Profit Based Control System Designmypages.iit.edu/~chmielewski/presentations/seminar/UIUC...Profit...
Transcript of Profit Based Control System Designmypages.iit.edu/~chmielewski/presentations/seminar/UIUC...Profit...
Profit Based Control System Design: A Globally Optimal Approach
Donald J. Chmielewski
Department of Chemical & Biological Engineering
Illinois Institute of Technology
*
BOP with
more profit
BOP with
less profit
OSSOP
EDOR’s due to
different controller
tunings
*
*
x
u
)(
)(
2
1
undesiredCB
desiredBA
k
k
iascapiscap
Rscap
Escap
iarm
Rarm
Larm
DC-DC
Converter
iabatibat
Rbat
Ebat
DC-DC
Converter
iafcifc
EfcDC-DC
Converter
Fuel
Cell
Power Bus
warm
Earm
kfc kbat kscap
SEPARATR
RISER
REGEN-TR
REG-CATY
FLUE-GAS
PRODUCTS
AIR
STEAM
STRP-STM
FEED-OIL
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Current Research Topics
• Fuel Cells – Modeling, Design, and Control
• PEMFC, SOFC and On-board Fuel Processors (ATR)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Current Research Topics
• Fuel Cells – Modeling, Design, and Control
• PEMFC, SOFC and On-board Fuel Processors (ATR)
• Stationary Power Plants – Modeling and Control
• Coal Fired Boilers with Oxygen Enrichment, IGCC
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Current Research Topics
• Fuel Cells – Modeling, Design, and Control
• PEMFC, SOFC and On-board Fuel Processors (ATR)
• Stationary Power Plants – Modeling and Control
• Coal Fired Boilers with Oxygen Enrichment, IGCC
• Control Theory
• Profit Based Controller Design
• Sensor and Actuator Selection
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Outline
• Motivating Example
• Controller Tuning
• Economic Based Tuning
• Robust Formulation
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Non-isothermal Reactor)
F
F
CA, T
AA
Apin
AAAinA
CTkr
rCHVTTFdt
dTV
VrCCFdt
dCV
)(
)/()(
)(
Increase F Increased production rate
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Non-isothermal Reactor)
F
F
CA, T
AA
Apin
AAAinA
CTkr
rCHVTTFdt
dTV
VrCCFdt
dCV
)(
)/()(
)(
Increase F Increased production rate
Decrease F Increase T Increase reaction rate
Increase production
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Limited Operating Region)
Process Limitations:
(max))( TtT
(max))( FtF
- Catalyst protection or
onset of side reactions
- Pump limit or limit on
downstream unit
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Limited Operating Region)
)()( )( spsp
c FTTKF
Process Limitations:
(max))( TtT
Possible Controller:
(max))( FtF
- Catalyst protection or
onset of side reactions
- Pump limit or limit on
downstream unit
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Performance in Time Series)
time
T(t)
F(t)
F(sp)
T(sp)
time
F(max)
T(max)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Performance in Phase Plane)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Elliptical Operating Region)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Elliptical Operating Region)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Limited Operating Region)
)()( )( spsp
c FTTKF
Controller:
),( )()( wTfF spsp
Steady-State Relation:
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Elliptical Operating Region)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Steady-State Operating Line)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Optimal Operating Point)
)(tF
)(tT
*
Decrease F
Increase T
Increase
conversion
Increase
production
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Optimal Operating Point: Another Possibility)
)(tF
)(tT
* Increase F
Increased
production rate
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Optimal Operating Point: Another Possibility)
)(tF
)(tT
*
Increase F
Increased
production rate
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example
(Suggests Different Controller Tuning)
)(tF
)(tT
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Less Aggressive Tuning)
time
F(t)
F(sp)
T(sp)
time
F(max)
T(max)
T(t)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Need for Automated Tuning)
)(tF
)(tT
*
*
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example (Need for Automated Tuning)
)(tF
)(tT
*
* *
* *
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Outline
• Motivating Example
• Controller Tuning
• Economic Based Tuning
• Robust Formulation
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Covariance Analysis (Open-Loop Case)
Process Model:
Steady State Covariance:
T
xz
T
w
T
xx
DD
GGAA
0
Plant
)()(
)()()(
tDxtz
twGtxAtx
)(tw )(tz
Gaussian white noise with covariance )(tww
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Expected Dynamic Operating Region (EDOR)
EDOR
defined by:
*
1z
2z
2
22
2
21
2
12
2
11
z
11
22
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Flexibility in EDOR Definition
a = 1 constraint
observance ~84% of time
a = 2 constraint
observance ~95% of time
a = 3 constraint
observance ~99.5% of
time
*
1z
2z
11a
22a
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Closed-Loop Covariance Analysis (Full State Information Case)
T
www
T
uxxuxz DDLDDLDD )()(
)()( tLxtu
0)()( T
w
T
xx GGBLABLA
wDuDxDz
wGBuAxx
wux
Process Model:
Controller:
Steady-State Covariance:
Plant )(tw
)(tx
L )(tu
)(tz
Department of Chemical and Biological Engineering
Illinois Institute of Technology
1z
2z
Closed-Loop EDOR
EDOR’s from different
controllers
* xLu 1
xLu 2
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Constrained Closed-Loop EDOR
*
1z
2z
Constraints
) ( izi za
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Constrained Closed-Loop EDOR
*
1z
2z
Constraints
) ( izi za
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Does there exist L such that:
00100 i
columnthi
Constrained Controller Existence
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
zii
T
izii niz 1/22
a
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Does there exist L such that:
00100 i
columnthi
Constrained Controller Existence
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
zi
T
izii niz 12
Department of Chemical and Biological Engineering
Illinois Institute of Technology
If and only if there exist X>0 and Y such that:
00100 i
Constrained Controller Existence (Convex Condition)
0)(
)(
XYDXD
YDXDDDT
i
T
ux
uxi
T
i
T
wwwii
0)()( T
w
T GGBYAXBYAX
zii niz 12
1 YXLAnd controller is constructed as: Lxu
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Implementation of (Pseudo-) Constrained Controller
*
1z
2z
EDOR based on
controller: Lxu
Constraints ) ( izi za
Department of Chemical and Biological Engineering
Illinois Institute of Technology
ziii
wux
TTT
ux
niztzz
wDuDxDtz
GwBuAxxts
dtRuuMxuQxx
1)(
)(
..
