Optimal control

Industrial applications
Flavio Manenti, Dept. CMIC Giulio Natta; Politecnico di Milano
Filip Logist, Jan Van Impe, Dept. of Chemical Engineering, KU
Leuven, Universityof Leuven
Flavio Manenti, Filip Logist KU-Leuven
2
Terms and conditions
Acronyms and notations
Advanced conventional process control
Advanced process control Model predictive control
Model predictive control
Based on linear/linearized models
Dynamic matrix control (DMC, LMPC, MPC)
Several commercial packages
Based on nonlinear models
Model predictive control (MPC, NMPC)
No commercial packages
Features of NMPC
A dynamic (convolution) model is used to foresee the future behavior of
the plant on a specific time horizon (prediction horizon, H_P) consisting
of p sampling times
Receding horizon methodology (moving horizon, not rolling horizon)
Manenti, Considerations on Nonlinear Model Predictive Control Techniques, COMPUTERS & CHEMICAL
ENGINEERING, 35(11), 2491-2509, 2011.
Flavio Manenti, Filip Logist KU-Leuven
3
Integration Pyramid
Plant
Management
Maintenance
and Production
Management
Enterprise
Management
Field
Conventional
Control
Advanced
Control
(MPC/NMPC)
Real Time
Dynamic
Optimization
Scheduling
Planning
Seconds
Minutes
Hours - Days
Weeks
Months - Years
| |
1 1
2 2
2
1
min ( ) ( ) ( ) ( 1)
p p p
k h k h k h
SET TAR
y react react T c c u c c
j k l k i k
T j T F l F F i F i e e e
+ + +
= + = =
( ( u = + +


Advanced
Control
(MPC/NMPC)
Real Time
Dynamic
Optimization
Flavio Manenti, Filip Logist KU-Leuven
4
Algorithm
Model Predictive Control MPC
Plant
Flavio Manenti, Filip Logist KU-Leuven
5
Receding horizon methodology
t
u
1,1
u
2,1
u
3,1
u
n,1
Set-point
Manipulated
Variable
PLANT
MODEL
Controlled Variable
u
1
H
P
H
C
u
2
u
3
H
D
1
H
P
H
C
u
1
u
2
u
3
2
H
P
H
C
u
1
u
2
u
3
3
Flavio Manenti, Filip Logist KU-Leuven
6
Industrial perspective
Outlier Detection
Robust methods
Linear/nonlinear Regressions
Performance Monitoring
Yield Accounting
Soft sensing
Data
Reconciliation
Mathematical
Modeling
Dynamic
Simulation
Model
Predictive
Control
Optimization
Model
Reduction
DCS, OTS, Plantwide control,
Soft sensing, process transients,
grade/load changes
Solvers
Planning
Scheduling
Dynamic optimization
Distributed predictive control
Nonlinear Systems
Optimizers
Differential systems
Stiff systems
ODE,DAE,PDE,PDAE
Efficiency
Decisions Raw Data
Parallel
Computing
Uncertainties
Optimal production
Optimal grade changes
Multi-objective
Real-time optimization
High accuracy
Reliable process control
Production improvement
Economy
Just in time
Market-driven
Logistics
Corporate
Supply Chain
Flavio Manenti, Filip Logist KU-Leuven
NMPC
PET plant
7
Manenti, Rovaglio
Integrated multilevel optimization in large-
scale poly(ethylene terephthalate) plants
Industrial & Engineering Chemistry Research
47(1), 92-104, 2008.
Flavio Manenti, Filip Logist KU-Leuven
8
Case Study: PET Plant
Flavio Manenti, Filip Logist KU-Leuven
9
Jacobian Matrix
Primary esterifier
Secondary esterifier
Low polymerizer
Intermediate polymerizer
High polymerizer
Solid state polymerizer
Resulting DAE:
1356 diff. eqs.
164 alg. eqs.
15 controls
2 controlled
16 constrained
Flavio Manenti, Filip Logist KU-Leuven
10
Frequent Grade Changes
Grade A: PET as textile fibres (melt process
I.V. = 0.55 0.65 dl/g)
Grade B: PET for bottles production (bottle
grade I.V. = 0.72 0.85 dl/g)
Grade C: PET for special fibres (tire-cord
resins I.V. = 0.95 1.05 dl/g)
Flavio Manenti, Filip Logist KU-Leuven
11
Comparison
0.440
0.445
0.450
0.455
0.460
0.465
0.470
0 500 1000 1500 2000 2500 3000
I
V

