FUNDAMENTAL MODELING AND CONTROL OF FALLING FILM EVAPORATORS by ZDRAVKO IVANOV STEFANOV, B.S., M.S. A DISSERTATION IN CHEMICAL ENGINEERING Submitted to the Graduate Faculty of Texas Tech University in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY Approved Chairperson of the Committee Accepted Dean of the Graduate School May, 2004
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
FUNDAMENTAL MODELING AND CONTROL OF
FALLING FILM EVAPORATORS
by
ZDRAVKO IVANOV STEFANOV, B.S., M.S.
A DISSERTATION
IN
CHEMICAL ENGINEERING
Submitted to the Graduate Faculty of Texas Tech University in
Partial Fulfillment of the Requirements for
the Degree of
DOCTOR OF PHILOSOPHY
Approved
Chairperson of the Committee
Accepted
Dean of the Graduate School
May, 2004
ACKNOWLEDGEMENTS
I am greatly indebted to my research advisor, Dr. Karlene A. Hoo, for her support
and advice. During the yeara apent under her guidance, ahe helped me to underatand
and appreciate the nature and the beauty of proceaa control and the value of mul
tivariate atatiatics. Dr. Hoo waa an excellent adviaor, and I would like to thank
her for the opportunity to work with her. I am also grateful for the advice and the
proofreading of my manuscripts and my disaertation. The experience with her was
one of the moat important experiences in my life.
Alao, I would like to thank Dr. Uzi Mann for being on my committee. Dr. Mann
waa very helpful with technical discussiona on the aapecta of the evaporator modeling.
I highly appreciate hia support for my laboratory inveatigations.
I would like to thank Dr. R. Tock and Dr. W. P. Dayawanaa for their willingnesa to
aerve on my comitee. Dr. Tock kindly asaisted me with my laboratory inveatigations.
Dr. W. P. Dayawanaa was extremely kind to provide advice. I would like to thank
him for the excellent experience I had in hia claaaroom.
I am grateful to Dr. D. Chaffin for his advice in the area of advanced computing. I
am alao grateful to Mr. M. Champagne and Tembec for providing me with a rewarding
graduate internahip at the Tembec Mill in St. Franciaville, LA. A special thanks to
Mr. Wayne McAdama, and Mra. Janice Hawley at the Tembec Mill in Crestbrook,
CA, for providing industrial data and valuable information used in the validation of
the evaporator models.
11
I would like to thank my fellow graduate studenta for their support and for the
friendly work environment.
I also appreciate the financial support of Petroleum Research Foundation and the
Dean's fellowship I received from the CoUege of Engineering in my first year.
Ultimately, I am alao indebted to my parents Lilia and Ivan for their deep under-
atanding and aupport.
Ill
CONTENTS
ACKNOWLEDGEMENTS ii
ABSTRACT xi
LIST OF TABLES xih
LIST OF FIGURES xv
I BACKGROUND AND INTRODUCTION 1
1.1 Evaporation 1
1.2 Evaporator typea 3
1.3 Modeling the evaporation procesa 4
1.4 Controlling the evaporation process 5
1.5 Evaporation of black liquor 6
1.6 Black liquor origin 6
1.7 Kraft pulp mill recovery cycle 7
1.8 Black liquor properties and the recovery boiler operation . . . 9
1.9 Falling film evaporators for black liquor concentration 10
1.10 Transport phenomena in falling films 11
1.11 Disaertation organization 12
II MODELING OF A SINGLE PLATE 14
2.1 Physical conditions 14
2.1.1 Mass flow rates of the liquor 14
2.1.2 Temperature and preaaure 15
iv
2.2 Hydrodynamica of the falling film at nominal conditions . . . . 15
2.3 Heat tranafer coefiicients of turbulent falling films 18
2.3.1 General description of turbulent flow in an open channel . 18
2.3.2 Heat transfer coefiicient of evaporation 22
2.3.3 Heat transfer coefiicient of condenaation 23
2.3.4 Heat transfer coefficient of senaible heating 24
2.3.5 Calculation of the dimensionless film thickness 24
2.4 Heat tranafer coefficients of wavy-laminar films 25
2.4.1 Transition from wavy-laminar to turbulent flow 25
2.4.2 Heat transfer coefficient of evaporation 27
2.4.3 Heat tranafer coefficient of senaible heating 32
2.5 Modehng of single plate 33
2.5.1 Dimenaionality 33
2.5.2 The physical phenomena . . . 36
2.5.2.1 Black hquor evaporation . . 37
2.5.2.2 Black hquor heating 44
2.5.2.3 Steam condenaation 46
2.5.2.4 Heat transfer at the wall 56
2.5.3 Dimensionleaa variables 59
2.5.4 Numerical Solution 60
2.5.4.1 Solution method 60
2.5.4.2 OCFE diacretization 62
v
2.5.4.3 Solver package 62
2.5.4.4 Phyaical parameter correlations 63
2.5.5 Validation of the aingle-plate evaporator model 63
2.5.5.1 Feed dry hquor concentration 64
2.5.5.2 Feed mass flow rate 66
2.5.5.3 Feed temperature 67
2.5.5.4 Steam pressure 70
2.5.5.5 Summary 71
2.6 Nomenclature 74
HI MODELING OF A SINGLE EVAPORATOR 77
3.1 Evaporator deaign 77
3.2 Liquor distributor 79
3.3 Plate atack 80
3.4 Evaporator inventory 82
3.4.1 Masa balance 82
3.4.2 Energy balance 84
3.5 In-line mixer 85
3.6 Splitter 86
3.7 Numerical aolution 86
3.7.1 Initialization 87
3.7.1.1 Caae 1-Evaporator cold startup 87
3.7.1.2 Caae 2-Evaporator hot atartup 87
vi
3.7.2 Calculation loop 87
3.8 ODE solver 88
3.9 Results and discussion 89
3.9.1 Feed flow rate 90
3.9.2 Feed concentration 91
3.9.3 Feed temperature 91
3.9.4 Steam pressure and vapor pressure 92
3.9.5 Summary 93
3.10 Nomenclature 99
IV CONTROL OF A SINGLE EVAPORATOR 101
4.1 Typical disturbancea 103
4.2 Senaor laaues 105
4.3 Decentralized control of a single evaporator 105
4.3.1 On aupply operations 106
4.3.2 Pairing of the manipulated and controlled variablea . . . . 108
4.3.3 Control structure 109
4.3.4 Closed-loop response of the evaporator 113
4.3.5 Limitationa of SISO loop controUera 118
4.4 Summary of decentralized control 119
4.5 Model predictive control (MPC) 120
4.5.1 Linear quadratic regulator 123
4.5.2 Dynamic matrix controller 125
vii
4.5.3 Quadratic dynamic matrix controller 129
4.6 MPC related modeling of single evaporator 132
4.6.1 Lumped model of a single evaporator 133
4.6.1.1 Plate atack 133
4.6.1.2 Bottom inventory 139
4.6.2 Validation of the nonlinear lumped parameter model . . . 139
4.6.3 A linear MPC deaign 139
4.6.3.1 Linearization 145
4.6.3.2 System-theoretic analyaia 148
4.7 Cloaed-loop reaults 151
4.8 MPC results 153
4.8.1 Reaulta 154
4.9 Nomenclature 172
V EVAPORATOR PLANT 173
5.1 Description 173
5.2 Modeling 177
5.3 Resulta and discusaion 179
5.3.1 Senaitivity to diaturbancea 180
5.3.1.1 Mass balance 180
5.3.1.2 Energy balance 189
5.3.2 Model vahdation 191
5.4 Nomenclature 192
viii
VI CONTROL OF MULTIPLE EVAPORATOR PLANT 193
6.1 Decentrahzed control 193
6.1.1 Evaporator inventoriea 193
6.1.2 Selection of controlled and manipulated variablea 196
6.1.3 Reaulta 197
6.2 Centralized control 201
6.2.1 Nonlinear ODE model 202
6.2.2 The effect of concentration diaturbancea 206
6.2.3 Linear model 210
6.2.4 Reaulta 211
6.2.4.1 Unconatrained and conatrained MPC controllers . . . 214
6.2.4.2 PI control 218
6.2.4.3 Control of the PDE system 219
VII SUMMARY AND FUTURE WORK 247
7.1 Modeling 247
7.2 Control 248
7.3 Decentrahzed control 249
7.4 Centralized control 250
7.4.1 Single evaporator 250
7.4.2 Multiple evaporator plant 251
7.5 Summary on the control reaulta 252
7.6 Future work 252
ix
7.7 Contributions 253
BIBLIOGRAPHY 255
APPENDICES 261
A: TURBULENT HEAT TRANSFER COEFFICIENTS 261
B: BULK FLOW MOMENTUM EQUATION 267
C: BLACK LIQUOR PHYSICAL PROPERTIES 271
D: COMPUTER PROGRAMS 274
D.l Matlab S-function for aimulation of aingle evaporator 274
D.2 Matlab S-function for simulation of multiple evaporators . . . 278
D.3 MAPLE code for linearization of single evaporator nonhnear ODE model 289
D.4 MAPLE code for linearization of multiple evaporator nonlinear
ODE model 291
E: OCFE DISCRETIZATION 307
ABSTRACT
Evaporatora are a common unit operation that can be found in many industries.
The evaporator plant, in the pulp and paper induatry providea a major role of regen
erating the proceaa chemicala from the fiber line waste liquor. The effectiveness of
the recovery, determinea the overall mill economy. Conaequently, the recovery cycle
must be fully operation becauae it ia unacceptable to diacard the waste liquor due to
ita highly negative effect on the aurrounding ecosystem.
The product of the evaporator plant, the concentrated black liquor aervea aa a fuel
to the recovery boiler. The recovery boiler is a combination of a chemical reactor and
a power boiler. The dry solida concentration of the black liquor affects the recovery
boiler performance not only from an economical point of view but also for aafety
reaaona. It is known, that if the dry sohds concentration of the black liquor falls
below a lower limit, there ia the poaaibility of an exploaion in the recovery boiler.
Evaporation of the waste hquor is usually accomplished in a multiple effect evap
orator plant. While there are more than one type of evaporator deaign, the most
modern and efficient deaign is the falling film plate evaporator. This design ia charac
terized with very high heat tranafer ratea at small temperature differences and high
reaiatance to acaling due to low reaidence times.
This research has two main objectives. The firat ia to develop a rigoroua distributed
parameter model of the falling film evaporator using the fundamental principlea of
masa, energy, and momentum conservation. The aecond is to synthesize an effective
XI
control atructure for the evaporator and the evaporator plant. A bench-acale ex
periment has ahown that one-dimenaional distributed model of the evaporator plate
is aatisfactory to describe the important transfer proceaaea on the plate accurately.
Additionally, it was confirmed by experimentation that two different hydrodynamic
regimes (turbulent and wavy-laminar) can exiat in the multiple effect black liquor
evaporator plants.
Inveatigationa into aimple and advanced control approachea have revealed that the
closed-loop performance of a proportional-integral-derivative (PID) controller design
in feedback with a aingle evaporator can provide satisfactory compenaation. How
ever, in the case of the entire evaporator plant, the advanced control approach of
model-predictive control (MPC) provides better control becauae the MPC central
ized controller can addreas multiple interactiona, input and output constraints, and
unmeasured diaturbancea.
This work preaents the development of the distributed parameter model and the
aynthesia of the control atructure; and demonatrates the performance of the cloaed-
loop aystem to meaaured and unmeasured disturbancea and parameter uncertainty.
Xll
LIST OF TABLES
2.1 Nominal parameters with respect to aingle plate 17
2.2 Predicted Reynolda numbers at the transition points for black hquor
at nominal conditions 26
2.3 Experimental conditiona for model dimenaionality experiment 35
2.4 Dimensionleaa Variablea 59
2.5 Operating Conditiona 63
3.1 Evaporator operating conditions 89
3.2 Step changea of evaporator operating condition 90
3.3 Evaporator aenaitivity to disturbances 90
4.1 Variable Selection and Clasaification 107
4.2 Control Loop Parametera 112
4.3 Evaporator nominal operating conditiona 113
4.4 Featurea of the proceaa closed loop reaponaes 119
4.5 Correlationa for Turbulent and Wavy-laminar regimea 136
4.6 Evaporator nominal operating conditiona 149
4.7 Tuning parametera for MPC 155
4.8 Tuning parameters for the PI controllers 155
5.1 Operating conditions of the multiple evaporator plant 179
5.2 Senaitivity of the dry aolids concentration 188
5.3 Changea in steam economy aa a function of the diaturbance 190
6.1 Operating conditiona of the multiple evaporator plant 199
6.2 Performance of the single-loop control atructure of the nonlinear mul
tiple evaporator plant 200
6.3 The effect of boiling point elevation to a 5% decrease in the feed dry
solids concentration 209
6.4 Timea to reach la hmits of 0.001 kg/kg of the multiple evaporator
plant product dry solids concentration 218
6.5 Tuning parameters of the MPC and PI controllers for multiple evapo
rator plant 220
6.6 Integral abaolute error of the controlled variablea 221
Xll l
B.l Dimensionleaa Variables 270
D.l Matrix A, columna 1 to 10 .
D.2 Matrix A, columns 11 to 20
D.3 Matrix A, columns 21 to 30
D.4 Matrix A, columna 31 to 32
D.5 Matrix B
D.6 Matrix C
D.7 Eigenvalues of Matrix A . .
300
301
302
303
304
305
306
XIV
LIST OF FIGURES
1.1 Schematic of a kraft pulp mill recovery cycle 13
2.1 Eddy diffuaivity of momentum for a falling water film 22
2.2 Heat transfer coefficient of wavy-laminar faUing fllm 31
2.3 Experimental deaign 34
2.4 Experimental aetup to determine flow distribution 47
2.5 Tracer patterna for water at 24°C, Re = 3557 48
2.6 Tracer patterns for water at 60°C, Re = 6970 48
2.7 Tracer patterna for SCMC at 24°C, Re = 0.699 48
2.8 Tracer patterns for SCMC at OO C, Re = 1.298 48
2.9 Preaence or absence of a heating zone 49
2.10 Differential volume of the evaporating film 50
2.11 Differential volume of the heating film 51
2.12 Differential volume of the condensing film 52
2.13 Wall differential volume element 56
2.14 The responae of the single plate evaporator to a 5% decrease in dry aolida concentration 64
2.15 The reaponse of the aingle plate evaporator to a 5% increaae in dry solids concentration 65
2.16 The reaponse of the single plate evaporator to a 5% decreaae in feed masa flow rate 67
2.17 The reaponae of the aingle plate evaporator to a 5% increaae in feed mass flow rate 68
2.18 The reaponae of the single plate evaporator to a 5% decrease in the feed temperature 69
2.19 The reaponse of the single plate evaporator to a 5% increaae in the feed temperature 70
2.20 The response of the aingle plate evaporator to a 5% decrease in the steam preaaure 71
2.21 The reaponae of the single plate evaporator to a 5% increaae in the ateam preaaure 73
3.1 Plate type evaporator aheme 78
3.2 Product dry aohda concentration to changea in the feed masa flow rate. 94
3.3 Product dry aolida concentration to changea in feed dry aohda concentration 95
XV
3.4 Product dry aohds concentration to changea in the feed temperature. 96
3.5 Product dry aohds concentration to changes in the ateam pressure. . . 97
3.6 Product dry aolida concentration to changea in the aecondary vapor aaturation temperature 98
4.1 Feedback control of the evaporator I l l
4.2 Closed-loop responae to a 5% increase in the feed flow rate 114
4.3 Closed-loop reaponae to a 5% decreaae in the feed dry sohda concentration 115
4.4 Cloaed-loop responae to a 5% decrease in the feed temperature. . . . 116
4.5 Cloaed-loop reaponse to a 10% increaae in the aet point of the dry solids concentration in the product stream 117
4.6 Evaporator reaponae to an increase of 10% in the dry solids concentration aet point 118
4.7 Nonlinear ODE model response to a 5% decrease in the feed flow rate. 140
4.8 Nonlinear ODE model response to a 5% increase in the feed flow rate. 140
4.9 Nonlinear ODE model response to a 5% increaae in the feed dry aolids concentration 141
4.10 Nonlinear ODE model reaponse to a 5% decrease in the feed dry solids concentration 141
4.11 Nonlinear ODE model reaponae to a 5% decrease in feed temperature. 142
4.12 Nonlinear ODE model responae to a 5% increase in feed temperature. 142
4.13 Nonlinear ODE model responae to a 5% increase in vapor preasure. . 143
4.14 Nonhnear ODE model reaponse to a 5% decrease in vapor preaaure. . 143
4.15 Nonhnear ODE model reaponse to a 5% increase in ateam preasure. . 144
4.16 Nonhnear ODE model reaponae to a 5% decrease in steam preaaure. . 144
4.17 Responses of the linear (LODE) and nonhnear (NLODE) ODE models to 5% decreaae in the steam pressure 150
4.18 Reaponaea of the linear (LODE) and nonlinear (NLODE) ODE models to 5% increase of the ateam preaaure 151
4.19 Reaponaes of the linear (LODE) and nonlinear (NLODE) ODE modela to 5% decreaae of the vapor pressure 152
4.20 Reaponsea of the linear (LODE) and nonlinear ODE modela to 5% increase of the steam preaaure 153
4.21 A feedback block diagram with the MPC controller 154
4.22 MPC performance in the presence of a 5% increase in the feed maaa flow rate 160
4.23 MPC performance in the preaence of a 5% decreaae in the dry aohda concentration of the feed 161
XVI
4.24 MPC performance in the presence of a 5% decrease in the feed temperature 162
4.25 Constrained MPC performance in the presence of a 5% increaae in the feed mass flow rate 163
4.26 Constrained MPC performance in the presence of a 5% decrease in the feed dry aohds concentration 164
4.27 Constrained MPC performance in the presence of a 5% decrease in the feed temperature 165
4.28 PI performance in the preaence of a 5% increase in the feed maaa flow rate 166
4.29 PI performance in the preaence of a 5% decreaae in the feed dry sohda concentration 167
4.30 PI performance in the preaence of a 5% decrease in the feed temperature. 168
4.31 Conatrained MPC performance in the preaence of a 6.6% decrease in the feed temperature 169
4.32 PI performance in the presence of a 6.6% decreaae in the feed temperature 170
4.33 Nonlinear PDE evaporator ayatem reaponse 171
5.1 Multiple falling fllm evaporator plant 176
5.2 Open-loop response of the PDE model of multiple evaporator plant to a 5% decrease in the flow rate of the feed 181
5.3 Open-loop reaponae of the PDE model of multiple evaporator plant to a 5% increase in the flow rate of the feed 182
5.4 Open-loop responae of the PDE model of multiple evaporator plant to a 5% decreaae in the dry aolids concentration of the feed 183
5.5 Open-loop reaponse of the PDE model of multiple evaporator plant to a 5% increeise in the dry aolida concentration of the feed 184
5.6 Open-loop reaponae of the PDE model of multiple evaporator plant to a 5% decreaae in the temperature of the feed 185
5.7 Open-loop reaponae of the PDE model of multiple evaporator model for 5% increase in the feed temperature 186
5.8 Open-loop reaponae of the PDE model of multiple evaporator plant to a 10% decreaae in the heat tranafer of the superconcentrator 187
5.9 Open-loop reaponse of the PDE multiple evaporator model for 10% decrease in the heat transfer of the superconcentrator 189
6.1 Syatem of three connected tanka 194
6.2 Decentrahzed feedback control atructure for the multiple evaporator plant 198
xvii
6.3 Closed-loop reaponae of the nonlinear multiple evaporator plant to 5% increase in the flow rate of the feed 200
6.4 Cloaed-loop responae of the nonlinear multiple evaporator plant to a 7% decreaae in the temperature of the feed 201
6.5 Cloaed-loop reaponae of the nonlinear multiple evaporator plant to a 5% decreaae of the dry aohda concentration of the feed 202
6.6 Reaponaea of nonlinear PDE (aolid line) and ODE (dotted fine) models of the multiple evaporator plant to a 5% decrease in the flow rate of the feed 203
6.7 Responses of nonlinear PDE (aohd line) and ODE (dotted line) models of the multiple evaporator plant to a 5% increase in flow rate of the feed. 204
6.8 Reaponaea of nonlinear PDE (aohd fine) and ODE (dotted line) models of the multiple evaporator plant to a 5% decrease in dry solids concentration of the feed 205
6.9 Reaponaea of nonlinear PDE (sohd fine) and ODE (dotted fine) models of the multiple evaporator plant to a 5% increaae in the dry sohda concentration of the feed 206
6.10 Reaponsea of nonlinear PDE (sohd hue) and ODE (dotted line) modela of the multiple evaporator plant to a 5% decreaae in the temperature of the feed 207
6.11 Reaponsea of nonlinear PDE (sohd line) and ODE (dotted line) modela of the multiple evaporator plant to a 5% increaae in the temperature of the feed 208
6.12 Reaponsea of the nonhnear (aohd line) and linear (dotted line) ODE models to a 5% decrease in the auperconcentrator steam pressurea. 212
6.13 Reaponaes of the nonhnear (aohd hue) and linear (dotted line) ODE models to a 5% increaae in the superconcentrator steam preaaurea. . 213
6.14 Reaponsea of the nonlinear (aohd line) and linear (dotted hue) ODE modela to a 5% decreaae in the ateam preaaure of the auperconcentrator evaporator SC-1 214
6.15 Reaponaea of the nonlinear (aohd line) and linear (dotted hue) ODE models to 5% increaae in the ateam pressure of auperconcentrator evaporator SC-1 ; 215
6.16 Reaponsea of the nonhnear (sohd line) and hnear (dotted hue) ODE modela to a 5% decrease in the aecondary vapor pressure of evaporator E-5 216
6.17 Responses of the nonhnear (sohd line) and linear ODE models to a 5% increase in the secondary vapor pressure of evaporator E-5 217
6.18 Closed-loop performance of the unconstrained MPC controller in the presence of a feed maas flow rate disturbance 222
6.19 Unconatrained MPC control actiona in the preaence of a feed maas flow rate disturbance 223
XVll l
6.20 Closed-loop performance of the unconatrained MPC controUer in the preaence of a feed dry aolida concentration diaturbance 224
6.21 Unconstrained MPC control actiona in the presence of a feed dry aolids concentration diaturbance 225
6.22 Closed-loop performance of the unconatrained MPC controller in the preaence of a feed temperature disturbance 226
6.23 Unconatrained MPC control actions in the presence of a feed temperature disturbance 227
6.24 Closed-loop performance of the unconstrained MPC controller in the presence of a heat tranafer diaturbance 228
6.25 Unconatrained MPC control actiona in the presence of a heat transfer disturbance 229
6.26 Cloaed-loop performance of conatrained MPC controller in the preaence of a feed mass flow rate diaturbance 230
6.27 Constrained MPC control actions in the presence of a feed mass flow rate diaturbance 231
6.28 Cloaed-loop performance of constrained MPC controller in the preaence of a feed dry solids concentration disturbance 232
6.29 Constrained MPC control actiona in the preaence of a feed dry solids concentration diaturbance 233
6.30 Cloaed-loop performance of conatrained MPC controller in the presence of a feed temperature disturbance 234
6.31 Constrained MPC control actiona in the presence of a feed temperature diaturbance 235
6.32 Cloaed-loop performance of constrained MPC controller in the preaence of a heat tranafer diaturbance 236
6.33 Conatrained MPC control actions in the presence of a heat tranafer disturbance 237
6.34 Closed-loop performance of the PI controUera in the preaence of a feed mass flow rate disturbance 238
6.35 PI control actions in the presence of a feed mass flow rate disturbance. 239
6.36 Closed-loop performance of the PI controUera in the presence of a feed dry solida concentration disturbance 240
6.37 PI control actiona in the presence of a feed dry sohda concentration diaturbance 241
6.38 Cloaed-loop performance of the PI controllers in the preaence of a feed temperature disturbance 242
6.39 PI control actions in the preaence of a feed temperature disturbance. 243
6.40 Closed-loop performance of the PI controUera in the presence of a heat transfer diaturbance 244
6.41 PI control actions in the presence of a heat transfer diaturbance. . . . 245
XIX
6.42 Cloaed-loop reaponae of the PDE model to a feed mass flow rate
diaturbance uaing the unconstrained MPC controller actiona 246
A.l A fluid element in the fully developed falling fllm 261
B.l Force balance on a fluid element in the faUing film 267
E.l OCFE discretization of the spatial domain. 307
XX
CHAPTER 1
BACKGROUND AND INTRODUCTION
Evaporation is a fundamental process operation found in many diverse induatriea.
In this chapter, the baaics of the evaporation proceaa will be diacuaaed, with some
specifics aa they relate to the evaporator units used in the pulp and paper industry.
There are more than one type of evaporator, however, this work will be restricted
to the falling film plate type evaporator and the process fluid of interest is the black
liquor that is the effiuent of the kraft proceaa of the pulp milla [1]. The important
tranaport phenomena that dominate the evaporation proceaaea aaaociated with the
evaporation of the black hquor process fluid wUl be discussed from both a theoreti
cal, experimental, and practical points of view. Understanding theae phenomena is
tantamount to the development of an accurate phenomenological model and stable
controller deaign to produce a product with proper quahty (dry aohda concentration-')
that is eaaential to the economica of the plant.
1.1 Evaporation
Evaporation ia the process of aolvent removal in the form of a vapor from either
solution, suspension or emulsion. The objective can be to concentrate the solution,
to regenerate the solvent, or both. In many caaea, however, the objective ia to con
centrate the solution. In these cases the solvent vapor is not regarded as product
and it may or may not be recovered [2, 3]. Evaporation has been used to recover
^The dry solids are all solids that remain after complete evaporation of the water in the black liquor
1
either diaaolved aohda (salts, organic material) or the pure hquid (deaahnation of sea
water) [3]; to concentrate diluted streama and thus to improve their value (milk and
juice concentration [3]); or to prepare a atream for future treatment (black liquor
concentration [1], Bayer alumina process liquor concentration [4]). There are areas
in the induatry, where the evaporation is preferred to other aeparation methoda, for
example the radioactive waste treatment [5, 6].
The process of evaporation is similar to other aeparation operationa such aa dia-
tillation, drying, and crystallization. All deacribed processes involve vaporization.
The evaporation process differs from the distillation process in that the task of the
former is not to separate the vapor components. In the case of a drying process, the
evaporation process ia different because the product is alwaya a liquid. Finally, the
evaporation process differs from a crystallization process because in the latter the
objective is crystal growth while in the former the objective is to concentrate the
aolution [2].
Uaually, the evaporative proceaa ia performed in unit operations caUed evaporators
that need to accomplish certain tasks [2]. These tasks include:
• To achieve a designed extent of separation, or in other words to provide a flnal
product with certain solids concentration.
• To provide large heat transfer with the smalleat possible heat transfer surface.
• To provide efficient energy usage, which is measured by the evaporator's utility
(eg. steam) economy. The utility economy is the maaa units of water evaporated
by one mass unit of utility per unit of time.
• To be compatible with the processed working fluid (hquor), which in many cases
can be corrosive, radioactive, etc.
1.2 Evaporator types
There are several types of evaporators that are commercially available [2, 3, 7].
Evaporators can be classified by the circulation type, design, and direction of the
flow. The type can be natural or forced; the construction, vertical or horizontal tube
bundles or plate stacks with respect to the heating elements; and the flow can be
either in the direction of the gravity or otherwise. Often, devices are installed in the
evaporator to improve the distribution of the process fluid. All these design features,
when combined, lead to a wide variety of evaporators. The most popular designs
however, are the rising fllm long tube vertical (LTV) evaporators and the falling film
tube or plate evaporators.
