jørgen bauck jensen optimal operation of refrigera- tion
TRANSCRIPT
Joslashrgen Bauck Jensen
Optimal Operation of Refrigera-tion Cycles
Doctoral thesisfor the degree of philosophiae doctor
Trondheim May 2008
Norwegian University of Science and TechnologyFaculty of Natural Sciences and TechnologyDepartment of chemical engineering
NTNU
Norwegian University of Science and Technology
Doctoral thesisfor the degree of philosophiae doctor
Faculty of Natural Sciences and TechnologyDepartment of chemical engineering
ccopy 2008 Joslashrgen Bauck Jensen
ISBN 978-82-471-7801-0 (printed version)ISBN 978-82-471-7815-7 (electronic version)ISSN 1503-8181
Doctoral theses at NTNU 200891
Printed by tapir uttrykk
iii
Abstract
The theme of this thesis is optimal operation of refrigeration cycles but the resultsalso applies to the reverse process generating heat
A simple refrigeration cycle has five steady-state degrees of freedom related tocompressor choke valve heat transfer in condenser heat transfer in evaporatorand ldquoactive chargerdquo One degree of freedom is used to control the load of thecycle eg the compressor It is usually optimal to maximize the heat transfer inthe condenser and in the evaporator (eg by maximizing the fan speeds) The tworemaining degrees of freedom may be used to control the degree of super-heatingand the degree of sub-cooling It is found that super-heating should be minimizedwhereas some sub-cooling is optimal in terms of cycle efficiency The degree offreedom related to active charge is often lost by specifying no sub-cooling by thedesign This gives a loss in the order of 2 Different designs for affecting theactive charge are discussed
By allowing for sub-cooling in the condenser of a sub-critical refrigerant cyclethere are no fundamental differences between a trans-critical cycle (eg withCO2
as working fluid) and a sub-critical cycle (eg with ammonia as working fluid)However the practical operation is quite different For a sub-critical ammoniacycle several simple control structures gives close to optimal operation For thetrans-criticalCO2 cycle on the other hand a combination of measurements is nec-essary
Refrigeration processes are often designed by specifying a minimum approachtemperature (∆Tmin) in the evaporator and in the condenser With this approachthe optimality of sub-cooling in the condenser will not be found In additionspecifying the areas found by designing with the∆Tmin-method and re-optimizingwithout the constraints on∆Tmin leads to a different operating point These twodeficiencies shows that the∆Tmin-method is not sufficient An alternative method(simplified TAC) is proposed and compared with the∆Tmin-method
Considering the large amount of work that goes into the design of LNG processesthere is surprisingly little attention to their subsequent operation This partly comesfrom the assumption that optimal design and optimal operation is the same but thisis generally not true
The PRICO LNG process is used as an example for optimal design and optimaloperation of LNG processes In the design phase there are a number ofconstraintsthat must be satisfied to get a feasible design The limitations imposed by theconstraints are discussed It is found that constraints related to the compressor per-
iv
formance is important to consider In operation the objective function is muchsim-pler than in design where also the equipment is part of the optimization Two mainmodes of operation are studied i) minimize compressor shaft work for given pro-duction and ii) maximum production (eg for maximum compressor shaft work)
An important issue for optimal operation and plantwide control is to find the de-grees of freedom available for optimization A previously published systematicapproach to determine the steady-state degrees of freedom is extended totake intoaccount the active charge and the refrigerant composition as possible degrees offreedom A number of case studies are used to illustrate the findings
v
Acknowledgement
Most of all I would like to thank my supervisor Sigurd Skogestad for givingmethe chance to work on this thesis His support and guidance has helped me throughthese four years I would also like to thank people at Statoil and Linde for arrangingmy stay and work at the offices of Linde Engineering in Munich The key personsresponsible for providing this opportunity are Jostein Pettersen and HeinzBauer
The Process Systems Engineering Group has been may daily colleagues and Iwould like to thank the people that has been working there for a nice workingenvironment and a lot of interesting discussions
