1. Ph.D. Summer school Process and Tools Integration Operability and Control for Process Integration 17. August 2005 Sten Bay Jørgensen CAPEC - Department of Chemical Engineering Technical University of Denmark, DK-2800 Lyngby, Denmark C A P E C

2. 8/8-2/9 2005 Operability and Control for Process Integration 2 Motivation for Process and Design Integration No recycle of information flow (arrow) – Integration possible ? Issue 1 Issue 2 Issue 1 Issue 2 Recycle of information flow – Integration possible? Sequential design of Heat integration Mass integration Control Integrated design of Heat and mass integration with control Requirement: Measures for dynamic consequences of integration to be used early in the design phase for control structuring and design

3. 8/8-2/9 2005 Operability and Control for Process Integration 3 Dynamics and Control of Integrated Plants Process dynamics and control – a recap! Transfer functions, dynamics and stability Process integration structures Effects of process integration on dynamics and control Analysis of linear behaviour Implications upon control Nonlinear behaviour Dynamic consequences of optimal operation How to configure control?

4. 8/8-2/9 2005 Operability and Control for Process Integration 4 Schedule for Operability and Control of integrated plants Lecture 1: Process dynamics and control recap 1 Lecture 2: Process dynamics and control recap 2 Lecture 3: Control of plants with units in series Lecture 4: Dynamics of integrated processes Lecture 5: Control effects of recycle Lecture 6: Effects of process integration and optimization

5. 8/8-2/9 2005 Operability and Control for Process Integration 5 Lecture 1: Process Dynamics and Control recap 1 Chemical Process Dynamics Simplified Material Balance Control

6. 8/8-2/9 2005 Operability and Control for Process Integration 6 Chemical Process Dynamics A → B A + B A B Heat Exchanger ReactorSeparator Standard process dynamics considers single simple standard units with linear dynamics expressed in transfer functions

7. 8/8-2/9 2005 Operability and Control for Process Integration 7 Material Balance P-Control exit flow

8. 8/8-2/9 2005 Operability and Control for Process Integration 8 Material Balance P-Control simulation

9. 8/8-2/9 2005 Operability and Control for Process Integration 9 Material Balance PI-Control

10. 8/8-2/9 2005 Operability and Control for Process Integration 10 Material Balance PI-Control simulation

11. 8/8-2/9 2005 Operability and Control for Process Integration 11 Material Balance P-Control inlet flow

12. 8/8-2/9 2005 Operability and Control for Process Integration 12 Material Balance PI-Control simulation

13. 8/8-2/9 2005 Operability and Control for Process Integration 13 Lecture 2: Process Dynamics and Control recap! Transfer functions and single loop control Internal model based control Performance limitations in single loop control Control of Production Rate in Chemical Plant Front end control (Push) On demand control (Pull)

14. 8/8-2/9 2005 Operability and Control for Process Integration 14 Transfer functions Local Transfer Function z i a zero in left half plane gives overshoot p j a pole in left half plane gives exponential decay Initially single variable transferfunctions are considered, i.e. all signals are scalars: g i (s) = n i (s)/d i (s) Transfer functions will also be divided into g(s)=g a (s)g m (s) where g m (s) is the minimum phase part and g a (s) is the allpass part, which contains all nonminimum phase components:

15. 8/8-2/9 2005 Operability and Control for Process Integration 15 Single Control Loop u d y gdgd g u y gdgd gcgc g d r Standard single variable open loop process: y = g u + g d d Significantly reduces sensitivity to disturbances at low frequences For high gain control the sensitivity to model uncertainty is significantly reduced Control performance is limited for RHP zeros, i.e. Nonminimumphase behaviour Standard single loop control:

16. 8/8-2/9 2005 Operability and Control for Process Integration 16 Internal Model Based Control Design u y gdgd g cIMC g d r -ĝm-ĝm The IMC regulator gives the closed loop: Thus the nonminimum phase in Ĝ a limits achievable performance!