The LTV evaporator's heating element ia a long tube bundle that enda with a
vapor space at the top and the process fluid inventory at the bottom [1, 2, 3]. The
feed is supplied at the bottom of the evaporator and the process fluid rises inside
the tubes. The tubes are heated by condensing steam. At a given spatial point, the
process fluid starts to boil. The vapors leaving the tubes cause the procesa fluid to
riae, climbing on the tube walls in a fashion similar to annual flow. In some cases,
a circulation stream is present depending on the amount of liquor that needs to be
processed.
The heating element of the faUing film type evaporator can be either a tube
bundle or a plate stack. In either case, the process fluid enters the heating element
from the top of the evaporator. Usually a distribution device ia neceasary to provide
uniform distribution of the process fluid. The process fluid flows by gravitational
forces in a manner similar to annual fllm flow or flow past a vertical flat plate. Since
the residence time on the plates is small (on the order of tens of seconds), forced
circulation ia almost always a necessity for this type of evaporator.
1.3 Modehng the evaporation process
A survey of the open literature produced lumped parameter steady-state models
of the evaporation process [2, 3, 7]. Also, there appeared to be a scarcity of dynamic
models of the evaporation process but a few dynamic modela of the evaporator system.
For example a fundamental lumped parameter dynamic model, published by Niemi
et al. [8, 9], was used to represent a multiple evaporator plant in the pulp and paper
induatry. Thia model waa extended by Ricker and SeweU [10] and Cardoao [11]. A
lumped model of flash type evaporators was pubhshed by To et al. [4].
The lumped parameter modehng has certain advantages, such as small compu
tation times and relative simplicity. However, since the evaporator system is a dis
tributed system, the lumped parameter model may not provide an accurate descrip
tion of the system phenomena in the spatial direction. For the evaporation process,
the most common evaporator types are the LTV and faUing fllm evaporators. Based
on their construction, these evaporators have a dominant distributed character. In
the case of the LTV evaporator well-mixedness and therefore a lumped parameter
representation can be assumed but only in the evaporator inventory and in the sec
tion of the tubes where the process fluid is below its boiling point. However, in the
rest of the tube space this assumption is not vahd. It is well known [12, 13, 14] that
up to seven different flow types can be observed in the rest of the tube space. A
distributed parameter model of a rising fllm evaporators was developed by [12].
Cardoso [11] modeled a falling film concentrator using a lumped parameter ap
proach. However, the physics of the process imphed a distributed character of the
system since the process fluid flows down onto the heat transfer surface and well-
mixedness cannot be assumed. In the work by Schutte et al. [15], a distributed
parameter model of a tube type falling film evaporator was described. However, no
model equations or detailed description of the modeling approach were provided.
1.4 Controlhng the evaporation process
Since the evaporation process is an important part of an industry such as the
pulp and paper industry, it is important to provide stability of the operation and
consistent quahty of the product. In certain cases, the stabihty of the operation
may be very significant issue. For example, in the evaporative processes used in
radioactive wastes the expectation is that near perfect separation should be achieved.
A survey of the open literature shows that the most widely employed control strategy
is the decentralized strategy of single loop feedback controllers implemented using
a proportional-integral-derivative (PID) control law [2, 3, 16]. In recent years, with
the rapid improvements in the computing industry, advanced control concepts such
as model based and model predictive control can be used to control these systems
[17, 18, 19]. An example can be found in the work of Gil et al. [20] where a multiple
evaporator plant is modeled using a lumped parameter model and controlled by a
linear quadratic regulator. An illustrative example of generalized predictive controller
can be found in [21] that employs a very simple lumped parameter model to develop
the GPC.
1.5 Evaporation of black liquor
The evaporation of the black liquor is a operation that is common in the pulp and
paper industry.
1.6 Black liquor origin
The black liquor is a spent hquor that originates from the pulp miU fiber hue.
The pulp miU's task is to produce hardwood or softwood pulp used in paper pro
duction, yarn production (rayon fibers) or other chemical processes (production of
carboxymethylceUulose, smokeless gunpowder production, food additives, etc.). The
pulp is produced from wood chips that are cooked in units called digesters. The
resulting unbleached pulp (brown stock) is washed from the cooking chemicals and
either used for paper production or bleached using a sequence of reactors and then
re-used for paper production or in other processes.
The black hquor is the spent pulp liquor that is collected from the washing process.
During the washing process, the cooking chemicals are transferred from the brown
stock to the black hquor. The color of the hquor is a result of the organic compounds
dissolved from the wood during the cooking process. The composition of the black
hquor depends on the cooking process type (sulphite, sulphate (kraft process)^, and
others), the cooking conditions, and the wood type.
1.7 Kraft pulp mill recovery cycle
The kraft process has several significant advantages when compared to the other
pulping processes. The two main advantages are: (i) improved physical properties
of the produced pulp and (ii) effective recovery of the cooking chemicals. The kraft
pulp process uses a solution of sodium hydroxide and sodium sulphide, called a white
liquor. The total alkali content is measured as units of sodium oxide (Na20) and
the sulphide content is about 15 to 40 % of the total alkali content. When the wood
is pulped and the non-cellulose organic part of the wood (hgnin) is dissolved, the
sulphur is transferred from the inorganic part of the liquor to the organic one. The
resulting pulp and cooking liquor is washed and the resulting black liquor is supplied
to the initial point of the recovery cycle (Figure 1.1), the black hquor evaporators.
The black hquor is then concentrated to a dry sohds concentration (from 0.65 to
0.75-0.80 kg/kg) and used as a fuel by the recovery boiler. In the recovery boiler, the
organic part of the black liquor is used as an energy source and steam is generated.
^The sulphate process is referred to as a kraft process, since the produced pulp has a higher mechanical strength when compared to the sulphite process. The word kraft has a German origin, which means strong.
The inorganic part faUs to the bottom of the recovery boiler and forms a bed of smelt.
In the smelt bed, the important recovery reactions occur in a reduction atmosphere.
In the boiler bed, the sodium hydroxide from the black hquor is transformed to
sodium carbonate and the sulphur is transformed to sodium sulphide. The resulting
smelt is discharged from the boiler to a special disaolving tank. Any ash carryover is
separated from the flue gases by the electrostatic precipitator (see Figure 1.1).
The smelt is dissolved in the dissolving tank and the resulting solution, called a
green liquor is transported to the causticizing process. In the causticizing process,
the green liquor is treated with calcium oxide (quick lime). Sodium carbonate is
transformed to sodium hydroxide, while calcium oxide is precipitated as calcium car
bonate. The suspension is separated and the resulting liquid is the white liquor that
is recycled back to the fiber line cooking process; at this point the recovery cycle is
closed. The hme mud that results from the separation is concentrated to a paste using
rotary drum filters and processed in the lime kiln to regenerate the initial calcium
oxide.
The above described recovery cycle is very effective with 90 to 95 % efficiencies.
However, such efficiencies are achieved when the proper operating conditions are
assured. The role of the recovery boiler cannot be underemphasized, it is the main
unit in the recovery cycle since it recovers not only the chemicals, but also energy
from the black liquor organic compounds. Without question, it is imperative that
stability of the recovery boiler operation must be guaranteed.
1.8 Black hquor properties and the recovery boiler operation
Since the black liquor is a fuel source for the recovery boiler, its quality affects
the boiler operation to a great extent [1, 22, 23]. The main black hquor characteristic
that affects the recovery boiler operation is the dry solids concentration in the hquor.
The dry solids concentration determines the amount of water to be evaporated in
the boiler combustion chamber before the combustion can begin. It is necessary to
achieve black liquor drying and stable firing simultaneously. The black liquor is fired
as small droplets produced by swirl or splash type burners. It has been shown that
the size distribution of the droplets determines how fast the droplets will be dried.
As the droplets are dried a solid (residue) is formed. It is important that only this
solid reach the boiler smelt bed. If instead, the wet droplet reaches the bed, the bed
temperature decreases which in turn reduces the efficiency of the chemical reduction.
In contrast, if the droplets are dried too far from the bed, the residue is carried out
of the combustion chamber with the flue gases resulting in chemical losses and fume
formation. The latter is extremely unwanted because the fumes accumulate on the
boiler heat transfer surfaces thereby reducing the heat tranafer rate.
The droplet size distribution has been shown to depend on the black liquor vis
cosity and consequently on the black hquor dry solids concentration. Therefore, it is
important to have a consistent dry solids concentration of the black liquor feed for
the recovery boiler.
1.9 FaUing fllm evaporators for black hquor concentration
The use of the falling film evaporators for black liquor concentration has numerous
advantages.
• FaUing fllm evaporators may be used even when small temperature differences
exist [1, 2]. The relatively high heat transfer coefficient makes them a suitable
choice and smaller temperature differences mean less energy consumption.
• Falling film evaporators can be operated at decreased loads up to 25% from
the nominal without experiencing scaling problems [1]. This permits a final dry
solids concentrations of 65-70%, which is significantly more, compared to the
LTV evaporators, where 50-52% is the usual final product dry solids concen
tration. The falling film evaporators are known to experience much less scaling
compared to LTV evaporators because the liquor resides on the plates for a short
amount of time and the evaporators operate at smaller temperature differences.
• Falling fllm evaporators provide very small carryover with the condensates [1].
When plate type evaporators are used the carryover is as low as 1 ppm Na20.
• Falling film evaporators are suitable for processing viscous liquids [2]. Concen
trated black liquor falls in this class of hquids with viscosities up to 3-5 Pa.s.
All these advantages determine the falling film evaporators as the most advanced
solution in the black liquor evaporation. Therefore this kind of evaporators is being
studied in the present work. Additional motivation is the lack of fundamental first
10
principles models in the open literature. Good fundamental model can be used for
control, design or optimization purposes. It was therefore decided to develop a first
principles distributed parameter model that wiU provide detaUed description of the
falling fllm evaporator. The distributed parameter representation was accepted to be
unavoidable, as already discussed above.
1.10 Transport phenomena in faUing films
A literature survey shows that the transport phenomena in faUing films have been
studied extensively. The work of Kapitza [24] is the first significant work in this area
and it is focused on describing the wavy nature of falling films and the important
transport phenomena that occur. Kapitza's work has been extended by Levich [25],
for example.
Since the work of Kapitza focused on wavy-laminar falling films and the major
ity of the industrial applications involve turbulent falling films, new experimental and
theoretical resulta appeared that cover both turbulent and wavy-laminar regimes. Im
portant experimental inveatigationa were contributed by Wilke [26]. His results were
used by Limberg [27], Chun and Seban [28, 29] and Seban and Faghri [30] to validate
their theoretical work. Chun and Seban provide additional experimental results to
improve the understanding of falling fllm phenomena. For instance that the falling
film structure is more comphcated [31, 32] than originally assumed [24]. Important
developments in the area of turbulent flows past flat surfaces are incorporated in the
faUing film theoretical framework [27, 33, 34].
11
Other relevant theoretical investigations to improve the understanding of the
faUing film phenomena include [35, 36, 37, 38, 39, 40]. In the case of experimen
tal work, see [41, 42, 43], the range of experimental conditions is extended to cover
more viscous liquids, which is the focus of this dissertation.
1.11 Dissertation organization
Thia diasertation ia organized as foUows. Chapter 2 develops a fundamental model
of the important processes that occur on a plate in a faUing film plate type evaporator.
Chapter 3 builds on the results of Chapter 2 by extending the model to represent one
complete falling film plate type evaporator. In both cases, the models are validated
using a set of expected changes in the feed parameters and utility stream. Chapter
4 introduces, develops, and demonstrates closed-loop control of a single falling film
evaporator. The performance of two control strategies are investigated and compared.
The first is a decentralized structure that consists of single-loop feedback controllers,
and the second is a more sophisticated centralized control strategy, model predictive
control. In Chapter 5, a fundamental model of a multiple effect falling film evaporator
plant (seven evaporators) is developed and validated. Subsequently, the control of
the multiple evaporator plant is examined in Chapter 6 uaing the same two control
strategies tested in Chapter 4. Lastly, Chapter 7 summarizes the work, lists the
contributions, and provides directiona for future work.
12
a >> o >,
> o o
a, 1—I
c3
a
zn
CO !-< pi
13
CHAPTER 2
MODELING OF A SINGLE PLATE
2.1 Phyaical conditions
In a multiple effect evaporation system, the term effect is frequently used to rep
resent an evaporator unit that has its own vapor and liquid space and its own heating
source.^ The physical conditions of the system are determined by the plant operating
parameters, i.e., pressures, temperatures, flow rates etc. Since the values of these pa
rameters change from effect to effect, their limits define the feasible operating regimes.
Below, a discussion of the limits and their combined effect on the physical phenomena
(eg. heating, evaporation, condensation) is presented.
2.1.1 Mass flow rates of the liquor
The total feed and respective product mass flow rates depend on the pulp miU
throughput and the evaporation intensity. For a pulp mill with throughput of approx
imately 250 000 tons/y, it is not unusual to process about 500 tons/hr weak black
liquor. The product mass flow rate can be in the range of 50 to 80 tons/hr. However,
in the presence of disturbances, or startup and shutdown procedures these values may
change.
^In this study, the term effect will be used even if the evaporator unit shares its vapor space with another evaporating unit. The other restrictions still hold.
14
2.1.2 Temperature and pressure
The temperature in the condensation and evaporation zone of each effect is de
termined mainly by the pressurea in theae zonea. In the case of condenaation, the
temperature is the saturation temperature of water vapor at the given pressure. In
the case of evaporation, the temperature is also affected by the boihng point rise
(BPR). The fresh steam pressure is usually in the range of 0.4-0.5 MPa. The pressure
of the secondary vapors of the last effect is in the range of 15 to 30 kPa. The values
of the pressures in the intermediate effects are between 15 kPa and 0.5 MPa.
2.2 Hydrodynamics of the falling film at nominal conditions
The hydrodynamics of the falling film is closely related to heat transfer effects.
Thus, it is critical to determine the hydrodynamic conditions of each effect to choose
the proper correlations to calculate the correct values of the heat transfer coefficients.
The hydrodynamics of the falling film can be described using three dimensionless
numbers, that describe different relations between the forces acting on the fllm
Reynolds number, Prandtl number and Kapitza number [43, 44]. The Reynolds
number, given by
4r Re=— (2.1)
represents the ratio between the inertial and the viscous forces acting on the fluid
where F is the mass flow rate per unit width of the channel and /x is the dynamic
15
viscosity. The Prandtl number, given by
Fr = ^ (2.2)
represents the ratio of momentum transfer due to convection and thermal conduction
where fi is the kinematic viscosity of the fluid, A is the fluid thermal conductivity and
c is the fluid heat capacity. Finally, the Kapitza number
4
Ka = ^ (2.3)
describes the competition between capillary (creation of waves) and viscous forces
(attenuation of waves) at the film surface where g is the gravitational acceleration
conatant, p ia the fluid denaity, and a is the fluid surface tension. The values of these
dimensionless numbers, together with the nominal conditions of the multiple effect
evaporator system are given in Table 2.1 where Pyap is the vapor pressure, Tyap is the
vapor temperature, Gpi is the black liquor mass flow rate at plates inlet and Xpi is
the black liquor dry solids at plates inlet.
From Table 2.1, it can be observed that the flow conditions change dramatically
from effect 5 to the last section of the auperconcentrator (SCI). The Reynolds num
ber decreases by about two orders of magnitude; the Prandtl number increases by
about the same order; and the Kapitza number increaaea by about eight orders of
magnitude. Thus, the flow characteristic in effect 5 is turbulent, with the transfer of
16
Table 2.1: Nominal parameters with respect to single plate.
Effect
5 4 3 2
SC3 SC2 SCI
P Pa ^ vap •) ^ ^
21372 40269 65768 115870 204996 204996 204996
T K
337 351 364 381 402 407 411
Gpi,kg/s
2.7 4.3 2.9 2.3 4.5 3.5 2.5
Xpi,kg/kg
0.2261 0.2152 0.2716 0.3804 0.5144 0.6527 0.7020
Re
2022 4793 2710 1297 1100 102 25.5
Pr
27.01 17.89 21.01 33.19 73.48 605 1697
Ka
9.2X-S 1.9X-8 3.9X-8
2.7x-^ 7.0X-6 3.4X-2
2.0
heat dominated by the eddy momentum transport. In SCI, the flow characteristic is
wavy-laminar [28]. The Prandtl number changes dramatically when going from effect
5 to SC3 to SC2, and to SCI. The values of the Kapitza number suggest that the
capillary forces are larger than the viscous forces in effect 5, resulting in wavy film
characteristics. In SCI, the viscous forces dominate and wave-like behavior of the
film is suppressed.
It is reasonable to conclude that two flow characteristics-wavy laminar, and
turbulent-may exist in a multiple evaporator plant. To account for this different be
havior, different correlations to calculate the heat transfer coefficients must be used.
The following discusses these three cases and the selected correlations to determine
the heat transfer coefficients properly.
17
2.3 Heat transfer coefficients of turbulent faUing films
2.3.1 General description of turbulent flow in an open channel
In turbulent flow, all transport processes, such as momentum, heat and mass
transfer, are dominated by the eddy viscosity (e), also known aa eddy viscosity of
Boussinesq [44]. The eddy viacoaity is the analog of the dynamic viacoaity term found
in Newton'a law of viacoaity. The expression for turbulent flow is given by [44]
_( tdvj; dVj; (2.4)
where r* is the Reynolds stress, p is the density of the fluid and e is the turbulent
equivalent of the kinematic viscosity of the fluid. Assuming the mechanics of the
motion of the eddiea are analogous to the motion of molecules of a gas, the following
expression for the wall shear stress was derived by Prandtl [45]
•yx = -pP
dvx
dy
dvx dy '
(2.5)
where I is the Prandtl mixing length.^ Combining Equations (2.4) and (2.5) and
transforming to dimensionless variables, gives the eddy diffusivity of momentum as
[43]
= /+2 dii+
dy^ (2.6)
^The Prandtl mixing length is similar to the average length of the free path of a gas particle.
18
where u+ is the dimensionless film velocity, y+ is the dimensionless distance normal
to the waU, and the other variables are as before. Therefore, to model turbulent
flow, expressions for the mixing length (l) and the velocity gradient [du-^/dy^) are
necessary. The derivation of these expressions ia closely related to the structure of
the turbulent flow in an open channel. The flow can be divided into three regions
According to Bird [44], the relationship for the mixing length in the wall boundary
layer region is linear and can be obtained from the following expression,
l^ = ky+, (2.7)
where k is the Von Karman constant, which is assigned a value of 0.4. The value
of Von Karman's constant is experimentally determined to be in the range of 0.36
< A; < 0.4 based on velocity profiles in round tubes [44]. In the turbulent core region,
the dependence deviates from linear behavior, and according to Limberg [27], this
deviation can be accounted for by multiplying the right hand side of Equation (2.7)
by the following function -CLy+
F = e 5+ , (2.8)
where 5"*" is the dimensionless film thickness and c is a constant to be determined.
Dampening of the turbulence at the wall boundary layer is accounted for by using
19
the Van Driest damping function [33]
D^ = l-e ^t , (2.9)
where A+ is aaaigned a value of 26 that correaponds to the best fit to the experimental
data used in [33].^ Thus, the expression for the mixing length in the waU boundary
layer and the turbulent core is given by
Z+ = ky+Dy,F. (2.10)
Combining Equations (2.10) and (2.6), gives an expression for the eddy diffusivity of
momentum
(2.11) - = k'y^'DlF' dy^
An expression for the velocity gradient is obtained by using the bulk flow mo
mentum equation for fully developed conditions and zero vapor shear, which after
integration^ is given by [43]
- ! ^ - ( ^ - ^ ) ^ - ( - ^ )
^In the same source [33], for another independent set of experimental data, the value of A+ is found to be 27. One may conclude that this parameter is almost constant.
"^The integration procedure is provided in Appendix B
20
Near the wall, this expression can be approximated by
l « ( l + - ) ^ (2.13)
y+ since close to the wall — <C 1. It follows that the expression for the eddy diffusivity
of momentum in the wall boundary layer and the turbulent core is then given by [43]
l = _ l + i y r f 4 ^ 2 ^ T 2 ^ p ^ . (2.14)
In the free boiindary layer region, Alhusseini et al. [43] developed an expression
for the eddy diffusivity of momentum, based on absorption data obtained by Chung
and Mills [48], and Won and Mills [49]. The expression is given by
e 1.199 X 10-16 Re ™ / ( ^ + - y +
where Af is an empirical function of the Kapitza number that corrects for the as
sumption of a smooth surface when the correct representation is that of a wavy
surface.
The final expression for the eddy diffusivity of momentum in the film is given by,
The differences in the predictions of the transition Reynolds number by these
correlation make the application of these correlations tenuous.
One alternative is to use the turbulent model of Alhusseini et al. [43] together with
an empirical correlation for the wavy-laminar regime [41]. The proposed expression.
Equation (2.25), uses the calculated heat transfer coefficients for both turbulent and
wavy-laminar regimes, thus providing a smooth transition
hE,totai = {h%^i + h%/^. (2.25)
It is shown in [41] that this expression fits the experimental data for all Reynolds
numbers between 30 and 15 000, including the transition region. A similar expression
26
for condensation is given by [50]
hE,total = {fh'E,i + h%t)h (2.26)
where / is a correction factor that has a value of 1.15. Bird et al. [44] propose expres
sions with the same structure for the combined Nusselt number when accounting for
free and forced convection heat transfer in the turbulent boundary layers. Since it is
common to determine the heat transfer coefficient by applying a superposition of the
combined effects (e.g., Equations [2.25] and [2.26]), it is reasonable to use Equation
(2.25), provided by Alhusseini et al., to calculate the heat transfer coefficients for
evaporation, condensation, and sensible heating over the entire range of operating
conditions.
2.4.2 Heat transfer coefficient of evaporation
The heat transfer of evaporating or condensing films, in the laminar region, has
been a well studied subject beginning with the work by Nusselt [44, 52]. In the case of
a vertical tube in which a saturated vapor is condensing, Nusselt derived the following
equation
where AH^v is the latent heat of vaporization, L is the channel length, and AT is
the temperature difference between the wall and the film. It was shown that this
expression underpredicts the value of the heat transfer coefficient by as much as 50%
27
[52]. The discrepancy was attributed to the fact that falhng films do not have smooth
interfaces, but rather wavy ones. The first significant attempt to describe the motion
and heat transfer in wavy falling films was made by Kapitza [24]. In that work, it
is assumed that the waves on the film surface are sinusoidal. The mass, momentum,
and energy balances are then solved to obtain approximate expressions (second order
approximations) for the velocity distribution, film thickness, wave ampHtude, etc.
Kapitza also proposed the following equation for the Reynolds number when the
flow character changes from laminar to wavy-laminar.
Ret,i = 2.43 ^ J 7 ^ . (2.28)
Additionally, Kapitza calculated that the heat transfer coefficient of sensible heating
should be 1.21 times larger than the value calculated at laminar conditions.
An experimental study by Kapitza confirmed the proposed theoretical develop
ments even though large errors (25-50 %) between the calculated and measured values
of the phase velocity and wavelength were found. It can be concluded that the theory
proposed by Kapitza does not represent the underlying phenomena rigorously. In
deed, the discrepancy between the theory and the experiments can be observed in the
photographs, taken by Kapitza, of the water and alcohol falling films. There, large
sohtary waves with non-sinusoidal form are visible. Moreover, at higher flow rates,
the solitary waves are seen to be foHowed by one or two smaU waves.
Other studies have taken a pure experimental approach. Chun and Seban [28] have
28
proposed the following correlation for the local heat transfer coefficient of evaporation
/ ^3 \ 5 / - pX -0-22
hEM = 0-QOG[-J) [-) (2.29)
Chun and Seban find that this correlation fits their experimental data satisfactory for
the evaporation of water and water-glycol mixtures. However, the above correlation
(Equation (2.29)) is vahd for a narrow range (1.77 to 5.7) of Prandtl numbers.
For a complex Hquid such as black liquor, it has been shown (see Table 2.2) that
Chun and Seban experimental correlations may not be appropriate.
The rapid advances in computing hardware have made solving very computational
intensive calculations possible. Thus, there are a number of accurate numerical so
lutions to the fundamental momentum, mass and energy balances for wavy films.
Alekseenko et al. [35] solve the Navier-Stokes equation for falling films using the
Karman-Poulhausen method of integral relations. This method assumes a parabolic
velocity profile (for details on the application of Von Karman integral balance equa
tions see Bird et al. [44]) for the falhng film. The results of Alekseenko et al. show
close similarity between the calculated and experimentally observed velocity and wave
profiles. A similar study by Chang et al. [36] provides another solution to the prob
lem. Here, too, good agreement was found with the experimental data provided by
Stainthorp and Allen [53].
A general characteristic of these studies is the occurrence of large irregular (e.g.,
varying amplitude, asymmetric) waves on the film surface. Research by [54, 55]
29
conflrmed this observation by direct measurements. Mudawar and Houpt [54] used
LASER-Doppler velocimetry to study water and water-glycerol falling films. Jayanti
and Hewitt [55] simulated, in great detail, different wave forms and confirmed known
experimental results about the increase of the local heat transfer coefficient. The
reason for the increase in the heat transfer is attributed to the transport of a large
part of the fluid in the observed solitary waves. Thus, the average film thickness
decreases, hence the increase in the value of the heat transfer coefficient.
Many of the studies have a common and significant drawback, they are computa
tionally intensive. For example, the study by Chang et al. [36] required about 500
hours of processing time on a Convex machine to calculate the film evolution on a
plate of approximately 2 meters and a duration of about 250 seconds. Such demand
ing computational loads continue to challenge even today's computing technology. A
compromise between the detailed models and the computing burden is to develop
more effective engineering correlations. Such an approach is found in the studies of
Miyara and Miyara and coworkers.
In Miyara and Uehara [38], wavy falling films were solved numerically from the
perspective of flow dynamics and heat transfer capabilities. This study confirmed
that the heat transfer enhancement is mainly due to a decrease in the average film
thickness. In a foUow-up study by Miyara [39] the influence of the Prandtl number
on the heat transfer coefficients of condensation and evaporation were investigated.
Since the range of Prandtl numbers studied was between 0.1 to 100, the results found
is of particular interest to the present study. Miyara concluded that for a Reynolds
30
number of 100, an increase in the Prandtl number in the range cited resuhed in an
increase of ^ 56% in the amount of heat transferred (see Figure 2.2).
1.6
1.35
o Miyara's data Power law fit, h*/h = 1.431Pr"
20 40 60 Prandtl number
80 100
Figure 2.2: Heat transfer coefficient of wavy-laminar falling film. The graph shows the relative increase of the heat transfer coefficient of evaporation/condensation found by Miyara [39].
In Miyara [40], a very useful comparison between the empirical correlations of
Kutateladze [56], Chun and Seban [28], and Uehara and Kinoshita [40] is provided.
These correlations, however, are functions of only the Reynolds number. However,
the Reynolds number neither accounts for the influence of the capillary forces nor
the heat conduction. Unfortunately, the study by [40] did not provide a correlation
that reflects the influence of the Prandtl or Kapitza numbers on the heat transfer
coefficient of evaporation.
31
In the present study, the correlation, obtained experimentally by Alhusseini et al.
[41, 42] and given by
hsM = 2.65i?e-°-i5«i^a°-°^63 (2.30)
will be used. This choice is dictated by the fact that this correlation accounts for the
influence of the Kapitza number even though it does not account for the influence
of the Prandtl number. The correlation is shown to be valid by Alhusseini et al.
[41] using experimental data in the Prandtl number range of 1.7 to 47. According
to Miyara (see Figure 2.2), the influence of the Prandtl number is significant in the
range of 0.1 to 10. For higher values, the influence is much smaller as can be seen in
Figure 2.2.
2.4.3 Heat transfer coefficient of sensible heating
The information on the sensible heating of wavy-laminar falling films in the open
literature is considerable less than the information on the condensation and evapo
ration. Heating without evaporation of wavy films is not well studied either exper
imentally or theoretically. One of the few reliable sources is the work of Kapitza
[24].