Finally I would like to thank my girlfriend Christina for her support and under-standing
Joslashrgen Bauck Jensen
vi
Contents
List of Figures xi
List of Tables xv
1 Introduction 111 Motivation 112 Thesis overview 213 Publication list 3
2 Introduction to vapour compression cycles 521 Introduction 5
211 Fundamentals 6212 Expansion device 8213 Condenser 9214 Evaporator 10215 Active charge 11216 Operation of simple cycles 12
22 Some design features 13221 Constant versus varying temperature loads 13222 Two-stage expansion and two pressure levels 14223 Internal heat exchange 16224 Cascaded cycles 16
23 Conclusion 17Bibliography 17
3 Simple cycles Part I 1931 Introduction 2032 Degrees of freedom in simple cycles 23
321 Design versus operation 23322 Active charge and holdup tanks 23
vii
viii Contents
323 Degrees of freedom for operation 2733 Discussion of some designs 28
331 Optimal designs 28332 Non-optimal designs 29
34 Optimality of sub-cooling 31341 Ammonia case study 31342 Explanation 34343 Discussion of sub-cooling Why not found before 35
35 Discussion 37351 Super-heating by internal heat exchange 37352 Selection of controlled variable 38
36 Conclusion 39Bibliography 39
4 Simple cycles Part II 4141 Introduction 4142 Selection of controlled variable 43
421 Linear analysis 44422 Combination of measurements 45
43 Ammonia case study 46431 Modelling 46432 Optimal steady-state operation 47433 Selection of controlled variables 47
44 CO2 case study 53441 Modelling 54442 Optimal operation 54443 Selection of controlled variable 55
45 Discussion 60451 Super-heating 60452 Heat transfer coefficients 61453 Pressure control 61
46 Conclusion 62Bibliography 62
5 Problems with the minimum approach temperature 6551 Introduction 6552 Motivating example Ammonia refrigeration cycle 68
521 ∆Tmin design-method 6953 Proposed simplified TAC method 71
531 Revisit of ammonia case study 7354 Other case studies 73
Contents ix
541 CO2 air-conditioner 73542 PRICO LNG process 75
55 Discussion 76551 ∆Tmin-method 76552 Other approaches 77553 Heat exchanger network design 77
56 Conclusion 78Bibliography 78
6 Optimal design of a simple LNG process 8161 Introduction 81
611 Optimal design 83612 Design constraints 85
62 Process description 8663 Results for optimal design 88
631 An alternative design optimization 9464 Discussion 95
641 Compressor 95642 Heavy extraction 96643 Feed pressure 97644 Large PRICO plants 97
65 Conclusion 98Bibliography 98
7 Optimal operation of a simple LNG process 10171 Introduction 10172 Process description 102
721 Nominal conditions 103722 Manipulated inputs 104723 Operational constraints 105
73 Model 106731 Compressor characteristic 106
74 Objective function 11075 Nominal optimum given production case (Mode I) 11176 Nominal optimum maximum production case (Mode II) 11277 Optimum with disturbances 114
771 Selection of controlled variables 11678 Discussion 120
781 Moving temperature profile 120782 Compressor characteristic 121783 Refrigerant composition 121
x Contents
784 Additional considerations 121Bibliography 121
8 Degrees of freedom for refrigeration cycles 12581 Introduction 12582 Degrees of freedom 126
821 Remark on active charge or ldquofeedrdquo for closed cycles 129822 Actual degrees of freedom for refrigerant cycles 130
83 Case studies 133831 Two-pressure level refrigeration 133832 Heat integrated distillation 134833 Small scale LNG process 137834 Propane pre-cooled mixed refrigerant (C3MR) 138835 Mixed fluid cascade (MFC) 141
84 Discussion 143841 Degrees of freedom 143842 Refrigerant composition 144843 Saturation in condenser 144
85 Conclusion 144Bibliography 145
9 Conclusion 147
A Optimal operation of a mixed fluid cascade LNG plant 149A1 Introduction 149A2 Process description 150A3 Degree of freedom analysis 151
A31 Manipulated variables (MVrsquos) 152A32 Constraints during operation 152A33 Active constraints 152A34 Unconstrained degrees of freedom 153
A4 Optimization results 153A5 Control structure design 155A6 Conclusion 156Bibliography 157
B Trade-off between Energy Consumption and Food Quality Loss 159B1 Introduction 160B2 Process description 160
B21 Degree of freedom analysis 161B22 Mathematical model 162
Contents xi
B23 Influence of setpoints on energy consumption 162B24 Influence of setpoint on food quality 162
B3 Problem formulation 164B4 Optimization 166
B41 Optimization 166B42 Optimization results 166B43 Trade-off between energy consumption and food quality loss167
