Title: Cybernetic Model Predictive Control for poly3hydroxybutyrate Production
1Cybernetic Model Predictive Control for
poly-3-hydroxybutyrate Production
- Timothy J. Crowley, Francis J. Doyle III
- University of Delaware, Newark, DE 19716
- Jeffrey Varner
- University of Minnesota, Minneapolis, MN 55455
2Outline
- Background motivation
- Case study
- cybernetic storage synthesis model
- linear model predictive control (MPC) with state
estimation - poly-?-hydroxybutyrate (PHB) productivity control
results - Summary
3Background
- Synthesis of PHB in Alcaligenes eutrophus
- polymer biodegradable, thermoplastic
- process green synthesis route for polyesters
- Potential applications
- medical (implants, drug delivery)
- film, packaging
- Focus PHB productivity control in continuous
culture - potentially higher productivity than fed-batch
operation
4Past Work
- Control of PHB in fed-batch cultures
- glucose concentration control (Kim et al., 1993)
- inferential control (Cornet et al., 1993)
- on-line optimal control (Lee et al., 1997)
- Control using cybernetic models
- data generator to identify model for MPC of a
bioreactor (Parker and Doyle, 1998)
5Motivation
- Cybernetic modeling
- captures complex cell regulatory mechanisms
(diauxic growth, storage synthesis) - model parameters identified from easily measured
quantities (substrate, biomass, etc.) - MPC with state estimation
- cannot measure PHB concentration on-line
- infer PHB from secondary measurements
- constraint handling
- multivariable structure
6Continuous Culture with Biomass Recycle
biomass recycle
dilution rate (glucose and ammonium feed)
PHB productivity
ideal separator (microfiltration)
reactor
glucose measurement
7Cybernetic Model Development(Varner, 1999)
- Lumped metabolic pathway
- S1, S2 glucose, ammonium sulfate
- ej enzymes
- p1, p2 metabolic precursors
- p3 storage polymer
- cr residual biomass
storage pathway model
p3
e5
e6
e3
e4
e2
e1
p2
p1
S2
S1
eg
cr
8Model Equations
Cybernetic variable regulating rate of p3
synthesis
9MPC Concept
- MPC prediction equation
- Calculation of control moves
past future
prediction horizon
10Inferential Linear MPC Algorithm(Lee et al.,
1992 Wisnewski and Doyle, 1999)
- Dual-rate state estimation
States disturbance states infrequent PHB
measurement frequent glucose measurement
11Simulation Case Study
- PHB productivity control
- process model/controller model mismatch
- unmeasured recycle drift, d
- measured glucose feed disturbance, ?
d
Process model
?
measurements
u
MPC with linearized controller model
PHB setpoint
12Process/Controller Model Mismatch
- Inhibition of biomass growth rate at high biomass
concentrations in process model
13Steady-state Operating Locus
- Controller model linearized at point on locus
- dilution rate 0.25 hr-1
linearization point
minimum phase
non-minimum phase
14Disturbances
- Step disturbance in glucose feed concentration
- Drift in recycle biomass concentration
15Disturbance Rejection and Reference Tracking with
Ideal Measurement Set
- Measurement of PHB every 2 hr with 1 hr delay
- Measurement of glucose and ammonium sulfate every
30 minutes
16Inferential Disturbance Rejection and Reference
Tracking with Glucose Measurement
- Measurement of glucose every 30 minutes
17Disturbance Rejection and Reference Tracking with
Off-line PHB measurement
- Measurement of glucose every 30 minutes
- Measurement of PHB off-line every 24 hours
18Wash-out Due to Persistent Recycle Drift
Disturbance
19Cybernetic Variable Profiles
- Cybernetic variables governing synthesis of
enzymes for PHB storage and biomass growth
20Cybernetic Variables Governing Enzyme Activity in
Batch Culture
- Cybernetic variables are non-smooth functions
- cybernetic variable modulating activity of
storage enzyme - Batch culture profile for v3
- Gradient based optimization?
21Open-Loop Batch PHB Production Optimal Ammonium
Initial Condition
- Optimization
- Optimization fails in 2-D search
22Summary
- Cybernetic model interfaced to dual-rate MPC
controller for PHB control in continuous culture - Steady-state offset observed for inferential
control case - Offset eliminated with one off-line PHB
measurement every 24 hours - Non-smooth cybernetic potentially problematic for
gradient based optimization
23Acknowledgments
- National Science Foundation (BES 9896061)
- Doyle group members