As mentioned in section 2.4.2, Kaptiza determined experimentally and theoreti
cally that the heat transfer coefficient of sensible heating is 1.21 times greater than
that for pure laminar conditions. If this information is to be used, it is necessary
to calculate the heat transfer coefficient of sensible heating for pure laminar regime.
This problem already has been extensively studied and solved initially by Pohlhausen
32
[57]. The analytical solution is based on the solution of the differential energy balance
of incompressible flow without heat generation [14, 44, 58]. The general expression
to calculate the heat transfer coefficient is given by.
1 „ 1 hH,y,i = 0.332Re-^Pr-^ (2.31)
It then follows that to account for wavy interface influence, the final correlation should
be,
hHM = 0.402Re'^Pr^. (2.32)
2.5 Modeling of single plate
2.5.1 Dimensionality
The problem of model dimensionality is of importance in any modeling application
especially when the system is of the class of distributed parameter systems. Thus,
it is necessary to determine how many spatial dimensions should be included in the
model to represent satisfactorily the behaviors of interest.
The models, describing fahing films usually are two-dimensional, nonlinear, and
very complex [35, 36, 40]. However, such systems of equations, while complete often
pose large computational overhead. In the modeling of falling film evaporators, there
is general agreement on neglecting the details of the film itself It is enough to
represent the spatial (distributed) behavior of the film but the dimensionality of the
its distributed nature (axial or plate direction or both axial and transverse directions)
33
must be established.
In this respect, an experiment was performed to determine the number of spatial
dimensions. The experiment employed a tracer to determine if there are any signifi
cant velocity components in the transverse direction. The experimental device, shown
in Figure 2.3 was constructed to introduce the tracer into heated or non-heated falhng
films. The device consists of a tin plate with a dimple. The size of the dimple was
Flow distributor
Temperature sensor
Lip actuators
Temperature controller
Yty- Relay
~ Power, 12 V, 10 A 0
Heating element
Tin plate
J
Rotameter F
Manual valve
Centrifugal pump
Figure 2.3: Experimental design.
34
chosen to be close to that found in real industrial evaporators. A flow distributor
is placed at the top of the plate to create a uniform falling film. The flow distribu
tor is equipped with a bending hp connected to six actuators for adjustment of the
opening of the lip, thus assuring film uniformity The liquid flows down the plate
and is collected into a tank where it is recirculated using a centrifugal pump (Teel
1P799A, Dayton Electric Mfg. Inc.). The flow rate is measured using a Brookfield
Instruments' rotameter with tube (R-8M127-1) and float (8-RS-14). The tin plate
is mounted onto a glass plate, heated from the opposite side by a resistance wire
heating element (power of about 1200 W at 12 volts). The temperature of the plate
is measured with a temperature sensor that is controlled by an on-off temperature
controller. The controller hysteresis was adjusted to assure ± 1 K deviation from the
set point. The dimensions of the tin plate and dimple are on Figure 2.4.
Table 2.3: Experimental conditions for model dimensionality experiment.
Temperature, °C Fluid F, kg/s.m p, mPa.s Re
24 60 24 60
Water Water SCMC SCMC
0.819 0.819 0.916 0.916
0.921 0.470 5237 2821
3557 6970 0.699 1.298
The experiment was performed at two different temperatures, 24°C and 60°C. Two
different liquids were used, water and a water solution of sodium carboxymethylcel-
lulose (SCMC). The solutions of SCMC are very viscous even at low concentrations.
The viscosity, depending on the brand of SCMC may reach 10 000 mPa.s for 1-2%
35
solutions^ The objective was to investigate the two limiting conditions observed in
the evaporation of a black liquor fluid (see Table 2.1). The experimental conditions
for each of the four tests are presented in Table 2.3. The viscosity of the water was
calculated fiom tabulated data and the viscosity of the SCMC at room temperature
was measured using Cannon-Fenske A919-500 viscometer. The viscosity of SCMC at
60°C was calculated using the data provided in [59].
The results of the experiment are shown in Figures 2.5 to 2.8.
It is clear, that the velocity component in the transverse (horizontal) direction
is negligible in the case of SCMC; there is no tracer spreading in the horizontal
direction. In the case of water, at Re = 3557, a very smaU dispersion of the tracer
in the transverse direction can be observed. At Re = 6970, there is a significant
dispersion in the transverse direction when the tracer is introduced at the edge of the
dimple. However, such high Reynolds numbers are not present in the system studied
here (see Table 2.1). Based on the experimental studies, it is concluded that a one-
dimensional falling film model is satisfactory to represent the behaviors of interest.
2.5.2 The physical phenomena
The two main phenomena that occur on the plate are evaporation of the black
liquor at the liquor side and condensation of water vapors at the steam side. De
pending on the temperature of the liquor that enters the evaporator, there are two
possibilities. If the temperature of the entering liquor is below the boiling point at
^When the water-SCMC solution was used, the flow rate was measured by using a standard volume vessel and a stopwatch
36
the operating pressure, then a heating zone will exist on the plate. In this zone, the
liquor is heated until it reaches its boUing temperature. However, if the temperature
is greater than the boihng point temperature at the operating pressure conditions, the
liquor that enters the evaporator flashes before the temperature decreases to the boil
ing point temperature. Thus, the hquor enters the plate at the boiling temperature.
These two possibilities are shown in Figure 2.9.
There are three main physical processes that occur on the plate evaporation and
heating of the black liquor and condensation of water. Additionally, heat transfer at
the wah cannot be neglected. The fohowing is adopted from Stefanov and Hoo [60].
2.5.2.1 Black liquor evaporation
The film evaporation is modeled using a differential volume approach. Figure 2.10
presents the differential volume in the black liquor film, where Gs is the mass flow
rate of the solids, G^ is the mass flow rate of the water, W is the mass flow rate
of the evaporated water, Q is the heat entering the differential volume, 5 is the film
thickness, a is the film(plate) width and Az is the differential volume length. The
following general assumptions have been made.
Assumption 2.5.1. The average film velocity in the vertical direction is independent
of the horizontal position parallel to the plate. Thus,
^ = 0. (2.33)
37
Assumption 2.5.2. There is no nucleate boiling.
The assumption is reasonable because high temperature differences are not favored
in the evaporation of the black liquor. When nucleate boiling is present at high
temperatures, it is hypothesized that the accompanying high mixing rate increases
collisions between the safl molecules permitting the formation of crystals [61]. This is
the primary reason for evaporator scaling, which decreases heat transfer significantly
[1, 61, 62].
Assumption 2.5.3. There are no significant changes in the black liquor parameters
in the differential volume.
Assumption 2.5.4. There is ideal mixing. Thus, the heat of dilution of the black
liquor can be excluded.
This assumption holds for a single plate, as it has been shown that the total error
in the evaporation energy requirements in the concentration of softwood black liquor
(from 20% to 65-90% dry solids) is about 4-6% [63, 64]. The change in the dry sohds
for a single evaporator is about 10%, thus, the influence of the heat of dilution can
be neglected.
Assumption 2.5.5. The pressure at the vapor and steam side does not change sig
nificantly due to friction losses.
According to Perry [65] section 11-110, the pressure drop in a falhng film evapo
rator is negligible; and the pressure over the falling film is essentially the same as the
vapor head pressures.
38
Energy balance of the differential volume
In the development of the energy balance, the mass flow rate of the black hquor
is divided in two parts mass flow rate of the solids, Gs and the mass flow
rate of the water, Gy, (see Figure 2.10). The overall differential volume element
energy balance can be written as
dF - ^ = Q-Ws + GsU {hs \z ^\vl + gz) + Gy, I {hy, I +\vl + gz)
Taking the limit a Az —>• 0, the energy balance is given by
f + „ . ^ = M ^ (2.59) ot oz pocp^i
The assumption that Cy^i w Cp^i (recall Cp — Cy = R, the universal gas constant)
is used; thus the difference between the heat capacities at constant volume and
constant pressure can be neglected.
2.5.2.3 Steam condensation
The model describing the condensation process is similar to that developed for
the evaporation process. The assumptions used above are presumed to be valid here.
The differential volume with the important flows and variables is shown in Figure
2.12.
46
102.5
72
344
Figure 2.4: Experimental setup to determine flow distribution. The dimensions of the dimpled plate are in mm. The size of the dimple is comparable with the size of the dimples in the real evaporator.
47
; , < I '
h ^ '•
- • 1 '
i t-
7
Figure 2.5: Tracer patterns for water at 24°C, Re = 3557. . . . . . ^
y 1
v ),wr
: >
Figure 2.6: Tracer patterns for water at 60°C, Re = 6970. it'' '
y
| ]
Figure 2.7: Tracer patterna for SCMC at 24°C, Re = 0.699
/ J 1
Figure 2.8: Tracer patterna for SCMC at 60°C, Re = 1.298.
48
'ledrculation < Tjat Iredrculation ^ 'sat
Condensation zone
Steam
13 :l' Liquor | inlet
Vapor
Flash zone ^ \
Evapqrator wall
Heating zone
Evaporation * - zone
. ^ Vapor
I Condensation " ^ zone
Steam
Liquor inlet
Evaporator wail I
• - Evaporation zone
Vapor
Plate Plate
Figure 2.9: Presence or absence of a heating zone.
49
X
^ W , Z ) ^ S , Z 5 V2
A
plate
r AZ~^ W
a
liquor film
' Gw,z+Az , G ,z Az 5 V
Figure 2.10: Differential volume of the evaporating film.
50
Tz+Az 5 Vz
Figure 2.11: Differential volume of the heating film.
51
L condensate film
W.
Gz+Az5 Vz
Figure 2.12: Differential volume of the condensing film.
• Energy balance of the differential volume element
The overall energy balance of the differential volume is given by
The substitution of these dimensionless variables in the model equations leads to
the following dimensionless systems of PDE's
ffP* dT* -L^<-^-^T{K-T*) dt* dz* (2.79)
_,+,*^^ = W,VzVc
59
and dG' .dG'
An analysis of the dimensionless parameter ??, shows that when Go ^ 0, 77 -^ oo
and when Go -^ 00, 77 ^ 0. In general, for very high feed flow rates there will be
no evaporation, while at very small feed flow rates the evaporation will be almost
instantaneous. This is consistent with the physics of the process. Also, W^ is a weak
function of Go. The influence of L, the length of the plate is clear. That is, there
will be no evaporation at zero plate length and very high evaporation rates at large
values of L.
In the case of rje, the mass flow rate at the top of the plate is zero because liquid
condensate is not present at the top. In this case Gc,o is taken to be a sufficiently
small number (0.001 kg/s) that approximates the zero mass flow rate at the top of
the plate.
2.5.4 Numerical Solution
2.5.4.1 Solution method
Solving of the mathematical models that describe the physical phenomena that
occur on a single plate is a challenging task. The system of equations are nonlinear,
partial differential and algebraic equations (PDAEs)with non-constant parameters.
For example, at the evaporation side, the film velocity, heat transfer coefficient, den-
60
sity, and heat capacity are functions of the black liquor mass flow rate, dry sohds
concentration, and black liquor feed temperature. Thus, it is very difficult to find an
analytical solution to the system of equations.
A number of methods exist for solving systems of PDEs with two independent
variables space and time. These methods include finite differences, finite element,
orthogonal collocation, Galerkin methods [67] to name a few. The choice of a suitable
method for the problem in this work was based on the considerations of computational
efficiency and the high probability of a sharp shock-like solution profile [36].
The efficiency of the method is a consideration because the final goal is to solve
a high dimensional system of PDAEs for an entire multiple effect evaporator plant.
The computational efficient constraint excluded the usage of the Galerkin method
(requires numerical integration at each step) even though this method is capable of
finding steep solution profiles [67]. The orthogonal collocation method is the method
of choice because it more computationally efficient.
Within the class of orthogonal collocation methods, the method of Orthogonal
Cohocation on Finite Elements (OCFE) is used because it has been proven to be
very suitable for solving problems with steep solutions [67]. This method, however,
has the inconvenience of reducing the system of PDEs to a system of differential-
algebraic equations (DAEs) and only a few commercial solvers are capable of solving
this type of problem accurately.
61
2.5.4.2 OCFE discretization
The OCFE scheme for this problem includes 10 finite elements with 2 internal
collocation points, which gives a total of 31 spatial points. In 21 points, the solution
is calculated by integration of 21 ODE's for each PDE in the PDE system. The
solution at the rest of the points is calculated by satisfying the algebraic conditions
for continuity of the solution's first derivatives at the element boundary points. So,
for the case of heating of the black liquor, the DAE system includes 93 equations and
for the case of evaporation the DAE system includes 124 equations. Of course, for
each spatial point the wall temperatures should be calculated, which adds another 31
algebraic equations to each case. The detailed discretization structure can be found
in Appendix E.
2.5.4.3 Solver package
The solver used to solve the discretized systems of PDEs is the LSODI (Livermore
Solver for Ordinary Differential equations. Implicit systems) solver, included in the
ODEPACK package by Painter and Hindmarsh [68, 69, 70]. The LSODI solver han
dles well-behaved DAE systems (low index). The system in this study has an index
of 1 (c/. ODE systems are of index 0). For more information on DAE properties and
the determination of their indices the reader is referred to the work of [71].
62
2.5.4.4 Physical parameter correlations
Accurate solutions of the model equations require reliable correlations for all phys
ical parameters. In the case of black liquor, only a hmited number of correlations exist
and the accuracy of their predictions may be difficult to establish against real mill
data since the latter themselves are variable as they are functions of the raw material
properties, composition, and cooking conditions. The correlations used in this study
correspond to the general behavior of kraft black liquor. The correlations can be
found in the Appendix C.
2.5.5 Validation of the single-plate evaporator model
The operating conditions for a single-plate evaporator system are given in Table
2.5.
Table 2.5: Operating Conditions
Parameter Nominal Value
Liquor feed mass flow rate 58 kg/s Liquor feed temperature 303.15 K Steam pressure 47000 Pa Vapor pressure 21372 Pa Plate mass flow rate 2.53 kg/s
To determine the validity of the single plate evaporator model, the system is
subjected to a set of expected and unexpected disturbances. These disturbances
include changes in the temperature, dry solids concentration, mass flow rate of the feed
and the heating source (steam) inlet temperature. The disturbances are introduced
63
as perfect step functions within the range of ± 5% of their nominal values.
The results are shown in Figures 2.14 to 2.21.
2.5.5.1 Feed dry liquor concentration
338
6 8 Plate length, m
Figure 2.14: The response of the single plate evaporator to a 5% decrease in dry sohds concentration. -I-: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit sohds concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
A change of the dry solids concentration in the feed stream results in changes in
64
6 8 Plate length, m
Plate length, m
10 12
Initial state Final state
12
Figure 2.15: The response of the single plate evaporator to a 5% increase in dry sohds concentration. -I-: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit sohds concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
the variables of interest (hquor mass flow rate, temperature, condensate mass flow
rate) with the largest effect on the dry solids concentration on the plate. A moving
fiont can be observed (see Figure 2.14) travelhng from the top to the bottom of
the plate. The residence time on the plate is approximately 7 seconds [60]. The
esponding change in the exiting dry solids concentration is approximately the corr
65
same as the change in the input dry solids concentration.
The effect of this disturbance on the other variables is very small. These results
are consistent with the physics of the process. The heating zone increases since the
amount of water to be heated to the boiling point has increased and water has a larger
heat capacity than the black liquor sohds. The decrease in the boihng point rise due
to the decrease in the inlet dry solids concentration leads to an overall decrease (small)
in the boihng point temperature of the hquor. The almost negligible effect of a change
in the feed dry solids concentration on the other variables is attributed to the natural
compensation by the system itself
2.5.5.2 Feed mass flow rate
Changes to the feed rate do not appear to affect the variables of interest signif
icantly. The temperature responses (Figures 2.16 and 2.17) show that the heating
zone increases for either a positive or negative change in the feed mass flow rate. An
explanation for this behavior is related to the properties of the system. If the mass
flow rate decreases, the heat transfer coefficients decrease resulting in an increase in
the length of the heating zone. In the opposite case, the increase in the amount of
liquor to be heated wih require an increase in the length of the heating zone. The
effect on the dry solids is very small. This is again attributed to a self-compensating
effect by the system. Note, that the single plate evaporator system appears to be
more sensitive to an increase in the feed rate as compared to a change of the same
size in the the feed dry solids concentration. Another consideration is that the evap-
66
orator recirculation ratio is about six, therefore any changes observed at the exit of
the plate may be amplified by the same factor or more.
JO
O
0.2
0.1
0 2 4
Plate length, m
1
6 1
8 1
10
-
1 Plate length, m
Figure 2.16: The response of the single plate evaporator to a 5% decrease in feed mass flow rate. +: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit solids concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
2.5.5.3 Feed temperature
Changes in the feed temperature show significant effects on the variables of
67
Plate length, m
Figure 2.17: The response of the single plate evaporator to a 5% increase in feed mass flow rate. -I-: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit sohds concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
interest. A 5% increase in the feed temperature leads to a complete disappearance of
the heating zone; flashing occurs before the feed reaches the plate. The corresponding
increases in the dry solids concentration and mass flow rate can be observed readily.
Note that the changes occur with almost the same magnitude in both the dry sohds
concentration and the mass flow rate.
68
When the feed temperature is decreased the variables of interest show more sen
sitivity as compared to an increase in the feed temperature primarily because the
heating zone has increased approximately one meter in length. Equivalently this
translates to approximately a 10% decrease in the heat transfer surface available for
Plate length, m
Figure 2.18: The response of the single plate evaporator to a 5% decrease in the feed temperature. +: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit sohds concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
69
337cp-
4 6 8 Plate length, m
10 12
Plate length, m
Figure 2.19: The response of the single plate evaporator to a 5% increase in the feed temperature. +: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit solids concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
2.5.5.4 Steam pressure
Steam pressure changes affect the variables of interest in almost the same magni
tude as the size of the change. The effects observed are in a direction consistent with
the physics of the process. There are no significant spatial effects since the change in
70
the steam pressure is distributed along the entire plate and not locally.
Plate length, m
Figure 2.20: The response of the single plate evaporator to a 5% decrease in the steam pressure. -I-: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit solids concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
2.5.5.5 Summary
To summarize, based on a set of expected disturbances in the range of ±5% of their
nominal operating values, the developed model described the single plate evaporator
71
system behavior in a manner consistent with the physics of the system. The most
important result observed is that the responses of the single plate evaporator system
are stable for all prescribed disturbances.
72
Plate length, m
Figure 2.21: The response of the single plate evaporator to a 5% increase in the steam pressure, -f: nominal response and o: non-ideal response. Liquor side variables (first three graphs): plate temperature, exit solids concentration, and exit mass flow rate. Steam side (bottom graph): condensate mass flow rate.
73
2.6 Nomenclature
Ai empirical wavy interface correction function
Ay, constant in Van Driest's damping function
a plate width, m
C constant in the turbulent Prandtl number formula
c specific heat, J/kg
Cp specific heat at constant pressure, J/kg
Cp^i liquor specific heat at constant pressure, J/kg
CL constant in the Limberg's function
Cy^i liquor specific heat at constant volume, [J/kg
Dy, Van Driest's damping function
Edy energy of a differential volume, J
Ey, energy of a wall differential volume, J
F Limberg's function
/ correction factor in the total heat transfer formulas
Gc condensate mass flow rate, kg/s
Gi liquor mass flow rate, kg/s
Gpi black liquor mass flow rate at plate inlet, kg/s
Gs solids mass flow rate, kg/s
Gyj water mass flow rate, kg/s
g gravitational acceleration, 9.81 m/s^
he heat transfer coefficient of condensation, W/m^-K
he enthalpy of condensate, J/kg
hE heat transfer coefficient of evaporation, W/m^-K
hE,i laminar heat transfer coefficient of evaporation, W/m^-K
hE,t turbulent heat transfer coefficient of evaporation, W/m^-K
hE,wi wavy-laminar heat transfer coefficient of evaporation, W/m^-K
hE,Nu turbulent heat transfer coefficient of evaporation calculated by Nusseh, W/m^-K
hn heat transfer coefficient of sensible heating, W/m^-K
74
hH,wi wavy-laminar heat transfer coefficient of sensible heating, W/m^-K
hg enthalpy of sohds, J/kg
hst enthalpy of steam, J/kg
hy enthalpy of water vapors, J/kg
hy, enthalpy of water, J/kg
AHeyap latent heat of vaporization, J/kg
AHe latent heat of condensation, J/kg
Ka Kapitza number, p'^g/pa^
k Von Karman's constant
L plate length, m
/ Prandtl mixing length, m
M mass of condensate, kg
Mdy mass of a differential volume, kg
Pyap vapor pressure. Pa
Pet turbulent Peclet number
Pr Prandtl number, Cpp/X
Pri constant in the turbulent Prandtl number formula
Prt turbulent Prandtl number
Q heat transition rate, J /s
Qew heat flux from condensate to wall element, J/s
Qyje heat flux from wall element to evaporating film, J/s
Re film Reynolds number, 4r/p
Ret transition film Reynolds number
Ret^i transition film Reynolds number proposed by Kapitza
Tr reference temperature, K
Pyap vapor temperature, K
Tf^e condensate film temperature, K
Tf^e evaporating film temperature, K
Tf^h heating film temperature, K
Tyj^c wall temperature at condensation side, K
Tyj^e wall temperature at evaporation side, K
Pw,h wall temperature at heating side, K
t time, s
75
AT temperature difference, K
U internal energy, J
u film velocity, m/s
Vx film velocity in x direction, m/s
W mass flow rate of evaporated water, kg/s
Wc mass flow rate of condensate, kg/s
Xpi black hquor dry solids concentration at plate inlet, kg/s
Xs dry sohds concentration, kg/s
X spatial coordinate, m
y spatial coordinate, m
2 spatial coordinate, m
yi distance from the wall to the film free interface, m
Greek Letters
F
5
Sw
e
V, Vc, VT
A
Atu
Ai
M* u
P T
'yx
e
black hquor mass flow rate per unit length, [kg/s- m
film thickness, m
wall thickness, m
eddy viscosity, m^/s
dimensionless variables
thermal conductivity, W/m- K
wall thermal conductivity, W/m- K]
dynamic viscosity. Pa- s
turbulent dynamic viscosity. Pa- s
kinematic viscosity, m^/s
density, kg/m^
residence time, s
Reynolds stress, N/m
dummy variable, m
Note: A bar over any variable represents time averaging and an asterisk or plus sign
represents a dimensionless variable.
76
CHAPTER 3
MODELING OF A SINGLE EVAPORATOR
3.1 Evaporator design
Falling film, plate evaporators consist of a shell, heating element, and a vapor
space. The position of the vapor space in a general falling film evaporator can be
either over [1, 2] or below the heating element [1, 2, 3]. There appears to be no
specific criterion to determine the position of the vapor space. In the present work,
the position of the vapor space is selected to be over the heating element (the plate
stack). This design is iUustrated in Figure 3.1.
The feed, the black liquor, enters the evaporator at the suction side of the recir
culation pump. The fluid is mixed with liquor that has been already processed in the
evaporator. A portion of the liquid product stream is recycled to the evaporator by
way of a distributor. The role of the distributor is to assure full usage of the available
heat transfer surface. As the fluid flows down the plates it becomes concentrated as
a result of heating (evaporation of the liquid in the fluid) by steam that is provided
to the opposite side of the plate. The steam itself condenses after the exchange of
energy. The released vapors from the fluid travels upwards and exists the plate stack.
For preventing of any liquor drops, entrained by the vapor flow to leave the evapora
tor, a separator is installed. The separator^ is a mechanical device that releases only
the vapors out of the evaporator and returns any entrained liquor back to the liquor
distributor. The concentrated liquor flows downwards existing the plate stack and
•'For more information on separators, see [2, 3].
77
Liquor distributor
Plate stacic
Steam inlet
Condensate outlet
Feed inlet
Liquor inventory
Secondary vapor outlet
Entrainment separator
Product outlet
Recirculation pump
Figure 3.1: Plate type evaporator sheme.
collecting in the bottom of the evaporator. The evaporator bottoms constitutes the
evaporator inventory.
From the evaporator design, the following distinct sections can be identified:
78
• Liquor distributor,
• Plate stack,
• Evaporator inventory,
• In-hne mbcer before the recirculation pump, and
• Splitter after the recirculation pump.
The types of processes that occur in each section are presented below.
3.2 Liquor distributor
The function of the liquor distributor is purely mechanical. The process of interest
is flashing by the liquor if its temperature is greater than its boiling point at the
operating conditions dictated by the vapor pressure and dry solids concentration.
Flashing is assumed to occur instantaneously. Moreover, when flashing occurs, the
amount of evaporated water and the changes in the liquor parameters (dry solids,
temperature and mass flow rate) should be calculated as they determine the input
conditions to the plate stack.
The steady state mass balance is given by
G^t = Gin - Wf, (3.1)
where Gout is the mass flow rate of the liquor out of the distributor, dn is the
recirculation mass flow rate and Wf is the mass flow rate of the water vapors.
79
Assuming that the vapor is saturated, the steady state energy balance is given by
GinCp^inlin = G outCp^outPboil + Wf{AHeyap + Cp^wTboil), (3-2)
where Cp^in and Cp^out are the heat capacities of the recirculating and flashed liquor,
respectively; Tin and Tbou are the recirculating and boiling temperature of the liquor,
respectively; AHeyap is the heat of evaporation of the liquor at T^on and Cp^y, is the
heat capacity of the water. Heat capacity is a function of the dry sohds and the
liquor temperature and the boiling temperature is a function of the dry sohds and
the saturation temperature. In this case, determination of the liquor parameters after
flashing requires solving a system of nonlinear algebraic equations. However, since
the boiling point rise is small enough [1] to be neglected^ the saturation temperature
can be used in place of of the boiling temperature. The change in the heat capacity
is negligible, thus the steady-state energy balance becomes
4.3.2 Pairing of the manipulated and controhed variables
Part of any controller configuration is to decide on how to pair the available
manipulated and controlled variables. One popular technique is to apply a Relative
Gain Array (RGA) analysis [73, 74]. The method, ahhough valid for steady-state
conditions, can provide a first approximation to pair controlled and manipulated
variables. The RGA analysis is by no means correct in every situation and may be
erroneous in strongly dynamic situations.. For multivariable problems, the elements
of the RGA matrix. A, are calculated as foUows [74],
Xij = kijhij i = l,...,Ny, j = l,...,Ny (4.2)
where A^ and Ny represent the number of controlled and manipulated variables and
H = ( K - 0 1\T (4.3)
The matrix K is a matrix of steady state gains between the controlled and manip
ulated variables. It is important to note that the multiplication in Equation (4.2)
corresponds to an element-by-element multiphcation and not the standard matrix
product.
In the current study, the steady state gain matrix is given by.
K Kx/Gp ^L/Gp
Kx/Ps KL/P,
-0.00427 0.000212
0.00903 -0.0000727
(4.4)
108
The corresponding RGA matrix is then given by.
A = -0.1936 1.1936
1.1936 -0.1936 (4.5)
The RGA guidelines are as follows:
Pair the controlled and manipulated variables so that the corresponding
relative gains are positive and as close to one as possible. [74]
Thus, the dry solids concentration should be controlled by the pressure of the steam
and the evaporator level should be controlled by the product flow rate.
4.3.3 Control structure
The SISO control strategy will be based on feedback control principles. Feedback
control involves measuring the control variable, comparing it to a pre-specified value
(set point) and applying a control law to determine the controller action that causes
the control variable to track the set point. When the set point is constant, the feed
back controller is said to provide disturbance compensation or regulation. When the
set point is a reference trajectory the feedback controller is said to provide servo con
trol. Feedback control is very common within the chemical industries in comparison
to for instance, feedforward control, because the action to compensate comes after
measuring the response. Thus, feedback control is often referred to as conservative
compensation.