2min0
,
Model Predictive Control
Department of Chemical and Biological Engineering
Illinois Institute of Technology
zi
wux
TTT
ux
nitz
wDuDxDtz
GwBuAxxts
dtRuuMxuQxx
1)(
)(
..
2min0
,
MPC LQG
Department of Chemical and Biological Engineering
Illinois Institute of Technology
..
2min0
,
GwBuAxxts
dtRuuMxuQxx TTT
ux
LQG
Given A, B, Q, R and M, then the unconstrained controller is
01 TT MPBRMPBQPAPA
TMPBRL 1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Does there exist Q, R and M such that:
Constrained LQG Controller Existence
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
zi
T
izii niz 12
01 TT MPBRMPBQPAPA
TMPBRL 1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Constrained LQR Controller Existence (Convex Condition)
???
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Does there exist L such that:
Constrained Controller Existence
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
zi
T
izii niz 12
Department of Chemical and Biological Engineering
Illinois Institute of Technology
such that:
Constrained Minimum Variance (CMV) Controller
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
zi
T
izii niz 12
i
iiL
dix
min
,,0
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Theorem 1 (Chmielewski & Manthanwar, 2004):
Controller Equivalence
The controller generated by CMV
is coincident with
the controller generated by some
LQG problem.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
such that:
Pareto Frontier Interpretation of Minimum Variance Control
T
www
T
uxxuxz
DD
LDDLDD
)()(
0
)()(
T
w
T
xx
GG
BLABLA
z
T
izii ni 1
i
iiL
dix
min
,,0
Achievable
Performance
Levels
Unachievable
2
1
Pareto frontier
Department of Chemical and Biological Engineering
Illinois Institute of Technology
zi
T
izii niz 12
such that:
Pareto Frontier Interpretation of CMV Control
T
www
T
uxxuxz
DD
LDDLDD
)()(
0
)()(
T
w
T
xx
GG
BLABLA
2
1 1<z12
2<z22
i
iiL
dix
min
,,0
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
V1 q2 V2 q1
q0
FT
FC
V1
q1(sp)
FT
FC
q2(sp)
0
1
11
01
00
00GBA
State Variables: x = [V1 V2]T Manipulated Variables: u = [q1 q2]
T
Disturbance w = [q0] Performance Output: z = [V1 V2 q1 q2]T
0
0
0
0
10
01
00
00
00
00
10
01
wux DDD
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
Objective Weights: d1 = d2 = d3 = d4 = 1 Performance Output: z = [V1 V2 q1 q2]T
Constraint: 4 < 2.252
-5 0 5
-150
-100
-50
0
50
100
150
Level in
Tan
k 1
(m
3)
Flow out of in Tank 1 (m3/min)
-2 -1 0 1 2
-150
-100
-50
0
50
100
150
Flow out of Tank 2 (m3/min)
Level in
Tan
k 2
(m
3)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
Objective Weights: d1 = d2 = d3 = d4 = 1 Performance Output: z = [V1 V2 q1 q2]T
Constraints: 4 < 2.252 and 1.52
-5 0 5
-150
-100
-50
0
50
100
150
Level in
Tan
k 1
(m
3)
Flow out of in Tank 1 (m3/min)
-2 -1 0 1 2
-150
-100
-50
0
50
100
150
Flow out of Tank 2 (m3/min)
Level in
Tan
k 2
(m
3)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
Objective Weights: d1 = d2 = d3 = d4 = 1 Performance Output: z = [V1 V2 q1 q2]T
Constraints: 4 < 2.252, 1.52 and 0.752
-5 0 5
-150
-100
-50
0
50
100
150
Level in
Tan
k 1
(m
3)
Flow out of in Tank 1 (m3/min)
-2 -1 0 1 2
-150
-100
-50
0
50
100
150
Flow out of Tank 2 (m3/min)
Level in
Tan
k 2
(m
3)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
Constraints: 4 < 2.252
041.0039.0
687.0726.0L
Constraints: 4 < 1.52
014.0013.0
700.0712.0L
Constraints: 4 < 0.752
003.0003.0
701.0734.0L
Department of Chemical and Biological Engineering
Illinois Institute of Technology
ziii
wux
TTT
ux
niztzz
wDuDxDtz
GwBuAxxts
dtRuuMxuQxx
1)(
)(
..
