Title: Integrated%20Process%20Networks:%20Nonlinear%20Control%20System%20Design%20for%20Optimality%20and%20Dynamic%20Performance
1Integrated Process NetworksNonlinear Control
System Design for Optimality and Dynamic
Performance
- Michael Baldeaa,b and Prodromos Daoutidisa
- aUniversity of Minnesota, Minneapolis, MN 55455
- bPraxair, Inc., Tonawanda, NY 14150
- Antonio C. Brandao Araujo and Sigurd Skogestad
- Norwegian University of Science and Technology
- NO-7491 Trondheim, Norway
2Chemical Plant
- Material recycle
- Heat integration
- Feedback interactions within the plant
3Control of Tightly Integrated PlantsChallenging
and Important!
- Decentralized control inherent limitations
- Fully centralized control generally impractical
- Size / complexity of dynamic models
- Ill-conditioning
-
- Efficient transient operation critical
- Moves across product slate due to frequent
- changes in market conditions and economics
- Going beyond regulatory control
- Accounting for network dynamics
4Research on Integrated Process Networks
- Dynamic Analysis
- Slow response, high sensitivity to disturbances,
instability - (Gilliland et al. 64, Denn Lavie 82,
Skogestad Morari 87, - Luyben 93, Mizsey Kalmar 96)
- Nonlinear dynamics
- (Morud Skogestad 94, 96, Bildea et al. 00,
Kiss et al. 06)
5Research on Integrated Process Networks
- Control
- Interaction of design and control for
reaction-separation networks - (Luyben 93, Luyben M. Floudas 94, Yin
Luyben 97) - Plant-wide control
- (Price Georgakis 93, Luyben et al. 97, Ng
Stephanopoulos 98, - Zheng et al. 99)
- Applications to benchmark problems
- (McAvoy Ye, 94, Ricker, 96, Ricker and Lee,
95, Larsson et al. 01, - Jockenhovel et al. 03)
- Self optimizing control (Morari et al. 80,
Skogestad 00) - Partial control
- (Shinnar et al. 96, Tyreus 99, Kothare et
al. 00) - Passivity based stabilization (Ydstie et al. 98,
99) - Dynamic optimization
- (Tosukhowong et al. 04)
- Time-scale analysis / nonlinear model reduction
and control - (Kumar and Daoutidis 02, Baldea et al. 06,
Baldea and Daoutidis 05 06)
6Present Work
- Combining time-scale analysis (dynamics)
- and self-optimizing control (steady state
- economics)
- Control structure design
- Nonlinear supervisory control
- Prototype reactor-separator-recycle network
7Plant-wide ControlHierarchy of
Decisions(Larsson and Skogestad, 2000)
Planning (months - years)
- I. TOP-DOWN
- Step 1. DEGREES OF FREEDOM
- Step 2. OPERATIONAL OBJECTIVES
- Step 3. CONTROLLED VARIABLES
- Step 4. PRODUCTION RATE
- II. BOTTOM-UP
- Step 5. REGULATORY CONTROL
- LAYER (PID)
- Step 6. SUPERVISORY CONTROL
- LAYER (MPC)
-
- Step 7. OPTIMIZATION LAYER
- (RTO) Can we do without?
8What should we control?
- Optimization level Solve
- Optimal solution usually at constraints
- most degrees of freedom are used to satisfy
active constraints - Control active constraints!
- Implementation usually simple
- What else should we control?
- Variables for remaining unconstrained degrees of
freedom acceptable losses in the presence of
disturbances and implementation errors
9Self-optimizing Control
Planning (months - years)
- Principle
- (Economically) acceptable operation (loss) should
be achieved using constant set points for the
controlled variables, - without the need to
- re-optimize when disturbances occur.
ccs
10Selection of Controlled Variables
11Integrated Process NetworkMultiple Time Scale
Dynamics
- Low single pass conversion - high recycle rate
- Impurities present in the feed small amount
- Impurities do not separate readily -small purge
stream
Baldea and Daoutidis, Comp. Chem. Eng., 2006.
12Dynamic Model
- scaled inputs large recycle loop
flowrates - scaled inputs medium flowrates
- scaled input small purge flowrate
- small parameter ratio of throughput to
recycle - small parameter ratio of purge to
throughput - states
- terms
stiffness, multiple time scales
13 Model ReductionTime Scale Decomposition
- Fast time scale (process units)
- Intermediate time scale (network)
- Slow time scale (impurity levels)
dimensional Equilibrium
manifold Manipulated inputs
dimensional
Equilibrium manifold Manipulated inputs
1-dimensional Manipulated input
14Hierarchical Controller Design
15Optimality and Dynamic Performance
- Self Optimizing Control
- economic insight
- selection of controlled variables
- Time Scale Analysis
- dynamic perspective
- selection of manipulated inputs
Combining control designs with inherent
optimality and good dynamic performance
16Case StudyGeneric Reactor Condenser Network
- Slow reaction , large recycle
- Product nonvolatile
- Volatile impurity present in the feed
- Degrees of freedom R (W), F, P, L
17Insights from Time-scale Analysis
- Control objectives
- vapor holdups (pressures) stabilization
(fast), - liquid holdup stabilization, X B
(intermediate) - impurity levels (slow)
- Available degrees of freedom
- R,F (fast)
- L, M RSP, M CSP (intermediate)
- P (slow)
18Hierarchical Controller Design (I)(Baldea and
Daoutidis CChE, 2006)
Time scale Controlled output Manipulated Input Controller
Fast Reactor holdup Reactor effluent Proportional
Fast Condenser vapor holdup Recycle rate Proportional
Intermediate Liquid holdup Liquid flowrate Proportional
Intermediate Product purity Reactor holdup Set point Nonlinear, model-based, cascade
Slow Inert levels in reactor Purge flowrate Nonlinear, model-based
19 Control Structure I
- Reactor pressure allowed to vary
- Compressor/pressure constraints?
20Insights from Self-optimizing Control
- Disturbances FO, yA,O, yI,O, xB,TR ,k1
- Cost function J pWW - pLL pPL
- Active constraints Reactor pressure, product
purity - Self-optimizing variable W
21 Self-optimizing Control Structure (II)
-
- No control of impurity
- Poor dynamics small purge controls product purity
22Hierarchical / Self-optimizing Controller Design
(III)
Time scale Controlled output Manipulated Input Controller
Fast Reactor holdup Reactor effluent Proportional
Condenser vapor holdup Recycle rate (compressor power) Proportional
Intermediate Liquid holdup Liquid flowrate Proportional
Product purity Condenser vapor holdup set point Nonlinear, model-based, cascade
Slow Compressor power Purge flowrate Proportional Integral
235 increase in purity setpoint
245 increase in purity setpoint
2520 increase in production rate
2620 increase in production rate
27Concluding Remarks
- Self-optimizing control / time-scale analysis
complementary perspectives -
- steady-state economics vs dynamics
- controlled variables vs manipulated inputs
- Control configurations that are self-optimizing
and have good closed - loop response
characteristics -
- Well-conditioned nonlinear supervisory
controllers - based on reduced order models
28Acknowledgements
- National Science Foundation
- MB partially funded by a University of Minnesota
Doctoral Dissertation Fellowship
29Integrated Process NetworksNonlinear Control
System Design for Optimality and Dynamic
Performance
- Michael Baldeaa,b and Prodromos Daoutidisa
- aUniversity of Minnesota, Minneapolis, MN 55455
- bPraxair, Inc., Tonawanda, NY 14150
- Antonio C. Brandao Araujo and Sigurd Skogestad
- Norwegian University of Science and Technology
- NO-7491 Trondheim, Norway