109
The feedback control law that wUl be used in this study is the proportional-integral
(PI) law. The velocity form of the PI law is given by [74, 75]
Ck - Ck-i = Kc {tk-ek-i) + ek Tl
(4.6)
where Ck is the control action at sample time k, e is the error between the set point
and the output at time k, TJ is the integral time constant or reset time, Ke is the
controller gain, and AT = k - {k - 1) is the sampling interval. Note this controller
is a two-parameter (gain and reset time) controller. Additionally, it is assumed that
these are fixed parameter controllers, that is, once the gain and the reset time are
selected (tuned) they are not adapted or modified. This may limit the performance
of the closed-loop system (process and controllers) to respond to any disturbance.
The control scheme is shown in Figure 4.1. The scheme shows two cascaded,
feedback control loops. The introduction of the cascade structure is necessary to
assure proper disturbance rejection. Disturbances in the steam pressure can result
from changes in the steam header and disturbances in the product flow rate can result
from variations in the re-circulation pump head that result from changes in the level
of the evaporator inventory, mechanical problems, etc. Cascaded, feedback control
loops are commonly used in many industries; what must be guaranteed for these
interconnected loops to function correctly is that the time constant for the inner loop
must be much faster that the time constant of the outer loop.
Actuators such as control valves do not respond immediately to the controller's
110
Vapor
Mass flow rate
Dry solids
Mass flow rate
Dry solids
Figure 4.1: Feedback control of the evaporator. There are two cascaded feedback control loops. The composition control (CC) loop uses steam to control the product purity and the level loop (LC) uses the product flow rate to control the evaporator level.
command. To account for the delay in the response, the dynamics of the steam and
product flow valves are modeled as first order systems
du Ty— - W = Cy{t), (4.7)
where Ty is the valve time constant, Cy is the control command, and u is the actual
position of the valve. The controller parameters and the valve constants are provided
in Table 4.2.
There are many methods to determine the values of the PI controllers tuning
parameters. Common ones include the method of Cohen-Coon, Integral of Time-
I l l
weighted Absolute Error (ITAE), Ziegler-Nichols open and closed-loop tuning rules,
relay method, and auto-tune [74]. Some of these methods require that the model
that approximates the input to the output be a first-order plus dead time (FOPDT)
system, given by
r'f^=K{t-e)-y, (4.8)
where K is the system gain, r is the system time constant, and 9 is the dead time
or transport lag. In the present work, the controller's parameters are found using a
trial and error procedure, (see [74], Chapter 13, pp. 298-309).
Table 4.2: Control Loop Parameters. FC: flow controller, PC: pressure controller, LC: level controller, and CC: composition controller
Loop
FC
PC
LC
CC
Gain {Ke)
0.09 s.%/kg
0.00009 %/Pa
-368 kg/s.m
1250000 Pa
Reset time
3.3 s
3.3 s
100 s
33.3 s
{ri) Valve
VlOl
V102
Time constant (r„ )
45 s
90s
Loop Time constant (r)
FC 58 s
PC 131 s
LC 435 s
CC 722 s
112
4.3.4 Closed-loop response of the evaporator
The closed-loop system of the evaporator and the feedback control strategy are
subjected to the following operating changes:
• Feed mass flow rate change of -f-5% fiom the nominal value,
• Feed dry solids concentration change of -5% from the nominal value,
• Feed temperature change of -6.2% from the nominal value,
• Product dry sohds concentration of -1-10% from the nominal value.
The nominal values of the process are hsted in Table 4.3.
The simulated results are shown in Figures 4.2 to 4.5. The criteria for satisfactory
closed-loop response include: the length of time it takes the controher to return the
value of the controlled variable to within ±lcr of the set point, the oscillation period,
the type of response, and the time to settle to the desired value. In this study the
113
acceptable value of a for the dry solids concentration is set to 0.001 kg/kg. This value
represents the error of the composition analyzer. Table 4.4 summarizes the responses
based on these criteria.
x10
E n S5
Time, min
Figure 4.2: Closed-loop response to a 5% increase in the feed flow rate. Top: the dry solids concentration in the products, bottom: the steam pressure. The dotted lines represent the ±la hmit of 0.001 kg/kg dry sohds.
The most important observation is that in all cases the closed-loop responses
are stable to the modifications in the inputs. The responses of the controlled and
manipulated variables are smooth and there are no violations of any constraints.
The pressure of the steam in each case is adjusted within a 3 to 6 kPa range. The
controlled variables are within their ±la limit within ~ 5 minutes (see Table 4.4).
114
. x 1 0
Figure 4.3: Closed-loop response to a 5% decrease in the feed dry solids concentration. Top: the dry solids concentration in the products, bottom: the steam pressure. The dotted lines represent the i l a limit of 0.001 kg/kg dry solids.
Comparing the performance of the evaporator among the first three input changes,
it can be concluded that the evaporator is approximately three times more sensitive
to changes in the feed dry solids concentration and feed temperature as compared
to throughput changes. This conclusion is based on the time it takes for the re
sponse to be within ±la limit of ±0.001 kg/kg. However, the extent of disturbance
compensation is comparable for all disturbances.
In the case of the fourth input change, which is an increase in the dry solids
concentration set point, the time required to stay within the ±la limit is comparable
to the disturbance rejection cases. The peak value of the change in the manipulated
variable (steam pressure) is ~ 40 kPa, which is an order of magnitude greater than
115
0.198
. x 1 0
0.
Figure 4.4: Closed-loop response to a 5% decrease in the feed temperature. Top: the dry solids concentration in the products, bottom: the steam pressure. The dotted lines represent the ±1CT limit of 0.001 kg/kg dry sohds.
the values obtained in the disturbance rejection cases. Regardless, the values of the
manipulated variables do not violate any constraints.
The performance of the feedback cascade controller, for servo control (higher prod
uct dry solids concentration), is not as satisfactory when compared to the disturbance
rejection cases. There are at least two possible reasons for this response. First, the
controUer parameters (gain and reset time) were selected based on disturbance re
jection and not servo control. It is possible that the controller parameters can be
adjusted to accommodate both regulation and servo control. Second, the set point
change requested is in the form of a perfect step, which may not be achievable for
this or any real, nonhnear, multivariable process.
116
30 40 50 60
xlO
E ni 3 to
Figure 4.5: Closed-loop response to a 10% increase in the set point of the dry sohds concentration in the product stream. Top: the dry solids concentration in the products, bottom: the steam pressure. The dotted lines represent the ±1(T limit of 0.001 kg/kg dry solids.
In general, when a system transitions from one set point to the next, the reference
trajectory is more likely to be a bounded ramp function rather than a step function.
Figure 4.6 shows the responses for the same set point change, an increase of 10%
in the dry solids concentration in the product stream, using bounded ramps. The
first ramp is to achieve the set point in 20 minutes while the second is to achieve
the same increase but in 10 minutes or in one-half the time of the first ramp. Not
surprisingly, the second graph shows less aggressive controller action for the slower
117
ramp as compared to the faster one.
0.22
^. 0.21
a 0.19
ro 0.22
0.21
0.19
Figure 4.6: Evaporator response to an increase of 10% in the dry solids concentration set point. The first two graphs, starting from the top, represent the closed-loop response and the controller command, respectively for a set point change that occurs over 20 minutes. The last two graphs represent the closed-loop response and the controller command, respectively for a set point change that occurs over 10 minutes.
4.3.5 Limitations of SISO loop controhers
The single-input single-output (SISO) feedback controllers that employ a proportional-
integral-derivative (PID) control law have several disadvantages. Firstly, the single
loops are susceptible to poor performance (offset, sluggish or aggressive responses)
when a high degree of interactions is present. This is especially true in highly in
tegrated, nonlinear processes. Secondly, there is no explicit means of incorporating
118
input and output constraints in the calculation of the controller action. This poses
the most significant drawback of the SISO PID controllers. There are some simple
techniques that can be used to ensure non-violation of constraints. These include sim
ple limit checking and detuning the controller. Lastly, the controller parameters are
fixed; this means that if the process dynamics change the fixed parameter controller
may not be appropriate to regulate the process in a timely and safe manner.
In this study, it has been already shown in the case of set point changes that
the two-parameter fixed point controllers exhibit a tendency of applying aggressive
changes in the control action. Although physical constraints have not been violated
or reached, there are no guarantees. Therefore, there is motivation to seek a different
control strategy or control law that may address some of these limitations.
Table 4.4: Features of the process closed loop
Performance
Feature
Disturbance
Gf Xf Tf
responses
Servo Control
Xp
Time to reach ±la (min) 3.85 4.81 6.99
Oscillation period (min) 5.25 5.62 6.03 5.45
% compensalsed 88.63 89.27 89.69
4.4 Summary of decentralized control
A decentralized control strategy that consists of multiple single-input single-output
configurations and a feedback control law is able to provide satisfactory disturbance
119
compensation. The disturbance set consisted of known levels of expected but un
measured disturbances. The controllers selected were the two-parameter family of PI
controllers tuned using a trial and error approach. The control strategy employed a
feedback control law. In addition to the typical characteristics one seeks in a closed-
loop response a ± l a limit of 0.001 kg/kg was also imposed.
It was found that satisfactory disturbance compensation can be achieved for step
changes in known input parameters. However, in the case of servo control, the con
troller actions were very large in comparison to those used in disturbance compen
sation. It was concluded that either the PI controller tuning parameters should be
adjusted or that the set point change be apphed as a bounded ramp rather than an
abrupt step change.
4.5 Model predictive control (MPC)
Some of the more important drawbacks of the PID feedback control law have been
discussed in Section 4.3.5. In response to these hmitations, a more advanced control
strategy, that of model-based control, will be investigated for regulation and servo
control of the evaporator. In particular, the study will focus on the class of model
predictive controllers (MPC). MPC is a control strategy that has several advantages.
One of the most important is that MPC formulation allows the exphcit handling of
constraints on the outputs (controlled variables), the inputs (manipulated variables)
and the rate of change of the inputs. Also, MPC address muhivariable systems
and their interactions. The technical literature contains many examples of applying
120
MPC successfully to control open-loop stable processes and even non-minimum phase
processes [21].
The MPC technology began with the successful implementation of Model Al
gorithmic Control (IDCOM) by Richalet et al. [17]. A subsequent manuscript by
Garcia and Morari explored the concept of internal model control (IMC) [18], which
was very similar in formulation to IDCOM. In 1979, a group of control experts at the
Sheh Co. implemented a control strategy that they referred to as dynamic matrix
control (DMC) [19]. The Sheh group also demonstrated that the theory could be
implemented in practice.
Some of the shortcomings of IMC and DMC were their inability to control open
loop unstable systems and to address constraints explicitly on the process inputs or
outputs. In response, a variation of DMC was developed, called Quadratic Dynamic
Matrix Control (QDMC) [76]. The QDMC formulation used constraint optimization
theory to determine the optimal control action at the current time and into the
future. The QDMC formulation appeared to be closely similar to the linear quadratic
regulator (LQR) [77].
In 1989, Garcia and Morari unified the area of model-based control with the
concept of Model Predictive Control (MPC) [78]. This led to other variations such
as General Predictive Control and Generic Model-Based Control to name a few [21,
79, 80]. This work focuses on two forms of MPC, DMC and QDMC to control the
multivariable falling film evaporator, which is formulated by non-linear PDE's.
121
Consider a general nonlinear system described by
X = f{x,u,e) (4.9)
y = 9{x,tp) (4.10)
^0 = x{0), (4.11)
dx where a; = -—-, /(-) is a vector valued function, / : X " x ?7™ x 0"- ^ A", h{-) is
the measurement function, h : X"" x U"^ x ip^'y ^ Y^, states a; G A" C 5?", outputs
y eY^ CRP, inputs u e C/"* C 3?™ are piecewise continuous functions and d E 3?""
and tp e 3?"" are vectors of parameters^. In the general, the objective of the controller
is to regulate (disturbance compensation) the process about a given nominal state.
The design of a controller based on the nonlinear system model is however difficult
because issues such as closed-loop stabflity and robustness for nonlinear controher
designs are not established in the general case. An alternative is to employ well
established linear control theory to design the controller and to apply the controller
at the vicinity of the desired operating conditions. There are many cases in the open
literature in which a linear controller is employed to regulate a nonlinear process
successfully. Careful study of these processes reveal that although the process is
nonlinear, the operation of the process is about a single nominal operating state.
Hence, if a model-based control strategy is to be employed, a linear representation of
the process about the nominal operating state suffices as a representative model of
^In the following discussion the scalars are represented by lowercase symbols, the vectors are bold lowercase symbols and matrices are bold uppercase symbols.
122
the nonlinear system. Thus, hnear control methods can be apphed.
The linear model can be in one of two forms, an input-output (convolution) model
such as a step or impulse response model; or a state-space model. The linear model
can be derived using either system identification methods [81] or by a first-order
Taylor series expansion about the nominal state. When system identification is used,
the system is excited by a suitable input signal (e.g. step, ramp, pseudo-random
binary sequence, etc.) of sufficient magnitude to give a reasonable signal to noise
ratio at the process output. Time series analysis is then used to fit a suitable model
to the input-and output data. When a nonlinear first principles model of the system
is available a linearization technique can be used to arrive at a state-space model [82].
4.5.1 Linear quadratic regulator
The simplest linear model based control algorithm is the Linear Quadratic Regu
lator (LQR) [77]. In the LQR formulation, a Linear Time Invariant (LTI) state-space
model is used. The form of the LTI model is given by
X = Ax + Bu (4.12)
y = Cx (4.13)
xo = x{0), (4.14)
123
where A e K"''" is the system dynamic matrix, B e 3?"^'" is the input matrbc, and
C eW'P is the output matrix. The state feedback control law is of the form
u = -Kx, (4.15)
where K G SR"" " represents the controller gain matrix and u is that input that
minimizes a performance measure given by
J = / {x'^Qx + u^Ru)dt + x'^{T)Mx{T), (4.16) Jo
where Q and M are typically positive-semidefinite matrices, H is a positive-definite
matrix,^ and T is the final time but x{T) is the final state. The terms x^Qx and
x'^{T)Mx{T) are a measure of control and terminal control accuracy, respectively;
and the term u^Ru is a measure of control effort.
In practice, not all the states of the system are measurable (i.e. p < n) and in
many cases only the measured values of the outputs are available. To address this
issue, an observer is designed and used to provide an estimate of the states based on
the output measurements. An example of a simple observer is given by
i; = Ax + Bu - L{y - xC) (4.17)
y = Cx, (4.18)
• A symmetric matrix Q is said to be positive semi-definite if its corresponding quadratic form x'^Qx > OVx and positive definite if x'^Qx > 0 and x'^Qx = 0 only when a; = 0.
124
where L is a feedback gain matrix. Thus
u = -Kx. (4.19)
For more discussion on the LQR and the related Linear Quadratic Gaussian (LQG)
controller, see [83].
4.5.2 Dynamic matrix controller
In digital control implementation, the output is sampled at some sampling interval,
A, and because of the nature of digital-analog converters, the input applied to the
system is constant over this interval. This necessarily results in a piecewise constant
control input with period A. Thus, the set that u{t) is a member of is the class of
piecewise constant control functions, such that u{t) = u[k] for each value oft between
successive sampling intervals, kA < t < {k + 1)A.
Through simple integration, the behavior of the system in Equation (4.13) at every
discrete interval k can be represented by
x{k + l) = ^x{k) + ru{k)
y{k) = Cx{k)
where
$ ,AAt
At
T = B fe^^^'-^^da
0
125
and $ contains the sampled-system dynamics. By successive substitution, the state
vector X can be eliminated, yielding a convolution model
y{k) = C^x{k-l) + CVu{k-l)
y{k) = C^^x{k-2) + C^u{k-2) + CTu{k-l)
N-l
y{k) = ^ C $ ^ - i r i t ( ; c - £ + l ) C $ ^ a ; ( f c - i V + l) .
The product of the terms multiplying the input u gives the impulse response coef
ficients. For stable systems and for A'' sufficiently large, the impulse response coeffi
cients are reasonably approximated by zero. Thus, a finite N is sufficient to describe
the system.
A model based on the response of the system to a step input is given by
AT
y{i\k) = yo + ^aiAu{k + l-i), (4.20) 1=1
where y{i\k) is the output I sample instances in the future with respect to the k"
time instance. This model is commonly used in dynamic matrix controller design and
can be obtained from the model given in Equation (4.5.2), since the step response
coefficient, a^ is the sum of the first i impulse response coefficients, and Au{k) —
u{k) — u{k — 1) represents the change in the manipulated variables. Equation (4.20)
126
can be re-written to separate the future and the past control actions
past future
N i
y{^\k) = yo + Y^ aiAu{k + i-i)-\-J2 aiAu{k + i ~ i) +d{i\k), (4.21) e+i i=i
where d{£\k) represents any unmodeled effects. The first term in this equation is the
effect of the past inputs if ah the future control actions are the same as u{k — 1), that
is, no additional control moves are made at sample instances /c, /c + 1 , The second
term represents the predicted behavior due to future inputs. If the linear convolution
model is a perfect representation of the process then d{i\k) represents only external
In summary, a system of ODEs (lumped description) of the plate stack is given
by Equations (4.33), (4.37), (4.39), and (4.44), re-stated here for convenience.
W = hEa{L
Lh
dX
~dt
-LI
h
dMst ^
dt ^^f''
- ^f'(Y
-Gout-W
' M
\)n{Ty,aU — Tboil{X, Tsat))
^•^evap
GfsCp{Tboil -Tfs)
.^n.n(T.. ^T^oil+Tfs).
138
4.6.1.2 Bottom inventory
The bottom inventory model is analogous to that described in Chapter 3. Thus,
the model equations are
dT ~dt ~
dL lit '
IS Gpip — GoutT
pLAt
_ NGp — Gout L dp
pAt pAt dt
dX _ NGp dt - pAtL^^' ^^
Tdp TdL T dcp
p dt L dt Cp dt
(4.45)
(4.46)
(4.47)
4.6.2 Validation of the nonlinear lumped parameter model
Figures 4.7 to 4.16 show a comparison between the nonlinear lumped and the
nonlinear distributed models. It is observed that the nonlinear ODE model represents
the main features of the nonlinear PDE model satisfactorily even though the process
temperature profiles show the largest deviations. Additionally, the steady-state gains
obtained from a ratio of the change in the outputs to the change in the inputs are
equivalent. It is concluded that the nonlinear lumped model is satisfactory for MPC
controller development.
4.6.3 A hnear MPC design
Although a nonlinear ODE model is available to design a nonlinear MPC con
troller, a linear, constrained MPC design is first investigated.
139
<
20 25 3D Time, min
Figure 4.7: Nonlinear ODE model response to a 5% decrease in the feed flow rate. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
X
<
20 25 30 Time, min
Figure 4.8: Nonlinear ODE model response to a 5% increase in the feed flow rate. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
140
0.015
20 25 30 Time, min
Figure 4.9: Nonhnear ODE model response to a 5% increase in the feed dry sohds concentration. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
<
xlO
-5
-10
-15 J I L_
10
ODE PDE
15 20 25 3D 35 40 45 50
20 25 3D 35 40 45 50 Time, min
Figure 4.10: Nonlinear ODE model response to a 5% decrease in the feed dry solids concentration. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
141
0.015
20 25 30 Time, min
Figure 4.11: Nonlinear ODE model response to a 5% decrease in feed temperature. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
X
<
5
0
-5
-ID
-15
xlD
10 15 20 25 30 35 40
PDE ODE
45 50
20 25 3D Time, min
Figure 4.12: Nonhnear ODE model response to a 5% increase in feed temperature. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
142
XlO
X
<
20 25 30 Time, min
Figure 4.13: Nonlinear ODE model response to a 5% increase in vapor pressure. Top: deviations in dry sohds concentration in the product, bottom: deviations in the process fluid temperature.
D.025
D.D2 CO
^ 0.015
J °° *3 0.005
10 15 20 25 3D 35 40
ODE PDE
45 50
20 25 30 Time, min
Figure 4.14: Nonlinear ODE model response to a 5% decrease in vapor pressure. Top: deviations in dry sohds concentration in the product, bottom: deviations in the process fluid temperature.
143
T3
a
<
is: •a o
1-^
<
0.01
0.008
D.OQB
D.004
D.D02
0
0.2
0,15
0.1
0.05
D 20 25 30
Time, min
Figure 4.15: Nonhnear ODE model response to a 5% increase in steam pressure. Top: deviations in dry sohds concentration in the product, bottom: deviations in the process fluid temperature.
XlO
X
<
E -0.05
20 25 30 Time, min
Figure 4.16: Nonlinear ODE model response to a 5% decrease in steam pressure. Top: deviations in dry solids concentration in the product, bottom: deviations in the process fluid temperature.
144
4.6.3.1 Linearization
Consider the following system of equations that represent a linear time-invariant
(LTI) state-space model (differential equation) and a companion measurement equa
tion (algebraic) [82]
x = Ax + Bu (4.48)
y = Gx + Du, (4.49)
where x is the vector of the system state derivatives, x is the vector of system states,
u is the vector of system inputs and y is the vector of system outputs, A G 3?"^",
B G 3??"'"", G G a?"''^ and D G 3R'"''''. In the general case. A, B, G, D are firnctions of
time. In the case of chemical systems, the matrix D is almost always a zero matrix,
i.e. there is no direct feed-through of the inputs to the outputs.
The procedure to obtain the above representation starting from a nonlinear sys
tem representation is as follows [82]: Consider the nonlinear state and measurement
equations given by
a;(t) = /(x(t) ,n(t)) x(to) = Xo (4.50)
y{t) = h{x{t),u{t)), (4.51)
where x = —, /(-) is a vector valued function, f : X^ x U^ -^ A", h{-) is the CJij
dx
'di measurement function, h:X^xU^^ F ^ states rr G A" C 3?", outputs yeY^cW
145
inputs u G {7™ C 3?"*, are piecewise continuous functions, to is a initial time and XQ is
the vector of the initial system states. Equation (4.50) can be solved for some nominal
input u{t) and initial state x(to) to obtain the nominal trajectory x{t). Consider small
perturbations from the nominal initial state and input given by
Xo = io + XQS
u{t) =u{t) + us{t).
Assuming that the resulting solution is close to the nominal solution then
X{t) ^X{t)+X5{t).
Thus, Equation (4.50) can be re-written as
| x ( t ) + | x , ( t ) = f{x{t) + xs{t), u{t) + Mt))
Xo = Xo + Xo5-
Applying a Taylor series expansion to the right hand side of this equation about the
nominal values and retaining only the first order terms gives
| x ( t ) + | x 5 ( t ) ^ f{x{t),u{t)) + ^{x,u)xs{t) + ^{x,u)us{t). (4.53)
Since
d i{t) = f{x,u),x{t,)=xo (4-54) dt
146
it foUows that the relation between xs and Uj is given by
xs{t) = Axs{t) + Bu5{t) xs{to) =xo- Xo, (4.55)
where
df A = -^{x{t),u{t)) (4.56)
B = ^{x{t),u{t)). (4.57)
Applying a similar expansion, the linearization of the nonlinear output equation.
Equation (4.51), is given by
C=^{x{t),u{t)). (4.58)
The linear approximation obtained by the Taylor series expansion can either be
obtained analytically or numerically. In this work, both methods are used. The
numerical function used is provided by Matlab® (Mathworks, Natick, MA)^. The
analytical solution is obtained using Maple'^^ (Waterloo, Canada). The Maple solu
tion is provided in Appendix D.
^The specific function is linmod.
147
4.6.3.2 System-theoretic analysis
This section apphes system-theoretic methods to analyze the properties of the
linear time-invariant model [82, 84].
Definition 4.6.1. Stability: A linear time invariant system is said to be exponentially
stable if and only if all eigenvalues of the matrix A have negative real parts.
Definition 4.6.2. State Controllability: A linear time invariant system is state con
trollable if the rank of the n x nm controllability matrix given by,
C = B AB •-- A^'-^B equals n.
Definition 4.6.3. Output Controllability: A linear time invariant system is output
controllable if the rank of the n x nm output controllability matrix given by, Co =
C CAB •-• CA^'-^B equals the rank of the matrix C .
Definition 4.6.4. Observability :A linear time invariant system is observable if the
r 1-^ rank of the np x n observability matrix given by, O — (j (jj[ ... CA^'^
equals n.
In the linearization of the lumped model it is assumed that the level of the evap
orator inventory is constant (1 meter). The the linear model has four states: (1) the
total mass of the black liquor on the plates, (2) the dry solids concentration in the
plate stack, (3) the dry solids concentration in the evaporator inventory, and (4) the
temperature in the evaporator inventory. The nominal conditions are given in Table
4.6. The operating conditions are provided in Table 4.3.
Analytical resuhs obtained for the matrices A G Sf? "" , B G ^'''^ and C G "^""^
are as follows:
A =
-0.10247 502.57 0 0
-0.10491 0.10942
0.01736 -0.021771
0.00005 3.0939 -3.8887 -0.019499
(4.59)
B =
-1.4692 1.6478
0.00011953 -0.00011581
0.016383
C =
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
(4.60)
The test of stability shows that all eigenvalues of the A matrix are real and neg
ative. By Definition 4.6.1 the LTI model is open-loop stable. The eigenvalues are
{-0.1156, -0.1042, -0.0180, -0.0018}. In Figures 4.17 to 4.20, comparisons are
149
shown between the nonlinear lumped and the LTI model when the same size step
changes are applied. It can be concluded that the LTI model represents satisfactory
the nonlinear system dynamics as the responses are qualitatively the same and the
differences in the steady state gains are small.
306D
LODE NLODE
10 2D 30 40 50
Time, min
BO 70
20 30 40
Time, mm
0.212
339.05
30 40 50
Time, min
Figure 4.17: Responses of the hnear (LODE) and nonlinear (NLODE) ODE models to 5% decrease in the steam pressure. Dashed line: LODE, sohd hne: NLODE.
The controUability matrix given in Definition 4.6.2 has a rank of four (same as the
number of states), therefore the LTI model is state controllable. Similarly, the rank
of the output controUabihty matrix, given in Definition 4.6.3, is the same as the rank
of the C matrix and therefore the LTI model is also output controllable. The rank
150
2-o
3160
3150
3140
3130
3120
3110
3100
3090
r^' i -•• LODE
NLODE
,
•
•
0.24
10 20 30 40
Time, min
50 60 70
iS 339.25
Time, mm
Figure 4.18: Responses of the hnear (LODE) and nonlinear (NLODE) ODE models to 5% increase of the steam pressure. Dashed line: LODE, solid hne: NLODE.
of the observabihty matrix (Definition 4.6.4) is also full, therefore the LTI model is
observable.
4.7 Closed-loop results
The development of the MPC controller applied to the evaporator is accomplished
using the Matlab® and associated toolboxes.^ The Simuhnk® tool provides several
built-in functions that designs the model predictive controller using a step response
model.
^Specifically, the functions nlcmpc, ss2step in Simulink® and Control® toolboxes.
151
3200
3180 O)
^ent
oty.
c
CL
3160
3140
3120
3100
3080
r^ ( • •
•
••-- LODE NLODE
, , 10 20 30 40 50 EO 70
Time, min Time, min
2r E
339.5
33B.5
ra 337,5
10 20 30 50 BO 70
Time, mm Time, min
Figure 4.19: Responses of the linear (LODE) and nonlinear (NLODE) ODE models to 5% decrease of the vapor pressure. Dashed line: LODE, solid hne: NLODE.
The controlled variables are the dry solids concentration and temperature of the
product stream. These two controlled variables correspond to the third and the fourth
states of the LTI model. The G matrix in the measurement equation has the form:
C = 0 0 1 0
0 0 0 1
The analysis for the controUed variables shows that the system is both state and
output controllable, and observable.
152
3140
3120
3100
3080
3040
3020
1 1 1
;
\ ^
1 1
- - • LODE NLODE .