17. 8/8-2/9 2005 Operability and Control for Process Integration 17 Control Performance Reducing Dynamics Local Transfer FunctionLocal Transfer Function Zero DynamicsZero Dynamics –Real zero in right half plane Singularities (due to sensitivity to uncertainty)Singularities (due to sensitivity to uncertainty) –Real pole into right half plane –Complex pole pair into right half plane

18. 8/8-2/9 2005 Operability and Control for Process Integration 18 Material Balance P-Control

19. 8/8-2/9 2005 Operability and Control for Process Integration 19 Plant Production Rate – Front End 1

20. 8/8-2/9 2005 Operability and Control for Process Integration 20 Plant Production Rate – Front End 2

21. 8/8-2/9 2005 Operability and Control for Process Integration 21 Plant Production Rate – Front End 3 Simple strategy

22. 8/8-2/9 2005 Operability and Control for Process Integration 22 Plant Production Rate - On Demand

23. 8/8-2/9 2005 Operability and Control for Process Integration 23 Lecture 3: Control of Plants with units in series Units in Series Disturbance effects Production rate – front end Production rate – on demand How to achieve changes in production rate Partial control Reactor control Examples of Production rate control

24. 8/8-2/9 2005 Operability and Control for Process Integration 24 Units in Series - No Recycle The plantwide control problem is greatly simplified when there is no recycle of mass or energy. The control system of each unit is configured individually to handle load disturbances. Separation Example Volatility order: A > B > C Direct Sequence: The lightest component is taken out of the top of the first column.

25. 8/8-2/9 2005 Operability and Control for Process Integration 25 Production Rate - Front End Disturbances propagate in the direction of mass flow

26. 8/8-2/9 2005 Operability and Control for Process Integration 26 Production Rate - On Demand Disturbances propagate in the opposite direction of mass flow

27. 8/8-2/9 2005 Operability and Control for Process Integration 27 Production Plant without recycles Production Plant without recycles An ideal abstraction since energy and rawmaterials are not used very efficiently!

28. 8/8-2/9 2005 Operability and Control for Process Integration 28 Production Rate Changes in production rate can be achieved only by changing the conditions in the reactor. Some variable that affects the reaction in the reactor must vary. Liquid Phase Reactors Hold-up Temperature Concentrations Gas Phase Reactors Pressure Temperature Concentrations

29. 8/8-2/9 2005 Operability and Control for Process Integration 29 Partial Control Often for reactors (and other units) the number of control objectives exceed the number of manipulated variables. We must assign manipulated variables to achieve the control objectives, which must be important for the operation of the plant and leave the rest of the variables uncontrolled.

30. 8/8-2/9 2005 Operability and Control for Process Integration 30 Plantwide Production Rate Control Production rate changes should be achieved by modifying the setpoint of a partial control loop in the reaction section. Separation section will not be significantly disturbed.

31. 8/8-2/9 2005 Operability and Control for Process Integration 31 Reactor Control Managing energy (temperature control) Keeping as constant as possible the composition and flow rate of the total reactor feed stream (Fresh feed and recycle).

32. 8/8-2/9 2005 Operability and Control for Process Integration 32 Units in Series - Production Rate How do we specify and control the plant- wide production rate of B, when there is a reactor in the plant? Reaction kinetics has to be considered! Sensitivities:

33. 8/8-2/9 2005 Operability and Control for Process Integration 33 Units in Series - Production Rate The production rate is controlled through partial control of the reaction rate. V controlled xA controlled T controlled (by ass.) Production rate may be changed by changing the setpoint to the reactor CC or the reactor LC. Reactor LC change will not change the composition fed to the distillation col. All three dominant reaction rate variables controlled => SMALL variance.

34. 8/8-2/9 2005 Operability and Control for Process Integration 34 Units in Series - Production Rate One dominant variable, xA, of the reaction rate is uncontrolled because reactor composition measurement is not possible. Reaction rate and production rate may fluctuate. Production rate may be changed by changing the setpoint to the reactor FC or the reactor LC. Rate set at front end. xA not controlled directly. This leads to larger variance in the production rate than in the previous configuration.

35. 8/8-2/9 2005 Operability and Control for Process Integration 35 Units in Series - Production Rate On-Demand: The production rate is specified by setting the FC of the bottom product in the distillation column. The disturbances propagates in the opposite direction of the mass flow. xA not controlled directly. This leads to larger variance in the production rate than in the first configuration.

36. 8/8-2/9 2005 Operability and Control for Process Integration 36 Lecture 4: Process Integration and Dynamics Process Integration Structures Series – has been covered Parallel Recycle Example Recycle Plant models Disturbance Sensitivity of Recycle plant

37. 8/8-2/9 2005 Operability and Control for Process Integration 37 Generic Production Plant Process integration is mandatory for energy and rawmaterial efficiency!