B5 Discussion 168B6 Conclusion 168Bibliography 169
xii Contents
List of Figures
21 A simple refrigeration cycle with pressure-enthalpy diagram 622 Schematic illustration of a Carnot 723 Expansion paths for turbine valve and a combination 824 Two basic types of condensers 925 Two basic types of evaporators 1026 A simple design with external fillingemptying system 1127 A possible control configuration for a simple refrigerant cycle 1328 Evaporator temperature profile 1429 Two stage expansion improves cycle efficiency 14210 Cooling at two pressure levels 15211 Internal heat exchange 16212 Cascaded cycles shown with internal heat exchange 17
31 A simple refrigeration cycle with pressure-enthalpy diagram 2032 Simple cycle with variable active charge 2533 Condenser with saturation at outlet (common but non-optimal) 2734 Evaporator with saturation at outlet (optimal) 2735 Two potentially optimal designs 2936 Three non-optimal designs 3037 Cold warehouse with ammonia refrigeration unit 3138 Pressure-enthalpy diagrams with and without sub-cooling 3239 Temperature profile in condenser 33310 Pressure-enthalpy diagram for a cycle with and without sub-cooling34311 Pressure-enthalpy diagram for infinite area case 36312 Internal heat exchange 38
41 Simple refrigeration cycle studied in this paper 4242 Optimal operation for the ammonia case study 4743 Ammonia case Compressor power and loss for disturbances 5044 Ammonia case Loss as function of implementation error 52
xiii
xiv List of Figures
45 Proposed control structure for the ammonia cycle 5346 TheCO2 cycle is trans-critical and has an internal heat exchanger 5447 CO2 case Temperature profiles 5648 CO2 case Compressor power and loss for different disturbances 5849 CO2 case Loss as function of implementation error 59410 Proposed control structure for theCO2 cycle 60411 Alternative cycle with liquid receiver on high pressure side 61
51 The effect of different values for∆Tmin 6652 An ammonia refrigeration system 6853 Temperature profile in the condenser for the∆Tmin-method 70
61 Simplified flowsheet of the PRICO process 8262 Further simplified flowsheet of the PRICO process used in this work8763 Illustrating the use of liquid turbine and valve as expansion device9364 Alternative locations of the NGL recovery 9665 Maximum LNG production as function of feed pressure 97
71 A simplified flowsheet of the PRICO process 10372 A cubic compressor characteristic curve 10873 Compressor map for the refrigerant compressor 10974 Nominal P-h diagram for mode II 11275 Nominal compressor operating point 11376 Temperature profiles for the maximum production case 11477 Shaft work as function of disturbances for mode I 11778 LNG production as function of disturbances for mode II 11879 ∆Tsup as function of disturbance inWmax
s 119710 LNG production as function ofN 120
81 A simple refrigeration cycle with 5 potential degrees of freedom 12882 A simple (not closed) refrigeration cycle 13083 Simple cycle with liquid receiver on the high pressure side 13184 Cooling at two pressure levelsNss= 4 13485 Two designs of heat integrated distillation 13586 A small scale LNG concept 13787 Flowsheet of the C3MR process 13988 Flowsheet of the MFC process 142
A1 Simplified flowsheet of the MFC process 151A2 Temperature profiles 154A3 Suggested control structure for the MFC process 155
List of Figures xv
B1 Simplified supermarket refrigeration system 161B2 Energy consumption under different setpoints 164B3 Fresh fish quality loss when stored at different temperatures 164B4 Optimization between food quality loss and energy consumption 168B5 Traditional operation (Case 1) 169B6 Optimal operation forTcabin= 1C (Case 2) 170B7 Optimal operation forT food = 1C (Case 3) 171B8 Optimal operation forQfood le 755 (Case 4) 172
xvi List of Figures
List of Tables
31 Structure of model equations 2132 Typical specifications in design and operation 2333 Optimal operation with and without sub-cooling 33
41 Structure of model equations 4642 Data for the ammonia case study 4743 Optimal steady-state for ammonia case study 4844 Linear ldquomaximum gainrdquo analysis for the ammonia case 4945 Conditions for theCO2 case study 5546 Optimal operation forCO2 case 5547 Linear ldquomaximum gainrdquo analysis for theCO2 case 57
51 Structure of model equations for the ammonia case study 6952 Ammonia case study 7153 CO2 air-conditioner 7454 PRICO LNG process 76