• - - , ,
2r
10 20 30 40 50 BO 70
Time, min
10 20 30 40 50 BO 70
Time, min
o
10 20 30 40 50 BO
Time, min
70 10 20 30 40 50 60 70
Time, mm
Figure 4.20: Responses of the linear (LODE) and nonhnear ODE models to 5% increase of the steam pressure. Dashed line: LODE, sohd line: NLODE.
4.8 MPC results
The design of the MPC controUer and its application to regulate the evaporator
are accomphshed using Matlab's® Simulink® tool. The most important process
variable to control is the dry solids concentration in the product stream. However, it
is possible that in some cases the temperature of the product must also be controUed.
To wit, in the MPC design, the temperature and the dry solids concentration in the
product stream are both controlled variable. From the analysis provided in Appendix
D, the manipulated variables are the temperature of the steam and the saturation
153
temperature of the process fluid. In reality these temperatures translate to set points
on pressure because MPC is configured as part of a supervisory control layer rather
than as a part of the regulatory layer.
The simulated responses of two different formulations of the MPC controUer are
compared to the SISO PI controller design. The formulations are unconstrained MPC
and constrained MPC. The actual feedback block diagram is presented on Figure 4.21.
I -
MPC Controller
U
NLPDE
NLODE
LODE
^ d
>\ ^'°^
+ ^ '
"-h
iM ^ -
^ - ( < - ^
r-t-
b
Figure 4.21: A feedback block diagram with the MPC controller.
4.8.1 Results
The resuhs from the simulations are provided in Figures 4.22 to 4.32. The tuning
parameters of the MPC and PI controhers are given in Tables 4.7 and 4.8. In Table
4.7, yi corresponds to the product dry sohds concentration, y2 corresponds to the
154
product temperature, ui corresponds to the steam saturation temperature and U2
corresponds to the secondary vapor saturation temperature. Note that aU values in
Table 4.7 are deviations from the nominal conditions (zero) since the linear model is
developed using deviation variables.
Table 4.7: Tuning parameters for MPC.
Prediction horizon, P 15
Control horizon, M 5
yi y2 Ul U2 Aui Au2
Weights 50 0.01 0.06 0.06
Minimum -0.001 kg/kg -0.2(1) K -10.96 K -18.23 K Maximum 0.001 kg/kg 0.2(1) K 6.04 K 4.77 K 3 K 3 K
Table 4.8: Tuning parameters for the PI controllers.
Ke Ti
XC loop 2000(1000) K 500(250) s
TC loop 30 300 s
The ±la limits for the dry solids concentration and the product temperature
are ±0.001 kg/kg and ±0.2 K, respectively Both limits are based on expected errors
associated with the analyzer or the sensor. The temperature limits may be considered
narrow; however, the narrow limits may provide better insights in the controller
performance.
155
The MPC controUer is tuned to provide satisfactory performance for disturbance
compensation. AU disturbances are modeled as step functions.
Disturbances of ±5% in the feed dry solids composition, mass flow rate, and
temperature are applied to the evaporator process. The responses of unconstrained
MPC controUer, constrained MPC controller and the PI controllers are monitored.
In each case, the controUer provides satisfactory disturbance compensation. The
responses are smooth without excessive overshoots or oscillations. The response of
the manipulated variables used to regulate the controlled variables do not exhibit
aggressive moves for each disturbance. All disturbances are compensated within ~35
minutes after the introduction of the disturbance. However, there are noticeable
differences in the performance (closed-loop response) of the MPC and PI controllers.
There are no significant differences between the performance of the unconstrained
and constrained MPC controller. This is attributed to a combination of the range
on the manipulated variables (Au and u) and the size of the disturbances. Small
differences in the manipulated variables can be observed in the presence of the feed
temperature disturbance (see Figures 4.24 and 4.27). In the constrained case, the
MPC controller is a httle more aggressive.
In the case of the feed temperature disturbance compensation it is observed that
the compensation is less demanding as compared to a feed concentration or flow
rate disturbance compensation. This is because temperature regulation is not as
high a priority (see the weights used in Table 4.7 as compared to the regulation the
product dry solids concentration. However, in other instances when tight control of
156
temperature is warranted, the weight on the temperature variable can be adjusted
appropriately to reflect its importance.
When the PI controhers responses are compared to both unconstrained and con
strained MPC controller responses it is found that the PI controllers provide tighter
control. This is attributed to the tuning parameters selected for the PI controhers,
which are unaltered for each disturbance. It is observed that the change in the ma
nipulated variables are more aggressive in all cases as a prioritization between the
controlled variables is not an option.
To examine the flexibility of the PI and MPC controllers, the magnitude of the
set of disturbances is increased by 1.6% and the performance of the constrained MPC
controller and the PI controllers is compared. In these tests, the tuning parameters
of the PI controller are changed to assure no violation of the input constraints. The
tuning parameters for this case are given in parentheses in Table 4.8. For disturbances
in the feed mass flow rate and dry solids concentration, no significant differences were
observed. However, in the case of a temperature disturbance, the closed-loop response
of constrained MPC is superior to that of the PI controllers. These responses are
shown in Figures 4.31 and 4.32.
For the additional 1.6% increase in the disturbances, the PI controllers exhibit a
decrease in closed-loop performance. For instance, the time for complete compensa
tion (time for which the controlled variable is driven back within ±lcr) is about 4
times longer (20 minutes) in comparison to that achieved by the constrained MPC
controher (5 minutes). Figure 4.31 also shows that constrained MPC maintains the
157
most important variable, the product dry solids concentration, within the ±lcr, which
is not achieved in the case of the PI controllers. The MPC controller provides this
critical performance feature while satisfying all other imposed constraints. If the PI
controllers are tuned to achieve the same performance with respect to the outputs,
they violate the rate of change limits on the input variables.
Since the MPC controller is successful in regulating the nonlinear ODE model,
the optimal controUer actions, calculated for a 5% decrease in feed temperature, are
apphed to the PDE model of the evaporator. The result of this test is shown in
Figure 4.33. From this result, it is clear that the dry solids concentration of the
product stream is regulated within ±lcr. The response of the product temperature
is not as accurate, the error is ~ 1 K. This test indicates that the approach to
design a model-based controller using a linear model obtained from the linearization
of a nonhnear, first-principles ODE model is vahd in this specific case. Clearly, no
generalization can be made to other systems.
The economic benefits of applying a model-predictive controller to regulate the
product dry solids concentration is difficult to quantify when only one evaporator is
used. In reality, an evaporator plant (more than one evaporator) is used to concentrate
the entire black liquor feed to the recovery boiler. If however, control is implemented
on the dry solids concentration in the product stream of the last evaporator in a
multiple-effect evaporator plant, then it is possible to quantify the economic benefits
of maintaining a constant dry solids concentration to the recovery boiler. Information,
provided in [85] provides an estimate of the difference in net annual economic loss/gain
158
if a black liquor with a concentration of 0.75 and 0.8 kg/kg dry sohds is produced as
compared to a base case with a concentration of 0.48 kg/kg. The difference between
the net annual economic loss/gain is approximately 600 000 US dollars per year. If
0.8 kg/kg is assumed to be the desired operating point and the process is maintained
within ±lo- limit for 80% of the processing time, but deviates within 0.01 kg/kg for
the rest of the processing time, this will result in a loss of 24 000 US doUars per year.
However, if the process is 80% of the time out of ±lcr limit, for the same desired
operating point, the loss increases up to 96 000 doUars per year or an increase of 400
%. This analysis includes only pure energy balance effects on the recovery boiler.
Variations in the recovery boiler operating conditions due to droplet size distribution
changes are difficult to quantify as they affect not only the boiler, but the entire
recovery cycle.
In all the simulated results present, it is assumed that the disturbances immedi
ately affect the outputs. This may lead to aggressive controller actions. To avoid this
in the case of a multiple-effect evaporator plant, the feed is not supplied to the first
evaporator in the suction line of the recirculation pump, but between the evaporator
inventory and the plate stack. However, this design change is valid only for the first
evaporator. AU other evaporators have the design shown in Figure 3.1 in Chapter 3.
159
CM I M CM < N CN
6>|/6>| •''°\
CO
—I o
ro ^ 00
IT) C O
CL CL
< <
E
LO t - LO - ! - O O o • O
O O O
o o o CM
O O O CN
,pojd
>l i LU'|8A9"| Bd'^^d V '^dV
9i -^
at ^
Tl OJ
4 - 1 4 - J
f ) 73 OJ
^ H fl)
+^ c« ^-1
•5 o
q=l
0)
> ni
O
ent
> Pl • •—{
-a ! - i
^ - t J
s " 03
- i d i-i
CU O H
fl CZ3
! - i O . . fl -d
•^ fl o o l O
03
fl CO CO CD
a S - l
o ^ 03
s ^ . fl O) CO
CU
.-fl
(U
fl o
t "" fl ;fl «i;
•-' o -.fl OJ CO -i-^
fl > ^ ^ .s I 0)
fl CO CO
•^ .fl Q;
CM ^ i " CM . CU
^ -§ :S S S .23 fl fl:! . ^ b fc
160
i n CO
LO CO
0.21
0.21
0.21
0.2
s^A •'°'\
334.
.poJd M i Bd ' d V d V
H fl a;
• > T3 fl .'^ ..
fl o
0 ) s-l
fl -1-3 03
03 (u
a S-H
I B I -o ^
^ o o a CO
f-l fl
-a o a; g
. . f l CO
.s § CO o3 03 !^
£ fl a <D
S-H
O a o3 >
-^ -l-i CO
. ^ a; OJ o fl - fl ^ T d
LO T d M l
° >.^ O T3 ^
(P O 03
a o fl t H CO
oj a ^ - ^ -t a r-i "3 ^
i -2 ^ a a ^ s - ^
•-f l l.^ CO !-i O .-H CD ,—I n ,
a ^ fl O <D fl3 PH -^ -^ ^ fl X5
- • Q ;
CO CO L ;
^ 2 :S • a ^ 0 S o 1:3 tn ,
[-iH ; f l 03
161
CM O
CM CD
CM CO ^ CO CM CO O
^ CM O CO CO CO CO CO CO O
LO O O
—
o o o o
o o o LO
O O O o LO
em''°\ .po jd
>i 1 LU '\dAQ-\ .A „,s „
Bd d V d V
-1-3 o3 ^ T3 fl .. CD M CO T 3
(-1 rH
CL> fl fl ro
<D >
>> !H O
+J fl
fl 03 -^ t-i .
03 .;-l S-l r f l Q; ':;3 a „ fl CD
•n <D
-^3 -(-= O
• T d
(U -.fl
CD
Td CD
.'^ ^—I
CD
fl + J 03 s-l CD
a 0
- t J
.fl O ^ ^ • ^ I fl O 03
xj "g a fa\ o O LO 03 D. ce "" a O g 03 03 -^3 CJ o3 fl i-i
03 03 ! - i
a
03 fl
0^ O .Jfl
-tJ 7 d hO
.a o gj CO r X ^
03 . - U CJ >i _ fl ± 1 ' ^ c3 "O fl fl - u 03
0 ° -a Q, 03
03 ! - l
fl CO CO
P H ra
CM CO 03
. f l a
S ^ .a
fc ^ ;fl
162
CO - r t
fl « . f l o3
LO CO
IS) >
CL CL < <
1 1
E
E
CM
CD
CM
O
CM LO
O CO
CO CO
CO CO CO
CM O
LO O O
O o o
o o o CM
O O O CM
iPOJd
>l X ,pojd
>1 1 UJ'|8A81 Bd d V d V
03
T 3 03 03 >
o
03
> 03 fl
- ( . ^ . - H
o3 . . '-' Td & • -O ^
^ ^ . CO 03 CO tH
o3 fl
•Td 03
^ a OJ 0^ .H U
.a fl Td o S-H
03
fl a en
03 T3 fl "
„ O >-'
5 ^ a o3
O 03 S - f l
03 O Cl 03 CO 03 ! - i
a Cl)
.d
ifi S-l
-M
fl 03 CD
rl o o
T3
-^ m
fl 1——1
-*-3
^ _bJJ
fl O 03 - f l CO _sH
03 > , -^^
fe 2 ^ a ^ Si o P H
a a o3 03
C O s - l
'S ^ ^ fl • fl 03
.•fl fl .Jfl o3 S -i-j SH ^ f l
X3 ^ CO M b "^ fl - -H 03 O - U fl D ^ ;fl O
03 .^ CM o3
^ -5 -a
hp a fl
163
0) >. rf l SH
6 /6>, ' "^V iPOJd >l 1 U!UJ'|8Ae-|
|A _ ,S
Bd dV dV
fl
O fl •-^ . f l +J - I 03 . . SH T d
-1-^ SH
« -fl 03 rfl o CJ
CO
SH T3
03
03 S-l fl
- J 03 •i-i CD
a a 03 C3 fl
0) T3 O
03
03 SH
S-, ^ a So
CO
. a ^ o
03 O CO 03 o3 CO 03 SH O 03
-Td
lO
03 SH
a S - l
O
a fl O 03
SH -(.^ fl • • 03 03
CO
o3 o HH fl Afl O O 03 rrH - ' H
S ^ T! CO M 03
a "O Td 03
r ^
4 J
,fl 03 O fl
- t J
o fl
prod
- t J
fl (Xl
ssur
e
Cl) s-l
03 CO a
a - fl •-IH • 03
s-l cr> •i^ 03 - t J CO
a 'g 03 o ; a ^ P H b CO
T3 03 fl
•I—I 03 S H
CO fl O
o
o) .a r f l
r ^ SH 03
CD CM
03 T 3
03 -SH r f l
a +
fl T3 03 - O a 03 fl -(.J
. - H O
fe -!d
u 03 > 03
164
CO -Td
fl « .fl o3
03 >
E CD
E
Bm ' "V ,pojd >1 1 LU |aA3" | Bd d V d V
-^^ O
0) r f l
SH O
fl 03 S-l
o3 b s-l fl 03 • ;g
a ^ 03
T d 03
s-l fl td SH
03 03 •- a 03
r f l a 03
•i^ - ( -= O fl
T3 O 03
SH O 0^ - ^
^ ^ fe? 8 LO 03 03 "
r v 02 • ^ to
03 SH
a SH O
a
03 -^^
fl O 03
O o3 fl ^ 03 03
^CO
03 fl
03 S - l fl to CO
a | - ^
^ :3 ;fl' . f l ' O 03
«3 r f l o:> ^ +=
a t ^ SH -3
a -^ SH O CD SH a a 03
O - ^ OH
• i 1 ^ fc o3
^ ' ^ 03 ^ CO CO
a •'fl s " T-H o;
r f l - ^ - ^ SH
03 CM
o; 03 ••
;: 03 + j
165
T—rr
CM CM CM CM CM O O O CO CD
CO CO
CO CO
LO (O C) ^>
CM O
—
6>)/6>| ,pojd, X
^ ,PO.d^ UJ ' |8A81
1.00
5
40
00
20
00
-200
0
Bd'^'d V '^dV
0) -i^ -^ 53 .4J o3 ^ Td
fl .. 0) rH CD T3
2i I ^ o SH 03 fl fl ^
T3 03
-1-3 -(-3 O
T3 03
r f l H
03
o fl
fl 03 ^
03 T3 !H SH
5 -_ 03~
fl
qfl CO CO o3
Td 03
.0)
03 r ^ -1-3
03 s-l 03
a a 03
4 J
fl -o -- o
s-l
a
03 S H fl CO CO 03 SH M T3 ^
" fl ^ a o SH
O vo O) Li
fe\ to g* lO _p > o3 . y r f l
2 CO
03 O . f l to 03 S H
a
fl O r f l
0) ! y ; f l •5 ' ^ 03 _ CO rfl
.,, S H - O 03 -r) rl p ^ a
s-l T3 O
's-l a 03
a ^ PH r "
(X) CM
^ S fl CO tZ2 03 S H
a i 03
03 rf l
fe +l;fl
166
CO 03 fl
Td fl 03
CD
E i -
CM
O CM
O CM
CD
CN
O CO CO
6x/6^ •'°\
CO CO
,po jd
o o o o o o o o o CO - q - CM
o o o CM
LU I8A81 A „ ,s „ dV dV
Td 03 03 >
-tJ 0)
o Td
03 -Jfl
o -1-3
fl H ^
.2 -d
2 rfl fl r, 0 3 0 3
S-l fl O fl O o3
SH
tZ3 T d
03
a
0 3 S-l fl
^ a O 0) CO +3
SH O
T d T d o 03 !H rJ 03 D , CO
03 T d b
rfl fl a ^ O !H
fl g a - CO a 03 r- > CO fl " S O 03
03 ! H CO
511 lO fl ^
O 4J 03 O r f l
^ ^ .Sf o ;f l —< „.. O CD 8 " fl >. ^ 0^ r r t T3 CO T3 ^ 03 M M - ^ 03
a g 03 T 3 - I fl ""
0 3 S-l fl CO 0 3 0 3 S-l
- ^ a fl QL. o3
fl K) tC
S .t5 03
^ a J SH . f l -tJ a b .-H
I I i-H 03
^ ^ ;f l rb:i
03 T d
03
03 CO 03
bp a 03
(35 CM
03
fe
rfl S H
fl
167
LO CO LO
CO
CD - ^
CM CM
O O
6>|/6>i
212
0.21
o ,pojd
X
336
• ^ CN CO CO CO CO
,pojd M i
330
1.02
E Q)' E i-
Oi |8A8"I
LO O O T —
o o o o o o o o o o O LO LO
" ^ ,A ,S
Bd dV dV
b T-H
^ 03
r f l - ) - : ' - t J
fl 03 CO 03 S-l
i ,c r ^
S H -o:
X3
r f l - M s-l fl
a o 03 SH
nes
T3 03
-t~3 - t J O
-a CD
r f l H
ai SH
atu
! H 03
a a 03
Td 03 03
t + H
03 . • ^
• ^ ^
fl . r - l
03 03
decrea
fe? L O
o3 M H
O
03 CJ fl 03 CO 03 S H
a 03
r ^
4 ^
fl •r—)
03 CJ fl o3
a s-l O
S-l
03
a h — 1
PL,
cii CO
^ a:> SH
fl bn
iZ ,
t + H
Td fl o3
03 > 03
>, S H
O - t^ fl 03
• • — (
TJ
03"" S - l
fl 4 J o3 SH 03
a a 0) 4 J
- ^ J CJ fl
Td o a
Td fl o CJ 03 03
. fl o
• ^
03 SH
- t J
fl 03 C3 fl O o
3 o 03
SH
Td
-^ O fl
-d o s-l
a
03 SH
JH
03
03 SH
fl CO CO 03
a E H O
a t 03
r f l - t J
CO
03
fl ; f l - M
j : ^ bC r H
03 r f l
-o fl
03
fl 03 CO 03 S-t
a a 03 03
CO
03 t^ r f l
a + J
CO
168
L O C O
o CO
03 - O
• — 1
'P, '03 03 ^
o , Td h SH
03 O
" 03 . >
03 fl S H " ^
fl .. +J Td 03 iH S-i . - 3 03 r f l a + fl 03"" 0 3 S-l
-^ fl 1 1
6 /6>1' "V UJ |8A8-| d V d V
^ 0 3
^ a rfl a 4 ^ 0 3
a ^ 03 jzi
S 2 fl r 2 •• 03
CO CD o3
O
a
0 ta 2 '-I 03 fl CO 0 3
fl^^ O 03
03 CJ . 3 O. f^ ^
a o ^ 03 " rfl
.Jfl T3 Sf -UJ .Zi i:ri o; C J
fl o3
SH 03
a O PH
Td 03 fl
•3 S-l
- l - ^ 0 3
fl o
O
o 0 3
C3
fl Td o S H
a 03
0 3
Td fl o3 03 ^-1 fl 03 03 CD SH
a
^ .a
03 03
03
03 rfl
03 r f l
CO
03 S H
fl bjO
• r—I
fa
S H
03 T3
fl
03 4 J S H S H
a fl 03 O S H '^-l
169
6 /6>l • ' ' "V UJ'|8A8-| EcJ ' cJ V ' d V
-^ ^ - ^ 03
03 (H 03 '13 SH rj
03 ^
03 a (H o3
-a <v
o
03 r f l
03 >
rSJ
SH O
-^J fl 03 > fl
03 SH . .
fl Td -I.J s-l
S :fl 0 3 -^•=
a 03
03 SH fl 03
o s-l 03 03
^ a
cz
E '8U
JII
r f l J_3 - ^ J _ , - tH
fl o • - ' fl . 03 Td OJ
decr
eas
nd:
pro
pre
ssui
^ O fH fe? CJ o ^ ^ & CD r- >
of a
at
ion
the
03 H -S le
nc
cen
ine
CQ r H 1 1
03 5 iL O -R
he p
: li
dc
ihg
h
-R O 0^ _ 03 r f l
•a >^:: - , !H Td 03 Td r< CJ S fl ie " 03 H 03 cl fl SH fl Td fl
irfo
r p
ro
ress
2< .. a a i ; "^ r " a PL| r'-' 03
PH 03 -R
CO + j 03
^ a ^ S ^ .22 S ) ^ 03
fa ^ ;fl
170
0) -R CO fl o3 03 03 03
6 /6>1 ' '"^V
s-l CJ 03
Td
03 S - l
a 03 fe? O)
l O 0^
^ ;fl
^ -R
Mrg 03 TJ >-< 03 O r f l ^ H
Td
i s ^ fl
CD SH 03
CO
ion
- R (•;
03
O SH
rt O CJ
era
a a 03 -R
C) fl Td o >H
a
03 R fl ° fl -R . rH .JHl -|H> O
a ^ 03
-fl
03 03 fl O
fl .2 ' - R 03 s-l
fl 03
. O
a fl CO O 03 CJ
aS 03 O
-R CO
>. h 03 !H SH " ^ O - R
- R O o3 ^
^ a 03 . .
W O Q H PH • ' ^ 03
S H SH cS ;3 03 4^ fl 03
Ifl >-<
^ a ^ a is
03 ."S ^ - a ^ ^ -^ ^ b 03 r, '"' Id r f l T l bO -+-" 03
fa . f l 4J
171
4.9 Nomenclature
Gin evaporator inlet mass flow rate, kg/s
Gp product mass flow rate, kg/s
Gout evaporator outlet mass flow rate, kg/s
Kc controUer gain
L evaporator inventory level, m
M control horizon
MPC model predictive control
P prediction horizon
Ps steam pressure. Pa
PID Proportional-Integral-Differential control law
As discussed in Chapter 4 the multiple effect falling film evaporator (MEV) plant
represents an important part of the kraft pulp mill chemical recovery cycle. Thus,
any instabilities in the operation of the MEV will have significant impact on the
overall mill operation and profit. Having a fundamental model that describes well the
MEV plant permits an evaluation of the performance (stability, robustness, flexibility)
of different control strategies. Two basic control strategies wiU be investigated: a
decentralized single-loop feedback proportional-integral-derivative (PID) control and
centralized constrained model-predictive control (MPC) control.
6.1 Decentralized control
The first step in designing a decentralized control system is to determine the con
trolled and manipulated variables and to pair them in a proper fashion (see Chapter
4).
6.1.1 Evaporator inventories
For the MEV plant studied in this work, there are seven evaporators, including
the three units mounted in a single body that constitute the superconcentrator. The
evaporator plant operates on supply mode. The levels of the liquor inventories of E-5
to E-2 are controlled using the product stream mass flow rate. The feed stream to the
superconcentrator (SC-3) cannot be selected to control any of the evaporator units,
193
since the evaporator plant operates on supply.
A similar example of three tanks in a parallel conflguration was studied by Rugh
[82].
Fin
1
V
A 1 X2
A
1 X3
V - ^
Figure 6.1: System of three connected tanks
Following the modeling approach by Rugh, the dynamic model of the three con
nected inventories (see Figure 6.1) is
Xl = Fin - qiV^i - ^2
CX2 = qi\/xi - X2 - q2y/x2 - X3 (6.1)
c i s = q2\/x2 - X3 - F^t,
where x ^ R^ represent the three levels. Fin is the inlet flow rate to the first in
ventory, Fout is the outlet flow rate and also the selected manipulated variable, c is
the tank cross-section area, and qj,j = 1 , . . . , 3 are parameters related to the valve
characteristics and the line.
194
To analyze suitable control conflgurations, a hnear model of the system is devel
oped, using Taylor series expansion about the nominal states, x,Fout- The resulting
linear-time invariant (LTI) system is expressed by the conventional state space rep
resentation [82] where for convenience all parameters are assumed to have the value
one,
S — •AcyX + JDcyU (6.2)
where ^ = x — ic is a vector of the deviation variables, u — Fcmt ~ Font is a vector of
manipulated variables and
•Ar-i, —
- 1 1 0
1 - 2 1
0 1 - 1
•'^r.t)
0
0
- 1
The companion measurement equation is given by
y = c^^, (6.3)
where y = ^ and
f-'CV
1 0 0
0 1 0
0 0 1
(6.4)
The LTI system is completely state and output controllable. However, controlhng
the level in each tank is not necessary because the material balances in tanks two
195
and three are not independent of each other. Thus, if the level in the last tank is
controlled using the outflow, then the other two levels have unique values for the
given operating conditions. Therefore, only the level in the last tank needs to be
controUed. In other words
0 0 1 (6.5)
This also means that even the level transmitters on the first two tanks can be omitted
because the observability condition is satisfied.
Based on this analysis it is decided to use the product flow rates of evaporators
E-5 to E-2 to control their levels and to use the final product flow rate of SC-1 to
control the level in the superconcentrator.
6.1.2 Selection of controlled and manipulated variables
The selection of the set of controlled and manipulated variables is first guided by
process knowledge provided by the control configurations of existing multiple effect
evaporator plants (Tembec Crestbrook Mill, Minton [2], Billet [3], and Bojov [16]),
and second by the previous work on control of a single evaporator. By making use
of existing control configurations performance comparison can be made to assess
improvements.
The dry solids concentration in the product stream is an important property and
therefore is selected as a controUed variable. With respect to the manipulated variable
that can be used to control the dry solids concentration in the product stream there are
four possible choices. They include: the steam pressures at the superconcentrator and
196
the secondary vapor pressure at the first evaporator (E-5). The final choice is based
on the manipulated variable with the shortest time delay and smallest time constant
(proximity rule). Thus, the steam pressure at the last superconcentrator evaporator
(SC-1) is selected to control the dry solids concentration in the product stream. In
practice, to provide robustness to disturbances in the steam header pressure, the
control of the dry solids concentrator involves using two cascaded feedback loops.
The inner loop is a pressure controUer that regulates the steam pressure of SC-1.
The outer loop is the concentration controller that determines the set point of the
inner loop. The other two steam pressures at the superconcentrator are assumed to
be constant. This assumption is also applied to the secondary vapor pressure at the
initial evaporator {E-5). The proposed control scheme is provided on Figure 6.2.
6.1.3 Results
To validate the performance of the proposed decentralized feedback control struc
ture, the multiple evaporator plant PDE model is subjected to changes in the operat
ing parameters (Table 5.1). The steam pressures at the superconcentrator for these
tests (open loop responses) have been decreased from the nominal values to provide a
wider range to observe the responses to different disturbances. The range of dry solids
concentrations at which the real plant operates is between 0.7 and 0.75 kg/kg. Large
deviations outside this range may render the empirical correlations of the black liquor
physical properties incorrect and even infeasible. The nominal operating conditions
for the multiple evaporator plant are provided in Table 6.1.
