The document discusses process simulation and the modeling of biochemical processes like ethanol production. It provides an overview of:
- Using process simulation to evaluate different process routes and operating conditions.
- Developing structured mathematical models to represent metabolic pathways and reaction kinetics.
- Modeling ethanol fermentation using a system of differential equations to describe substrate consumption and product formation over time.
- Extending these models to systems with immobilized cells or continuous bioreactors.
Scaling API-first – The story of a global engineering organization
Experience on System Integration and Simulation
1. Universidade Estadual de Campinas- UNICAMP
School of Chemical Engineering
EXPERIENCE ON SYSTEM
INTEGRATION AND SIMULATION
Professor RUBENS MACIEL FILHO
•Laboratory of Optimization, Design and Advanced Process
Control
•Department of Chemical Processes, School of Chemical
• Engineering, State University of Campinas, Campinas - Brazil
e-mail maciel@feq.unicamp.br
VIRTUAL SUGAR CANE BIOREFINERY-
CTBE - August 2009
2. MOTIVATION
• Process Simulation
– Evaluation of several possible routes –
routes discrimination
–Investigation of different scenarios
- Process understanding
- Impact of operation variables on process
performance
3. Process Simulation (cont.)
-Preliminary evaluation of costs, water and
energy consumption
-Studies of variable interaction and process
dynamics
-Operator Training
-Dynamic simulation- process control strategies
may be evaluated
Design of Equipments and plant conceptual
design
4. PROCESS MODELLING
Steady State Model
Dynamic Model
Simplified versus Detailed Model
Physico-Chemisty based Models (Deterministic) versus
Empiric and or Statistical Models
Hybrid Model
Single Unit Models
Large Scale Plant Model
5. Process Simulation
System –can be seen a set of subsystem depending
upon of required investigation
Interaction among subsystems – made through mass
and heat transfer parameters
Subsystem 1– an important component of the process,
inside an equipment where the phenomena are
intrinsically taking place- for instance catalyst particle,
bagasse to be hydrolyzed and microorganism in
biotechnological process. When considered explicitly a
heterogeneous model is formulated.
6. Subsystem 2 - Equipment - peace of the plant where the
changes (reactions, mixtures or separations) are
occurring. In this category it may be place reactors,
separation columns, fermentors, etc.
Subsystem 3 – large scale plant or a set of equipments in
which there exist interest to study
Subsystem 1 and 2 – normally require software
development if detailed representation are desired.
Subsystem 3 – simulators, including the commercial
ones (Hysis , Aspen, Gproms etc)
7. System Integration
There exist an incentive for high
operational performance operation
Process optimization begins with better
process control
Large Plant Optimization and control
RTO: Integrate economic objectives and
control
Stability, controllability and safety
8. System Integration
Large Plant Optimization and Control
RTO (Real Time Operation): Integrate economic
objectives and control
Stability, controllability and safety- may be
expressed as plant restriction
Refinery process ⇒large scale units, high
products output, monitoring difficulties,
data reconciliation
9. Optimization Strategies
Two main strategies are to be
implemented:
One layer approach
two layers approach
Hybrid approach may be necessary
10. One layer approach
Economical optimization problem is solved
together with the control problem
very sensitive to model mismatch
dimension of the optimization problem can
be very large
( on line applications can be restrictive)
use of simplified model may not be suitable
11. controller/
optimizer
Estimation
block
measured
outputs
measured inputs
Process
non-measured
non-measured outputs
inputs
One layer approach
12. Two layers approach
hierarchical control structure where there
is an optimization layer that calculates set-
points to the advanced controller
the optimization layer is composed of an
objective function and a process steady-
state model
13. Optimizer
setpoints
Controller
Estimation
block
measured inputs Process measured outputs
non-measured