61 Maximum head per compressor wheel 8662 Design constraints based on data from Price and Mortko (1996) 8963 Optimal design results for nine different cases 89
64 Optimal design withC0 = 110middot103kgs-1(m-2)065
95
71 The nominal operating point for mode I and II 11172 Nominal minimum and maximum values for the disturbances 11473 Structure of model equations 12374 Data for the PRICO process 12375 Optimal operation with disturbances 124
81 Potential operational degrees of freedom for typical process units 12782 Actual degrees of freedom for refrigeration cycles 129
xvii
xviii List of Tables
A1 Optimal operation of a MFC process 155
B1 Model equations 163B2 Some data used in the simulation 163B3 Traditional operation and optimal operation for different constraints166
Chapter 1
Introduction
11 Motivation
The demand for energy is increasing rapidly and it is expected to increaseevenfaster in the years to come At the same time the amount of oil is decreasing soalternative transportable energy sources has gained more attention Oneof thesealternatives is liquefied natural gas (LNG) which is natural gas (mainly methane) inliquid state at atmospheric pressure and aboutminus160C The process of liquefyingnatural gas requires large amounts of energy so a lot of work has beendone ondesign of LNG processes However it seems that the subsequent operation is lessstudied at least in the open literature This is a bit surprising considering that evensmall improvements will give large savings due to high throughput
An important issue for optimal operation is to find the steady-state degrees offree-dom available for optimization (Nopt) This number is important for several rea-sons First it determines the degrees of freedom available for solving theoptimiza-tion problem However more importantly in terms of operation it determines thenumber of steady-state controlled variables that need to be selected Optimalop-eration is normally implemented by keeping those variables at constant setpointsThe selection of controlled variables is therefor also an important issue The ob-jective is to achieve ldquoself-optimizingrdquo control where a constant setpoint for theselected variable indirectly leads to near-optimal operation Note that the selectionof a good controlled variable is equally important in an ldquoadvancedrdquo controlschemelike MPC which also is based on keeping the controlled variables close to givensetpoints
Our goal was to study optimal operation of refrigeration cycles used for liquefac-
1
2 Introduction
tion of LNG but as a start we needed to understand operation of simpler refrig-eration cycles We therefore started out by studying basic refrigerationcycles andmoved in the direction of refrigeration processes used for liquefaction ofnaturalgas
12 Thesis overview
We discuss degrees of freedom for simple refrigeration cycles in Chapter 3 Asimple ammonia refrigerator model is used for numerical results The main focusis to show that allowing for sub-cooling in the condenser gives one extra steady-state degree of freedom related to ldquoactive chargerdquo This shows that there are nofundamental differences between a sub-critical cycle and a trans-critical cycle (egtheCO2 cycle presented in Chapter4) We also show that some sub-cooling isbeneficial in terms of cycle efficiency
Chapter4 is dedicated to selection of controlled variables for simple refrigerationcycles We present two different case studies a conventional sub-critical ammoniacycle and a trans-criticalCO2 cycle We find that the ammonia cycle has severalcontrol structures that will give close to optimal operation For theCO2 cyclehowever a combination of controlled variables is necessary to give acceptableperformance
We address the problem of specifying the minimum approach temperature (∆Tmin)as a method of designing processes with heat exchangers in Chapter5 A simpledesign method for preliminary designs are presented and compared with the∆Tmin
method
In Chapter6 we discuss design optimization of a simple LNG process the PRICOprocess Nine cases are presented to show the effect of different constraints onthe optimum The PRICO process is studied further in Chapter7 where the themeis optimal operation and selection of controlled variables We present differentmodes of operation and find a self-optimizing control structure
Chapter8 discuss operational degrees of freedom for refrigeration cycles and sum-marize some of the important results for degree of freedom analysis Several illus-trative process examples are presented to illustrate our findings These cases areincluding heat integrated distillation and two complex processes for liquefactionof natural gas the mixed fluid cascade (MFC) process from Statoil-LindeLNGTechnology Alliance and the propane pre-cooled mixed refrigerant (C3MR) pro-cess from Air Products