2min0
,
Model Predictive Control
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Inverse Optimality
01 TT MPBRMPBQPAPA
TMPBRL 1
Theorem 2 (Chmielewski & Manthanwar, 2004):
If there exists P > 0 and R > 0 such that
0
RPBRL
PBRLPAPARLLTT
TTT
PAPARLLQ TT )( PBRLM T
0
RM
MQT
and P and L satisfy
Then and are such that
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 1: Surge Tanks
Constraints: 4 < 0.752
003.0003.0
701.0734.0L
160160
160161Q
4448.81
8.81299R
8.588.20
8.602.20M
Thus, weights are available for MPC implementation.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
GwBuAxxts
dtRuuMxuQxx TTT
ux
..
2min0
,
CMV Control
zi
T
izii niz 12
such that:
T
www
T
uxxuxz
DD
LDDLDD
)()(
0
)()(
T
w
T
xx
GG
BLABLA
i
iiL
dix
min
,,0
Achievable
Performance
Levels
Unachievable
2
1
Pareto frontier
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 2: Inventory Control
V1
Starts
Inventory Demand
Delivery delay
IC
Inventory
Set-point
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 2: Inventory Control
V1
Starts
Inventory Demand
Delivery delay
IC
Inventory
Set-point
Specific scenario:
Delivery Delay = 5 days
Demand Variance = 102
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Inventory Control (Pareto Frontier)
2 4 6 8 10 1224
25
26
27
28
29
30
31
32
33
Std
. D
ev
. In
ve
nto
ry
Std. Dev. Starts
A
B
C
-10 -5 0 5 10-40
-30
-20
-10
0
10
20
30
40
Inv
en
tory
Starts
C
B
A
Point A (the point of minimum inventory variance)
is the basis of classic “safety stock” analysis
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Multi-Echelon Inventory Control
(from Seferlis & Giannelos, 2004)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Decentralized Control)
V1
Starts
Inventory
1 Demand IC
Inventory
Set-point
1
V1
Starts
IC
Inventory
Set-point
2
Inventory
2
Delivery delay = 3 Delivery delay = 5
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Centralized Control)
V1
Starts
Inventory
1 Demand
Delivery delay = 3
Inventory Set-
point 1
V1
Starts Delivery delay = 5
IC Inventory Set-
point 2
Inventory
2
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Pareto Frontier)
0 5 10 15 20 25 3024
25
26
27
28
29
30
31
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Std. Dev. Inventory Tank 1
Decentralized
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (EDORs)
-10 -5 0 5 10-30
-20
-10
0
10
20
30
Inv
en
tory
Starts
Tank 2
Decentralized
-15 -10 -5 0 5 10 15-20
-15
-10
-5
0
5
10
15
20
Inv
en
tory
Starts
Tank 1
Decentralized
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Pareto Frontier)
Small “safety stock” increase at I2, leads to large “safety stock” decrease at I1
0 5 10 15 20 25 3024
25
26
27
28
29
30
31
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Std. Dev. Inventory Tank 1
Decentralized
Centralized A
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (EDORs)
Small “safety stock” increase at I2, leads to large “safety stock” decrease at I1
-15 -10 -5 0 5 10 15-20
-15
-10
-5
0
5
10
15
20
Inv
en
tory
Starts
Tank 1
Centralized A
Decentralized
-10 -5 0 5 10-30
-20
-10
0
10
20
30
Inv
en
tory
Starts
Tank 2
Decentralized
Centralized A
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Pareto Frontier)
Small “safety stock” increase at I2, leads to large “safety stock” decrease at I1
0 5 10 15 20 25 3024
25
26
27
28
29
30
31
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Std. Dev. Inventory Tank 1
Decentralized
Centralized A
Centralized B
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (EDORs)
Small “safety stock” increase at I2, leads to large “safety stock” decrease at I1
but starts variance size trend is unexpected
-15 -10 -5 0 5 10 15-20
-15
-10
-5
0
5
10
15
20
Inv
en
tory
Starts
Tank 1
Centralized A
Decentralized
Centralized B
-10 -5 0 5 10-30
-20
-10
0
10
20
30
Inv
en
tory
Starts
Tank 2
Decentralized
Centralized A
Centralized B
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Example 3: Hybrid Fuel Cell Vehicle
FC Kfc
iafcifc
Vfc
iab
ib
Vb
Rb
Eb
ia
Ra
La
VaKb
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Classic Control
FC Kfc
iafcifc
Vfc
iab
ib
Vb
Rb
Eb
ia
Ra
La
VaKb
Vehicle
kbatPmot
+-
kfc
x
Pfc
+-
x
Pbat
Pmot(sp)
Pbat(sp)
VfcFUEL CELL
VOLTAGE
CONTROLLER
Vfc(sp)
PI
PI
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Separation of Time-Scales
FC Kfc
iafcifc
Vfc
iab
ib
Vb
Rb
Eb
ia
Ra
La
VaKb
5 10 15 20 25 30 35
-200
0
200
400
600
800
1000
Power Profles [W]
time, sec
Pload
(sp)
Battery
Fuel Cell
Armature
Vehicle
kbatPmot
+-
kfc
x
Pfc
+-
x
Pbat
Pmot(sp)
Pbat(sp)
VfcFUEL CELL
VOLTAGE
CONTROLLER
Vfc(sp)
PI
PI
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Dynamics of PEMFC
Cooling
Air In
Jacket
Exhaust
Anode
In
(H2, H2O)
Ecell
H2
Cathode
In
(air)
Cathode
Exhaust
O2
H2O
N2
Solid Material Current Collector
Insulator
H2O
Anode
Exhaust
Polymer Membrane
Catalyst LayersGas Diffusion