197
03 03 03
o (X
03 r — 1
'o
o o
13 > 03
o -u cd ^ o a (Tl
> 0^ 03
Cu - ; j
;=! S 03
^ -u SH
n !-l-l 03 ! - l
Pl - t J 1:3 ^ S-l
- U OJ
.—H
(5 s-l
- t J
a 0 0
,^ 0 oj
r.Q -D 03
M-l
T l 03 N
CO SH
l-l 03 C3 03
Q
CM
CO
03 SH
Pl br fc
! H
03
n - t J
PI 0 0
i=l 0
• 1—1 •l-J
en 0 a B n r i
0 0 s-T OJ
0 SH
- tJ
rt 0
198
Table 6.1: Operating conditions of the muhiple evaporator plant
Parameter
Gfeed
Xfeed
Tfeed
Psteam ( S C - 1 tO SC-3)
^condenscT
116
Value
kg/s, 50/50 split
0.145 kg/kg
303.15 K
460 kPa
21.4 kPa
The disturbances applied to the system are a 5% decrease in the dry solids con
centration in the feed stream, a 5% increase in the mass flow rate of the feed stream,
and a 7% decrease in the temperature of the feed stream. The tuning parameters of
the single-loop controllers are the same values as those used in the feedback control
of a single evaporator (see Table 4.2). The closed-loop responses of the nonlinear
multiple evaporator plant (nonlinear system of partial differential equations) to the
change in the operating conditions are shown in Figures 6.3 to 6.5.
The results show that in all cases the disturbances are compensated within 40
to 50 minutes. The responses of both the controlled and manipulated variables are
smooth and the size of the control action does not imply excessive changes. The
length of time it takes to bring the dry solids concentration of the product to within
the iblcr limit of 0.001 kg/kg are listed in Table 6.2.
In Figure 6.5, to compensate for the decrease in the dry sohds concentration
the steam pressure is decreased. This is in contrast to the results of the the closed
199
1 0 , 7 3 ;
Figure 6.3: Closed-loop response of the nonlinear multiple evaporator plant to 5% increase in the flow rate of the feed. Top: dry solids concentration in the product stream, bottom: steam pressure of SC-1. The dotted lines represent the ±lcr limit of 0.001 kg/kg.
Table 6.2: Performance of the single-loop control structure of the nonlinear multiple evaporator plant
Disturbance
5% increase of Gfeed
5% decrease of Xfeed
7% decrease of Tfeed
Time to ±1(T
27.8 min
20.6 min
20.4 min
loop tests provided in Chapter 4 where the pressure was observed to increase as
the dry solids concentration in the product stream decreased. It is also observed
that initially after the disturbance in the dry solids concentration is applied, the dry
solids concentration of the product stream increases. This behavior while unexpected
can be explained by the physics of the process. That is, when the evaporator plant
produces a product with dry solids concentration in the range of 0.7 to 0.75 kg/kg,
the boiling point elevation increases significantly (see Appendix C, Equation (Cl) ) .
200
^ i w 0.735
5
^ 0.73
'
\
1
0 10
fl 650
& S 600
s S. 550
5 500
u 10
'
20
'
, 70
30
'
1
30
'
4 0
'
1
40 Time, mli>
'
50
1
50
1
60
'
60
^
7 0
1 -
7 0
.
8
BO
Figure 6.4: Closed-loop response of the nonhnear muhiple evaporator plant to a 7% decrease in the temperature of the feed. Top: dry solids concentration in the product stream, bottom: steam pressure of SC-1. The dotted lines represent the ±lcr limit of 0.001 kg/kg.
For such operating conditions, when the dry sohds concentration in the feed stream
is increased, any consequent increase in the dry solids concentration of the product
stream is eliminated by the increase in the boiling point elevation. The increased
boiling point elevation decreases the total available temperature difference (driving
force) of the multiple evaporator plant. Thus, the total effect of the increase in the
dry solids concentration of the feed is a decrease of the dry solids concentration of
the product. More details on this behavior of the system is provided in Section 6.2.2.
6.2 Centralized control
In this control strategy, a model predictive control (MPC) formulation is inves
tigated, see Figure 4.21 in Chapter 4. The performance of the MPC controller is
demonstrated on the nonlinear ODE and PDE representations of the multiple evap
orator plant. The strategy reUes on a model of the system to determine the control
201
^ 0.7421-
«• "•'* " • 0.738 -
§ 0,736 -
o
^ 0 734 -
« 480
Figure 6.5: Closed-loop response of the nonlinear multiple evaporator plant to a 5% decrease of the dry solids concentration of the feed. Top: dry solids concentration in the product stream, bottom: steam pressure of SC-1. The dotted lines represent the ±la limit of 0.001 kg/kg.
actions now and in the future. As discussed in Chapter 4, this formulation uses a
linear-time invariant (LTI) model (ordinary differential equations) or an input-output
(convolution) model. To generate the LTI model, first a nonlinear ordinary differential
equation (ODE) description of the multiple evaporator plant is developed and vali
dated. Second, the validated nonlinear ODE model is used to develop the LTI model.
Third, the LTI model is transformed to an input-output (step response) model, which
is used to develop the MPC controller.
6.2.1 Nonlinear ODE model
The development of the nonlinear ODE or lumped model of the multiple evap
orator plant follows the development of the lumped model of a single evaporator
discussed in Chapter 4 (see Equations (4.33), (4.37) and (4.44) to 4.47)). The model
developed in Matlab's®: Simulink ® : environment. A computer program is was
202
provided in Appendix D.
To validate the lumped ODE model, the responses of the distributed PDE model
and the lumped ODE model are compared for the same changes in the operating
conditions. The responses are shown in Figures 6.6 to 6.11.
.XlO' Evaporator 5
B)
X
<
0 20 40 60 80 100 120
0.02
0,01
0
-0.01
-0.02 0 20 40 60 80 100 120
Time, min
0.04
i 0.02 o
< 0.01
n
Superconcentrator 1
^ '
-
PDE . ODE ,
1
^ 0.5 •D
< -0.5
-1
0 02
0.01
E
< -0.01
-0,02
0
0 20 40 60 80 100 120
20 40 60 80 100 120
"0 20 40 60 80 100 120
Time, min
Figure 6.6: Responses of nonlinear PDE (solid line) and ODE (dotted line) models of the multiple evaporator plant to a 5% decrease in the flow rate of the feed. Left column: Evaporator E-5, right column: superconcentrator SC-1. Top: dry sohds concentration of the product stream {AXprod), middle: temperature (ATj^od), and bottom: inventory (AL).
The responses show that the nonlinear ODE and PDE models have similar trends.
The profile of the responses of the most important variable, the dry sohds concen-
203
i <
0
-2
-4
-6
1
0,5
0
-0.5
-1
0.02
.XlO'" Evaporator 5
\ \ ''' ' r /-" ^
• . . .
PDE " ODE ,
E _ i "
<
0.01 L
0
-0 01
-0 02 0 20
0 20 40 60 80 100 120
r
0 20 40 60 80 100 120
40 60 80 Time, min
100 120
§, -0,02
'•a
X^ -0,04 <
-0,06 0
E -J <
Superconcentrator'
V-,. N. " - - - . . . . .
PDE ODE
\ ^ ^
20 40 60 80 100 120
20 40 60 80 100 120
20 40 60 80 100 120
Figure 6.7: Responses of nonlinear PDE (solid line) and ODE (dotted line) models of the multiple evaporator plant to a 5% increase in flow rate of the feed. Top: dry sohds concentration of the product stream {AXprod), middle: temperature {ATprod), and bottom: inventory (AL).
tration are similar and the steady state gains are on the same order. The responses
to feed temperature disturbances show that the nonlinear ODE model is less sensi
tive to this particular disturbance, especially in the case of the first evaporator, E-5.
The temperature disturbances effect on the superconcentrator evaporator SC-1 are
represented with satisfactory accuracy.
There are noticeable differences in the profiles of the temperature and level re
sponses in some cases (feed flow rates and feed concentration changes). This may be
204
40 60 80
Time, mm
100 120
X <
10
5
0
-5,-
XlO"
E
<
PDE ODE
20 40 60 80 100 120
20 40 60 80 100 120
Time, min
Figure 6.8: Responses of nonlinear PDE (solid line) and ODE (dotted line) models of the multiple evaporator plant to a 5% decrease in dry solids concentration of the feed. Top: dry solids concentration of the product stream {AXprod), middle: temperature (ATprod), and bottom: inventory (AL).
amplified because the controller parameters of the level controllers are different. Oth
erwise, it is always possible to tune the level controllers in the both cases to achieve
a closer correspondence. This was not done to demonstrate the effect of level on the
performance of the models.
In can be concluded, that the nonlinear ODE (lumped) model is suitable to develop
a centrahzed control strategy for the multiple evaporator plant.
205
X
<
.XlO"" Evaporator 5
E
<
-0 02
/<^"^~' ^^^^""""^
/
• I I I
^LZZlZ^lllZ^lZ,!^
PDE ODE •
•
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Time, min
XlO' Superconcentrator 1
100 120
100 120
"0 20 40 60 80 100 120
Time, min
Figure 6.9: Responses of nonlinear PDE (sohd line) and ODE (dotted line) models of the multiple evaporator plant to a 5% increase in the dry solids concentration of the feed. Top: dry sohds concentration of the product stream {AXprod), middle: temperature {ATprod), and bottom: inventory (AL).
6.2.2 The effect of concentration disturbances
The nonlinear ODE model can be used to study the operation of the multiple
evaporator plant. It also has an advantage in that the ODE model can be integrated
in about 20-30 seconds on a Pentium''''^III 1.33 GHz machine). In comparison, the
nonhnear PDE model reqmres 24 hours on 7 1.2 GHz AMD™Athlon® nodes of a
Rocks cluster^ for the same duration.
^ Rocks is an open source high performance Linux® cluster solution. For more information, see www. rocksclusters .org
206
X <
Evaporator 5 Superconcentrator 1
PDE ODE -i
X <
PDE ODE
0 20 40 60 80 100 120 -0,03
0 20 40 60 80 100 120
0 20 40 60 80 100 120 20 40 60 80 100 120
0,02
0.01
-0 01
-0 02
o f ^
0 20 40 60 80 100 120
Time, min
0.02
0,01
0
-0 01
-0 02 0 20 40 60 80 100 120
Time, mm
Figure 6.10: Responses of nonlinear PDE (sohd line) and ODE (dotted line) models of the multiple evaporator plant to a 5% decrease in the temperature of the feed. Top: dry solids concentration of the product stream {AXprod), middle: temperature (ATprod), and bottom: inventory (AL).
The performance of the nonlinear ODE model of the multiple evaporator plant
is evaluated at different operating conditions and different boiling point elevation
coefficients {K). The results are summarized in Table 6.3 and are given as deviations
from the final dry solids concentration values in the product stream.
The results show that in the range 6.5 < A" < 7 that is associated with the
calculation of the boiling point rise, depending on the value of the steam pressure,
there is a point at which the effect of increases to the feed dry sohds concentration
207
Ul
^
X <
XlO Evaporator 5 Supeconcentrator 1
/ •
PDE ODE
•
0 20 40 60 80 100 120
kg/k
g
•D O
a. X <
0,03
0,02
0,01
0
-0.01
. ^^-^^^^
• • ' ' '
PDE . ODE
, 20 40 60 80 100 120
y-<
E
<
E
<
20 40 60 80
Time, min
20 40 60 80 100 120
Time, min
Figure 6.11: Responses of nonlinear PDE (solid line) and ODE (dotted hne) models of the multiple evaporator plant to a 5% increase in the temperature of the feed. Top: dry sohds concentration of the product stream (AXprod), middle: temperature (ATprod), and bottom: inventory (AL).
on the product dry solids concentration is reversed. That is, for different increases
in the feed dry solids concentration, the product dry solids concentration increases
at some operating conditions but decreases at others (see Table 6.3). At low dry
solids concentrations (< 0.63 kg/kg), the response of the muhiple evaporator plant
to a decrease in the dry solids concentration if the feed stream leads to a decrease
in the dry solids concentration in the product stream. This resuh was observed in
the open loop tests of the PDE model (see Figures 5.4 and 5.5). However, ff the
208
Table 6.3: The effect of boiling point elevation to a 5% decrease in the feed dry sohds concentration.
Steam pressure Xsci Boihng point rise AXprod/^XPDE,% at SC-1, 2 and 3, kPa coefficient, K
410
390
380
360
410
390
380
360
360
340
320
300
0.6560
0.6330
0.6209
0.5954
0.6747
0.6514
0.6391
0.6129
0.7188
0.6877
0.6533
0.6160
7
7
7
7
6.5
6.5
6.5
6.5
4
4
4
4
18
5
-5
-15
20
8
?»0
-17
-5
-25
-49
-75
boihng point elevation is less sensitive to changes in the dry solids concentration
(e.g., a boiling point rise coefficient of i<' = 4) then a dry solids concentrations > 0.7
kg/kg in the product stream does not result in a reverse response, i.e., for a decrease
in the feed dry solids concentration there is a decrease in the product dry solids
concentration. The reduced pressures in the case of K = 4 are necessary to keep the
dry sohds concentration in the product within feasible limits so that the correlations
used to calculate the physical properties of the process fluid remain valid. It suffices
to conclude that the boiling point elevation of the process fluid determines to a large
209
extent the response of the evaporator plant to changes in the dry sohds concentration
of the feed
6.2.3 Linear model
The linear-time invariant state-space model of the multiple evaporator plant has
been calculated analytically, using Maple'^'^ The procedure is similar to the one used
in Chapter 4; the Maple code is presented in Appendix D. The resulting state-space
representation matrices A,B, and C also are given in Appendix D.
In the development of this linear model, the equation that models the heating zone
length is omitted (Equation (4.44 in Chapter 4)). AU levels are assumed constant
with a value of 1 meter. The manipulated variables are the same as in the case of the
decentralized control - the steam pressures at the superconcentrator (SC-1 to SC-3)
and the secondary vapor pressure at E-5.
The real part of all thirty-two eigenvalues of the dynamic constant coefficient
matrix A are negative; hence open-loop stability is assured [82, 84]. Two of the
eigenvalues are complex conjugates, which would imply some oscihations in the open
loop responses. However, the magnitude of the imaginary parts of these eigenvalues
is very small (order 10""* when compared to the largest eigenvalue of the system
(-2.1465). Thus, the oscihations do not dominate the response. In general, the
system has a dominant mode that corresponds to the largest eigenvalue. All the
other eigenvalues are much smaller between one to three orders of magnitude, see
Appendix D).
210
The controUability analysis shows that the system is not state controllable [84];
the rank of the controUabflity matrix is four but the number of states is thhty-two.
Since only four of the states are assumed measurable, output controUability analysis
can be used to establish the controUabihty Suppose that the measured variables are
the dry solids concentrations of E-5, SC-3, SC-2 and SC-1 in their respective output
streams. Controllability analysis shows that the system is not output controllable
(the rank of the output controUabflity matrix is 3). Another potential set of control
variables include ah three superconcentrator dry solids concentrations in the recircu
lating streams and the multiple evaporator plant product (SC-3) temperature. For
this set of four controlled variables, the output controUabihty condition is satisfied,
the rank of the output controllability matrix is four.
The responses of the linear model to changes in the manipulated variables are
shown in Figures 6.12 to 6.17.
The responses show that the linear model represents the nonlinear model well.
The steady state gains are comparable. Further analysis shows that there are smaller
differences between the response of the linear and nonlinear models of the evaporator
that is closer to where the change is implemented.
6.2.4 Results
The closed-loop performance of the centralized unconstrained and unconstrained
MPC controllers, and the decentralized single-loop (PI) controllers are compared for
disturbance rejection. In the PI control scheme, the dry solids concentrations at each
211
ID
o, -2
o .4
XlO
\ \
1 •
NLODE LODE
' ^—' ^ 50 100
Time, min
150 50 100
Time, min
150
.x10
S
50 100
Time, min
50 100
Time, mm
Figure 6.12: Responses of the nonlinear (sohd hne) and linear (dotted line) ODE models to a 5% decrease in the superconcentrator steam pressures.
superconcentrator unit is controlled by the corresponding steam pressure; the product
temperature is controlled by the secondary vapor pressure at E-5. The parameters of
the MPC and PI controller are listed in Table 6.5. In this table, yi is the dry solids
concentration at SC-3, 2/2 is the dry solids concentration at SC-2, ys is the dry solids
concentration at SC-1, and 2/4 is the temperature of SC-1. Also, Ui,U2 and U3 are the
steam pressures at SC-3, SC-2, and SC-1, respectively; and u^ is the secondary vapor
pressure at E-5. Note that all values in Table 6.5 are deviations from the nominal
conditions since the linear model is developed using deviation variables.
The responses of the controlled variables to the different disturbances are shown
in Figures 6.18 to 6.41. The mass flow rate, dry solids concentration, and the temper-
212
NLODE LODE
50 100
Time, mm
150 50 100 150
Time, mm
XlO
50 100
Time, min
150 50 100 150
Time, mm
Figure 6.13: Responses of the nonlinear (solid line) and linear (dotted line) ODE models to a 5% increase in the superconcentrator steam pressures.
ature of the feed have been decreased by 5% each and the heat transfer coefficients
of the evaporators have been decreased by 10%. The performance in all cases is as
sessed by the length of time each response reaches ±lcr limits of 0.001 kg/kg, and the
integral of the absolute deviation between the set point (y^p) and the measured value
{y{t)).
I A E = / \ysp - y{t)\dt. Jo
(6.6)
The time to reach ±la limit is for each disturbance is given in Table 6.4. The values
of the lAE are given in Table 6.6.
213
XlO
/
,
NLODE LODE
50
Time, min
100 150
0
0,2
04
06
0,8
-1
1 1
XlO •
i •
NLODE LODE -
^ " :
50 100
Time, min
50 100
Time, mm
150
Figure 6.14: Responses of the nonlinear (solid line) and linear (dotted line) ODE models to a 5% decrease in the steam pressure of the superconcentrator evaporator SC-1.
6.2.4.1 Unconstrained and constrained MPC controUers
All the closed-loop responses of the unconstrained and constrained MPC con
trollers are stable. There are no offsets in the responses of the controUed variables
and the approach to set point is continuous without excessive oscillations or discon
tinuities. Similar descriptions can be stated for the controller actions (manipulated
variables).
The closed-loop responses show a greater sensitivity to throughput (feed mass flow
rate change) changes. This is reflected in the length of time the response takes to
reach ± l a limit of 0.001 kg/kg (see Table 6.4) and the values of the lAE (see Table
214
, X 1 0
I O
<
100 150
s
<
/\
3
2
1
0
-1
XlO
'
NLODE LODE
. Time, mm
50 100
Time, mm
150
. X l O
50 100 150
Time, mm
50 100
Time, mm
150
Figure 6.15: Responses of the nonlinear (sohd hne) and linear (dotted line) ODE models to 5% increase in the steam pressure of superconcentrator evaporator SC-1.
6.6).
The ± l a limit is used as a constraint in the MPC formulation of only the final
product stream of SC-1. Constraints on the other two dry solids concentration limits
are not as stringent since only the final dry solids concentration should be maintained
in a narrow range. The temperature of the product stream of SC-1 is constrained
to within ± 3 K from the desired set point. The limits on the manipulated variables
are selected to satisfy reasonable process limits such as maximum on the steam con
sumption or a minimum on the secondary vapor pressure at E-5. The rate of change
for the manipulated variables is at most 30 kPa/min.
The closed-loop responses to changes in the temperature of the feed and the heat
transfer parameters disturbances are similar. The response with respect to feed dry
215
0,01
0 008
•a 0 006
5^" 0 004 <
0,002
°0
. — . ^ /
/ / • "
,•••'
.
— NLODE LODE
. 50 100
Time, mm
50 100
Time, min
150
150
>r
, x 1 0
/ " :
/ / " "
''•
,
NLODE LODE
50 100 150
Time, min
50 100 150
Time, min
Figure 6.16: Responses of the nonhnear(solid line) and linear (dotted hne) ODE models to a 5% decrease in the secondary vapor pressure of evaporator E-5.
solids concentration is less sensitive when compared to the responses for the other
disturbances. The dry solids concentration in the product stream does not violate
the ±la limit for either MPC formulations.
It takes approximately 13 minutes for the closed-loop responses to reach ±lcr
limit for all disturbances, except in the case of the change in the feed dry sohds
concentration where the product dry solids concentration never leaves these limits.
Also, it is observed that the effect of a change in the dry solids concentration at SC-3
is more significant than at SC-2 and SC-1 based on the larger control actions (steam
pressure Pscs) acting on SC-3.
Data from a pulp mfll operated by Tembec, Inc. (St. Francisvihe) show that
216
X <
-10
xlO
K —" 1
NLODE LODE •
-
\
50 100
Time, min
150
0
-0,002
•a "° °°''
b -0,006
< -0,008
-0,01 •
-0,012
NLODE LODE
0 50 100 150
Time, min
50 100
Time, mm
150 50 100 150
Time, min
Figure 6.17: Responses of the nonlinear (sohd line) and linear ODE models to a 5% increase in the secondary vapor pressure of evaporator E-5.
variations with amplitudes as large as the ±2cr limit do not affect the stability of the
recovery boiler. The effect of deviations in this dry sohds concentration range can
manifest themselves only in the longer term operation of the boiler where changes in
the burning characteristics of the black liquor can be detected.
The results show no constraint violation on the inputs, outputs, and rate of change
of the inputs in all cases. The controller actions (maximum of rate of change of 30
kPa/min is applied in 1 minute or 15 kPa every 30 seconds, which is the sample time)
are more aggressive to compensate for feed mass flow rate, feed temperature, and heat
transfer disturbances on SC-1 (see Figures 6.27, 6.31, and 6.33). The lAEs calculated
for the constrained MPC controller are larger when compared to the unconstrained
217
Table 6.4: Times to reach la limits of 0.001 kg/kg of the multiple evaporator plant product dry solids concentration.
MPC PI
Disturbance Unconstrained Constrained
5% increase of Gfeed 13.74 min
5% decrease of Xfeed
5% decrease of Tfeed 12.97 min
10% decrease of hs 12.78 min
case. This is not unexpected whenever constraints are active.
31.1 min
59.92 min
30.34 min
6.2.4.2 PI control
In the case of the decentralized strategy that employs single PI loops, the responses
are oscillatory with a period of approximately 12.5 minutes. The length of time to
reach ±la limit in the range of 30 to 60 minutes. This is longer when compared
to the performance of the both MPC controUers. The ±2a limit is violated in the
case of temperature and heat transfer disturbances. This duration of the violation
is not excessive, however this shows that the unconstrained MPC controller achieved
better results as a function of the selected tuning parameters. The sensitivities to all
disturbances are similar to those observed in the case of MPC control.
The most important comparison between the MPC and PI controllers is in the
values of the lAEs. The lAE values of the closed-loop response under PI Control
show that the dry solids concentrations at SC-3 and SC-2 are more tightly controlled
218
when compared to the values obtained in the case of the MPC controllers. This
can be advantageous if tight control at SC-3 and SC-2 is necessary However, since
the most important variable to be controlled is the dry sohds concentrations of the
product stream at SC-1, and since it is not as tightly controlled by the PI controher
when compared to the responses of both MPC controllers, it can be concluded that
the performance of both MPC controllers is better than the performance of the PI
controher.
6.2.4.3 Control of the PDE system
Closed-loop control of the PDE system that describes the multiple effect evapora
tor plant is carried out in the same manner as in Chapter 4. That is, the manipulated
variables found from the control of the nonlinear ODE description of the multiple ef
fect evaporator plant are implemented on the PDE system when disturbance in the
feed mass flow rate are present. The closed-loop response of the PDE system is shown
in Figure 6.42. The response shows that the MPC controller actions provide excellent
compensation. The dry solids concentration in the product stream of SC-1 is regu
lated to within ±la limit in about 20 minutes. Also the temperature of the product
stream is maintained at the desired set point. It can be concluded that nonlinear
ODE representation and the subsequent linear ODE derived from it is satisfactory to
develop an effective MPC controller for the actual multiple effect evaporator plant.