38. 8/8-2/9 2005 Operability and Control for Process Integration 38 Dynamic consequences of process integration g 1 (s)g 2 (s)g 3 (s) g 4 (s) Plant as an integration of different unit processes Relate behaviour of integrated plant to behaviour of individual units structure of interconnections Thereby existing knowledge of unit behaviour can be exploited, for the analysis of linear behaviour Hangos (1991) and Jacobsen (1999)

39. 8/8-2/9 2005 Operability and Control for Process Integration 39 Interconnection Structures g 1 (s)g 2 (s) g 1 (s) g 2 (s) SeriesParallel g 2 (s) g 1 (s) Recycle Zeroes and poles are the union of those of units Zeroes are moved Poles are the union of those of units Zeroes are the union of those of n 1 and poles of d 2 Poles are moved!

40. 8/8-2/9 2005 Operability and Control for Process Integration 40 Summary: Process Integration Structures Series and parallel interconnections: Realtively simple to deduce overall behaviour from unit behaviours (only zeros are affected in parallel interconnections). Recycle interconnections: More complicated relation between overall behaviour and unit behaviours (poles are moved).

41. 8/8-2/9 2005 Operability and Control for Process Integration 41 Simple Recycle Example (1)

42. 8/8-2/9 2005 Operability and Control for Process Integration 42 Simple Recycle Example (2) Laplace Transformation

43. 8/8-2/9 2005 Operability and Control for Process Integration 43 Simple Recycle Example (3)

44. 8/8-2/9 2005 Operability and Control for Process Integration 44 Simple Recycle Example (4)

45. 8/8-2/9 2005 Operability and Control for Process Integration 45 Simple Recycle Example (5)

46. 8/8-2/9 2005 Operability and Control for Process Integration 46 Simple Recycle Example (6) Both the time constant and the steady-state gain has been dramatically changed by the recycle stream Unit Step Response

47. 8/8-2/9 2005 Operability and Control for Process Integration 47 Snowball Effect Observation: Recycle systems has a large tendency to exhibit large variations in the magnitude of the recycle flow. Snowball effect: sensitivity of recycle flow rates to small disturbances

48. 8/8-2/9 2005 Operability and Control for Process Integration 48 Snowball Effect – Static analysis Snowball effect: sensitivity of recycle flow rates to small disturbances F, x F A=>B R, x R V B, x B L D Only show composition of reactant A, i.e. x All A is removed in Distillate, i.e. x B =0 and x D =1: Total balance: Component balance around reactor: Thus if Da = Vk/F approaches x F then R can become very large!

49. 8/8-2/9 2005 Operability and Control for Process Integration 49 Control Implications of the Snowball Effect Set the production rate at the front end, I.e. by setting U. If the snowball effect is dominant, K2*K >> 1, small changes in U lead to large changes in X2. Large changes in X2 implies that the recycle valve goes either fully open or closed. As X2 is large, X1 is also large and this may overload the separation section. Production rate can typically NOT be set at the front end for mass recycle systems.

50. 8/8-2/9 2005 Operability and Control for Process Integration 50 Snowball Effect - Example Isothermal reactor operation (perfect temperature control) Produce pure B Be able to manipulate the production rate of B Select a control structure that will meet these objectives

51. 8/8-2/9 2005 Operability and Control for Process Integration 51 Snowball Effect - Example All flows in recycle loop set by level controllers A small change in the production rate set front-end leads to large changes in the recycle loop flow rates. No plantwide control of inventory of A. SMALL flexibility index regarding production rate.

52. 8/8-2/9 2005 Operability and Control for Process Integration 52 Snowball Effect - Example We cannot manipulate production rate directly by manipulating the fresh feed flow The setpoint to the reactor LC is used to control production rate No snowball effect due to FC in recycle loop System inventory of A is controlled by the reactor LC. This improves the flexibility index.

53. 8/8-2/9 2005 Operability and Control for Process Integration 53 Snowball Effect - Example To prevent the snowball effect, the mass recycle loop must have a flow controller. The plant inventory of A must be controlled. It is not sufficient to control the individual unit inventories of A. In the upper flow sheet any disturbance that increase the total inventory of A in the process will produce large increases in the flowrates around the recycle loop.