13 Publication list 3
AppendixA presents some results from optimization on the MFC process Weassumed that the degree of super-heating was controlled at 10C and optimizedthe remaining 13 unconstrained degrees of freedom (including 9 compositions)The paper only presents the nominal operating point so it still remains to find aself-optimizing control structure
In AppendixB we consider a refrigerator that is used to refrigerate food cabinetinside a store Optimal operation of this refrigerator is studied with respect tobothfood quality and energy consumption by using a simple model for the food qualityand a simple model for the refrigerator performance We compare the traditionalway of controlling such refrigerators with different improved schemes The mainsavings are from letting the condenser pressure vary with the ambient temperatureHowever there are also some savings related to reducing the temperature inthefood cabinet during the night when the refrigerator has higher efficiency
13 Publication list
Journal papers
Chapter 3
J B Jensen and S Skogestad (2007) ldquoOptimal operation of simple refrigerationcycles Part I Degrees of freedom and optimality of sub-coolingrdquo Comput ChemEng 31 712-721
Chapter 4
J B Jensen and S Skogestad (2007) ldquoOptimal operation of simple refrigerationcycles Part II Selection of controlled variablesrdquo Comput Chem Eng 31 1590-1601
Chapter 5
J B Jensen and S Skogestad ldquoProblems with specifying∆Tmin in design of pro-cesses with heat exchangersrdquoAccepted for publication in Ind Eng Chem Re-search
4 Introduction
Chapter 8
J B Jensen and S Skogestad ldquoDegrees of freedom for refrigeration cyclesrdquoSub-mitted to Ind Eng Chem Research
Conferences
Chapter 4
J B Jensen and S Skogestad ldquoControl and optimal operation of simple heatpump cyclesrdquoEuropean Symposium on Computer Aided Process Engineering -15 Barcelona Spain 2005 Published by Elsevier ISBN 0-444-51987-4 p 1429-1434
Chapter 7
J B Jensen and S Skogestad ldquoOptimal operation of a simple LNG processrdquoInternational Symposium on Advanced Control of Chemical Processes GramadoBrazil 2006
Appendix A
J B Jensen and S Skogestad ldquoOptimal operation of a mixed fluid cascade LNGplantrdquo 16th European Symposium on Computer Aided Process Engineering (ES-CAPE) and 9th International Symposium on Process Systems Engineering (PSE)Garmisch-Partenkirchen Germany 2006 Published by Elsevier ISBN0-444-52969-1 p 1568-1574
Co-authored papers
Appendix B
J Cai and J B Jensen and S Skogestad and J Stoustrup ldquoBalanceEnergy Con-sumption and Food Quality Loss in Supermarket Refrigeration SystemrdquoSubmit-ted for publication in the proceedings of the 2008 American Control ConferenceSeattle USA
Chapter 2
Introduction to vapourcompression cycles
21 Introduction
In this chapter it is given an overview of some basic vapour compressioncycledesigns The goal is not to cover all designs but we wish to introduce thefeaturesused later in the thesis
The vapour compression cycle is the most common process used in both refriger-ation systems and heat pumps The two processes operate in the same manner Asimple flowsheet and pressure enthalpy diagram with the nomenclature is given inFigure21 For refrigeration it is the cooling dutyQC that is of interest whereas itis the heating dutyQH that is of interest for a heating cycle When a heat pump isused in heat integration both the heating and cooling duty is utilized
Some typical areas of usage are
Household Refrigerators air-conditioners and heat pumps
Automotive Air-conditioners and refrigerators
Industry Refrigeration of process streams heat pumps for heat integration
5
6 Introduction to vapour compression cycles
QC
QH
12
3 4
Wsz
Ph
Pl
Evaporator
Condenser
Chokevalve
Compressor
(a) Flowsheet
12
3 4
P
h[Jkg-1]
Ph
Pl
∆Tsub
∆Tsup
(b) Pressure enthalpy diagram
Figure 21 A simple refrigeration or heat pump cycle with a corresponding typicalpressure-enthalpy diagram indicating both sub-cooling and super-heating
211 Fundamentals
The basic refrigeration (or heating) cycle has four states (denoted 1234 in Figure21) and operates in the following manner