219
a
s « s o
03 Ss
^ o o
o
o3 SH O
a OJ
3
a cn SH a;
a o o
c (33
o O H
a SH 03
bjO
fl
LO
i o
ft faO el
•3 3
d o o
O
O
o O 10
PH S
o
o
s
I
Ss
N O N
II a -H o o .a a
PH
O
o
bO
O
O
r i 4
O O
bO r i 4
O
O
bC
bjO Ai
O
o
ft 3
O
o LO
o o C<l
o o
o o CM
a
s s
<
a S
H
a
s
LO
AJ
bO
a 'a p
o a o o
l-H OH
3000
L
oop
XC
3
T-H
Loo
p T
CI
cS PH j ^
m + 3
a ^ 4 J
1=1 ft
1—1
0 0)
4 J 03 I H
- t J
3 ft
03 PH
-^ cn + j
1 CU bC fl
cha
o o o o o o 1>- CO
ft 0 0
HJ
l -H
0 X
ft 0 0
1-J CN 0 X
220
CQ
a>
iab
3 >
-d
)lle
SH
cont
th
e I
o SH O
err
'o CQ -O o3
egra
]
fl 1—1
CO
CO
OJ
Tab
1—1
a.
o OH
S
ined
tr
a
cn fl O
O
o OH
ned
M
• 1—1
03
cons
fl
• ^
Si
S
C<l 5 5
S
S
CO ; = i
cq S i
55
^
S
C-1 SJ
I-H
S i
CN 1—1
CN ^ H
<d>
^ o CD
I - H
CN
CO
CO
o CN
^ O
CO
CX) 0 0
CO
CN
1-H
O
o CN
CO
o CO
C i
0.8
1.76
T3
^ U
'* 05 6? LO
+
LO CN
CN
o CO
CO CO
CO
LO
o CO
o CO
CO I-H
0.0
CN 1—1
CO
I—1 CO
o CO
CO 1—1
o o
CN ^ H
CO
r-H
CO
t J u <u
, ' ^ >< 6? L O
1
L O I-H I-H
N -r-H
CO
LO o CO
N -CN
o
LO CN CN
CO
0.0
CTi
CO
t^
i-H
o CN CN
LO
o CO
CN
0.9
1.79
•a w <u
- • ^
E-i
L O
1—1
CO I-H
CN I—1
CO
^ o CO
CN oq CO
LO CO CN
CO
0.0
1—1
CN
CO CO CN
CO CO
CO
I - H
I—1
CN
K)
-c; fe? o I-H
1
221
CD O fl 03
SH
OJ ft
6>i/6>i • los sa V 6>i/6>feos'sossav M ' I O S I V tJU'tOS ja^a-i
_ CD fl - ^
+3 4-i CD CJ
^ ^ CD O
S .tl
co" I
o CO -d fl 03
S
T3 CD
U-l
03 CN
°6 a CO o , ^
go
ft.2 fl OJ
—• CJ SH fl (D O
1=1 O
2 ^
CJ CO
o b PH T3
' s o) o fl CJ
. f l CD d M SH
+^ Cl cn >H
1=1 .2 O 4^ O 03 C SH ^ ^ CD <D
^ 8 8:2
S b • cS -c V S ^ O ft ^ ^ ftl'S
-^ ^. ^ CD ' ^ fl cn y^ O) - O >
fl
00 -S
^ g I S) £ fl
. .H . r t _^j M H &-I 03
cn O
O
222
B>|/6>( ,|.DS B>|/6>| •^^ '^d ' " ^ d
,puoo Bed d
s o3 CD
- | J
cn
' f t O
- ^ j
CD - f l 4 J
s o
^ • ^
cn SH
fe ai o fl
s-l fl
- ^ J
cn
^ <D
+ J 03 SH
^ o cfl cn cn ai g
X) <D
. 1 ^ U H
c6 M H
O <D o fl CD cn <D SH
ft CD
- f l
_fl
cn fl
_o +3 CJ 03
rol
ont
c; D PL,
-a CD fl '3
SH
nst
o o fl
^ OT , -H
CO CD s-l
fl bO
fo
LTD
H <+-! o <D SH
fl cn cn 0) SH
ft ; H
O ft 03
> >. s-l 03
X) fl O CJ OJ cn
a SH
^ •IS
.. CO 1
O CO -o fl 03
CN
6 CO t lH G cn CD S H
fl cn 73 CD SH
ft
i CO
-o fl O O CD
cn i-H
1
CO H H
O ID SH
fl cn cn CD SH
ft
223
p
CD
cn CD
.tos, B>l/B>|"-'^SaV B>|/6>| ' "^sa V ' "^sa V M ' l O S l V LU '10SI3A3-1
o3 SH
+ J fl CD
one
o cn
T3
'o cn
413 + j
Co' 1
O CO -a fl o3
CN
6 >> CO SH
- d
- d
fee
03
( 4 H O CD CJ
fl CD cn <U s-l ft <D
Xi -i-j
rt SH
a>
trol
fl o o
OH
S
ined
03 SH
-»-= cn fl O O fl
the
u U-l O CD CJ
B 03
a SH
c2 SH <D
ft ft O o
Td <D cn O
O
ctj CO
QJ SH
fl
fa
^(H
O t/3 fl _o o3 SH
+3 fl CD CJ fl o CJ
cn r2 'o cn >, SH
T3
T3 fl o O <D
cn
fl" _o 03 SH
- IJ fl CD O fl O CJ
cn -o ^ O cn
SH
CJ fl
x) o SH
ft
^ U ^
ft o -t-3
(D ,.fl
a o
^ -f^ cn SH
E CD CJ
fl 03
rO SH
fl cn
1 — ! 1
O CO t4H O > i SH O
- t ^ fl CD
> fl - > — 1
xi S H
fl o ^ H
! H
fl - t J 03 SH (D ft 6 CD
- t J
+ i CJ fl
T3 O s-l
ft
224
Bd'^^^d B>|/6)| '^^^d '^^^d ,puoo
Ed d
cn SH
fa CD CJ fl o3
Xi S H
fl - cn ^ fl
.2 ' • ^
cfi
entr
CJ
fl O o cn S "o cn
>. S H
T3
X) CD (D
^ o3 ^4H
o OJ o fl 0) cn (D S H
ft CD
-fl 4-3
,fl cn fl _o '•w CJ 03
ntro
l
o CJ
O PH
S X3 OJ
fl •3
SH - ^ J
cn fl 0 CJ
fl '^ I-H CN
CO
a; SH
fl , _bfl
Lfi
W • ^ H
0 CD SH
fl cn cn CD SH
ft ! H 0 ft
> SH Cfl
X ) fl 0 0 QJ m
XJ SH
s - ^ J
0 0 " 1
0 CO X3 fl o3
CN
0 CO SJH
0 cn CD S H
fl CO
cn OJ SH
ft
a CD
cn
XJ fl 0 CJ CD cn
I-H"
6 CO <4H
0 CD S H
fl cn cn 0) SH
ft
a o3 OJ
- ^ J
cn
ft;
225
fN O a
o CN o o o
6>l/6>l S a V B)(/B>) •"^sa V '"^sa V >1 ' lOSiV UJ 10S|3A31
cn
^ fa (D CJ
fl 03 JD SH
fl cn
X !
fl)
fl - tJ 03
CD
oT JH
fl en SH
m ft
a CD - M
fl X! o SH ft
X f >H
0'-^ a CD
X5 CD
03 M H f5
QJ
enc
UJ ID SH
ft (11
4 ^
CO
O CO T l
fl cri
CM
o ( / J 1*H
o rn
fl xi .2
CD (D 1=1 CJ O fl SH O
CJ X3
o cn
O PLH
•xs X3 CD fl 03 SH
cn fl O CJ fl fl CD
O CD CJ
a SH
,o
Xi fl O CJ CD cn
_o o3 SH
^ OJ CJ fl o CJ cn
O
ft SH
ft^ o 1
X3 CD cn O
O CM CM
O CO
CJ fl
X) 2 5 ft >,
s-l
O ft o
CO OJ
Xi CD +^
bjO p , f - l
fa
rfl SH fl
.O
226
03 CD
ft O
Bd'^^^d
• f CD (N 1 -•sr ^ ' ^ - ^
Bd '"^d ''""d
^ 2.2 T - CN
,PU03 Bd d
CD
o
CD V LO
^
• I-H
X )
ture
o3
per
S OJ
• « CD
t+H
o3 14H
O <U CJ
sen
CD SH
ft OJ
Xi - M
_fl CC fl
_o +3 CJ o3
O SH
-tJ fl o o
O fa S X3 CD fl
•3 SH
cn
con
fl
CO ( N
CO
QJ SH
fl bO
•1—I
fa
fl CC
pres
s-l 0 ft o3 P SH 03
•X} fl 0 0 CD cn
XJ SH
2 r .
CO
6 CO X ! fl o3
CN 1
0 CO M H 0
CO CD SH
fl CC cn CD S H
ft a OJ
- (J CD
XJ fl 0
sec
I-H" 1
0 CO <4H
0 OJ S H
fl cn CO 0) SH
ft
227
6>I/B>I' tOS 'X V B>|/6>|'£os'SOSXV >1 ' lOSiV uu ' l O S |8A31
CO
OJ CJ fl OJ cn OJ SH
ft CD
-fl 4 J
.a SH
,S 'o SH •h^
fl O CJ
O fa S X 3 CD fl
•3 S H
CO
fl 0 CJ fl fl OJ
- f l - t J
C4H 0
CD CJ
fl 03
a S H 0
t4H SH
CD
ft
03
CN 1
0 CO ' J H
0 cn C _o o3 SH
- t J
fl CD CJ fl 0 CJ
cn
73 3 cn
S H
T 3
XJ fl 0 CJ OJ cn
fl""
_o VJ o3 SH
4 ^ fl CD 0
fl 0 CJ
cn
r—t
0 cn
SH
ft X3
-loo
X ! OJ CO 0
0
'^ CM
CO
OJ s-l
fl fcuO Si
CJ
fl T 3 0 SH
ft
' f t 0
- t J
CD
r f l 4^
a 0
- M
1—t
1
CO
^ >. SH
0
fl <D
> _fl
r f l SH
fl 0
228
03 OJ
Bd '^^^d Bd '^^^d PUB ^^^d ,puoo
Bd d
ft o - ^ J
CD r f l - i J
a o
^ - t J cn s-l
fa ai o fl o3
Xi SH
fl - tJ cn
^ s-l
.CD t 4 H cn fl 03 SH
• ^
heat
o3 «4H
O OJ
enc
03 CD S H
ft CD
Xi - u
fl • -^ cn fl .2 '-^3
% •—^ o s-l - t J
fl O CJ
O fa S X3 OJ fl
•3 - f j
cn fl 0 CJ fl U)
LCD
fa u-i 0 OJ
fl cn cn OJ SH
ft S H
0 ft o3 P S H
Cfl
-d fl 0 CJ CD cn
X3 S H
13 -^^ CO~
6 CO
X3 fl a
CN
6 CO ^ H
0
cn CD SH
fl cn cn <D SH
ft
a 03 CD
- u cn
xJ fl 0 CJ OJ CO
^ r H
0 LO CO CM 'JH
CO
0) S H
fl bT)
fa
0
OJ S ^
fl CO GO CD SH
ft
229
Bt/BH' los sa V
O CD 9
B>|/6>|'EOS'20SSaV >l ' lOSiV <JU '1.0S |3A31
CO ! H
fa CD CJ fl o3
r Q S H
fl 4 J CO
^ CD
4J o3 s-l
^
fio
CO
^ a
-Td CD (D
t+H
03
C4H O aj CJ r^ 03
res
ft OJ
r f l 4 J
fl • ^
SH CD
trol
l
fl o CJ
O fa S XJ ID fl
•3
str
fl 0 CJ
t4H 0 CD CJ
fl cS
a s-l 0
<a S H
CD
ft
OJ SH
fl
S H CD ft
a 03
-1-^ 0 fl
X3 0 SH
ft
xi SH
2 4 J . CO
1
U CO X3 fl 03
CN
6 CO u-l 0 1/3 fl
, 0
03 SH
- H fl QJ CJ fl 0 CJ
CD
' 0 CC
>, S H
X!
-id fl 0 CJ OJ cn fl .2 + 3 o3
4 J fl QJ CJ
fl 0 CJ CO
X) ' 1
0 cn
s-l ft X3
-loo
X3 CD CC
0
0
CO ( ^ CO QJ SH
fl bO
• I-H
fa
0 fl
X3 0 S H ft
' f t 0
-4J
QJ
rfl -^J
a 0
l-H 1
CO f 1 1
0
>> SH
0 •g QJ
> .a r f l - * J
SH
fl -2
230
• ^
Bd '^^^d B d ' ^ = ^ d ' " ^ d
,puoo Bd d
i ft o CD
Xi
o , S H
cn S H
QJ CJ L O
-2 ^ 03 SH fl CO CC QJ
CC
o3 S H
o qfl 03
r-i '- '
X) QJ
o3
-o fl O ^ ^
QJ I f l
^ CO"
1^ ft CO
. 3 CN
Q CO
CJ
a ro
l
fl o CJ
O fa ^ X )
ine
03 SH
nst
o
r-rsi CO QJ s-l
fl bC
CJ
re CI) S H
fl cn
res
ft
a 03 03
c/>
X3 fl O o 03 CO
1-H
U CO <4H
o QJ
9 C/J
QJ
fa ft
231
CJ
fl X! O
B>l/B>|''-'''saV B^/B^ '"^SQ V '^^^sa v >I'L0S1V Ul LOS |9A31
fl .2 a
CO
X3
CO
Td fl ai
s-l CO
X) ° QJ cn
.Q:" fl
tjH g
03 *4J
- 2 fl QJ o fl O CJ QJ
r\ cn ft -^ CD ^
-fl O • ^ cn rH > .
fli x i fl o
SH
o CJ
O fa
o CJ QJ cn fl'~
_o '-13 03
Td P! S QJ
o3 O
§;H CJ - Q
^ H CO
a rS ^ r x l 'JH o o o
^ SH . !H ft p QJ ^ ^ SH
ft •• o ft p a o ° s; o fl
-Tl 03 . f l X ) - ^ . . Q "^ ^
^ a O r- ^
p^ O fl
OO -Jf 03"
^ S 1 QJ . S-l
s!i « Q
^ a a fa ,o .5
232
Bd'^^^d B>|/B>| ' " ^ d ' " ^ d ,puoo
Bd d
ft c 4 J
QJ
r f l - ^ J
a o
^ - t ^ 03 S H
fa QJ CJ
fl 03
r O S H
fl - t J CC
^ c O
ati
entr
CJ
fl o CJ 03
rs "o 03
>> S H
X) Td
QJ QJ
• 4 5
o3 t + H
O QJ CJ
fl QJ
cn QJ SH
ft QJ
-fl - t J
fl • r H
cn
ion,
- t i CJ
cS
"o ( H
^ o CJ
U fa S X)
QJ
fl •3
SH
^
1 0
Ei M H
0 QJ SH
fl 03 03 QJ S H
ft S H
0
9-£? > SH
03
X )
fl G C3 QJ cn
Td S H
3 - I J
co" 1
0 CO
X3 fl o3
C N 1
0 CO •4-1
0 03 03 SH fl 03 03 QJ S H
ft
a o3 QJ
4 J 73
Td fl 0 C3 QJ CC
I -H 1
n fl ^ 0
0
ai CN CO
QJ
igur
r -PH
o QJ U
fl cn 03 03 S H
ft
i 03 4 J 03
233
B>i/6>)''-^^sav 6)(/B>i'^^^sa V'^^^sa V M ' l -os iV UJ 'LOS (S'lSI
4 J
SH
fa
n> o fl 03
r Q I H
^ rn
T d
QJ S H
fl
rat
CD U (
a 03 4 J
T3 Q) OJ
t4H
03
M H
O CD CJ Cl Q)
res
o, 03
r f l 4 J
fl ''—1
SH Cl)
lizj
'. S H
fl 4 J Cfl
OJ ft
fl OJ
4 J CJ fl
X3 O S H
ft
ird
r l - t J
., CO
C) rn x i fl 03
CN 1
CO '4-1
o 03
fl o
rat
4 J fl C13 CJ fl C> CJ
c ; S H
4 ^
fl o CJ
O fa S T d 03 fl
•3 ^ 4 J cn fl 0 0
M H
0 QJ CJ fl a a SH
0 <4H SH QJ ft
ft 0
^ 1
XJ OJ 03 0
1 1
0
CO CO CO
CD S-l
fl bfl
fa
cn T d 'Xi 0 03
>, ^ T3
T d fl 0 CJ QJ 03
r .
fl _o '-3 03
SH - t J
fl QJ CJ fl 0 CJ
03
73 0 cn
> v S H
2 ^ fl CO
p 0 S H
ft >. S H
0 CD
-^ > OJ fl
r f l - ^ -t^ . . fl - ^
a t, 2 3 t-B 0
234
o CO
ft o
a o
- f l ft o3 SH
bO
fa P OJ cn O 03
• QJ s-l
ft fl 03
Xl O
cfl SH QJ ft OJ
03 Td fl G o
03 LO
OJ . . ,QJ - f l "^ ft
o3 o3
Q "2 o .fl fl rfl OJ 4 J 03 CD 03 SH OJ
ft ^ QJ M
rfl 03 4J O J
s-l
.a ^ 03 S fl 03 o OJ
o3 CO
G - g Cfl SH
"S CN G f^
" 6 O CO
fa ^ S ft « bO '^ X i
fl 03 SH
Bd'^^^d B d ' ^ ^ ^ d ' ^ ' d Bd'P"°=d
03 fl G
o CO ^
03 CO
QJ SH fl
SH
ft
i fa ^
235
4d ^
P
6>l/6)| • LOS SO V 6>|/6>|'EOS'SOS s a V > i ' l O S i V UJ LOSIS'^BI
QJ O fl 03
r Q
QJ SH fl
- tJ o3 I H
QJ ft
CO u
S H O QJ SH
• J H S 03 ft fl .. g Td fH U
4. .f l I f l ^
03 O ^ H ' ^ G - o QJ fl CJ o3 fl CN QJ V OJ ' w ' !H CO ft 4 H QJ G
^ 03 - t ^ fl
fl .2
r2 -t
R 8 O X3 fa fl^
XI b
o3 - O
03 O
fl ' O '^ r, cn
QJ CJ fl 03
fl _o 4 J o3 SH ^ QJ C3 fl O C3
ft 03
ftS 2 ^ O 03 X) QJ 03 G
o CM CO
CO
> 5 S H
X3 - tJ C3 fl
X) G S H
ft
QJ ft b c) fl -* bO 03
fa S
o CO '^-l
O
236
o3 QJ
O CO
Bd '^^^d Bd '^^^d pue ^^^d ,puoo
Bd d
ft O
r f l ft 03 S H
bjO
OJ CJ fl 03
X SH
fl 03
'-B SH
03 =4H 03 fl Cfl S-l
-IS
-1-^
a QJ r f l
03 <4H
G 03 CJ CJ 03 03 QJ SH
ft 03
Xi • ^
fl
03
fl _o 4 J CJ
cfl
.__, G SH
4 J fl G C3
o fa ^ X ) QJ fl o3 S H
cn fl O
o CO CO
CO
QJ S-l
fl hO
fa
3 cn 03 03 SH
ft S H
G
ft 03 > >. SH 03
X3 fl G CJ QJ 03
LO
fa i5 ft o3 SH
bO
X3 SH
'S 4 J
.. cn 03 SH
fl 03 03 QJ SH
ft
a 03 QJ
- ^ J 03
CO
Td fl 03
CN 1
o CO
r f l ft Cfl SH bO
X I fl G CJ QJ 03
Qj"" SH
fl 03 CO QJ SH
ft
237
B>j/6>|'l'='SsaV B>|/B>| ' " ^ s a '^^^sa V > l 'LOSiV ui'LOSISAsn
the
a o
<4H
+ J C/J
fa
nee
03 r O
71 4 .3 OJ
rth:
fl o M H
QJ SI 1
03
QJ
ft
tern
4H CJ fl
X I O
03 X i
^ o cfl
mas
s
X l QJ Cl)
o3 =4-1 O 03 CJ c"! QJ ro Cl>
ft QJ
r f l
r f l
CO
SC-
X I
fl 03
CN
o CO
'^-l
n cn
fl O
03 SH
fl
G SH
4J fl G CJ
CJ fl O CJ
SH
XJ
fa -o QJ fl
- f l G 4 J CJ •41 ' ^ ^ 03
tu fl" fl .2
o fi <4H QJ
I H CJ
S fi ft O
ft "
§3 Td o 5 cn cn 1^ S ^ ^ o -« 6 .. tJ '^
'^ fl '^ CO Td
« 2 „ OJ ft G
fa 5 .a
SH
238
03 IH fl 03 03 QJ
03 03
Bd '^^^d
^ CN
Bd'"^d'"=d
•sr
2.15
2.1 in CN o
(N
,puoo Bd d
r f l ft Cfl IH
c
£ Q)"
'
QJ ^ CJ ffi y QJ CD SH
r Q ft I H fl >-•
S G 03 ft
^ ^ -^3 i>^
I H CO
^ ^ G G
'fl y OJ O] 03
d l O
a fa Td . . 9. -fi
^ ft ^ cS cfl I H
^ H bO
° X3
ence
th
ir
^ I ft S QJ 03
fl ft
fl o3 O QJ
• S 4-= 4-3 03 CJ 03 CO
I'B - t^ cd
§ ^
" 6 fa^
• ^ - ^
i2 ft CO 03
^ bi) QJ SH fl bO
xd fl
u*J u fa 03
239
6>l/6>|''-^^SaV' B>i/B>i '^^^sa ' " ^ s a V > I 'L0S1V LU 'LOS is '^a i
03 03
• - fl fa S cfl
. S H
QJ CD
fl ^ O QJ u -^ fl 4^
4 J C3 CO ; 3
^ X i
fl 2 o ft o3 SH
• ^
fl 03 CJ fl O C3
03
'o 03
X) tH
Td QJ
. 0 3
O CO
Td c 03
CN
'6 C O
^ 03 QJ h
fl ^ s g 03 3 QJ g SH O
ft CJ
^§ .3 "" ^ T l
r2 ' o Td '-I fl
o " R QJ CJ CC
fa fi ^ G
QJ
QJ O
Ifl tH -^ fl QJ CJ fl G CJ
^ ^ QJ CO
ft >>
^^ ^ fl CO
" ^ ^ H
2 ^ ft >.
.. ^ ^ CO o S CO - ^ >
c d ^ . a QJ -^^
T d 03 03
S a bO p
fa
rfl -t3 (H fl
240
o3 QJ
o CO
Bd'^^^d B d ' " ^ d ' " ' d ,PU03
Bd d
a G
G . IH
rfl ft 03 IH bO
o a
IZ •p
ime,
1
1—
o LCI
t i l
QJ CJ
fl o3
Xi ^ fl
4-3 03
'-B fl .2 4-3 03
ncen
tr
O CJ
03
3 "o 03
SH
Td Td QJ QJ
<4H
03
QJ I H
fl
ess
I H
ft SH
G ft
f SH 03
Xi
seco
n
l O
H
rfl ft 03 SH
bO Td
SH
3 43
CC QJ
44 iH O fl CJ QJ
IH
ft
ft o3 QJ
QJ 4J rfl CO
- ^ CO
.a XI CO 3 fl "
.2 ' 5
1^ 03 CO
-a ft fi 03 G IH O bO
hH r^
fa fl G
CO CC
^ Qj" OJ H iH fl rH 03
to S •— IH
fa ft
241
QJ
-fl 4 3
a o
<i3 4-3 1/1 S H
• rH
fa QJ CJ
fl Cfl
tur
rn
x i QJ S H
fl Cfl SH QJ
ft
a 4-3
x i 03
,CD C4H
r f l I H
fl n <4H
r .
S 4 3
era
ft
a 03 - ^ J
+ 3 C)
fl -d u I H
ft
Td IH
-fl - ^ 3
CO 1
O UJ
X I fl 03
03 <N
B>l/B>| • '-'^sa V 6))/6>| '"SgQ V '^^^sa V >i'^==lV uJ'LOSiaASl
242
fl CO 03 QJ
o3 QJ
Bd'^^^d
'S - CN ^
B d ' " ^ d ' " ^ d
3.8
2.2 T - CN
CN
,puoo Bd d
u CO
ft G
rfl
S-l
o o —
£
'£ Q)' £
1-
o un
o
. fl CD CO CJ cn S QJ 03 ' r v
- Q ^ s-l I H
fl o -^ ft .23 Cfl Td > QJ ^ I H h
fi iS 4 3 T d 03 S
03 8 ft OJ
a " OJ LO -t3 1
Td W QJ . .
QJ Xi ' ^ ft o3 o3
SH
^ bO
OJ ' 2 CJ . f l fl rfl QJ 43 cn
pre
res,
r f l 03 4 J 03 _ I H
.3 ^ fi 1 O QJ
4 3 - t^ -g cn 03 CO
"o -^ ti fl ^ ' O CN
fa ^ ,:..• r f l
s ft CO 3^ ^ bb
Figu
re
seco
nd
243
> ft O fl
QJ
6>l/6>1 ' LOS s a V B v B V E O S ' S O S S a V > l ' L O S i V iiJ'L0S|3Aa-|
o -^ ih IH
c i fl G
4 3 <44 cn
• ti CD fa ^ QJ CJ fl o3
rO SH
fl
Td
o3
QJ
ft
CJ
fl Td G QJ
03 ft fl
03 -^^ OJ
r f l CO • ^ I
a O •44 CO G r ^ 03 fl CJ 03 fl 03 cn r\
fel ^ ft "44 QJ G
r f l cn 43 fl
fl -2
CN
03 03 s-l
O 8 j3 fi o " O 03
HH : 2
QJ tn r f l > . 4 3 SH
G ..
fl § g QJ a to
t l .2 03 4 3
ft g
& fi ° 8
fl G C3 03
-fl O 03
Xl QJ 03
_o
^ SH
fl-5 o bO o ^
•rH j _ | ^_ |
fa ft G
244
B d ' l O S d Bd EOSd 'SOSd Bd 'puood
X! fl G C3 QJ 03
Qj"
03 03 03
o3 03
O CO
ft O
O , S H
r f l ft Cfl S-l
bJ3
fa *-^ QJ
• SH OJ fl CJ 03 fl cn
3 S SH O ft
>> SH 03
Td fl O O QJ 03 Cfl
QJ r f l
cfl fa H H
QJ -5 tJ S* fi S
ft .a ^ 5 - ^ 03" H QJ
2 S ! H 03 G QJ
• •g ft r^ a O 03 SH 03
4 3 4 3
fl tn
8 ^ '-' "2 fa CN •^ 6 1 ^ CO
QJ . .
fi ft bO ifl
fa bO
245
o L O
x L
6>|/6>|'''°\v .POJd — >\ i V LU |8A91
O S H
o CJ
O fa
xi QJ
fl
03
fl G O fl fl 03 <4H
-fl O • ^ cn bC fl fl O
CJ
fl Td O s-l
a XJ iH
-^3
co" I
O CO X)
I
O CO
03
fl QJ CJ fl o3
r Q
cfl S H
^ OJ C3 fl O C3
Td 03
o3
O 03
Cfl ' d 03 03 03
Td QJ
,QJ
QJ X3 G
fl QJ CJ fl O o
fl 03
Q ^ fa " QJ b
rfl Td 2 O
03 X l 03 o
CO
fl G ft 03 03
O
. . S H
- ^ G ft -S o fi I ft 43
r2 -fl 1 4 J
o § b r2 ±J fl
6i 2 ^ fa CO ^
QJ 52 S H fl
fl o QJ ft
bO -43 a rT' 2 tD M H o3 4 3
246
CHAPTER 7
SUMMARY AND FUTURE WORK
7.1 Modeling
This work developed a fundamental model of a multiple effect falling film evap
orator plant that processes a black liquor feed stream normally found in pulp and
paper mills. The model strived to represent the important transport phenomena of
evaporation and sensible heating of the falling film, as well as water film condensation
were modeled in detail. It also accounted for turbulent and wavy-laminar hydrody
namic regimes found in these processes. No attempt was made to describe the wavy
nature of the falling films or other small scale phenomena. Using a simple laboratory
experiment, laboratory tracer experiment on the falling film past a vertical dimpled
plate, it was shown that a one spatial dimension model adequately described the
phenomena of interest at the necessary scale.
The resulting system of partial differential equations was subsequently solved us
ing orthogonal cohocation on finite element (OCFE). The model was validated by
examining the response to a set of expected disturbances in the feed stream and
changes in the utility stream (steam) pressure. The results showed that feed mass
flow rate had the least eflFect on the dry solids concentration in the product while feed
temperature disturbances had the largest effect.
The single evaporator PDE model was used to develop the models for the multiple
effect evaporator (MEV) plant. The multiple evaporator plant consisted of seven
247
single evaporators interconnected in a countercurrent design. The pressure dynamics
were modeled by system of four ordinary differential equations. Also the condensate
flashing was modeled using a steady state assumption on the flashing process.
The resulting model of the MEV was integrated using parallel processing, in this
case MPI (Message Parsing Interface), where each evaporator's equations are inte
grated on a separate processing node. The model was validated using an expected
set of disturbances to the feed stream and steam utilities. The responses showed
that the system physics are well represented. To further validate the MEV model,
steady state data from an industrial plant were used. The results showed excellent
agreement - steady state values of the product dry solids concentration showed very
small diflterences of 0.006 kg/kg or about 0.8%.
7.2 Control
To study control, a nonlinear model in the form of a system of ordinary differential
equation (ODE) was developed. Validation of the ODE and PDE models showed
that for the same set of disturbances the ODE model satisfactorily represented the
distributed parameter system. The nonlinear ODE model was then approximated by
a linear ODE model which permitted linear control theory to be applied to develop
a suitable controller. Two sets of linear controllers were applied, decentralized single
loop feedback control law (proportional-integral controller) and a centralized model
predictive controller.
248
7.3 Decentralized control
A decentralized control scheme was tested on the single evaporator represented by
the PDE model. The control scheme includes two cascaded feedback (proportionally-
integral or PI) loops controhing the level in the evaporator inventory with the flow
rate of the evaporator product mass flow rate and the dry sohds concentration of
the product with the steam pressure. The evaporator was controlled for on-supply
operations. The results showed that although not optimally tuned, the PI controllers
provided satisfactory performance for disturbance compensation. The control action
was more aggressive in the case of set point changes (servo control), however if the
changes were applied as ramps for one or half of the system time constant aggressive
control actions may not occur.
A similar decentralized control scheme was tested on the multiple evaporator
plant represented by the PDE model. The control scheme included six cascaded
feedback (proportionally-integral or PI ) loops controlling the level in five of the
evaporator inventories with the flow rate of the corresponding evaporators product
streams. The dry solids concentration of the product of the last evaporator was
controlled with the steam pressure supplied to the evaporator. The results showed
satisfactory disturbance compensation in the presence of feed mass flow rate, feed
temperature, and feed dry solids concentration disturbances. The response time in
which the product dry solids concentration was within ±lcr limits were satisfactory.
249
7.4 Centralized control
7.4.1 Single evaporator
Unconstrained and constrained MPC formulations were implemented using the
nonlinear ODE models. In this case the controlled variables were the product dry
solids concentration and the product temperature. The evaporator level inventory
was controhed using a cascaded single loop feedback (PI) controller. The closed loop
system was subjected to different disturbances and the performance of the system
was evaluated. The unconstrained MPC controller was able to provide successful
disturbance compensation in the presence of a feed mass flow rate, dry solids concen
tration and temperature changes and to provide regulation of the product dry solids
concentration within ±lcr hmit in less than flve minutes.