Layers (GDLs)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Dynamics of PEMFC
Cooling
Air In
Jacket
Exhaust
Anode
In
(H2, H2O)
Ecell
H2
Cathode
In
(air)
Cathode
Exhaust
O2
H2O
N2
Solid Material Current Collector
Insulator
H2O
Anode
Exhaust
Polymer Membrane
Catalyst LayersGas Diffusion
Layers (GDLs)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Dynamics of PEMFC
Cooling
Air In
Jacket
Exhaust
Anode
In
(H2, H2O)
Ecell
H2
Cathode
In
(air)
Cathode
Exhaust
O2
H2O
N2
Solid Material Current Collector
Insulator
H2O
Anode
Exhaust
Polymer Membrane
Catalyst LayersGas Diffusion
Layers (GDLs)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Double Storage Configuration)
iascapiscap
Rscap
Escap
iarm
Rarm
Larm
DC-DC
Converter
iabatibat
Rbat
Ebat
DC-DC
Converter
iafcifc
EfcDC-DC
Converter
Fuel
Cell
Power Bus
warm
Earm
kfc kbat kscap
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Supervisory Control
Vehicle
Power
System
+ -Pscap
(sp) Pscap
kscap
Supervisory
Controller
Pmotor
+ -Pbat
(sp) Pbat
kbat
+ -Pfc
(sp) Pfc
kfc
PI
PI
PI
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Supervisory Model)
maxminmax
maxminmax
maxminmax
0
0
0
scapscapscapscapscapscapscap
batbatbatbatbatbatbat
fcfcfcfcfcfcfc
PPPEEPE
PPPEEPE
PPPPPPP
scapbatfcmot PPPP
fc
ratefcfc
sc
ratescsc
bat
ratebatbat CPPCEPCEP maxmaxmaxmaxmaxmax
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Supervisory Model)
maxminmax
maxminmax
maxminmax
0
0
0
scapscapscapscapscapscapscap
batbatbatbatbatbatbat
fcfcfcfcfcfcfc
PPPEEPE
PPPEEPE
PPPPPPP
scapbatfcmot PPPP
fc
ratefcfc
sc
ratescsc
bat
ratebatbat CPPCEPCEP maxmaxmaxmaxmaxmax
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Supervisory Model)
maxminmax
maxminmax
maxminmax
0
0
0
scapscapscapscapscapscapscap
batbatbatbatbatbatbat
fcfcfcfcfcfcfc
PPPEEPE
PPPEEPE
PPPPPPP
scapbatfcmot PPPP
fc
ratefcfc
sc
ratescsc
bat
ratebatbat CPPCEPCEP maxmaxmaxmaxmaxmax
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Supervisory Model)
maxminmax
maxminmax
maxminmax
0
0
0
scapscapscapscapscapscapscap
batbatbatbatbatbatbat
fcfcfcfcfcfcfc
PPPEEPE
PPPEEPE
PPPPPPP
scapbatfcmot PPPP
fc
ratefcfc
sc
ratescsc
bat
ratebatbat CPPCEPCEP maxmaxmaxmaxmaxmax
Colored noise disturbance
(modeled from drive cycle data)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Drive Cycle Simulation 1)
0 200 400 600 800 1000-20
-15
-10
-5
0
5
10
15
20
25Power Out/ Motor In
Pow
er
(kW
)
time (sec)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (EDOR Case 1)
-30 -20 -10 0 10 20 30-1.5
-1
-0.5
0
0.5
1
1.5x 10
4
Fu
el
Ce
ll P
ow
er
(W)
Fuel Cell Power Rate of Change (W/s) -3000 -2000 -1000 0 1000 2000 3000
-1
-0.5
0
0.5
1
x 107
En
erg
y i
n B
att
ery
(J
)
Power from Battery (W)
-4000 -3000 -2000 -1000 0 1000 2000 3000 4000-5000
0
5000
En
erg
y i
n S
up
er
Ca
p (
J)
Power from Super Cap (W)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (EDOR Case 2)
-30 -20 -10 0 10 20 30-1.5
-1
-0.5
0
0.5
1
1.5x 10
4
Fu
el
Ce
ll P
ow
er
(W)
Fuel Cell Power Rate of Change (W/s) -3000 -2000 -1000 0 1000 2000 3000
-1
-0.5
0
0.5
1
x 107
En
erg
y i
n B
att
ery
(J
)
Power from Battery (W)
-8 -6 -4 -2 0 2 4 6 8
x 104
-1
-0.5
0
0.5
1
x 105
En
erg
y i
n S
up
er
Ca
p (
J)
Power from Super Cap (W)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Hybrid Fuel Cell Vehicle (Drive Cycle Simulation 2)
0 200 400 600 800 1000-20
-15
-10
-5
0
5
10
15
20
25Power Out/ Motor In
Pow
er
(kW
)
time (sec)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Outline
• Motivating Example
• Controller Tuning
• Economic Based Tuning
• Robust Formulation
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Motivating Example
)(tF
)(tT
*
*
* *
Department of Chemical and Biological Engineering
Illinois Institute of Technology
EDOR
Constrained Operating Region
MV’s
CV’s
*
Constraints
Steady-State Operating Point
Department of Chemical and Biological Engineering
Illinois Institute of Technology
EDOR
Real-Time Optimization
*
*
Optimal
Steady-State
Operating Point
(OSSOP)
MV’s
CV’s Constraints
Steady-State Operating Point
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Real-Time Optimization
),,(),,( pmshqpmsfs
Original Nonlinear Process Model:
(s,m,p,q) ~(state, mv, dist, performance) ~ (x,u,w,z)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Real-Time Optimization
),,(),,( pmshqpmsfs
Original Nonlinear Process Model:
maxmin
,,
),,(),,(0
s.t. )(min
iii
qms
qqqpmshqpmsf
qg
Real-Time Optimization (minimize profit loss):
(s,m,p,q) ~(state, mv, dist, performance) ~ (x,u,w,z)
RTO solution denoted as (sossop,mossop,possop,qossop)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Backed-off Operating Point (BOP)
EDOR
*
* *
Backed-off
Operating Point
(BOP)
MV’s
CV’s
Optimal
Steady-State
Operating Point
(OSSOP)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Steady-State BOP Selection (Bahri, Bandoni & Romagnoli, 1996)
0
),,(
),,(0s.t.