non-measured
inputs
outputs
Two layer approach
14. Advanced Controllers
• CONTROLADORES LINEARES
• NON LINEAR CONTROLLERS
• PREDICTIVE CONTROLLERS
• ROBUST CONTROLLERS
• ADAPTIVE CONTROLLERS
• HYBRID CONTROLLERS (NEURAL
NETWORK AND FUZZY COUPLED WITH MODEL
BASED CONTROLLER)
16. STRUCTURED MATHEMATICAL
MODEL
FOR ETHANOL PRODUCTION
Possible to handle with substrate to drive
the fermentation
17. STRUCTURED MATHEMATICAL MODEL
Representative Metabolic Route (F. Lei et al. Journal of Biotechnology 88 (2001) 205-221)
18. Mass balance equations and reaction rate
of the model
∂S glu cos e
= −(R1 + R7 )X + D (S feed − S glu cos e )
∂t
s glu cos e s glu cos e s glu cos e
R1 = k1l X a + k1h X a + k1e
s glu (K 1i s acetaldehyde + 1) + K 1e
s acetaldehyde X a
s glu cos e + K 1l s glu cos e + K 1h
s glu cos e
R7 = k 7 Xa
s glu cos e + K 7
∂S pyruvate
= (0.978 R1 − R 2 − R3 )X − D (S pyruvate )
∂t
s pyruvate 1
R2 = k 2 Xa
s pyruvate + K 2 K 2i s glu cos e + 1
s4
R3 = k 3
pyruvate
Xa
s 4
pyruvate + K3
19. ∂S acetaldehyde
= (0.5 R3 − R4 − R6 )X − D (S acetaldehyde )
∂t
s acetaldehyde
R4 = k 4 X a X Acdh
s acetaldehyde + K 4
s acetaldehyde − k 6 r s ethanol
R6 = k 6 Xa
s acetaldehyde + K 6 + K 6 r s ethanol
∂S acetate
= (1.363R4 − R 5 − R8 )X − D(S acetate )
∂t
s acetate s acetate 1
R5 = k 5 X a + k 5e Xa
s acetate + K 5 s acetate + K 5e K 5i s glu cos e + 1
s acetate 1
R8 = k 8 Xa
s acetate + K 5e K 5i s glu cos e + 1
20. ∂S ethanol
= (1.045 R6 )X − D(S ethanol )
∂t
∂X
= (0.732 R7 + 0.619 R8 )X − D( X )
∂t
∂X a
= (0.732 R7 + 0.619 R8 − R9 − R10 ) − (0.732 R7 + 0.619 R8 )X a
∂t
k9
s glu cos e s ethanol 1 s glu cos e
R9 = + k 9e X a + k 9c Xa
s K s
s ethanol + K 9e 9i glu cos e + 1 s glu cos e + K 9
glu cos e + K 9
s glu cos e s ethanol
R10 = k10 X a + k10e Xa
s glu cos e + K 10 s ethanol + K 10e
21. ∂X Acdh
= (R9 − R11 ) − (0.732 R7 + 0.619 R8 )X Acdh
∂t
R11 = k11 X Acdh
• Mass balance equations → 8
• Kinetic parameter → 37
• Parameter adjust → Genetic Algorithm
X → biomass; Xa → active cell material;
XAcdh → Acetaldehyde dehydrogenase; D → dilution rate;
Ki → rate constant; Ki → affinity constant;
Kji → inhibition constant
27. Some Chemical Products via fermentation
Acetaldeído
Ácido acético
Anidrido acético
FERMENTATION
CHEMICAL SYNTHESIS
Etanol Acetato de etila
Ácido acético Acetato de vinila
Sugar Ácido lático Crotonaldeído
Glycose Acetona Butanol Etanol Paraldeído
Sacarose Butanol
Acetato de butila
Piridina
Nicotinamida
Glicol
Butadieno
Glioxalato
Produtos químicos produzidos por fermentação
28. Other Products to be obtained from biomass
Etileno
Etanol
Acetaldeído
FERMENTATION
Ácido acético
HYDROLYSIS
Propano
Propileno
BIOMASS Sugar
Ácido acrílico
Glicose
Glicerol
Sacarose
Ácido lático
Xilose
Butadieno
Arabinose
Butanodiol
Ácido succínico
Produção de novos produtos químicos a partir de biomassa
29. Fermentation process – piuvirate is formed in glycolysys
GLICOSE
ATP
ADP
Glicose 6-fosfato
Frutose 6-fosfato
ATP
ADP
Frutose 1,6-bifosfato
NAD+ NADH
+Pi +H+
Gliceraldeído 3-fosfato 1,3-Difosfoglicerato
ADP
ATP
3-fosfoglicerato
Gli cos e + 2 NAD + 2 Piruvato + 2 NADH + 2 H +
→
2-fosfoglicerato ∆G10 = −146kJmol − 1
'
Fosfoenolpiruvato
ADP
2 ATP + 2 Pi 2 ATP + 2 H 2 O
→
∆G10 = 61kJmol − 1
ATP
'
PIRUVATO
Processo de glicólise
30. GLICOSE
Rota (EMP)
10 reações sucessivas
2 Piruvato
Condições anaeróbias
Condições anaeróbias
O2
2 Etanol + 2CO2 Condições aeróbias 2 Lactato
CO2
2 Acetil CoA
2 Ácido Acrílico + 2H2O
O2 Ciclo do
ácido TCA
4 CO2 + H2 O
Rota glicolítica
32. STRUCTURED MODEL WITH IMOBILIZED CELLS
Structured Models based on the work of Lei et al. (2001) e Stremel (2001).
Model of Lei et al. (2001) -a structured biochemical model
that describes the aerobic growth of Saccharomyces
cerevisiae in a medium limited to glucose and / or ethanol.
Model of Stremel (2001) -alternative structured model to
represent the dynamic simulation of a tubular bioreactor
with immobilized cells of Saccharomyces cerevisiae for
alcoholic fermentation.
33. Para desenvolvimento deste modelo foi considerado:
Continuous isothermal process
heterogeneous model ;
biomass composition: CH1,82O0,576N0,146;
spherical particles ;
heterofermentative process
production associated with cell growth;
axial dispersion .