I
P

(
d
l
/
g
)
Time (min)
0.600
0.610
0.620
0.630
0.640
0.650
0.660
0.670
0.680
0 500 1000 1500 2000 2500 3000
I
V

H
P

(
d
l
/
g
)
Time (min)
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
0 500 1000 1500 2000 2500 3000
P

H
P

(
m
m
H
g
)
Time (min)
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00
0 500 1000 1500 2000 2500 3000
P

I
P

(
m
m
H
g
)
Time (min)
NMPC
NMPC
Flavio Manenti, Filip Logist KU-Leuven
12
Comparison
0.640
0.650
0.660
0.670
0.680
0.690
0.700
0.710
0 500 1000 1500 2000 2500 3000
I
V

P
H
C
R

(
d
l
/
g
)
0.760
0.770
0.780
0.790
0.800
0.810
0.820
0.830
0.840
0 500 1000 1500 2000 2500 3000
I
V

S
S
P

(
d
l
/
g
)
NMPC
NMPC
Time (min)
Time (min)
Flavio Manenti, Filip Logist KU-Leuven
NMPC
Ethylene splitter
Eni, Italy
13
Flavio Manenti, Filip Logist KU-Leuven
14
The C2-splitter
Column design:
Total tray number : 110
Feed : tray #55
(*)
Ethylene cut : tray #104
(*)
Feed composition
(**)
:
C
2
H
4
79%
C
2
H
6
19%
Others 2% (H
2
, CO,
CO
2
, CH
4
, C
3
H
8
, C
3
H
6
)
(*) bottom-up numeration
(**) molar basis
Flavio Manenti, Filip Logist KU-Leuven
15
Validation
Reflux flowrate change effect on overhead
composition and on temperature at tray #5
0,0
100,0
200,0
300,0
400,0
500,0
600,0
700,0
800,0
900,0
1000,0
89,40 89,60 89,80 90,00 90,20 90,40 90,60
O
v
e
r
h
e
a
d

i
m
p
u
r
i
t
i
e
s

[
p
p
m
]
Refluxflowrate[ton/h]
Refluxflowratechangeeffectsonoverhead
streamimpurities
Simulation
Plant
41,00
40,00
39,00
38,00
37,00
36,00
35,00
89,60 89,80 90,00 90,20 90,40 90,60
T
r
a
y

t
e
m
p
e
r
a
t
u
r
e

C
]
Refluxflowrate[ton/h]
Refluxflowratechangeeffectsontemperatureat
tray#5
Simulation
Plant
0,0
200,0
400,0
600,0
800,0
1000,0
1200,0
96,60 96,80 97,00 97,20 97,40 97,60
O
v
e
r
h
e
a
d

i
m
p
u
r
i
t
i
e
s

[
p
p
m
]
Boilupflowrate[ton/h]
Boilupflowratechangeeffectsonoverhead
streamimpurities
Smulation
Plant
41,00
40,00
39,00
38,00
37,00
36,00
35,00
34,00
96,60 96,80 97,00 97,20 97,40 97,60
T
r
a
y