The constrained MPC controller showed improved performance, however, no sig
nificant difference between the unconstrained and constrained implementations can
be observed. The results of both unconstrained and constrained MPC controllers
were compared to equivalent decentralized control scheme. The performances were
found to be similar for disturbance compensation. However, when the magnitude of
the disturbances are increased, the performance of the constrained MPC controller
is superior to that of the decentralized scheme. This demonstrates that the MPC
controller is a more robust controller.
The final validation of the MPC controller was the application of the controller
actions found for the nonhnear ODE model to regulate the distributed parameter
model (PDE model). The results showed that the most important variable, the
250
product dry solids concentration was well controlled and regulated within ±la limit
in approximately seven minutes.
7.4.2 Multiple evaporator plant
Similar to control of a single evaporator, unconstrained and constrained MPC for
mulations were tested using the nonlinear ODE model. The controlled variables were
the last three evaporators (SC-3, SC-2 and SC-1) product streams' dry solids con
centrations and the last evaporator (SC-1) temperature. The manipulated variables
were the steam pressures at the last three evaporators (SC-3, SC-2 and SC-1) and the
secondary vapor pressure at the first evaporator (E-5). The closed-loop system was
subjected to feed mass flow rate, feed dry solids concentration, feed temperature and
heat transfer disturbances. The feed dry solids concentration disturbance was well
compensated by both constrained and unconstrained MPC. In the presence of the
other disturbances, the performance of constrained MPC was found to be superior to
that of unconstrained MPC. For unconstrained MPC, the response of the product dry
solids concentration remained within the r t lu limit within ten to thirteen minutes
after the introduction of the disturbance, or about one fifth of the open-loop system
time constant.
When compared to the equivalent decentralized control scheme, the advantages of
the MPC formulation were more obvious than in the case of single evaporator. The
responses of the decentralized scheme showed oscillatory character with the response
of the product dry solids concentration remaining within ±a limits in about twenty
251
to thirty minutes.
The final validation of the MPC controller involved the apphcation of the controher
actions found for the nonlinear ODE model to the distributed parameter model. The
results showed that the most important variable, the product dry solids concentration
was well controlled as in the case of single evaporator.
7.5 Summary on the control results.
The results of the application of the different control strategies showed that both
decentralized and centralized control can be used to control the single evaporator
and multiple evaporator plant successfully. However, constrained MPC provided ro
bustness. Therefore, the control strategy applied should be determined by evaluating
the quality requirements on the controlled variables. For tight constraints on the
controlled and manipulated variables the centralized control strategy should be used.
7.6 Future work
The following are areas in which future development may be very useful. First,
it is necessary to build at least a single laboratory scale falling film evaporator and
to refine the PDE model using dynamic data. Laboratory scale experiments are
preferable as there will be less uncertainty in the measured parameters. The model
should be validated using real plant data if available. It is possible to use different
working fluids. Suitable choice can be a carboxymethylceUulose solution, which is
non-corrosive and its dry sohds concentration can be measured easily through the
252
solution conductivity.
Second, there is room for improvements in the numerical techniques used to solve
the PDE model. The lengthy calculation times found in this work obstructed more
detailed investigation of the PDE model.
Lastly, additional models that describe the scaling process or possible crystahiza-
tion in the evaporators can be developed and incorporated in the PDE model. If
the computation times are reduced by optimization of the numerical solution method
a long term operation of the evaporators can be simulated in order to predict how
scaling affects the performance of falhng film evaporators. This knowledge can be
used to optimize the scheduling of the washing cycles that are common when high
dry solids concentration is to be achieved in the final evaporators/concentrators.
7.7 Contributions
The contributions of this work are fohows:
• A fundamental first principles distributed parameter model of a falling film plate
type evaporator was developed and
• Two control strategies were developed to regulate the dry solids concentration
in the product stream of evaporator, which is essential to provide a stable feed
to the recovery bofler.
• A fundamental model of a multiple evaporator plant (seven evaporators) was
developed and validated.
253
• An effective MPC strategy was developed to regulate the dry sohds concentra
tions in three evaporators and the temperature of the final product stream.
• Stefanov, Z.I. and K.A. Hoo (2003) "A Distributed Parameter Model of Black
Liquor Falling Fflm Evaporators. Part I. Modehng of a Single Plate," Industrial
& Engineering Chemistry Research, 42, pp. 1925-1937.
254
BIBLIOGRAPHY
[1] G. Hough. Chemical Recovery in the Alkaline Pulping Processes. TAPPI Press, Atlanta, GA, 1st edition, 1985.
[2] P. E. Minton. Handbook of Evaporator Technology. Noyes Publications, New Jersey, NJ, first edition, 1986.
[3] R. Billet. Evaporation Technology. VCH Pubhshers, New York, NY, first edition, 1974.
[4] L. C. To, M.O. Tade, and M. Kraetzl. Robust Nonlinear Control of Industrial Evaporation Systems. World Scientific, New Jersey, NJ, first edition, 1999.
[5] Y. Yamamoto, N. Mitsushi, and S. Kadoya. Design and operation of evaporators for radioactive wastes. Technical Report 87, IAEA, Vienna, Austrua, 1968.
[6] K. Vijayaraghavan, A. Nikolov, and T. D. Wasan. Radioactive waste evaporation: Current methodologies employed for the development , design and operation of waste evaporators at the Savannah River site and Hanford waste treatment plajit. The 9th International Conference on Radioactive Waste Management and Environmental Remedation,Examination School, Oxford, England, icemOS-4515(WSRC-MS-2003-00050, Rev.O), 2003.
[7] W. L. McCabe, J. C. Smith, and P. Harriot. Unit Operations of Chemical Engineering. McGraw-Hih, Inc., New York, NY, fifth edition, 1993.
[8] A. Niemi and R. Koistinen. Dynamic Modeling of Multiple-Effect Evaporator Plant. Paper och Tra, 54(2):297-306, 1972.
[9] A. Niemi, E. 0 . Hartikainen, and R. Koistinen. Dynamics and Control of Multiple-Effect Evaporator and Drum Washer Plants. Proceedings of 4ih IFAC/IFIP International Conference on Digital Computer Applications to Process Control,Zurich, pages 503-519, 1974.
[10] L. N. Ricker and T. Seweh. Dynamics of Multiple-Effect Evaporators: Computer Simulation and Plant Tests. AIChE Symposium Series: Applications of Chemical Engineering in the Forest Products Industry, 80:87-94, 1984.
[11] A. J. L. Cardoso. Modelizagdo e Controlo Robusto de uma Cadeia de Evaporagdo. Master of science, Universidade de Coimbra, Coimbra, Portugal, 1995.
[12] M. G. Kelly and L. L. Edwards. Two-Phase Flow and Heat Transfer in Rising Film Evaporators: A Distributed Parameter Model. AIChE Symposium Series: Applications of Chemical Engineering in the Forest Products Industry, 80:95-109, 1984.
[13] F. P. Incropera and D. P. DeWitt. Introduction to Heat Transfer. John Wiley & Sons, Inc., New York, NY, third edition, 1996.
255
[14] M. Kaviany. Principles of Heat Transfer. John Wiley k Sons, Inc., New York, NY, first edition, 2002.
[15] P. Schutte, A. Gorak, and L. Jeromin. Stationare und dynamische Simulation eines Fallfilmverdampfers. Chem.Ing.Tech., 64(5):468-469, 1992.
[16] I. S. Bojov. Basics of Automatic Control. Technika, Sofia, Bulgaria, third edition, 1992.
[17] J. A. Richalet, A. Rault, J. D. Testud, and J. Papon. Model predictive heuristic control. Automatica, 14:412, 1978.
[18] E. C. Garcia and M. Morari. Internal model control: a unifying review and some new results. Industrial Engineering Chemical Process Design and Development, 21:308-323, 1982.
[19] C.R. Cutler and B. L. Ramaker. Dynamic matrix control a computer control algorithm. AIChE 86th Annual Meeting, Paper WP5-B, 1979.
[20] P. Gil, H. Ramos-Duarte, and A. Dourado. Dynamic Modelling and Optimal Hierarchical Control of a Multiple-Effect Evaporator-Superconcentrator Plant. 17th IFIP Conference on System Modeling and Optimization, Prague, Czech Republic, 1995.
[21] E.F. Camacho and Bordons C. Model Predicitve Control in the Process Industry. Springer Verlag, New York, first edition, 1985.
[22] W. J. Frederick, T. Noopha, and M. Hupa. Combustion behavior of black liquor at high sohd firing. TAPPI Journal, 74(12):163-170, 1991.
[23] R. Karvinen, P- Hyoty, and P. Siiskonen. The effect of dry solids content on recovery bofler furnace behavior. TAPPI Journal, 74(12): 171-177, 1991.
[24] P.L. Kapitza. Collected Papers of Kapitza. Pergamon Press, Oxford, United Kingdom, 1 edition, 1965.
[25] V. G. Levich. Physicochemical Hydrodynamics. Prentice Hah Inc., Englewood Cliffs, NJ, 1962.
[26] W Wilke. Warmeiibergang an rieselfilme. VDI Forsh.Heft, 490, 1962.
[27] Limberg. H. Warmeiibergang an turbulente und laminare rieselfilme. Int. J.Heat Mass Transfer, 16:1691-1702, 1973.
[28] K. R. Chun and R. A. Seban. Heat transfer to evaporating hquid films. J. Heat Transfer, 93:391, 1971.
[29] K. R. Chun and R. A. Seban. Performance prediction of falhng-film evaporators. J. Heat Transfer, 94:432, 1972.
[30] R. A. Seban and A. Faghri. Evaporation and heating with turbulent falling liquid films. J. Heat Transfer, 98:315-318, 1976.
256
[31] H. Brauer. Stromung und warmeiibergagn bei rieselfilmen. VDI Forsh.Heft, 457 1956.
[32] A. S. Telles. Liquid Film Characteristics in Vertical Two-Phase Flow. Doctor of phylosophy. University of Houston, Texas, USA, 1968.
[33] E. R. vanDriest. On turbulent flow near a wah. J.Aeronaut.Sci., 23(11):1007-1011, 1956.
[34] D. K. Chung and A. F. Mflls. Heat Transfer Across Turbulent Falling Films. Int. J. Heat and Mass Transfer, 16:694-696, 1976.
[35] S. V. Alekseenko, V Ye. Nakoryakov, and B. G. Pokusaev. Wave formation on a vertical falhng liquid film. AIChE Journal, 31:1446-1460, 1985.
[36] H. Chang, E. A. Demekhin, and E. Kaladin. Simulation of noise driven wave dynamics on a fahing film. AIChE Journal, 42:1553-1568, 1996.
[37] E. Stuhltrager, A. Miyara, and H. Uehara. Flow dynamicas and heat transfer of a condensate film on a verical wall-ii. Flow dynamics and heat transfer. Ind.J.Heat Mass Transfer, 38(15):2715-2722, 1995.
[38] A. Miyara and H. Uehara. Flow dynamics and heat transfer of a condensate film on a vertical wall - ii. flow dynamics and heat transfer. Int. J.Heat.Mass. Transfer, 38(15):2715-2722, 1995.
[39] A. Miyara. Numerical analysis on flow dynamics and heat transfer of falling liquid films with interfacial waves. Heat and Mass Transfer, 35:298-306, 1999.
[40] A. Miyara. Flow dynamics and heat transfer of wavy condensate film. Transactions of ASME, 123:492-500, 2001.
[41] A. A. Alhusseini. Heat and Mass Transfer of Falling Film Evaporation of Viscous Liquids. Doctor of phylosophy, Lehigh University, Bethlehem, Pennsylvania, 1994.
[42] A. A. Alhusseini, K. Tuzla, and J. C. Chen. Falhng film evaporation of single component liquids. Int. J .Heat Mass Transfer, 41:1623-1632, 1997.
[43] A. A. Alhusseini and J. C. Chen. Transport phenomena in turbulent falling films. Ind.Eng.Chem.Res., 39:2091-2100, 2000.
[44] R. B. Bird, W. E. Stewart, and E. N. Lightfoot. Transport Phenomena. John Wiley & Sons, New York, NY, 2 edition, 2000.
[45] L Prandtl. Zeits.f.angew.Math.u.Mech, 5:136-139, 1925.
[46] F. Blangetti. Lokaler Warmeiibergang bei der condensation mit iiberlagerter Zwangskonvection im vertikalen Rohr. Doctor of phylosophy. University of Karlsruhe, Karlsruhe, Germany, 1979.
257
[47] G.L. Hubbard, A.F. Mflls, and D.K. Chung. Heat transfer across turbulent falling film with concurrent vapor flow. J. Heat Transfer, 98:319, 1976.
[48] D. K. Chung and A. F. Mills. Experimental study of gas absorbtion into turbulent falling films of water and ethylene glycol mixtures. Int. J. Heat and Mass Transfer, 19:51, 1976.
[49] Y. S. Won and A. F. Mills. Correlation of the effects of viscosity and surface tension on gas absorbtion rates into freely falling turbulent liquid films. Int. J. Heat and Mass Transfer, 25:223, 1982.
[50] K. Stephan. Heat Transfer in Condensation and Boiling. Springer-Verlag, Berlin, Germany, 1 edition, 1992.
[51] W. M. Kays and M. E. Crawford. Convective Heat and Mass Transfer. McGraw-Hfll, New York, NY, 2nd edition, 1980.
[52] W.L. Badger and W.L. McCabe. Elements of chemical engineering. McGraw Hill Book Company, Inc., New York, London, second edition, 1936.
[53] P.P. Stainthorp and J.M. Allen. The development of ripples on the surface of hquid films flowing inside a vertical tube. Trans. Inst. Chem. Eng., 43:85, 1965.
[54] I. Mudawar and R.A. Houpt. Measurement of mass and momentum transport in wavy-laminar falling liquid fllms. Int. J.Heat.Mass.Transfer, 36(17):4151-4162, 1993.
[55] S. Jayanti and G.F. Hewitt. Hydrodynamics and heat transfer of wavy thin film flow. Int.J.Heat.Mass.Transfer, 40(1):179-190, 1997.
[56] S.S. Kutateladze. Fundamentals of Heat Transfer Academic Press Inc., New York, NY, translated from second edition, 1963.
[57] E. Pohlhausen. Z.angew.Math.u.Mech, 1:115, 1921.
[58] C O . Bennett and J.E. Myers. Momentum, Heat and Mass Transfer. McGraw Hfll, Inc., New York, NY, second edition, 1974.
[59] A.A. Khahd, S.R. Hosahahi, G. Doyon, and C. Toupin. Effects of concentration and temperature on carboxymetylcehulose rheology. Int. J.Food Sci.&Tech., 29:243-253, 1994.
[60] Z. I. Stefanov and K. A. Hoo. A Distributed-Parameter Model of Black Liquor Falling Film Evaporators. Part I. Modeling of a Single Plate. Ind.Eng. Chem.Res, 42:1925-1937, 2003.
[61] T. N. Adams. Sodium salt scaling in black liquor evaporators and concentrators. TAPPI Journal Peer Rewieved Paper, 84(6): 1-18, 2001.
[62] S. J. Severtson, R Y. Duggirala, P. W. Carter, and P. E. Reed. Mechanism and chemical control of calcium carbonate scaling in the kraft process. TAPPI Journal, 82(6): 167-174, 1999.
258
[63] A.L. Pricke and M.A. Stoy. Development of a method for measuring the heat of dilution of kraft black hquor and water. TAPPI Journal, 77(8): 169-174, 1994.
[64] A.L. Fricke and M.A. Stoy. Enthalpy concentration relations for black hquor. TAPPI Journal, 77(9):103-110, 1994.
[65] R. H. Perry, Green D. W., and Maloney J. 0 . Perry's Chemical Engineer's Handbook. McGraw-Hill, New York, NY, 7th edition, 1997.
[66] J.C. Patterson, T. Graham, W. Schopf, and S.W. Armfield. Boundary layer development on a semi-infinite suddenly heated vertical plate. J.Fluid Mech, 453:39-55, 2001.
[67] B. A. Finlayson. Nonlinear analysis in chemical engineering. McGraw-Hih, 1980.
[68] A.C. Hindmarsh. Odepack, a systemized collection of ode solvers. Scientific Computing, I of IMACS Transactionson Scientific Computing:55-64, 1982.
[69] A.C. Hindmarsh and L.R. Petzold. Algorithms and software for ordinary differential equations and differential/algebraic equations-part i. Computers in Physics, 9(1):34-41, 1995.
[70] A.C. Hindmarsh and L.R. Petzold. Algorithms and software for ordinary differential equations and differential/algebraic equations-part ii. Computers in Physics, 9(2): 148-155, 1995.
[71] C. Tischendorf. Solution of index-2 differential algebraic equations and it's application in cirquit simulation. Doctor of phylosophy, Humboldt-Universitat, Berhn, Germany, 1996.
[72] Black Liquor Recovery Boiler Advisory Comitee. Safe firing of black liquor in black liquor recovery boilers, www.blrbac.org, 10 2003. Freely available to the public.
[73] E. H. Bristol. On a measure of interactions for multivariable process control. IEEE Trans. Auto. Control, AC-11:133, 1966.
[74] D. E. Seborg, T. F. Edgar, and D. A. Mellichamp. Process Dynamics and Control. John Wiley and Sons, New York, first edition, 1989.
[75] G. Stephanopoulos. Chemical Process Control. Prentice-Hall, NJ, fhrst edition, 1984.
[76] C. E. Garcia. Quadratic dynamic matrix control (QDMC) of nonlinear processes: An apphcation to a batch reaction process. AIChE National Meeting, San Francisco, CA, November 1984.
[77] G. F. Franklin, J. D. Powel, and M. L. Workman. Digital Control of Dynamic Systems. Addison Wesley, Boston, MA, 1994.
[78] E. C. Garcia, D. M. Prett, and M. Morari. Model predicitve control: Theory and practice a survey. Automatica, 25(3):335-348, 1989.
[79] D. W. Clarke, C. Mohtadi, and P. S. Tuffs. Generalized predictive controkl.the basic algorithm. Automatica, 23(2): 137, 1987.
[80] D. W. Clarke, C. Mohtadi, and P. S. Tuffs. Generalized predictive con-trokll.extensions and interpretations. Automatica, 23(2):149, 1987.
[81] L. Ljung. System Identification. Prentice Hall Inc., Englewood Cliffs, NJ, 1962.
[82] W. J. Rugh. Linear System Theory. Prentice Hall, Upper Saddle River, NJ, second edition, 1996.
[83] P. Dorato, V. Cerone, and C. Abdaflah. Linear Quadratic Control: An Introduction. Prentice Hah, Englewood Cliff, NJ, 1995.
[84] P J. Antsaklis and A. N. Michel. Linear System. McGraw-Hih, New York, NY, 1997.
[85] J. L. Clement and J. W. Blake. Economics of high solids black hquor combustion. TAPPI Engineering Conference, pages 243-252, 1996.
[86] I. A. Mudawar and M. A. El-Masri. Momentum and heat transfer across freely-falling turbulent liquid films. Int.J.Heat.Mass.Transfer, 12(5):771-790, 1986.
[87] J. C. Slattery. Advanced Transport Phenomena. Cambridge University Press, New York, NY, first edition, 1999.
[88] J. Guihchsen, J. Guihchsen, and H. Paulapuro. Chemical Pulping. Fapet Oy, Atlanta, GA, 1st edition, 1999.
[89] P. Ramamurthy, A. R. P. van Heiningen, and G. J. Kubes. Viscosity and thermal conductivity of black liquor. TAPPI Journal, 76:175-179, 1993.
260
APPENDIX A
TURBULENT HEAT TRANSFER COEFFICIENTS
The derivations of the formulas for the heat transfer coefficients are taken from
[86]. Consider a vertical wah with a fluid flowing down the length of the wah and
that the flow is fully developed (see Figure A.l). The side of the wall opposite the
Figure A.l: A fluid element in the fully developed falhng film.
flowing fluid is heated uniformly. Suppose that the transport of energy by convection
is neghgible in comparison to the energy transport due to conduction. Then, the heat
flux is given by
q = -{X + Xt) dT
dy' (A.1)
where A is the fluid thermal conductivity. At is the turbulent thermal conductivity, T is
the fluid temperature, and y is the spatial coordinate normal to the wall. Accordingly,
261
the turbulent thermal diffusivity {at) is defined as
at = - ^ , (A.2) pcp
where p and Cp are the density and the heat capacity of the fluid, respectively.
By analogy with the molecular Prandtl number, the turbulent Prandtl number
is defined ag
Prt = - , (A.3) at
where e is the eddy viscosity. Using the quantities defined by Equations (A.3) and
(A.2), Equation (A.l) can be re-written as
^ . Pr e\ dT , , ,
It is useful to make the heat flux equation dimensionless using the following
change of variables listed in Table A.l. It then follows that the dimensionless heat
flux equation is given by
1 / Pr e\ dT+
262
For a completely developed non-evaporating film, the temperature profile in the di
rection of the flow {x) is uniform, thus
d ( Tn - T w
dx \ Ty, - T M = 0 (A.6)
j^udy
where TM is the mean film temperature.
Table A.l: Dimensionless Variables
Transformed Definition Variable
pCpU*{Ty,-T) T. . , , , T^ = — Liquid temperature
qw
Spatial coordinate normal to the wall
Liquid velocity
Fflm thickness
u* = ^/g5 Friction velocity
For constant flux at the wah, Qy,, the term {Ty,—TM) is uniform in the x direction.
Therefore
dT _ dTyj _ dTM / . o\ dx dx dx
y+
u+
6+
=
=
"^Z.
u'y
U
u*
u
263
The energy equation for a Newtonian fluid with constant thermal conductivity and
density is given by [87]
pCp dT dT dT dT g,+^^^ + ^y dy dz
, ,'d^T d^T d^T\
dx'^ dy'^ dz"^ dux
dx + dUy
dx
dUy
dx
+/i \dy dx J {dz dx J \dz dy J
+ PQ, (A.9)
where Q is the external radiant energy transmission rate per unit mass and equals
zero in the studied case. As the velocity distribution in the studied case is assumed
to be
Ux = Ux{y) (A.IO)
and the axial conduction is assumed to be negligible, using Fourier's law [87]
A— dy
(A.ll)
the energy balance equation (A.9) reduces to the form
dT pCp u-
dx dy (A.12)
264
Using Equations (A.5), (A.7) and (A.12) the dimensionless form of the heat transport
equations can be obtained as
T+ =
/ u'^dy'^
Re
T q
Pr Prt V
/ T+u+dy+ Jo
Re
T
(A.13)
(A. 14)
(A.15)
The integral in Equation (A.15) can be further simplified using integration by parts.
Thus, the dimensionless heat transfer coefficient of sensible heating is given by
h*H qw'''
2/3 S+V3pr
{Ty,-TM)\g"^ T+
5+V^Pr
(
&+
1 - Jo u'^dy'^
\ -dy^. (A.16)
Re
T /
Pr Prt V
In the case of an evaporating or condensing film, the energy supplied through the
wall almost equals the energy release at the film surface as the heat of evaporation.
The wall heat flux is then preserved across the film {q = g^) and the temperature
265
distribution is obtained by direct integration of Equation (A.5). That is
y"*" I
r-=/^ X 7 X I ' '^•'" Pr Prt u
The heat transfer coefficient of evaporation is then given by
(5+'Pr h-E = —. - ^ • {A.18)
/ X-X7I\*^ Pr Prt Vi/
266
APPENDIX B
BULK FLOW MOMENTUM EQUATION
Consider a vertical wall with a film of fluid flowing down the length of the wall and
that the flow is fufly developed (see Figure B.l).
\
I dz
5
- 1 J
xdz 1
pg(5-
^
•y)dz
'
7
'
Figure B.l: Force balance on a fluid element in the falling film.
The following holds for the time averaged velocity components:
Vx =0
v„ = 0
Vz = Vz{y).
(B.l)
(B.2)
(B.3)
267
For incompressible, steady flow, and the conditions of Equations (B.l) to (B.3)
the momentum balance (Navier-Stokes equations) for turbulent flow is given by
dP d , , cl^vz ,^ , • ^ = - P ^ ( . . t ; . ) + / . ^ , (B.4)
where P is the global pressure acting on the film and the term lyOJ represents the
fluctuations of the velocity components about the average velocities in the y and z
directions.
The force balance over a differential volume of thickness y, width A, and length
dz is given by
Ay dP = -TA dz, (B.5)
where r is the shear stress. Using Equation (B.5), the integration of Equation (B.4)
in the y direction gives
dvz
/ -dy = -p d{u^) •+ p, yd Jo y Jo Jo dy
or
, dVz ,.r> a\ T = -pVyVz + p - ^ . (B.6)
268
Here the term, -pVyV^, is the Reynolds stress, rt (see for example Bird et al. [44]).
The Reynolds stress can also be expressed as
Tt = pe--. (B.7 dy
The shear stress in the film is a function of y an d (see [58]) is given by
= r. (l - I) , (B.8)
where Ty, = p6g is the shear stress at the wall and 6 is the film thickness. Combining
Equations (B.6) to (B.8) results in the momentum balance
piy \ oJ \ uJ dy
or
^9 (^ y\ fs I ^\ ^'"^ ?(-D-('-;)f- i - ' dy
The dimensionless form of Equation (B.9) is given by
e\ dvt
using the variable transformations given in Table B.l.
269
Table B.l: Dimensionless Variables
Transformed Definition Variable
y+ = Spatial coordinate normal to the wall
(J+ = ^ Fflm thickness
Vz
f;+ = —p= Liquid velocity Tg
270
APPENDIX C
BLACK LIQUOR PHYSICAL PROPERTIES
For the properties below all units are in SI system.
Density [1]
p = 1007 - 0.495(r - 273.15) + 6X
303K <T < 450K
0 kg/kg < X < 85 kg/kg,
where T is the black liquor temperature and X is the dry solids concentration.
Kinematic viscosity [88]
exp(6.1373 x 10^ - 42.178 x WX + 335.12 x lO^X^ - 349.23 x lO^X^)
The original correlation was derived for temperature range from 0 to 100 centi
grade, however the in this study the correlation is extrapolated to the temper
ature range indicated above.
Heat capacity [1]
=(0.98 - (-9.122 X 10~^r + 0.793)X)4190
2,Q2K <T < 450K
0 kg/kg < X < 85 kg/kg
Surface tension
The information available in Hough [1] and Guihchsen [88] shows, that the
surface tension of the black liquor does not change significantly with changes in
the dry solids concentration and temperature. In this work it has been assumed
to be an average value of 0.033 N/m.
272
Boiling point rise [1]
BPR = K- ^ 1 - X
0 kg/kg < X < 80 kg/kg,
where ii' is a constant that varies from 6.5 to 7.5 K. In this work an average
value of 7 is used.
273
APPENDIX D
C O M P U T E R P R O G R A M S
D. l Mat lab S-function for simulation of single evaporator
function [sys,yO,str,ts] = csfunc(t,y,u,flag) a = *; %Plate width, m% Kw = *; %Wall heat transfer coefficient, W/m^.K% Kc = *; %Heat transfer coefficient of condensation, W/m^KVo pi = *; %Number of plates% L = *; %Plate length, m% Gpump = 58*7; if t > tstep %Introducing of step disturbance%
yO = [* * * . . . ] ; VoEnter the nominal steady state values here.% YoFor initialization start with close, feasible values.% %str is an empty matrix.%
str = []; %Initialize the array of sample times; in this case the sample% %time is continuous, so set ts to 0 and its offset to 0.%
yO = [* * * . . . ] ; %Enter the nominal steady state values here.% %For initialization start with close, feasible values.%) %)Str is an empty matrix.%o
str = 0; %)Initialize the array of sample times; in this case the sample%i %)time is continuous, so set ts to 0 and its offset to 0.%,
ts = [0 0]; %)End of mdllnitializeSizes.%)
%omdlDerivatives% %Retums the derivatives for the continuous states.Vo
function sys= mdlDerivatives(t,y,u,a,Kw,Kc,pl,L,Gpump,Gfeed,Xfeed,...