}{ maxmax s.t. )(mini],[,, maxmin
ii
iipppqms
pmshq
pmsf
qqqg
Solve the following Semi-infinite Programming Problem
Extensions:
- Dynamic version in Bahri, et al, (1995)
- Linearized version in Contreras-Dordelly & Marlin (2000)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Linearized Perspective
maxmin
),,(
),,(
)(
iii
bop
qqq
pmshq
pmsfs
qg
Nonlinear Linear wrt OSSOP Linear wrt BOP
Deviation Variables w.r.t. OSSOP:
s’ = sbop – sossop m’ = mbop - mossop
p’ = pbop - possop q’ = qbop - qossop
maxmin '''
''''
''''
')(
iii
wux
q
ossop
qqq
pDmDsDq
GpBmAss
qgqg
Deviation Variables w.r.t. BOP:
x = s– sbop u = m - mbop
w = p - pbop z = q - qbop
maxmin
iii
wux
zzz
wDuDxDz
GwBuAxx
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Assume controller L is given and calculate i :
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
z
T
izii ni 1
Stochastic BOP Selection (Loeblein & Perkins, 1999)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
min2/1max2/1
maxmin
',','
''''
')''('
''0 s.t. 'min
iiiiii
iiiuxii
qqms
qqqq
qqqmDsDq
BmAsqg
Assume controller L is given and calculate i :
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
z
T
izii ni 1
Solve the following Linear Program:
Stochastic BOP Selection (Loeblein & Perkins, 1999)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
min2/1max2/1
maxmin
',','
''''
')''('
''0 s.t. 'min
iiiiii
iiiuxii
qqms
qqqq
qqqmDsDq
BmAsqg
Assume controller L is given and calculate i :
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
z
T
izii ni 1
Solve the following Linear Program:
Stochastic BOP Selection (Loeblein & Perkins, 1999)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Stochastic BOP Selection (EDOR within Constraint Set)
iiiiii qqqq '''' max2/1min2/1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Stochastic BOP Selection (EDOR within Constraint Set)
iiiiii qqqq '''' maxmin
Department of Chemical and Biological Engineering
Illinois Institute of Technology
q'1 max
EDOR BOP
q'1 min
q'2 min q'2 max
q'2 - q'2min q'2
max - q'2
Stochastic BOP Selection (EDOR within Constraint Set)
iiiiii qqqq '''' maxmin
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Fixed Controller BOP Selection
Controller is fixed EDORs have fixed
sizes and shapes
Loeblein and Perkins (1999):
OSSOP
*
* *
x
EDOR
u
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Peng et al. (2005):
Variable Controller EDORs have variable
sizes and shapes
*
* *
OSSOP EDOR
u
x
Variable Controller BOP Selection
Department of Chemical and Biological Engineering
Illinois Institute of Technology Peng et al. (2005)
*
BOP with
more profit
BOP with
less profit
OSSOP
EDOR’s due to
different controller
tunings
*
Profit Control (Simultaneous BOP and Controller Selection)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Profit Control (Simultaneous BOP and Controller Selection)
min2/1max2/1
maxmin
,,,',','
''''
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
Lqms
qqqq
qqqmDsDq
BmAsqg
zxi
T
www
T
uxxuxz DDLDDLDD )()(
0)()( T
w
T
xx GGBLABLA
z
T
izii ni 1
Peng et al. (2005)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
min2/1max2/1
maxmin
,,',','
''''
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
YXqms
qqqq
qqqmDsDq
BmAsqg
i
0)(
)(
XYDXD
YDXDDDT
i
T
ux
uxi
T
i
T
wwwii
0)()( T
w
T GGBLAXBYAX
Profit Control (Simultaneous BOP and Controller Selection)
Peng et al. (2005)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
min2/1max2/1
maxmin
,,',','
''''
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
YXqms
qqqq
qqqmDsDq
BmAsqg
i
0)(
)(
XYDXD
YDXDDDT
i
T
ux
uxi
T
i
T
wwwii
0)()( T
w
T GGBLAXBYAX
Peng et al. (2005)
Computational Aspects of Profit Control
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Reverse-Convex Constraints
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
2
1
max
11 )''( qq 2min
111 )''( qq
1'q
1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Global Solution
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
Based on Branch and Bound algorithm
1'q
1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Combinatorial Growth of Branch and Bound
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q-2 -1.5 -1 -0.5 0
0
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q-2 -1.5 -1 -0.5 0
0
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q-2 -1.5 -1 -0.5 0
0
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
Region 1
Region 2
Region 3
Region 4
Region 5
1'q
Combinatorial Growth of Branch and Bound
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Heuristic Scheme
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Combinatorial Growth??