Solution by orthogonal collocation
36. Reaction Rates
S S
R1 = k1 X a + k1a Xa
S + K1 S + K1a
S 1
R2 = k 2 Xa
S + K2 L
1+
K
2i
P
R3 = k 3 Xa
P + K3
L
R4 = k 4 Xa
L + K4
L 1
R5 = k 5 1+ K S X a
L + K5
5i
S L 1
R6 = k 6
S +K +
L + K 6a K AA + 1 X a
6 6i
S AA
R7 = k 7
S+K X a + k 7a
X a
7 AA + K 7 a
37. Mass Balances for the solid phase
Glicose
∂S D AS 1 ∂ 2 ∂S
− (R1 + R2 )e A X
− K AA
= r
∂t R 2 r 2 ∂r ∂r
Piruvato
∂P D AP 1 ∂ 2 ∂P
+ (0,978 R1 − R3 )e A X
− K AA
= r
∂t R 2 r 2 ∂r ∂r
Lactato
∂L D AL 1 ∂ 2 ∂L
+ (1,023R3 − R4 − R5 )e A X
− K AA
= r
∂t R 2 r 2 ∂r ∂r
Ácido Acrílico
∂AA D A( AA ) 1 ∂ 2 ∂AA
+ (0,8 R4 − R7 )e A X
− K AA
= r
∂t R 2 r 2 ∂r ∂r
Células
∂X X − K A `AA
= (0,732 R2 + 0,821R5 )X 1 − e − kd X
∂t
X sat
Células ativas
∂X a
= (0,732 R2 + 0,821R5 − R6 − R7 ) − (0,732 R2 + 0,821R5 )X a
∂t
Enzima lactato desidrogenase
∂X LADH
= R6 − (0,732 R2 + 0,821R5 )X LADH
∂t
38. Mass Balance for the Fluid Phase
Glicose
Piruvato
∂S
dt
∂ 2 S ∂S 1 − ε
= Daz − u −
∂z 2 ∂z ε
[
η (R1 + R2 )e − K A AA X ]
∂P ∂ 2 P ∂P 1 − ε
= Daz
∂z
− u +
2 ε
[
η (0,978R1 − R3 )e − K A AA X ]
dt ∂z
Lactato
∂L ∂ 2 L ∂L 1 − ε
= Daz
∂z
− u +
2 ε
[
η (1,023R3 − R4 − R 5 )e − K A AA X ]
dt ∂z
Ácido Acrílico
∂AA
dt
∂ 2 AA ∂AA 1 − ε
= Daz − u
∂z 2 ∂z +
ε
[
η (0,8R4 − R7 )e − K A AA X ]
43. Detailed modeling
where: A = Parallel flow in the baffles holes
B = Flow near the baffle end
C = Parallel flow in the space between bundle of tubes and shell
D = Flow between baflles and shell
E = Cross Flow in the window zones
48. REACTION SYSTEM
AdsorbtionCellulase on cellulose and lignin, β-Glucosidase on lignin
R1Cellulose to Cellobiose (Catalized by cellulase adsorbed on cellulose)
R2Cellulose to Glucose (Catalized by cellulase adsorbed on cellulose)
R3Cellobiose to Glucose (Catalized by non-adsorbed β-Glucosidase)
Enzymes
(Cellulase,
β-glucosidase)
Adsorption R3
R1
R2
49. EXPERIMENTAL DATA AND
MASS BALANCES
G-1% G-3% G-5% G2-1% G2-3% G2-5%
Cellulose
25
Glucose [G] - Cellobiose [G2]
dC
20
= −r1 − r2
dt
15
(g/L)
10 Cellobiose
5 dG2
= 1.056r1 − r3
0 dt
0 12 24 36 48 60 72
Time (h)
Glucose
Fig. 1 Observed time course of glucose (G) and
cellobiose (G2) profiles. Enzymatic hydrolysis
dG
of AHP-pretreated sugarcane bagasse at different = 1.111r2 + 1.053r3
initial solid loadings (% w/w). dt
50. REACTION SCHEMES
Three reaction Scheme
(General)
Two reaction Scheme
(No direct glucose formation
from cellulose)
One reaction Scheme
(Nor direct glucose formation
from cellulose neither
cellobiose accumulation)
51. MATHEMATICAL MODELING
Enzyme adsorption on cellulose and lignin
• One site Langmuir isotherm
Non-mechanistical, fit experimental data,
• Two sites Langmuir Isotherm most used in the literature
Enzyme inhibition by cellobiose and cellulose
• Competitive
• Non-competitive Both are used in the literature. There is no consensus
Recalcitrance
• Substrate reactivity α(S/S0)n+cte (S:substrate)
• Substrate susceptibility v=v0Exp(-Krec(1-(S/S0))) (v0:adsorbed enzyme)
Enzyme deativation (Thermal, mechanical)
• First order kinetic Very important for design of continuous reaction
systems at industrial scale
52. EXPERIMENTAL PROCEDURE AND KINETIC
PARAMETER ESTIMATION
Adsorption Enzyme Loading
• Enzyme adsorption on pretreated substrate 5 FPU – 500 FPU –
• Enzyme adsorption on hydrolyzed substrate CBU/g CBU/g
• Enzyme adsorption on lignin cellulose cellulose
Hydrolysis Substrate Loading
• Hydrolysis of pretreated substrate
1%(W/W) 8%(W/W)
• Hydrolysis of partially hydrolyzed susbtrate
• Hydrolysis with backgrond sugars (Cellobiose, glucose)
• Fed batch (enzyme and susbtrate) hydrolysis
Parameter estimation with global and local optimization techniques
• Genetic algorithms + quasi Newton
• Simulated annealing + quasi Newton
• Particle swarm method + quasi Newton
Model validation
53. CONTINUOUS REACTION SYSTEMS I
Goals
•Subs conc.
