From Process Control to Business Oriented Operation

1
From Process Control toBusiness Oriented
Operation

• Ben Betlem
• University of TwenteFaculty of Science and
  Technology

2
Outline of presentation
MPC demands - integration

• traditional MPC
• principles
• four generations MPC
• current commercial products
• overview current state

• requirements
• market
• process dynamics of operation

• integration
• new structures
• problems
• future developments

3
Principes MPC
MPC demands - integration

• controlled
• variable/trajectory

manipulatedvariable
4
Problem definition MPC
MPC demands - integration

• Modelingmulti-variable, non-linear,
  constrained, changing dynamic behavior
• Optimumeconomic optimum usually on
  constraints? keep system on constraint without
  constraint violation

5
Hierarchy of Control System Functions
MPC demands - integration
MPC move the plant from one constrained steady
state to another while minimizing constraint
violations
6
4 Generations MPC
MPC demands - integration

• before 1972 Linear Quadratic Gaussian Controller
• 1st 1970/1980 (industrial development)
• hierarchy of optimization and control
• predicted output closely to first-order path
  (L-MPC), or closely to setpoint (DMC)
• impulse response (L-MPC), linear step response
  (DMC)
• ad-hoc constraints check
• heuristic solution (L-MPC), least-square solution
  (DMC)
• 2nd 1980/1985
• objective function re-written in quadratic
  programming form (QDMC)
• explicit way to implement soft input and output
  constraints

7
4 Generations MPC
MPC demands - integration

• 3rd 1985/1995
• fault tolerance recovering from infeasible
  sub-plant solutions
• avoiding large objective functions defining
  ranked hard and soft constraints and
  multi-objective functions
• introduction state space models with state
  estimation and model for unmeasured disturbances
  (SMOC)
• 4th 1995/
• windows
• multiple optimization levels with economic
  objectives
• robust control design consideration of model
  uncertainties

8
MPC Current Technology
MPC demands - integration

• drives process from one steady state to another
• MPC multivariable compensation
• MPC feedforward compensation
• prevent violation of constraints
• dynamic output optimization
• dynamic input optimization
• fault tolerance control as much of the plant as
  possible

9
MPC Calculation Procedure
MPC demands - integration

• read mvs, dvs, cvs values from process
• feedback by ad-hoc bias level adaptation
• state estimation only available in two products
• determine controlled subset mvs and cvs
• remove ill-conditioning
• mv move suppression by conditioning matrix
  inversion
• local steady state optimization
• steady state cvs as closely as possible to
  targets from local economic optimization
• as optimal targets change due to disturbances
• with exception of 2 products
• dynamic optimization
• output mvs

10
Commercial Products (Qin Badgwell)
MPC demands - integration
11
Linear MPC Commercial Products
MPC demands - integration
12
nonlinear MPC Commercial Products
MPC demands - integration
nonlinear optimization sequence of iterations of
linearized version
13
Requirements
MPC demands - integration

• Market Economic
• increase of capital turnaround for given margins
• minimal stocks
• direct coupling between productionand market
  supply
• direct coupling with raw material market
• Automation and Organization
• higher degree of automation
• operator task from PID-loop controller to quality
  controller to economic controller
• degrees of freedom and information assigned to
  the persons responsible to realize objectives

14
Goals
MPC demands - integration

• truly integrated optimization and control
• improvement of process economics
• matches process behavior
• respect imposed operating constraints
• for dynamic operation
• external adjustment
• to time varying supply chain and market economics
• to maximize added value
• internal adjustment
• to current state
• regarding changing process behavior

15
Levels of Process Dynamics
MPC demands - integration

• stationary
• scheduled or fluctuating load changes
• scheduled product quality changes
• polymer molecular weight distribution
• long-term process change
• (semi-)batch operation
• crystallization,
• cyclic (semi-) batch with limited exhaustion
• distillation, bio-reaction (variable final time)
• combinations of continues - batch operation

16
Dynamic Optimization
MPC demands - integration

• unscheduledload change
• scheduledquality change
• batch procesoptimal final time
• batch withrecyclesoptimal exhaustion

17
Integration Structures
MPC demands - integration
Conventional minimal coupling
Integrated separation of co-ordination
18
Integrated Structure Design
MPC demands - integration

• Criteria of centralization ? distribution of
  goals
• advantages overall above distributed
  optimizationdetermined by mutual dependency
  between units
• advantages dynamic above momentary
  optimizationdetermined by market and process
  dynamics
• Model, model adaptation, model consistency
• hybrid dynamic models using as much a-priori and
  realt-time knowledge as possible e.g. fuzzy
  modeling (van Lith)first principles dynamics
  empirical model components
• models derived from each other
• Quality determination
• dynamic partial least square

19
MPC demands - integration
20
Co-operation (van Brempt)
MPC demands - integration
yopt
uopt
mutual triggering possible
Dy
Du
y
u
21
Academic Results
MPC demands - integration

• Hybrid modeling and model reduction
• Plant wide optimization
• Development of new numeric methods for large
  prediction horizons
• Non-linear MPC for processes with large range of
  dynamics (stiff systems)
• Efficient numerical techniques available

22
Future Commercial MPC
MPC demands - integration

• introduction of dynamic optimization (batch)
• multiple objective functions
• integrating economic objectives
• infinite prediction horizon
• process models
• non-linear models from combined first principles
  and experimental data
• robust(accuracy/uncertainty versus control
  application)
• improved adaptation strategies
• algorithms
• inherently stable, avoiding extensive tuning
• robust numerical methods (from academics)

23
Literature

• Qin, S.J. and T.A. Badgwell (2003)A survey of
  industrial model predictive controlControl
  Engineering Practice, in press
• Incoop - Integrated control and dynamic
  optimization for the proces industry, Workshop
  23-24 jan. 2003, Düsseldorf
• Brempt, W.van, T. Backx, J.Ludlage, P.van
  Overschie, B. de Moor and R. Tousain (2001)A
  high performance model predictive controller
  application on a polyethylene gas phase
  reactorControl Engineering Practice, 9, pp
  829-835
• Lith, P.F. van, B.H. L. Betlem and B. Roffel
  (2002)A structured modeling approach for dynamic
  hybrid fuzzy-first principles models, Journal of
  Process Control, 12, pp 605-615

24
Stelling

• Als RT-procesoptimalisatie en MPC worden
  geïntegreerd om beter te kunnen inspelen op de
  supply chain, dan
• dient de operator bewaker te worden van de
  (locale) economische optimalisatie.