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
1'q-2 -1.5 -1 -0.5 0
0
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
2'q
-2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
3'q -2 -1.5 -1 -0.5 00
0.2
0.4
0.6
0.8
1
zss,i
i
(zss,i
+dmin,i
)2 (z
ss,i+d
max,i )
2
Feasible Region
4'q
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Mass-Spring-Damper Example
Mass
wf
rmax
r
rmin
160and11 fr
System Model:
wfv
r
v
r
1
0
1
0
32
10
force edisturbanc theis
and (MV) forceinput theis
velocity, theis position, mass theis
w
f
vr
System Constraints:
where
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Mass-Spring-Damper Example
Mass
wf
rmax
r
rmin
160and11 fr
System Model:
wfv
r
v
r
1
0
1
0
32
10
force edisturbanc theis
and (MV) forceinput theis
velocity, theis position, mass theis
w
f
vr
System Constraints:
where
Mas
s P
osi
tio
n
Upper Bound on Position
time
} EDOR
Backed-off Operating Point (BOP)
*
OSSOP
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Mass-Spring-Damper Example
Mass
wf
rmax
r
rmin
160and11 fr
System Model:
wfv
r
v
r
1
0
1
0
32
10
force edisturbanc theis
and (MV) forceinput theis
velocity, theis position, mass theis
w
f
vr
System Constraints:
where
Mas
s P
osi
tio
n
Upper Bound on Position
time
} EDOR
Backed-off Operating Point (BOP)
*
OSSOP
Mas
s P
osi
tio
n
Upper Bound on Position
time
}
EDOR
Backed-off Operating Point (BOP)
*
OSSOP
Department of Chemical and Biological Engineering
Illinois Institute of Technology
* BOP
Mass-Spring-Damper Example (Phase Plane)
r
f
Steady-State Line
*
OSSOP
Constraints
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Mass-Spring-Damper Example (Phase Plane Solution)
4 6 8 10 12 14 16-0.5
0
0.5
1
Mass
Po
siti
on
(m
)
Input Force (N)
FSI Case
PSI Case
OSSOP
FSI Case:
Full State Information:
Controller is
PSI Case:
Partial State Information:
One Velocity Sensor.
Controller is
)()( tLxtu
)(ˆ)( txLtu
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Mass-Spring-Damper Example (Impact of Constraints )
6 8 10 12 14 16 18 20
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
Mass
Posi
tion (m
)
Input Force (N)
Case A
Case B
Case C
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Discrete-time Simulation (Scatter Plot)
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force (N)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
ziii
wux
TTT
ux
niztzz
wDuDxDtz
GwBuAxxts
dtRuuMxuQxx
1)(
)(
..
2min
maxmin
0,
MPC and the EDOR
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force (N)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
0
1)(
)(
..
2min
maxmin
0,
i
ziiiii
wux
TTTT
ux
s
nisztzsz
wDuDxDtz
GwBuAxxts
ssdtRuuMxuQxx
Soft Constraints
Department of Chemical and Biological Engineering
Illinois Institute of Technology
0
1)(
)(
..
2min
maxmin
0,
i
ziiiii
wux
TTTT
ux
s
nisztzsz
wDuDxDtz
GwBuAxxts
ssdtRuuMxuQxx
Soft Constraints
Department of Chemical and Biological Engineering
Illinois Institute of Technology
MPC with Soft Constraints
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)
gm = 107 gf = 103
gm = 103 gf = 103
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Flexibility in EDOR Definition
a = 1 constraint
observance ~84% of time
a = 2 constraint
observance ~95% of time
a = 3 constraint
observance ~99.5% of
time
*
1z
2z
11a
22a
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Impact of EDOR Definition
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force (N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Input Force (N)
Mass P
osit
ion
(m
)
a = 1 a = 2
Department of Chemical and Biological Engineering
Illinois Institute of Technology
MPC with Soft Constraints EDOR = 2 std dev’s
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)
gm = 107 gf = 103
gm = 103 gf = 103
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Impact of EDOR Definition (Reduced Sensitivity to Soft Weights)
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force(N)0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Mass P
osit
ion
(m
)
Input Force (N)
0 5 10 15
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Input Force (N)
Mass P
osit
ion
(m
)
gm = 107 gf = 103
gm = 103 gf = 103
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Fluidized Catalytic Cracker (FCC) Example
Regenerator and Separator (dynamic):
Riser (pseudo steady state):
(adapted from Loeblein & Perkins, 1999)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
FCC Example
Process Constraints:
Profit Function:
Fgs Fgl and Fugo are product flows
(gasoline, light gas and unconverted oil).
(adapted from Loeblein & Perkins, 1999)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Fixed LQG Controller
wDuDxDzGwBuAxxts
dtDzz
wux
T
ux
..