CSTR
•Subs conv.
•Enzy consump. •Continuous substrate and
•Power Consump. enzyme feeding
•Resid time n-CSTR
Continuous substrate and
enzyme feeding at the first tank
n-CSTR with distributed feeding
•Ad hoc distributed feeding
strategy of substrate and/or
enzyme
•Model-based distributed
feeding strategy of substrate
and/or enzyme
54. CONTINUOUS REACTION SYSTEMS II
λ λ
Goals PFR with or without side feeding
•Subs conc.
Bafled PFR with or without side feeding
•Subs conv.
•Enzy consump. •Continuous substrate and enzyme feeding
•Power Consump.
•Resid time •Ad hoc side feeding strategy or model-based
•Overcome viscosity side feeding strategy of substrate and/or
limitations enzyme
56. REACTOR MODELING
n-CSTR Microfluid model PFR
VRi S (i −1) − Si dVR dS h
τi = = ϕ
=−
ϕ r ( Si ) r (S h )
n-CSTR Macrofluid model
CFD based model
• Ideal residence time distribution
•Virtual tracer
t n −1
E (t ) = e −t / τ i Experiments
(n − 1)!τ in
•Virtual
• Substrate conversion determination of
t →∞
sh
1 − X sh = ∫ s
E (t )dt RTD
t =0 h 0 Batch •Application of
macrofluid model
57. RESULTS FOR n-CSTR
Macrofluid Model
120
110 NR=1 NR=2 Fig. 2 Total mean hydraulic
residence time (tao=τ) as a
NR=3 NR=5
100
NR=20 PFR
90
80
function of cellulose conver-
sion (Xc) predicted by the
tao[h]
70
60
50
macrofluid and microfluid
40 model.
30
20
10
0,650 0,670 0,690 0,710 0,730 0,750 Microfluid Model
Xc 120
110 N=1 NR=2
NR=3 NR=5
100
NR=20 PFR
90
Initial bagasse concentration 80
tao[h]
70
ST0=50 g/L; 60
initial cellulose concentration 50
40
SC0=40g/L. 30
20
10
0,650 0,670 0,690 0,710 0,730 0,750
Xc
58. CFD APPLIED TO REACTOR DESIGN I
ANSYS CFX (of Ansys Inc., EUROPE)
xy velocity field Modeling approaches
Pseudo-homogeneous
suspension with apparent
rheological properties
‘or’
Multiphase
•Eulerian-Eulerian
approach
•Eulerian-Lagrangian
approach
59. CFD APPLIED TO REACTOR DESIGN II
Baffled PFR
Mesh details and
Pipe geometry
60. CFD APPLIED TO REACTOR DESIGN II
Baffled PFR
2.
1.
2.
1.
Predicted solids volume fraction distribution (1)
and solid velocity (2)
62. The hydrodesulfurization (HDS), hydrodenitrogenation
(HDN), hydrodeoxygenation, hydrocraking and saturative
hydrogenation of middle distillates has been studied in this
work.
An adiabatic diesel hydrotreating trickle bed packed
reactor was simulated numerically by a heterogeneous
model in order to check up the behaviour of this specific
reaction system. Alternative design is proposed
The model consists of mass and heat balance equations for
the fluid phase as well as for the catalyst particles, and take
into account variations in the physical properties as well as
of the heat and mass transfer coefficients. Heterogeneous
model is developed
70. System for Adsorption
Process
Different modelling approach
Different operational
Different numerical Different equilibrium parameters,
methods relationships and adsorbent
characteristics
71. Column parameters:
dimensions
bed porosity
Feed Conditions: Arrangement of the columns: Equilibrium isotherms
single adsorbate fixed Adsorbent type and
binary or multicomponent in sequency characteristics
continuos or pulse simulated moving bed Mass transfer model
73. In the developed software:
1
• different numerical methods
1 different isotherms
•
1
•
were carried out in order to be possible to
take decisions in relation to:
1 the evaluation of an operating adsorber
1 the possibility to apply this separation process for
recovering a given component from a mixture
74. Model and Solution
Simulation of packed bed adsorption columns
using the pore diffusion model, in which two mass
transfer processes were considered:
the external mass transfer from the bulk
liquid phase to the particle surface
internal pore diffusion within the adsorbent
particle itself
75. In the model formulation the following
assumptions were made
• Diffusion coefficients independent of the
mixture composition
• Spherical particles with equal sizes
• Constant temperature and porosity
• Not including axial dispersion
• Solution Procedure: orthogonal
collocation method coupled with the DASSL
routine
78. STATE UNIVERSITY OF CAMPINAS BRAZIL
Department of Chemical Engineering
SOFT SENSOR FOR MONITORING AND CONTROL OF AN
INDUSTRIAL POLYMERIZATION PROCESS
OBJECTIVE:
To develop a Soft Sensor for polymer viscosity of
an industrial PET Process.
