 10/15/20
  intervals equally spaced or not with tight or
  wider bounds for each interval.

33
Optimal Grade Transition in an FBR using
gPROMS-gOPT Simulation-Optimization Package

• Time domain is divided into 10/15/20 non-equally
  spaced intervals. The transition under the
  optimum policy of the manipulated variables is
  compared with the transition under a simple step
  change of the manipulated variables.

34
CONCLUSIONS - PolyPROMS

• An easy-to-use environment with a variety of
  interfaces including a fully interactive process
  flow diagram (PFD).
• Capability to simulate a wide range of
  polymerization mechanisms and reaction media.
• Full access to kinetic parameters, species
  properties/concentrations, process conditions.
• Consistent thermodynamic modeling and data
  throughout.
• A non-linear steady-state solver for the
  evaluation of steady states for interactively
  specified values of flowsheet parameters and the
  sensitivity analysis of the process output
  variables with respect to variations of model
  parameters.
• General-purpose linear and non-linear programming
  algorithms for the off-line and on-line parameter
  and state estimation in polymerization process
  models.
• General optimization tools for parametric
  steady-state optimization of a selected
  polymerization process.
• Dynamic optimization methods for the
  determination of the time optimal control
  policies to improve product quality, maximize
  reactor throughput, or/and minimize the off-spec
  amount of polymer during a grade transition.
• Non-linear model based predictive control (NMPC)
  algorithms for the feedback control of the
  polymerization processes.