min0
,
),,( MRQRicL
DDDMDDDRDDDQ
ID
u
T
xu
T
ux
T
x
Department of Chemical and Biological Engineering
Illinois Institute of Technology
785 790 795 800 805 810 815
990
992
994
996
998
1000
Cyclo
ne T
em
pera
ture
(K
)
Separator Temperature (K)
0.0125 0.013 0.0135 0.014 0.0145 0.015
990
992
994
996
998
1000
Reg
en
era
tor
Tem
p (
K)
Coke Fraction in Separator
Fixed Controller
5 5.5 6 6.5 7 7.5 8 8.5
x 10-3
280
300
320
340
360
380
400
Cata
lyst
Flo
w (
kg
/s)
Fraction of Coke in Regenerator0 5 10 15 20
x 10-4
25
26
27
28
29
30
31
32
Inle
t A
ir (
kg
/s)
Oxygen Mass Fraction
Fixed Controller FCC (Loeblein & Perkins, 1999)
(adapted)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
0 5 10 15 20
x 10-4
25
26
27
28
29
30
31
32
Inle
t A
ir (
kg
/s)
Oxygen Mass Fraction5 5.5 6 6.5 7 7.5 8 8.5
x 10-3
280
300
320
340
360
380
400
Cata
lyst
Flo
w (
kg
/s)
Fraction of Coke in Regenerator
0.0125 0.013 0.0135 0.014 0.0145 0.015
990
992
994
996
998
1000
Reg
en
era
tor
Tem
p (
K)
Coke Fraction in Separator
Fixed Controller Free Controller
785 790 795 800 805 810 815
990
992
994
996
998
1000
Cyclo
ne T
em
pera
ture
(K
)
Separator Temperature (K)
Free Controller FCC (Profit Control)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
FCC Profit
Gross Profit Diff from OSSOP
($/day) ($/day) OSSOP $36,905 - Fixed Control $35,768 - $1,137
Department of Chemical and Biological Engineering
Illinois Institute of Technology
FCC Profit
Gross Profit Diff from OSSOP
($/day) ($/day) OSSOP $36,905 - Fixed Control $35,768 - $1,137 Profit Control $36,160 - $745
Department of Chemical and Biological Engineering
Illinois Institute of Technology
FCC Profit
Gross Profit Diff from OSSOP ($/day) ($/day) OSSOP $36,905 - EDOR = 1 std. dev. Fixed Control $35,768 - $1,137 Profit Control $36,160 - $745 EDOR = 2 std. dev. Fixed Control $34,631 - $2,274 Profit Control $35,416 - $1,489
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
CB
BA
k
k
2
1
(adapted from de Hennin & Perkins, 1991)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
(adapted from de Hennin & Perkins, 1991)
211, ccMF QQQQ
Manipulated Variables:
Disturbances:
inA
in
C
Tw
,
2
2
10
03w
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
(adapted from de Hennin & Perkins, 1991)
KT 3501 KT 3502
smQQ MF /8.0 3
1
KTKT outcoutc 300330 ,2,1
smQsmQ MF /05.0/05.0 33
1
21
1
21
01.0
1.01.0
)(10
CQQ
MF
BMF
aj
incja
aj
incja
UQ
TTQUq
UQ
TTQUq
2,
,222,
2
1,
,111,
12
)(2
2
)(2
Some Process Constraints:
Profit Function:
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
300 305 310 315 320 325 330294
295
296
297
298
299
300
301
Tem
pera
ture
fro
m J
acket
2 (
K)
Temperature from Jacket 1 (K)
344 345 346 347 348 349 350
325
330
335
340
345
350
Rea
cto
r 2
Tem
per
atu
re (
K)
Reactor 1 Temperature (K)
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
0.4
Feed
Flo
w (
m3/s
)
Makeup Flow (m3/s)
0 0.2 0.4 0.6 0.8
0
0.2
0.4
0.6
0.8
Jacket
Flo
w 1
(m
3/s
)
Jacket Flow 2 (m3/s)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
(adapted from de Hennin & Perkins, 1991)
Disturbances:
a
inA
in
U
C
T
w ,
2
2
2
021.000
010
003
w
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
300 305 310 315 320 325 330294
295
296
297
298
299
300
301
Tem
pera
ture
fro
m J
acket
2 (
K)
Temperature from Jacket 1 (K)
No HEX Fault
HEX Fault
344 345 346 347 348 349 350
325
330
335
340
345
350
Rea
cto
r 2
Tem
per
atu
re (
K)
Reactor 1 Temperature (K)
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
0.4
Feed
Flo
w (
m3/s
)
Makeup Flow (m3/s)
0 0.2 0.4 0.6 0.8
0
0.2
0.4
0.6
0.8
Jacket
Flo
w 1
(m
3/s
)
Jacket Flow 2 (m3/s)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
Disturbances:
a
inA
in
U
C
T
w ,
2
2
2
021.000
010
003
w Assume Ua is not white noise, but highly correlated colored noise
(slowly varying).