The working fluid is evaporated and possibly super-heated (3rarr 4) by heat ex-change with the cold source (eg air inside the refrigerator) Energy is addedin a compressor (eg as electricity) to increase the pressure of the working fluid(4 rarr 1) The high pressure vapour is de-super-heated condensed andpossiblysub-cooled (1rarr 2) by heat exchange with the hot source (eg air in the room)The liquid is then expanded through an expansion device (choke valve) (2rarr 3) togive a low temperature two-phase mixture at the evaporator inlet
The efficiency of a vapour compression cycle is often reported in terms ofldquocoef-ficient of performancerdquo (COP) The COP for for a heating and cooling process isgiven by
COPh =h1minush2
h1minush4=
m(h1minush2)
m(h1minush4)=
QH
Ws(21)
and COPc =h4minush3
h1minush4=
m(h4minush3)
m(h1minush4)=
QC
Ws(22)
respectively
The vapour compression cycle for heating or cooling has some similarities withcyclic processes for generating mechanical work from heat eg steamturbine cy-cles These work generating cycles were studied extensively during the1800 and
21 Introduction 7
Nicolas Leonard Sadi Carnot studied the theoretical cycle later called the Carnotcycle He found that the maximum work that may be extracted from a given hotsource is only depending on the hot source temperatureTH and the cold sink tem-peratureTC This may be derived using the 2nd law of thermodynamics∆Stotalge 0where the equality holds for an ideal reversible process
QH
QC
WsIdeal cyclic machine
Cold sink (TC)
Hot source (TH)
Reversible entropy change
∆SH = minusQHTH
∆Smachine= 0
∆SC = QCTC
Figure 22 Schematic illustration of a Carnot machine generating mechanicalwork from a hot source
The Carnot machine is illustrated in Figure22 The cyclic machine will have nonet change in entropy so∆Smachine= 0 If we assume reversible heat transfer andthat the hot source and cold sink has constant temperature we get∆SH = minusQH
THand
∆SC = QCTC
Thus for the reversible case we get
QH
TH=
QC
TC(23)
Taking an energy balance around the machineW = QH minusQC we may express theabove as follows
W =
(
1minusTC
TH
)
QH (24)
Here 1minus TCTH
= ηCarnot is named the Carnot efficiency This is the theoretical max-imum fraction of the heatQH that may be converted to mechanical workWs for an
8 Introduction to vapour compression cycles
ideal reversible process CorrespondinglyTHTC
minus1 is the minimum fraction of thecooling dutyQC that must be added as mechanical workWs Thus for a work gen-erating process a large difference betweenTC andTH is beneficial For a coolingprocess however the opposite is true
212 Expansion device
In household refrigerators a fixed expansion device called capillary tube is oftenused This is a thin long pipe that gives the necessary pressure drop from the con-denser to the evaporator The operational characteristics of systems withcapillarytubes is discussedDossat(2002 page 356)
For larger systems with more variations in operating conditions it is desirable tohave a variable expansion device The most common is a choke valve where thepressure is reduced without doing any work (isenthalpic expansion)
P[bar]
h[Jkg-1]
Phase envelope
(ii) (iii) (i)
(a) Different expansion paths
Turbine for liquidexpansion
Valve with smallpressure drop
Tank withsaturation
Valve for two-phase expansion
(b) Liquid turbine and choke valve path (iii)
Figure 23 Expansion in turbine increase the system performance The left figureshows different expansion alternatives in a pressure enthalpy diagram i) chokevalve ii) turbine and iii) combination of liquid turbine and choke valve shown inright figure
The choke valve may be replaced by a turbine to improve the efficiency of thecy-cle The pressure enthalpy diagram in Figure23(a)illustrates the different pathsfor (i) isenthalpic expansion (choke valve) and (ii) isentropic expansion(ideal tur-bine) The vapour fraction (and specific enthalpy) into the evaporator islower
21 Introduction 9
with a turbine than for a choke valve so the COPc is higher becauseh4 minus h3 islarger without affectingh2minush1 A practical problem of utilizing a turbine is thatsome of the liquid becomes vapour during expansion and this may cause wearinthe turbine A solution to this problem is to use a combination of a liquid tur-bine and a choke valve illustrated in Figure23(b)(Barclay and Yang 2006) Theturbine will then expand the liquid down to a pressure slightly higher than the sat-uration pressure and the choke valve will handle the remaining pressure drop intothe two-phase region see path (iii) in Figure23(a) This is only useful if the liq-uid is sub-cooled before the expansion device otherwise there will be nopressuredrop for the liquid turbine