t
e
m
p
e
r
a
t
u
r
e

C
]
Boilupflowrate[ton/h]
Boilupflowratechangeeffectsontemperatureat
tray#5
Simulation
Plant
Boil-up flowrate change effect on overhead
composition and on temperature at tray #5
Flavio Manenti, Filip Logist KU-Leuven
16
Servo-mechanism problem
9000
9200
9400
9600
9800
10000
10200
10400
10600
0,0 5,0 10,0 15,0 20,0 25,0 30,0
R
e
f
l
u
x

f
l
o
w

r
a
t
e

[
l
b
m
o
l
/
h
]
Time[h]
Refluxflowrate
PI
DMC
NMPC
0,9600
0,9650
0,9700
0,9750
0,9800
0,9850
0,9900
0,0 5,0 10,0 15,0 20,0 25,0 30,0
E
t
h
y
l
e
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethylenemolarfractionincutstream
PI
DMC
NMPC
SPDistillate
0,8600
0,8700
0,8800
0,8900
0,9000
0,9100
0,9200
0,9300
0,9400
0,9500
0,0 5,0 10,0 15,0 20,0 25,0 30,0
E
t
h
a
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethanemolarfractioninbottomstream
PI
DMC
NMPC
SPBottom
43,20
43,40
43,60
43,80
44,00
44,20
44,40
44,60
44,80
45,00
0,0 10,0 20,0 30,0 40,0 50,0 60,0
R
e
b
o
i
l
e
r

t
h
e
r
m
a
l

d
u
t
y

[
1
.
E
+
0
6

B
T
U
/
h
]
Time[h]
Reboilerthermalduty
PI
DMC
NMPC
Flavio Manenti, Filip Logist KU-Leuven
17
Servo-mechanism problem
0,8600
0,8700
0,8800
0,8900
0,9000
0,9100
0,9200
0,9300
0,9400
0,9500
0,0 5,0 10,0 15,0 20,0 25,0 30,0
E
t
h
a
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethanemolarfractioninbottomstream
PI
DMC
NMPC
SPBottom
39,50
40,00
40,50
41,00
41,50
42,00
42,50
43,00
43,50
44,00
44,50
45,00
0,0 5,0 10,0 15,0 20,0 25,0 30,0
R
e
b
o
i
l
e
r

t
h
e
r
m
a
l

d
u
t
y

[
1
.
E
+
0
6

B
T
U
/
h
]
Time[h]
Reboilerthermalduty
PI
DMC
NMPC
9000
9200
9400
9600
9800
10000
10200
10400
10600
0,0 5,0 10,0 15,0 20,0 25,0 30,0
R
e
f
l
u
x

f
l
o
w

r
a
t
e

[
l
b
m
o
l
/
h
]
Time[h]
Refluxflowrate
PI
DMC
NMPC
0,9600
0,9650
0,9700
0,9750
0,9800
0,9850
0,9900
0,0 5,0 10,0 15,0 20,0 25,0 30,0
E
t
h
y
l
e
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethylenemolarfractionincutstream
PI
DMC
NMPC
SPDistillate
0,9800
0,9850
0,9200
0,9250
Flavio Manenti, Filip Logist KU-Leuven
18
Regulation problem (feed
composition disturbance)
0,9890
0,9895
0,9900
0,9905
0,9910
0,9915
0,9920
0,0 10,0 20,0 30,0 40,0 50,0 60,0
E
t
h
y
l
e
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethylenemolarfractionincutstream
PI
DMC
NMPC
SPCut
0,8800
0,8900
0,9000
0,9100
0,9200
0,9300
0,9400
0,9500
0,9600
0,9700
0,0 10,0 20,0 30,0 40,0 50,0 60,0
E
t
h
a
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethanemolarfractioninbottomstream
PI
DMC
NMPC
SPBottom
9950,00
10000,00
10050,00
10100,00
10150,00
10200,00
10250,00
10300,00
10350,00
0,0 10,0 20,0 30,0 40,0 50,0 60,0
R
e
f
l
u
x