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR’s in Series
300 305 310 315 320 325 330294
295
296
297
298
299
300
301
Tem
pera
ture
fro
m J
acket
2 (
K)
Temperature from Jacket 1 (K)
No HEX Fault
HEX Fault White Noise
HEX Fault Highly Correlated
Noise
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
0.4
Feed
Flo
w (
m3/s
)
Makeup Flow (m3/s)
0 0.2 0.4 0.6 0.8
0
0.2
0.4
0.6
0.8
Jacket
Flo
w 1
(m
3/s
)
Jacket Flow 2 (m3/s)
344 345 346 347 348 349 350
325
330
335
340
345
350
Rea
cto
r 2
Tem
per
atu
re (
K)
Reactor 1 Temperature (K)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR Profit
Gross Profit Diff from OSSOP ($/day) ($/day) OSSOP $2,486 - Ua = 0 $2,342 - $144
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR Profit
Gross Profit Diff from OSSOP ($/day) ($/day) OSSOP $2,486 - Ua = 0 $2,342 - $144 Ua - white noise $2,176 - $310
Department of Chemical and Biological Engineering
Illinois Institute of Technology
CSTR Profit
Gross Profit Diff from OSSOP ($/day) ($/day) OSSOP $2,486 - Ua = 0 $2,342 - $144 Ua - white noise $2,176 - $310 Ua – slow varying $2,115 - $371
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Outline
• Motivating Example
• Controller Tuning
• Economic Based Tuning
• Robust Formulation
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Uncertainty Characterization
Plant in the polytopic set:
W=co{Aj, Bj, Gj, Dxj, Duj j = 0 … NW}
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Uncertainty and Stability
Plant in the polytopic set:
W=co{Aj, Bj, Gj, Dxj, Duj j = 0 … NW}
Sufficient condition (quadratic stability):
$ P > 0 s.t. AjP + PAjT < 0 j = 0…NW
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2min2max
maxmin
,,',','
)''()''(
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
YXqms
qqqq
qqqmDsDq
BmAsqg
i
0)(
)(
XYDXD
YDXDT
i
T
ux
uxii
0)()( T
w
T GGBLAXBYAX
Profit Control (Simultaneous BOP and Controller Selection)
Department of Chemical and Biological Engineering
Illinois Institute of Technology
W
Njniqqqq
niqqqmDsDq
mBsAqg
ziiijiiij
ziiiuxii
q
YXqms
ij
...1...1)''()''(
...1')''('
''0 s.t. 'min
2min2max
maxmin
00
00
,,',','
W
Nj
ni
XYDXD
YDXDz
T
i
T
ujxj
ujxjiij
...1
...10
)(
)(
W
Nj
G
GYBXAYBXA
w
T
j
j
T
jjjj...10
)()(1
Profit Control with Robust Performance Conditions
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Profit Control with Robust Performance Conditions
Robust BOP
OSSOP
EDOR’s
due to
uncertain
plant
model *
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Inventory Control with Yield Uncertainty
V1
Starts
Inventory Demand
Delivery delay
IC
Inventory
Set-point Specific scenario:
Delivery Delay = 5 days
Demand Variance = 102
Yield Rate [0.8, 1.0]
Lost Orders
Department of Chemical and Biological Engineering
Illinois Institute of Technology
0.8 0.85 0.9 0.95 124.4
24.5
24.6
24.7
24.8
24.9
25
25.1
25.2
25.3
Yield
Std
. D
ev
. In
ve
nto
ry
Yield = 1.0
Yield = 0.8
Robust
Inventory Control with Yield Uncertainty
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Delivery Delay Uncertainty)
V1
Starts
Inventory
1 Demand
Delivery delay = 3
Inventory Set-
point 1
V1
Starts Delivery delay [4, 5]
IC Inventory Set-
point 2
Inventory
2
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2-Echelon Inventory Control (Delivery Delay Uncertainty)
V1
Starts
Inventory
1 Demand
Delivery delay = 3
Inventory Set-
point 1
V1
Starts Delivery delay [4, 5]
IC Inventory Set-
point 2
Inventory
2
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Inventory Control with Delivery Delay Uncertainty
0 0.2 0.4 0.6 0.8 125
26
27
28
29
30
Alpha
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Robust
Alpha = 0
Alpha = 1
Department of Chemical and Biological Engineering
Illinois Institute of Technology
0 0.2 0.4 0.6 0.8 125
26
27
28
29
30
Alpha
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Robust
Alpha = 0
Alpha = 1
0 5 10 15 20 25 3024
25
26
27
28
29
30
31
Std
. D
ev
. In
ve
nto
ry T
an
k 2
Std. Dev. Inventory Tank 1
Centralized A
Robust
Alpha = 1
Alpha = 0
Inventory Control with Delivery Delay Uncertainty
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2min2max
maxmin
,,,',','
)''()''(
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
YXqms
qqqq
qqqmDsDq
BmAsqg
i
a
0)(
)(
XYDXD
YDXDT
i
T
ux
uxii
Profit Control (Peak-to-Peak Formulation)
0)()(
IG
GXBYAXBYAXT
T
a
a
Department of Chemical and Biological Engineering
Illinois Institute of Technology
2min2max
maxmin
,,,',','
)''()''(
')''('
''0 s.t. 'min
iiiiii
iiiuxii
q
YXqms
qqqq
qqqmDsDq
BmAsqg
i
a
0)(
)(
XYDXD
YDXDT
i
T
ux
uxii
Profit Control (Peak-to-Peak Formulation)
0)()(
IG
GXBYAXBYAXT
T
a
a
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Conclusions
• Relationship between control system performance
and plant profit quantified.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Conclusions
• Relationship between control system performance
and plant profit quantified.
• Enables profit guided control system design.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Conclusions
• Relationship between control system performance
and plant profit quantified.
• Enables profit guided control system design.
• Globally optimal search algorithm as well as
heuristic scheme proposed.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Conclusions
• Relationship between control system performance
and plant profit quantified.
• Enables profit guided control system design.
• Globally optimal search algorithm as well as
heuristic scheme proposed.
• Applicable to a broad set of applications from
a variety of disciplines.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Conclusions
• Relationship between control system performance
and plant profit quantified.
• Enables profit guided control system design.
• Globally optimal search algorithm as well as
heuristic scheme proposed.
• Applicable to a broad set of applications from
a variety of disciplines.
• Extendable it robust and peak-to-peak framework,
but conservatism a concern.
Department of Chemical and Biological Engineering
Illinois Institute of Technology
Acknowledgements
• Students and Collaborators:
Amit Manthanwar Jui-Kun Peng
Syed K. Ahmed Benjamin P. Omell
Wai-Kit Ong (UG)
Prof. Professor Durango-Cohen (IIT Business School)
Dr. Aymeric Rousseau (ANL)
• Funding:
Argonne National Laboratory
Illinois Clean Coal Institute
Undergraduate Research Center, IIT
Graduate and Armour Colleges, IIT
Chemical & Biological Engineering Department, IIT