The inclusion of a liquid turbine is considered in Chapter6
213 Condenser
The design of the condenser is important for the possible extra degree offreedomrelated to active charge Two basic types of condensers are shown in Figure24Figure24(a)shows a case where the liquid drains into a tank below the heat trans-
To evaporator From evaporator
(a) Saturation at outlet
To evaporator From evaporator
∆Tsub
Air fan
(b) Plug-flow with possible sub-cooling
Figure 24 Two basic types of condensers
fer zone With this design there will be no sub-cooling in the condenser sothereis no degree of freedom related to active charge since additional charge will onlychange the level in the tank below the condenser An alternative design would beif the liquid covers part of the heat transfer zone but this is not considered here
Figure24(b)shows a condenser with plug-flow The refrigerant is first desuper-heated then condensed and finally sub-cooled Here the degree of freedom related
10 Introduction to vapour compression cycles
to active charge may be available (if we have means of changing it) becausechang-ing the charge will directly affect the pressure and sub-cooling in the condenser
Remark The plug-flow condenser may also give no sub-cooling by adding a liquid re-ceiver after the condenser
214 Evaporator
The design of the evaporator influences the degree of super-heating before thecompressor No super-heating is achieved with the flooded evaporator shown inFigure25(a) Here a float controller may adjust the choke valve to maintain aconstant liquid level but other strategies are also possible Figure25(b)shows aplug flow evaporator where we may have super-heating at the outlet The super-heating may be controlled by a thermostatic expansion valve (TEV) The TEV willadjust its opening to give a certain degree of super-heating out of the evaporatorThe TEV requires a certain degree of super-heating to be able to measureit and toassure that no liquid is fed to the compressor also dynamically The super-heatingis therefor typically controlled at 10C Langley(2002 page 43)
From condenser To condenser
(a) Saturation at outlet
From condenser To condenser
∆Tsup
Air fan
(b) Plug-flow with possible super-heating
Figure 25 Two basic types of evaporators
Remark The plug-flow evaporator may also give no super-heating by adding a liquidreceiver after the evaporator
21 Introduction 11
215 Active charge
Consider first the simple case where both the condenser and evaporatorare plug-flow heat exchangers Figure24(b)and25(b)respectively There is no additionalliquid level in the cycle and we neglect the holdup in the piping compressor andvalve We then have that the total charge in the system ismtot = mevaporator+mcondenser An illustration is given in Figure26 The compressor controls the loadof the cycle and the choke valve controls the degree of super-heating We assumethat the flow of hot and cold source are kept constant so there are no degrees offreedom in the cycle However we have indicated an external fillingemptyingsystem that may be used to adjust the charge in the system
TC
Super-heatcontrol
(a) Initial charge gives saturation at condenseroutlet
TC
Super-heatcontrol
(b) Increased charge gives sub-cooling at con-denser outlet
Figure 26 A simple design with external fillingemptying system illustrating thedegree of freedom related to active charge
In Figure26(a)the charge is just enough to give saturation before the choke valveIf the charge is increased by filling from the external tank we will get the situationillustrated in Figure26(b) The explanation is as follows
Since the degree of super-heating out of the evaporator is controlled (by the chokevalve) there will be no room for additional charge in the evaporator (except theincrease due to higher density in vapour phase for higher pressure)The chargewill therefor accumulate in the condenser where it will increase the pressure (andtherefore the heat transfer) and give sub-cooling at the outlet This illustrates thatthere is an extra degree of freedom related to the active charge This degree offreedom is lost if the condenser design in Figure24(a)is used
12 Introduction to vapour compression cycles
The active charge may also be controlled without the external tank This is dis-cussed in Chapter3
Remark If we remove refrigerant from the cycle in Figure26(a)we may get to a pointwhere the condenser pressure is to low to get full condensation This illustrates why sepa-rate tank with a variable liquid level is desirable