81. The variables, related to intrinsic viscosity, used for the
neural net training are given in Table 1.
Table 1- Variables for neural net training
Input variable Name
1 PE temperature T-1
2 SE temperature T-2
3 Temperature of the LP second stage T-4
4 Vacuum of the LP first stage P-1
5 Vacuum of the LP second stage P-2
6 HP temperature T-5
7 HP Vacuum P-3
8 Additive flow rate (catalyst). F-1
Output variable
1 Measured viscosity by viscometer V-1
82. Viscosimeter Soft-Sensor
1,020
1,010
Viscosity
1,000
0,990
0,980
0 4 8 13 17 21 25 29 33 38
Tim e (h)
Figure 4 Viscosimeter versus Soft-Sensor (real time measurements
- normalized values)
83. Polymer viscosity Set-point
1,050
1,025
Viscosity
1,000
0,975
0,950
0 4 8 13 17 21 25
Tim e (h)
Figure 5. Process controlled using viscosity values estimated by Soft-
Sensor (normalized values)
93. •Usual existing processes: 3 or 4 tanks in series
•Alternatives processes are under tests as flocculation and extractive
Extractive alcoholic fermentation process
94. Ff
Vapour
Flash
Tf Pf
Feed
Return
T D pH Tb Fermentor Filter
Purge Permeate
EXTRACTIVE FERMENTATION PLANT
95. Extractive Process
• This process was build up and validated for bioethanol production in
bench scale by Atala (2004);
96. Development of Real-time State Estimators for
Extractive Process - Introduction
- On-line monitoring by SS:
- Allow real time monitoring of key variables of processes;
- Off-line monitoring:
- Leads to time delay between sampling and results;
- Requires advanced analytical instruments (including
near infrared spectrophotometers) → difficult to calibrate
due to presence of CO2 in the media.
97. Software Sensor
• Software sensor: an algorithm where several measurements are
processed together. The interaction of the signals from on-line
instruments can be used for calculating or to estimate new quantities
(e.g. state variables and model parameters) that cannot be measured
in real-time. POTENTIAL INPUT VARIABLES
• On-line measurements (input): Pf Ff Tf T D pH Tb
- Temperatures;
- Dilution rate;
- pH; ANN-BASED ANN-BASED
SOFT-SENSOR (1) SOFT-SENSOR (2)
- Turbidity in the fermentor;
- Pressure;
- Feed flow rate in the flash vessel. ESTIMATED ESTIMATED
Pferm Pflash
• Off-line measurements (output):
ethanol concentration in the fermentor and in the condensed stream from
the flash vessel.
98. ANN Structure Selection
• Multilayer Perceptron (MLP) Neural Networks :
- One of the most common ANN used in engineering;
- understandable architecture and a simple mathematical form;
• This NN consists of: input, output and one or more hidden
layers.
• Numbers of neurons are N, M and K
Input layer Hidden layer Output layer
θ1
θj
w11
+ f1(•)
...
...
...
...
x1 w1N x1 wj1
θ2 β1
w21 W11
x2 wj2 + f(•) yj
...
...
...
...
+ f2(•) W12 + F1(•) g1
...
...
...
...
...
...
...
...
...
...
...
...
xN
w2N
...
W1M
θM
wM1
xN wjN
+ fM(•)
...
...
...
...
wMN
(a) (b)
99. Results and Discussion
250
Pf (mmHg)
200
• Even using on-line (input) data 150
100
with different levels of noise 50
210
→The software sensor described 198
Ff (L/h)
185
accurately the ethanol 173
160
35.5
concentrations. 34.8
Tf ( C)
34.0
o
33.3
32.5
34.5
34.0
T ( C)
33.5
o
33.0
32.5
0.5
0.3
D (h )
-1
0.2
0.1
0.0
4.4
4.3
pH
4.2
4.1
4.0
31
28
Tb (%)
25
22
19
200 250 300 350 400 450
Time (h)
101. '
Kalman filter
training weight
adjustment
Error
Kalman filter
(NLSTC)
+
-
RNN
N
Substrate
Air flow State measurement
Penicillin
process
The proposed non-linear Self-tuning controller scheme
102. 35
30
Biomass concentration (g/l)
25
20
Process
15 Kalman filter
10
5
0 20 40 60 80 100 120
Time (h)
Estimation of the biomass concentration
103. 14000
Penicillin concentration (g/l) 12000
10000
8000
6000
4000
2000
Process
0 Kalman filter
-2000
0 20 40 60 80 100 120
Time (h)
Estimation of the Penicillin concentration with the multiple extended
Kalman filter algorithm
104. Fractional Brownian motion as a model
for an industrial Air-lift Reactor
fBm (Mandelbrot,
1968) BH(t+τ)-BH(t) é
estatisticamente igual
ao [BH(t+τr)-BH(t)]/rH
fGn: definido como
derivado do fBm:
fGn = BH(t+1)-BH(t)
105. Comparação entre o sinal de
pressão e o ruído Gaussiano
fracionário (fGn)
3.32 4
3.3 3
2
3.28
1
3.26
0
3.24
-1
3.22
-2
3.2 -3
3.18 -4
0 500 1000 1500 2000 25000 500 1000 1500 2000 2500
Industrial Air-Lift Reactor Data Fractional Brownian Model
with H = 0.7
106. Synthesis of a fuzzy model for linking
synthesis conditions with molecular
characteristics and performance properties
of high density polyethylene
107. Cognitive Dynamic Model
y(k)- prediction by linear equation – Takage Sugeno
approach:
y(k) = w0i + w1iu1(k-τu1) + w2iu1(k-τu1 -1) +...+ wp1iu1(k-τu1-p1)+
w(p1+1)iu2(k-τu2) + w(p1+2)iu2(k-τu2 -1) +...+ w(p1+p2)1iu2(k-τu2-p2)+
w(p1+p2 +1)iy(k-1) + w(p1+p2 +2)iy(k-2) +...+w(p1+p2 +m)iy(k-m).