54. 8/8-2/9 2005 Operability and Control for Process Integration 54 Snowball Effect - Example Consider a 20% production rate increase of B. In the first control structure the separation section must handle the entire load, as xA must change with 20%. The feed to the distillation column changes, as well as the feed rate. In the second control structure both reactor composition and volume changes. So the separation section sees a smaller load disturbance Production rate can only be changed by changing the conditions in the reactor!

55. 8/8-2/9 2005 Operability and Control for Process Integration 55 Disturbance Sensitivity of single loop control u d y gdgd g Standard single variable process: y = g u + g d d u y gdgd gcgc g Standard single loop control: g d g g c y = ----------d + ----------- r 1 + g g c 1 + g g c Significantly reduces sensitivity to disturbances at low frequences What happens with process integration? d

56. 8/8-2/9 2005 Operability and Control for Process Integration 56 Disturbance Sensitivity with process recycle yd gdgd g rec g = g d /(1 – g d g rec ) = S g d The Sensitivity function S = 1/ (1 – g d g rec ) catches the effect of recycle upon disturbance sensitivity. Instability is induced by recycle if g d g rec is stable and | g d g rec (iω c )| > 1 and φ(g d g rec (iω c )) = n 2π where ω c is the critical frequency Note feedback may be positive or negative Control is based upon negative feedback Recycle introduces positive feedback

57. 8/8-2/9 2005 Operability and Control for Process Integration 57 Feedback effects on Disturbance Sensitivity Negative feedback if | g d g rec (0)| < 0 Static Sensitivity |S(0)| < 1 Hence disturbance sensitivity is reduced at low frequences The critical frequency ω c > 0 – Increasing the loop gain will yield a pair of complex poles crossing the imaginary axis. The closed loop response usually is faster Positive feedback if | g d g rec (0)| > 0 Static Sensitivity |S(0)| > 1 Hence disturbance sensitivity is increased at low frequencies The critical frequency may be at ω c = 0 – thus a real pole crosses the imaginary axis for | g d g rec (0)| > 1, i.e. static multiplicity. Or at ω c = n 2π where a complex pair crosses. Thus the recycle loop response usually is slower if not unstable

58. 8/8-2/9 2005 Operability and Control for Process Integration 58 Example Plant Mixer ReactorSeparator F, x Fi xFxF A+R=>2R xRxR V B =R, x B L D=F, y D Note autocatalytic reaction, e.g. bioreactor Main disturbance: x Fi Objective: Maintain y D constant

59. 8/8-2/9 2005 Operability and Control for Process Integration 59 Example Plant – Unit models M Mixer – static: Reactor: Separator:

60. 8/8-2/9 2005 Operability and Control for Process Integration 60 Example Plant – Block Diagram 1-k grgr GDGD k L yDyD xFxF xRxR xBxB x Fi

61. 8/8-2/9 2005 Operability and Control for Process Integration 61 Example Plant: Disturbance Sensitivity 1-k grgr k yDyD xFxF zFzF xBxB x Fi g D12 g D22 Effect of x Fi on y D : Sensitivity S = 1/(1-kg r g D22 ) Static loop gain: kg r (0) g D22 (0) = 1.32 k thus positive feedback Unstable for k > 0.76 (R/F > 3.1)

62. 8/8-2/9 2005 Operability and Control for Process Integration 62 Summary on Sensitivity effects of Recycle Recycle of material or energy introduces positive feedback which increases low frequency disturbance sensitivity induces slower dynamics or instability Thus recycle implies a stronger need for control to reduce the effect of disturbances and also to stabilize the plant How to handle the increased disturbance sensitivity?