f
l
o
w

r
a
t
e

[
l
b
m
o
l
/
h
]
Time[h]
Refluxflowrate
PI
DMC
NMPC
43,20
43,40
43,60
43,80
44,00
44,20
44,40
44,60
44,80
45,00
0,0 10,0 20,0 30,0 40,0 50,0 60,0
R
e
b
o
i
l
e
r

t
h
e
r
m
a
l

d
u
t
y

[
B
T
U
/
h
]
Time[h]
Reboilerthermalduty
PI
DMC
NMPC
Flavio Manenti, Filip Logist KU-Leuven
19
Regulation problem (feed
composition disturbance)
0,9890
0,9895
0,9900
0,9905
0,9910
0,9915
0,9920
0,0 10,0 20,0 30,0 40,0 50,0 60,0
E
t
h
y
l
e
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethylenemolarfractionincutstream
PI
DMC
NMPC
SPCut
0,8800
0,8900
0,9000
0,9100
0,9200
0,9300
0,9400
0,9500
0,9600
0,9700
0,0 10,0 20,0 30,0 40,0 50,0 60,0
E
t
h
a
n
e

m
o
l
a
r

f
r
a
c
t
i
o
n

]
Time[h]
Ethanemolarfractioninbottomstream
PI
DMC
NMPC
SPBottom
9950,00
10000,00
10050,00
10100,00
10150,00
10200,00
10250,00
10300,00
10350,00
0,0 10,0 20,0 30,0 40,0 50,0 60,0
R
e
f
l
u
x

f
l
o
w

r
a
t
e

[
l
b
m
o
l
/
h
]
Time[h]
Refluxflowrate
PI
DMC
NMPC
43,20
43,40
43,60
43,80
44,00
44,20
44,40
44,60
44,80
45,00
0,0 10,0 20,0 30,0 40,0 50,0 60,0
R
e
b
o
i
l
e
r

t
h
e
r
m
a
l

d
u
t
y

[
B
T
U
/
h
]
Time[h]
Reboilerthermalduty
PI
DMC
NMPC
0,9900
0,9905
0,9500
0,9550
0,9500
0,9550
Flavio Manenti, Filip Logist KU-Leuven
Self-adaptive NMPC
Deisobutanizer
Mongstad Refinery, Norway
20
Dones, Manenti, Preisig, Buzzi-Ferraris
Nonlinear Model Predictive Control: a Self-
Adaptive Approach
Industrial & Engineering Chemistry Research
49(10), 4782-4791, 2010
Flavio Manenti, Filip Logist KU-Leuven
Unit
21
Flavio Manenti, Filip Logist KU-Leuven
Spoiled Jacobian
22
Flavio Manenti, Filip Logist KU-Leuven
Results
Use of compartmental models
Model self-adaptation
To the problem
To the dynamic
To the computational effort
Benefits
Accurate when needed
Fast when possible
(viceversa)
23
93
93.5
94
94.5
95
95.5
0 500 1000 1500 2000 2500 3000
c
o
n
c
e
n
t
r
a
t
i
o
n

i
C
4

t
o
p

[
%
]
time [s]
NMPC with full dynamic model
NMPC with 5-dynamic-trays model
ANMPC
360
365
370
375
380
385
390
395
400
0 500 1000 1500 2000 2500 3000
r
e
f
l
u
x

s
t
r
e
a
m

[
m
o
l
/
s
]
time [s]
NMPC with full dynamic model
NMPC with 5-dynamic-trays model
ANMPC
4
6
8
10
12
14
16
0 500 1000 1500 2000 2500 3000
n
u
m
b
e
r