together with cognitive information
108. Implementations
• Du PONT Polymerization Process
• Rhodia Nylon-6,6 Process
• High Non Linear Process – large scale plant
Deterministic model – difficult to assembly
109. Copolymer molar fraction
0,75
PLANTA
MODELO
0,70
0,65
0,60
Yap
0,55
0,50
0,45
0,40
-200 0 200 400 600 800 1000 1200 1400 1600 1800
tempo (h)
Teste para a fração molar do copolímero
110. Polymer Molecular Weight
PLANTA
37000 MODELO
36000
Mpw (kg/kmol)
35000
34000
33000
0 200 400 600 800 1000
tempo (h)
Validação para o peso molecular do copolímero
111. Nylon-66 Molecular weight
38000
par de dados da planta e do modelo
37000
Mpw (kg/kmol) - modelo
36000
35000
34000
33000
33000 34000 35000 36000 37000 38000
Mpw (kg/kmol) - planta
113. Condição 1 2 3 4 5 6
Ordem das entradas 23 17 7 23 17 7
Ordem do estado interno 1 1 1 2 2 2
Regras 7 7 5 7 7 5
Fator erro indexado (J) 1.19e-3 1.2 e-3 1.22 e-3 1.17 e-4 1.19 e-4 1.21 e-3
1,05 1,05
Temperatura adimensional dos reagentes
J = 1,21E-3
Temperatura adimensional dos reagentes
Modelo determinístico J = 1.2E-3 Modelo determinístico
Modelo Cognitivo
Modelo Fuzzy Modelo Cognitivo
Modelo Fuzzy
1,00 1,00
0,95 0,95
0,90 0,90
0,85 0,85
0,80 0,80
100 200 300 400 500 600 700 100 200 300 400 500 600 700
Tempo Tempo
Ordem 7 para a entrada e Ordem 17 para a entrada e 1 para
1 para estado interno estado interno
114. 1,05
Modelo determinístico
Temperatura adimensional dos reagentes
Modelo Cognitivo
J = 1,19E-3 Modelo Fuzzy
1,00
0,95
0,90
0,85
0,80
100 200 300 400 500 600 700
Tempo
Ordem 23 para a entrada e 1 para estado interno
115. Properties Correlations
Molecular
Crystallinity Weight molecular Weight Density Melt index
distribution
Correlation
Fuzzy model
Mecanical Thermic Tensile Reologic
Properties Properties Properties properties
116. Properties Product modelling from
operationals dates throght Fuzzy Logic
Output variables
control in deterministic Performance
model properties
Thermic
properties
Product Density
Fuzzy Fuzzy
Model Model
Rheologic
Properties
MI
Mechanical
Properties
Weight
molecular
Tensile
Properties
Plant
Fuzzy model
117. Properties Product modelling from
operationals dates throght Fuzzy Logic
Monomer Performance Properties
Co-monomer
Stifness
CAT Impact Strength
CO-CAT Conversion Hardness
Rate Melt Strength
Solvent Produc
production
t Stress Crack
H2 Mn Resistance
PFR PFR - trimer Mw Tensile Strength
T PFR Density Tm
CSTR Fuzzy
T CSTR Pd Tc
Model -
P system MI Tg
type C
SE crystallization percent
Feed Lateral Process melt swell
softening Point
Fuzzy Model - type A
Fuzzy Model - type B
118. Results – Fuzzy model type A
Type A. Such model considers the linking of the property of flow stress exponent (SE)
versus the variables of the synthesis process. The SE of a polymer is a measure of
melt viscosity and is a direct measure of molecular weight distribution. The Stress
Exponent, determined by measuring the flow (expressed as weight, in grams) through a melt index
approaches (ASTM D 1238).
120. UFBA
Optimization Based Polymer
Resin Development
121. Introduction
Output Conditions
Input Conditions Temperature
Concentration 0.60
Temperature
0.50
Polymerization Conversion
SE (dim.)