63. 8/8-2/9 2005 Operability and Control for Process Integration 63 Lecture 5: Control of Recycle Plants Feedback Control of Recycle Plants Control of variable in recycle path Control of variable not in recycle path Summary of control effects of recycle Conclusions on linear dynamics and control of Process Integrated Plants

64. 8/8-2/9 2005 Operability and Control for Process Integration 64 Feedback Control SISO versus recycle variable d y g gdgd Standard single variable process: y = g u + g d d Perfect rejection of disturbance requires: u = - (g d / g ) d u d y g gdgd Control of variable in recycle loop : y = ( gu + g d d)/(1-g d g rec )= S(gu +g d d) Perfect rejection of disturbance requires: u = - (g d / g ) d Thus required input unaffected by recycle u g rec

65. 8/8-2/9 2005 Operability and Control for Process Integration 65 Feedback Control of variable not in recycle 1 d x g 21 g 22 Control of variable not in recycle loop : u g rec y g 11 g 12 Thus the transfer function from u to y is affected by recycle! But how? u2u2

66. 8/8-2/9 2005 Operability and Control for Process Integration 66 Feedback Control of variable not in recycle 2 Recycle affetcs the static behaviour such that: 1. It will have more poles in the RHP than g 11 if g 22 (0)g rec (0) >1 and λ 11 (0) ≠1 2. It will have more zeros in the RHP than g 11 if g 22 (0)g rec (0)/λ 11 (0) >1 and λ 11 (0) ≠1. The above two conditions are sufficient for moving a real pole or zero into the RHP. Thus if g 11 is stable and nonminimum phase the above two conditions imply that the recycle system has RHP poles and RHP zeros respectively. In Conclusion: Closing a control loop from y to u will most certainly be affected by the dynamics introduced through recycle! The recycle transfer function:

67. 8/8-2/9 2005 Operability and Control for Process Integration 67 Plantwide Control Structure Design Procedure (Luyben et al.) Establish control objectives Determine control degrees of freedom Establish energy management system Set production rate Control production quality and handle safety, environmental and operational constraints Fix a flow in every recycle loop and control inventories Check component balances Control individual unit operations Optimize economics and improve dynamic controllability

68. 8/8-2/9 2005 Operability and Control for Process Integration 68 Summary on control effects of recycle Control of variables within the recycle loop Input required to reject a disturbance is unaffected by recycle Control of variable not within the recycle loop Input required to reject a disturbance is affected by recycle in fact the effect of control inputs relative to disturbance may decrease significantly. Recycle may introduce RHP zeros If acceptable control is not possible then redesign such that recycle loop gain decreases

69. 8/8-2/9 2005 Operability and Control for Process Integration 69 Conlusions on linear dynamics and control Plant dynamics may be strongly affected by recycles Recycle usually gives positive feedback increases low freqency sensitivity renders response slower or causes instability Controllability for variables outside the recycle loop may be severely reduced by recycle, i.e. reduced efffect of control inputs possibly combined with RHP zeros Recycle may significantly increase model uncertainty for units in plant compared to that of individual units (not shown). Remedy: Redesign loop to decrease loop gain. Often that means modify reactor design!

70. 8/8-2/9 2005 Operability and Control for Process Integration 70 Lecture 6: Effects of Process Integration on nonlinear behaviour The Control Hierachy and degrees of freedom Profit Optimizing Control Operational Implications Example: Continuous cultivation of yeast Analysis Experiment Example with Optimal operation of process integrated plant Ammonia reactor with feed-effluent heat exchange

71. 8/8-2/9 2005 Operability and Control for Process Integration 71 Profit Optimizing Control Productivity in Continuous Process:Productivity in Continuous Process: Optimality requires : Max JOptimality requires : Max J

72. 8/8-2/9 2005 Operability and Control for Process Integration 72 Gain Changes for X prod vs. F Output MultiplicityOutput Multiplicity –Dynamic Consequence: Instability when (dX prod /dF)<0 Input MultiplicityInput Multiplicity –Dynamic Consequence: May be a zero in RHP, i.e. unstable zero dynamics.

73. 8/8-2/9 2005 Operability and Control for Process Integration 73 Control Performance Reducing Dynamics Local Transfer FunctionLocal Transfer Function Zero Dynamics - input multiplicityZero Dynamics - input multiplicity –Real zero in right half plane Singularities - output multiplicitySingularities - output multiplicity –Real pole into right half plane –Complex pole pair into right half plane

74. 8/8-2/9 2005 Operability and Control for Process Integration 74 Process Analysis: Operational Implications of Optimality Complex behaviour may be encountered near an optimal operating point Optimised process integrated design increases the likelihood of complex behaviour Theorems based upon induction:

75. 8/8-2/9 2005 Operability and Control for Process Integration 75 Continuous Cultivation of Yeast Bifurcation analysis reveals: –Hysteresis curve, multiple steady-states at maximal biomass productivity! f 0.30.320.340.360.380.4 5 10 15 Biomass [g/L] Dilution rate [1/hr] Chemostat, Sf = 28g/L Stable Unstable