o
f

d
y
n
a
m
i
c

t
r
a
y
s

u
s
e
d

i
n

t
h
e

m
o
d
e
l
time [s]
Flavio Manenti, Filip Logist KU-Leuven
D-RTO
Olefins plant
Invensys, USA
24
Manenti et al.
Process Dynamic Optimization Using ROMeo
Computer Aided Chemical Engineering
29, 452-456, 2011
Flavio Manenti, Filip Logist KU-Leuven
25
Industrial perspective
Outlier Detection
Robust methods
Linear/nonlinear Regressions
Performance Monitoring
Yield Accounting
Soft sensing
Data
Reconciliation
Mathematical
Modeling
Dynamic
Simulation
Model
Predictive
Control
Optimization
Model
Reduction
DCS, OTS, Plantwide control,
Soft sensing, process transients,
grade/load changes
Solvers
Planning
Scheduling
Dynamic optimization
Distributed predictive control
Nonlinear Systems
Optimizers
Differential systems
Stiff systems
ODE,DAE,PDE,PDAE
Efficiency
Decisions Raw Data
Parallel
Computing
Uncertainties
Optimal production
Optimal grade changes
Multi-objective
Real-time optimization
High accuracy
Reliable process control
Production improvement
Economy
Just in time
Market-driven
Logistics
Corporate
Supply Chain
Dynamic Optimization
Flavio Manenti, Filip Logist KU-Leuven
26
Tools
Outlier Detection
Robust methods
Linear/nonlinear Regressions
Performance Monitoring
Yield Accounting
Soft sensing
Data
Reconciliation
MATHEMATICAL
MODELING
Dynamic
Simulation
Dynamic
Optimization
Optimization
Model
Reduction
DCS, OTS, Plantwide control,
Soft sensing, process transients,
grade/load changes
Solvers
Enterprise-wide
Planning
Scheduling
Nonlinear Systems
Optimizers
Differential systems
Stiff systems
ODE,DAE,PDE,PDAE
Efficiency
Decisions Raw Data
Parallel
Computing
Supply Chain
Management
Uncertainties
Optimal production
Optimal grade changes
Multi-objective
Real-time optimization
High accuracy
Reliable process control
Production improvement
Just in time
Market-driven
Conscious MGM
Mathematical
Modeling
Dynamic
Simulation
Optimization
DYNSIM
ROMeo
Flavio Manenti, Filip Logist KU-Leuven
27
Tools
Outlier Detection
Robust methods
Linear/nonlinear Regressions
Performance Monitoring
Yield Accounting
Soft sensing
Data
Reconciliation
MATHEMATICAL
MODELING
Dynamic
Simulation
Dynamic
Optimization
Optimization
Model
Reduction
DCS, OTS, Plantwide control,
Soft sensing, process transients,
grade/load changes
Solvers
Enterprise-wide
Planning
Scheduling
Nonlinear Systems
Optimizers
Differential systems
Stiff systems
ODE,DAE,PDE,PDAE
Efficiency
Decisions Raw Data
Parallel
Computing
Supply Chain
Management
Uncertainties
Optimal production
Optimal grade changes
Multi-objective
Real-time optimization
High accuracy
Reliable process control
Production improvement
Just in time
Market-driven
Conscious MGM
Is it possible?
DYNSIM
Mathematical
Modeling
Dynamic
Simulation
Dynamic
Optimization
Optimization
ROMeo
Flavio Manenti, Filip Logist KU-Leuven
28
The Idea
On-line Optimization
Nonlinear MPC
Data Reconciliation
( ) ( )
( ) ( )
T
min
. .: 0; 0
R
M R M R
i i i i
i
x x x x
s t f g

= s

x
W
x x
( )
( )
,
min
. .: , 0
, 0
;
n m
Z Profits Costs
s t f
g
R N
=
=
s
e e
x b
x b
x b
x b
( )
( ) ( )
( ) ( )
1 2
, ,
min ... ...
. .: 0; , 0
0; , 0
; ;
n
n p m
Z
s t f f
g g
R R N
=
= =
s s
e e e

x u b
x x x
x x x
x u b

Dynamic Optimization
( ) ( )
( )
( )
T
min
. .: , 0
, 0
R SET R SET
i i i i
i
x x x x
s t f
g