0.40
0.30
Concentrations process model 0.20
0.10
Flow Rate Polymer
0.00 0.20 0.40 0.60 0.80
Reactor Length (dim.)
1.00
Properties
Improve Quality
Optimization Design of new products
model
Goal: Determine optimal operating policies in order
to produce pre-specified polymer resins
122. Braskem Ethylene continuous polymerization
in solution with Ziegler-Natta catalyst-
Industrial Plant
Stirred Configuration
Ethylene
Tubular Configuration
Hydrogen
Product
Solvent PFR2
Ethylene Ethylene
Hydrogen Hydrogen
Solvent Solvent
PFR1
CSTR
H2
CAT CC CAT CC
123. Mathematical Model
Stirred Configuration
Product W1 W r-1 Wr W R-1
PFR2 W0
CSTR1 .... CSTRr .... CSTRR
B2 Br Br+1 BR
Monomer FZ 1 FZ r FZ R
H2
Solvent
CSTR WR W out
PFRJ+1
CAT CC
Tubular Configuration
W1 Wj WJ
PFR1 PFRj .... PFRJ
Product
Monomer PFR2
Fj FJ
H2
Solvent
PFR1 W1 W r-1 Wr W R-1
WP
CSTR1 .... CSTRr .... CSTRR
CSTR B2 Br Br+1 BR
H2
CAT CC WR W out
PFR J+1
124. Polymer Specification
Melt Index (MI): MI = α ⋅ (MW )
β
w
•
1
SE =
• Stress Exponent (SE): α + γ ⋅ exp(β ⋅ PD )
DS = α + β ⋅ log(MI ) + γ ⋅ SE
• Density (DS): Embiruçu et al. (2000)
• Specification at the end of reaction (z=zf)
Desired polymer properties
end-point constraints of the optimization
125. Objective Function
• Different operating policies can yield the same
resin
Maximize Profit
• Objective Function
Φ = a ⋅ WPE − (bM ⋅ WM + bH ⋅ WH + bCAT ⋅ WCAT + bCC ⋅ WCC + bS ⋅ WS ) €/h
where
a: polyethylene sales price (€/kg)
b: reactant costs (€/kg)
W: mass flow rates (kg/h)
126. Decision Variables
Stirred Configuration Tubular Configuration
M PFR2
Ws PFR2
H2,0 Tin
M Tin Pin
Wt
H2,0 Pin
PFR1
Wt
CSTR
CSTR
H2,j
CAT CC
CAT CC
• Side Feed (Ws) • Monomer Input Concentration (M) • Lateral Hydrogen injection point (j)
• Hydrogen Input Concentration (H2,0) • Lateral Hydrogen Concentration (H2,j)
• Catalyst Input Concentration (CAT)
• Inlet Temperature (Tin)
• Inlet Pressure (Pin)
• Total Solution Rate (Wt)
127. Multi-stage Systems
• Discontinuities ⇒ new stage system
DAE
Event Event Event f k (x k , x k , y k , u k , p, z ) = 0 , z ∈ [z k -1 , z k ]
k = 1,...,nk
g k ( x k , y k , u k , p, z ) = 0
f(1) f(2) f (n k )
x 0 ( z0 ) − x 0 = 0
( n k −1)
z(0) z(1) z(2) z z (n k ) = z (f)
Stage Transition
J (j k ) (x ( k ) , x ( k ) , y ( k ) , u ( k ) , x ( j ) , x ( j ) , y ( j ) , u ( j ) , p, z ) = 0
• Examples
– Injection of mass along a tubular reactor
– Reactor switch
129. Multi-stage Process
Steady-state DAE (axial coordinate)
Analogy: axial coordinate ⇔ time
Tubular configuration Stirred configuration
f 4 ( z )f 4 ( z 4 (fz4)( z )
f)
PFR2 PFR2
g3
g3
f1 ( zf1 ( f1)(f 2)( z ) ( z )
) z z f2
g3
PFR1 g3
CSTR
CSTR
H2
CAT CC CAT CC
z
f1 ( z ) f2 ( z) g3 f4 ( z)
f k (z ) : differential equation
g k : algebraic equation
k : stage number
z : axial coordinate
Dynamic Optimization Techniques for multi-stage systems
130. Results – Stirred Configuration
0.20 1.0
0.8
Concentration (dim.)
H 2,0
H2,0
CAT
0.15
Profit (dim.)
0.6 Ws
Ws
M
0.4
0.10
0.2
0.05 0.0
0.240 0.260 0.280 0.300 0.320 0.240 0.260 0.280 0.300 0.320
SE (dim.) SE (dim.)
0.80
0.80
Revenue, Cost (dim.)
0.75
Q, WPE(dim.)
0.70
0.70
Q Revenue
0.60
W PE
WPE 0.65 Cost
0.50 0.60
0.40 0.55
0.24 0.26 0.28 0.30 0.32 0.24 0.26 0.28 0.30 0.32
SE (dim.) SE (dim.)
131. Results – Tubular Configuration
One H2 injection point at a pre-specified length (4 stages)
0.20 1.0
Concentration (dim.)
0.8
0.15
Profit (dim.)