76. 8/8-2/9 2005 Operability and Control for Process Integration 76 Adaptive Model Predictive Control

77. 8/8-2/9 2005 Operability and Control for Process Integration 77 Response to Etanol Setpoint Changes

78. 8/8-2/9 2005 Operability and Control for Process Integration 78 Ethanol Concentration vs. Dilution Rate

79. 8/8-2/9 2005 Operability and Control for Process Integration 79 Ammonia Reactors Operating point: Feed temperature Feed concentration Feed flow rate Pressure No automatic control of inlet temperature 3-bed quench reactorsimple reactor

80. 8/8-2/9 2005 Operability and Control for Process Integration 80 Energy Integrated Ammonia Reactor 0123 5 10 15 20 Inlet Ammonia Mole Fraction [%] Outlet Ammonia Mass Fraction [%] Stable Steady State Unstable Steady State Hopf Bifurcation Stable Limit Cycle Unstable Limit Cycle  Subcritical Hopf bifurcation from the upper steady state  Stable limit cycle coexists with the upper stable steady state !Safer to operate in region with no stable limit cycle ! IIIIIIIVVVII

81. 8/8-2/9 2005 Operability and Control for Process Integration 81 Dynamic Simulation Inlet Ammonia Mole Fraction [%] Operate at ignited steady state and increase inlet concentration: –Passing Hopf at 2.3 mole% –Large amplitude oscillations Decrease inlet concentration –Passing cyclic fold at 2.1 mole% –Stable steady state 050100150200 1.8 2.0 2.2 2.4 2.6 2.8 Dimensionless time 300 350 400 450 500 550 Bed Outlet temperature [C] Hopf Cyclic fold

82. 8/8-2/9 2005 Operability and Control for Process Integration 82 Conclusions on nonlinear analysis New process design tools should be developed to account for possible nonlinear behaviours To operate near optimal operating points reliable model identification and nonlinear control is desirable - a profit margin of 3% has been estimated! Is a combined process and nonlinear control design optimization formulation solvable - to exploit the nonlinearity?

83. 8/8-2/9 2005 Operability and Control for Process Integration 83 General Plantwide Control Structure Design Procedure Top down analysis –Define operational objectives –Manipulated variables and degrees of freedom for control –Select primary controlled variables (given ¨via design goal) –Production rate: determine where to set this in the plant, often at some interior position –Investigate possible nonlinear complex behavioours near optimal operation Bottom up design –Regulatory control layer Stabilization Local disturbance rejection –Supervisory control layer Keep controlled outputs at optimal setpoints –Optimization layer identify active constraints and determine optimal setpoints –Validation simulations Extention of Skogestad (2004)

84. 8/8-2/9 2005 Operability and Control for Process Integration 84 Conclusions on Dynamics and Control of Process Integrated Plants Linear Analysis explains large sensitivity of recycle plants especially for control of variables not in recycle path. Optimizing Operation exploits nonlinearities, therefore nonlinear analysis is recommendable. Nonlinear Analysis explains specific cases – it is therefore difficult to generalise. It is however important to understand how to avoid occurrence of potentially serious problems.

85. 8/8-2/9 2005 Operability and Control for Process Integration 85 References and Further Reading Luyben, Tyreus, Luyben: Plantwide Process Control, McGraw-Hill (1998), chap. 1-3 Jacobsen, E.W.: On the dynamics of integrated plants – non-minimum phase behaviour. Journal of Process Control 9 (1999) 439-451 Skogestad, S. : Plantwide control: the search for the self- optimizing control structure: Journal of Process Control 10 (2000) 487-507 Skogestad, S.: Control structure design for complete chemical plants. Comp. and Chem. Engineering 28(2004)219-234.

86. 8/8-2/9 2005 Operability and Control for Process Integration 86 Monographs Buckley: Techniques of Process Control, Wiley (1964) Shinskey: Process Control Systems, McGraw-Hill (1988) Rijnsdorp: Integrated Process Control and Automation, Elsevier (1991) Luyben, Tyreus, Luyben: Plantwide Process Control, McGraw-Hill (1999) Ng, Stephanopoulos: Plant-wide control structures and strategies, Academic Press (2000)