=
s

u
W
x x
x x

Flavio Manenti, Filip Logist KU-Leuven

29
Example
Series of three ideal CSTRs
Open-loop
Closed-loop
P-7
P-16
P-21
Flavio Manenti, Filip Logist KU-Leuven
30
Open-loop in C++
Key-component molar flow
exiting the reactor:
#1
#2
#3
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 1 2 3 4 5
[
m
o
l
/
s
]
Time
Flavio Manenti, Filip Logist KU-Leuven
31
Open-loop in DYNSIM
UAM MODELS inserted
into the ICON PALETTE
(C++ dynamic library)
Flavio Manenti, Filip Logist KU-Leuven
32
Open-loop in DYNSIM
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 1 2 3 4 5
[
m
o
l
/
s
]
Time
Flavio Manenti, Filip Logist KU-Leuven
33
Using BzzMath in DYNSIM
No changes at the
DYNSIMs interface
Flavio Manenti, Filip Logist KU-Leuven
34
Closed-loop in C++
0.398
0.4
0.402
0.404
0.406
0.408
0.41
0.412
0.414
0.416
0.418
0 10 20 30 40 50
[
m
o
l
/
s
]
Time
0.198
0.199
0.2
0.201
0.202
0.203
0.204
0.205
0 10 20 30 40 50
[
m
o
l
/
s
]
Time
0.0982
0.0984
0.0986
0.0988
0.099
0.0992
0.0994
0.0996
0.0998
0.1
0.1002
0 10 20 30 40 50
[
m
o
l
/
s
]
Time
CV
(SP: 0.1 mol/s)
Flavio Manenti, Filip Logist KU-Leuven
35
Full Integration (All-in-one)
Flavio Manenti, Filip Logist KU-Leuven
36
Drag & Drop
DYNSIM
Flavio Manenti, Filip Logist KU-Leuven
37
Drag & Drop
DYNSIM
Flavio Manenti, Filip Logist KU-Leuven
38
Drag & Drop
ROMeo
Flavio Manenti, Filip Logist KU-Leuven
39
Drag & Drop
ROMeo
Flavio Manenti, Filip Logist KU-Leuven
40
Smart Dynamic Simulation with
ROMeo
Flavio Manenti, Filip Logist KU-Leuven
41
D-RTO with Multiple Shooting
Flavio Manenti, Filip Logist KU-Leuven
42
Friendly Interface for D-RTO
Possibility to give the user to enter
any kind of data for D-RTO
Specific
D-RTO
TAB
Flavio Manenti, Filip Logist KU-Leuven
43
Preliminary Comparison
RTO vs D-RTO
0.095
0.1
0.105
0.11
0.115
0.12
0.125
0 10 20 30 40 50 60 70 80 90
'checkdrto.ris' u 1:4
'check.ris' u 1:4
Traditional approach
Two-shooting
Multiple-shooting
Flavio Manenti, Filip Logist KU-Leuven
44
Validation Case (Olefins)
Cracking Furnace (SPYRO-based D-RTO)
Flavio Manenti, Filip Logist KU-Leuven
45
Steam
Cracking
Furnace
Feed
Stack
Damper
Breeching
Convection
Section
Radiant
Section
Burners / Air Blowers
Coil Outlet
Temperature
(COT)
Steam
Transfer Line
Exchanger (TLE)
High
Pressure
Steam
Main
Factionator