0.6 H 2,0
H 2,0
CAT
H 2,j
2,j
0.4
M
0.10
0.2
0.05 0.0
0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75
SE (dim.) SE (dim.)
132. Benefits of the developed tool
Development of a potential tool able to improve
the polymer quality or to create new resins in a
simple and quick manner.
– Better customer satisfaction.
• Robust approach
– Use of Dynamic Optimization algorithms for a
stationary multi-stage process.
• Versatile tool, since other polymerization
processes can be used as basis.
135. RESULTS
700 Through CENPES/PETROBRAS Molecular
Distillation of
Through Molecular Distillation
600
500
the Alfa
Temperature (oC)
400
petroleum
obteined 10 %
300
200
100
of distillate
0 acumullated
0 20 40 60 80 100
% Distillate accumulated (% w)
The distillation curve was determined from the
temperature and the percentage of distillate
obtained experimentally through molecular
distillation and using ASTM D1160.
137. Product Industrial data (ton/day) Simullation Result (ton/day) Error (%)
Fuel Gas 360.0 360.4 0.11
LPG 1167.0 1191.4 2.09
Gasoline 3534.0 3436.2 2.77
LCO 667.0 677.0 1.50
Slurry 1107.0 1067.5 3.57
Products recovery: industrial data and simulation results.
138. Green Ethyl Acrylate
SUBSTRATOS
Glicose Vários
Lactose Sacarose
C5 e C6
O 1 2
- O Fermentação
H3C C O -
H3C C O 1) Fermentação de ácido
HC Láctico (ex. Lactobacilli,
HC 3 O
+ - Bacilli Streptokokki).
NH3 H3C C O
OH 2) Fermentação de ácido
L-Alanina Lactato CH2 Propiónico.
Propianato 3) Redução Direta (ex.
Clostridium propionicum).
4 5 6 4) Desidratação
5) Conversão Química
6)Caminho Oxidativo (ex.
O Pseudomonas aeroginosa)
-
H2C C O
CH
Ácido Acrílico
139. ETHANOL R EC 3
R EC 2
D ISTIL 1
STRIPPER EXTR ACT R AF
AC ID
TOPO AC R YLATE
C OOLER
FEED
C OOL
R EAC TOR
D ISTIL 2
R EC 1 EXT
WASTE
WATER
Conceptual Plant design for Green Ethyl Acrylate
140. Reactor Mathematical Model
Equações adimensionalizadas
Balanço de Massa para o Ácido Acrílico
∂G ∂G ∂ 2G
= B1 .
∂u + u. ∂u 2 + B2 .rA
∂z ad
Balanço de Energia no Tubo
∂θ l ∂θ l ∂ 2θ l
= B3 .
∂u + u. ∂u 2 + B4 .rA
∂z ad
Balanço de Energia do Fluido Térmico
= B5 .(θ NT − θ F )
dQ
dz ad
Queda de Pressão
dP ad
= B7
dz ad
Solução por Colocação Ortogonal
141. Reactor simulation
Conversion for several temperatures Tubular reactor 5,0 meters long
0,7
0,6
0,5
Conversão
0,4
0,3
0,2 Conversão @ 75 C
Conversão @ 80 C
0,1 Conversão @ 85 C
0,0
0,0 0,2 0,4 0,6 0,8 1,0
Coordenada Axial
142. Conceptual Plant design for Green Ethyl Acrylate
ETHANOL R EC 3
R EC 2
D IS TIL 1
STRIPP ER EXTR ACT R AF
AC ID
TOPO AC R YLATE
C OOLE R
FE ED
C OOL
R EAC TOR
D IS TIL 2
R EC 1 EXT
WAS TE
WATER
FEE REC1 TOP COO WAT RAF EXT ACRYL REC2 WAST REC3
Vazão(kmol/h)
D L E
Ácido Acríl 20,82 20,82 0,00 0,00 0,00 0,00 0,00 0,00 0,0000 0,0000 0,0000
0,00 20,82 20,82 0,00 0,00 20,82 0,00 0,0000 0,3790 20,440
Etanol 20,82
9
0,00 29,18 29,18 20,00 4,59 44,58 0,00 4,5999 36,380 8,1995
Água 29,18
4
Acril de Etila 29,18 0,18 29,00 29,00 0,00 22,64 6,36 19,74 2,9003 0,00 6,3595
Total (kmol/h) 100,0 21,00 79,00 79,00 20,00 27,23 71,70 19,74 7,5000 36,76 35,00
1.518 4.388 4.388 360 2349 2.399 1976 373,53 672,91 1726,1
Total (Kg/h) 5.907
1
Temp. (ºC) 78,0 140,5 79,0 25,00 25,0 29,9 29,3 99,4 82,42 97,10 77,71
143. Green Acrylic Acid – Unicamp/CTC/Braskem
Seleção das rotas Cana –
metabólicas fonte de açúcar
Seleção de microrganismos Otimização do meio de cultura
Fermentação
Ácido Láctico
Separação/purificação
Cinética
Ácido desidratação redução Ácido
Processo Propiônico
Acrílico
Cinéticas Modelagem Otimização dos Controle dos processos
processos