>800C 400C
COT
Olefins
TC
FC
Fuel
Air
PV
PV
OUT
OUT
SP
Temperature
Controller
Flowrate Ratio
Controller
RADIANT SECTION
PV
Flavio Manenti, Filip Logist KU-Leuven
46
D-RTO
- ROMeo (OPERA) -
Software Integration
All-in-one tool for SPYRO-based smart dynamic simulation and
optimization of olefins plant
SPYRO
- FORTRAN -
SPYRO (FORTRAN)
Mixed-language (FORTRAN-C++)
Cracking furnace SPYRO-based dynamic model
(C++)
Very performing ODE/DAE solver (BzzOde,
BzzDae, BzzDaeSparse BzzMath)
Smart dynamic simulation (grade change,
DYNSIM)
Full integration in DYNSIM
Dynamic real-time optimization (multiple
shooting, ROMeo)
Full integration and OPERA synchronization
BzzMath
- C++ -
Dynamic Model
- DYNSIM -
Flavio Manenti, Filip Logist KU-Leuven
47
SPYRO-based (Smart) Dynamic
Simulation
C3H6/C2H4 Severity Change from 0.62 to 0.67
FUEL FLOWRATE
[kg/h]
DUTY [kcal/h] CH4/C3H6
SEVERITY
COIL OUTLET
TEMPERATURE (COT)
[C]
WALL
TEMPERATURE
[C]
C3H6/C2H4
SEVERITY
Initial Severity
Arrival Severity
1.34e+007
1.35e+007
1.36e+007
1.37e+007
1.38e+007
1.39e+007
1.4e+007
0 20 40 60 80 100 120
Time [min]
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
0 20 40 60 80 100 120
Time [min]
0.61
0.62
0.63
0.64
0.65
0.66
0.67
0.68
0.69
0 20 40 60 80 100 120
Time [min]
786
788
790
792
794
796
798
800
802
0 20 40 60 80 100 120
Time [min]
1088
1090
1092
1094
1096
1098
1100
1102
1104
0 20 40 60 80 100 120
Time [min]
3450
3500
3550
3600
3650
3700
3750
3800
0 20 40 60 80 100 120
Time [min]
Flavio Manenti, Filip Logist KU-Leuven
48
Severity Change Convergence
3450
3500
3550
3600
3650
3700
3750
3800
0.9 0.92 0.94 0.96 0.98 1 1.02 1.04
3450
3500
3550
3600
3650
3700
3750
3800
0.61 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69
Fuel Flowrate vs
CH4/C3H6 Severity
Fuel Flowrate vs
C3H6/C2H4 Severity
SPYRO-based (Smart) Dynamic
Simulation
Flavio Manenti, Filip Logist KU-Leuven
49
SPYRO-based (Smart) D-RTO
As per DYNSIM, UAM
inserted into the ICON
PALETTE (C++ dynamic
library)
Flavio Manenti, Filip Logist KU-Leuven
50
8-shoots flowsheet
Flavio Manenti, Filip Logist KU-Leuven
51
32-shoots flowsheet
No changes at the
ROMeos interface
Flavio Manenti, Filip Logist KU-Leuven
52
2000
3000
4000
5000
6000
7000
8000
9000
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05
Converging Path
F
U
E
L

F
L
O
W
R
A
T
E
4 SHOOTS
16, 32 SHOOTS
TRADITIONAL
STARTING
POINT
OPTIMUM
C3H6/ C2H4 SEVERITY
No changes at the
traditional control level
Flavio Manenti, Filip Logist KU-Leuven
53
3200
3250
3300
3350
3400
3450
3500
3550
3600
3650
3700
3750
0.6 0.65 0.7 0.75 0.8 0.85 0.9
High Benefits, Few Shoots
F
U
E
L

F
L
O
W
R
A
T
E
32 SHOOTS
16 SHOOTS
C3H6/ C2H4 SEVERITY
Flavio Manenti, Filip Logist KU-Leuven
54
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
0 20 40 60 80 100 120
Time [min]
0.5
0.6
0.7
0.8
0.9
1
1.1
0 20 40 60 80 100 120
Time [min]
Market dynamics
Market dynamics (the current market condition is a higher demand of
propylene, thus higher price) imposes a severity change in ethylene/propylene
production:
TRADITIONAL
4 SHOOTS
TRADITIONAL
4 SHOOTS
CH4/ C3H6 C3H6/ C2H4
16, 32 SHOOTS
16, 32 SHOOTS
Flavio Manenti, Filip Logist KU-Leuven
55
2000
3000
4000
5000
6000
7000
8000
9000
0 20 40 60 80 100 120
Time [min]
700
720
740
760
780
800
820
0 20 40 60 80 100 120
Time [min]
Severity change
Coil outlet temperature [C] of the
radiant section of the cracking furnace
Fuel flowrate [kg/h] entering the
cracking furnace
TRADITIONAL
4 SHOOTS
TI ME [min]
TRADITIONAL
4 SHOOTS
Supposed practical
upper bound 16, 32 SHOOTS
16, 32 SHOOTS
Flavio Manenti, Filip Logist KU-Leuven
56
3000
3200
3400
3600
3800
4000
0 50 100 150 200 250 300
Quantitative comparison
Traditional RTO
TIME [min]
32-shoots D-RTO
2000
2500
3000
3500
4000
4500
5000
0 50 100 150 200 250 300
D-RTO off-spec time
RTO off-spec time
TIME [min]
F
U
E
L

F
L
O
W
R
A
T
E
To operate at the optimum conditions dictated by the market, the RTO requires
more than 2h to accomplish the severity change, whereas the D-RTO requires
about 1h.
Consider that severity changes are not only imposed by market dynamics, but
even by feedstock changes, load changes As a result, frequent severity
changes are required in each coil of each cracking furnace of each olefins plants
1-step traditional RTO
2-steps traditional RTO
32-shoots D-RTO
F
U
E
L

F
L
O
W
R
A
T
E
Flavio Manenti, Filip Logist KU-Leuven
57
Industrial feasibility
D-RTO with ROMeo
Feasible
D-RTO is more stable than RTO
D-RTO halves off-spec periods
On-line feasibility for the industrial scale
Computational times are comparable
No visible changes to the user in ROMeo environment
No changes to the existing control scheme
Easy-to-use when implemented (few parameters to be defined)
SEVERITY
2000
3000
4000
5000
6000
7000
8000
9000
0.5 0.6 0.7 0.8 0.9 1 1.1
Upper bound
F
U
E
L

F
L
O
W
R
A
T
E
Flavio Manenti, Filip Logist KU-Leuven
Corporate level
Case: Eni Versalis
17 sites, European area (large-scale)
58
Flavio Manenti, Filip Logist KU-Leuven
Corporate optimal control
Air separation units
Air Liquide, Italy
Linde Gas, Italy
59
Manenti et al.
Raising the decision-makinglevel to improve the
enterprise-wide production flexibility
AIChE J ournal
59(5), 1588-1598, 2013
Flavio Manenti, Filip Logist KU-Leuven
Just a premise
High flexibility/operatibility is useless without corporate control
The case of Linde Gas, Ternis site:
Test preliminare (offline Munich-Arluno-Terni)
60
Single-site Corporate
Flavio Manenti, Filip Logist KU-Leuven
Industrial viewpoint
61
Outlier Detection
Robust methods
Linear/nonlinear Regressions
Performance Monitoring
Yield Accounting
Soft sensing
Data
Reconciliation
Mathematical
Modeling
Dynamic
Simulation
Model
Predictive
Control
Optimization
Model
Reduction
DCS, OTS, Plantwide control,
Soft sensing, process transients,
grade/load changes
Solvers
Planning
Scheduling
Dynamic optimization
Distributed predictive control
Nonlinear Systems
Optimizers
Differential systems
Stiff systems
ODE,DAE,PDE,PDAE
Efficiency
Decisions Raw Data
Parallel
Computing
Uncertainties
Optimal production
Optimal grade changes
Multi-objective
Real-time optimization
High accuracy
Reliable process control
Production improvement
Economy
Just in time
Market-driven
Logistics
Corporate
Supply Chain
Flavio Manenti, Filip Logist KU-Leuven
62
Available for questions:
[email protected]