216 Operation of simple cycles
The simplest (and probably most well known) process using vapour compressioncycle is the household refrigerator The evaporator is mounted inside the refrig-erator and the remaining equipment (compressor condenser and chokevalve) ismounted on the outside of the refrigerator
Household refrigerators are usually controlled with a thermostat that switches thecompressor on and off depending on the temperature inside the refrigerator Theexpansion device is usually fixed for example a capillary tube Natural convec-tion heat exchangers (no fans) are used Thus it is only one manipulated variablenamely the onoff switch for the compressor Design and characteristics ofrefrig-erators is discussed byDossat(2002)
Onoff control of the compressor is reasonable for a refrigerator since the com-pressor will operate at the design point with high efficiency most of the time (notduring start-up) The use of a constant expansion device is also reasonable sincethe conditions are more or less constant (only small variations in the refrigeratorand room temperature) However for other applications it is necessarywith con-tinues capacity control The compressor may then have variable speed andthereare larger variations in the cycle such that a fixed valve position is inefficient Itis therefor normal to also have a variable valve In addition it is common to haveadjustable fans on the heat exchangers to improve heat transfer This gives fourmanipulated variables but as discussed above there may be a fifth manipulatedvariable This fifth manipulated variable is related to the active charge in the cycleand depends on the design of the cycle
The compressor is normally used to control the cooling load The valve may beused to control the liquid level in the evaporator (or the degree of super-heating)As shown in AppendixB it is close to optimal to have the fans at constant speedsWe are then left with one degree of freedom that should optimize the operationGood controlled variables are found in Chapter4 and include the degree of sub-cooling (∆Tsub) and the temperature difference at the condenser outlet A possiblecontrol configuration is shown in Figure27
22 Some design features 13
TC
Super-heatcontrol
Sub-coolingcontrol
Figure 27 A possible control configuration for a simple refrigerant cycle Thefans are at constant speed
22 Some design features
The description in this section is based on refrigeration processes but the sameapplies for heating cycles
221 Constant versus varying temperature loads
The choice of refrigerant and configuration depends heavily on the cooling loadIf the cooling load has a constant or close to constant temperature throughout theevaporator a pure refrigerant will give a good temperature match Therefrigerantwill evaporate at a slightly lower temperature than the cooling load (TC) Thisis illustrated in Figure28(a) where there is some super-heating (∆Tsup) Notethat this super-heating may be removed if the evaporator is designed as a floodedevaporator We have assumed constant temperature loads in Chapter3 and4 bythe use of cross-flow heat exchangers Other cases of constant temperature loadsare single component condensation or evaporation
Often the cooling load change temperature as it is being cooled If the temperaturevaries a lot a single pure refrigerant will give large temperature differences inthe warm end of the evaporator illustrated in Figure28(b) This large temperaturedifference gives a loss in terms of cycle efficiency COP so it may be economicallyattractive to consider alternatives to a single pure refrigerant cycle
14 Introduction to vapour compression cyclesT
[C
]
Position[-]
Cooling load (TC)
Pure refrigerant
∆Tsup
(a) Constant temperature load
T[
C]
Position[-]
Cooling load (TC)
Pure refrigerant
Mixed refrigerant
(b) Variable temperature load
Figure 28 Evaporator temperature profile
Using a multicomponent refrigerant will give a gliding temperature also on therefrigerant side in the evaporator This is illustrated in Figure28(b) The refriger-ation composition may be adjusted to optimize the performance of the system
222 Two-stage expansion and two pressure levels
(a) Flowsheet
Large vapour fraction LargeWs
High temperatureHHHHj
P[bar]
h[Jkg-1]
(b) Pressure enthalpy diagram dotted cycle is con-ventional one-stage expansion
Figure 29 Two stage expansion improves cycle efficiency
If a high pressure ratioPR= PhPl
is required (because of large difference inTH andTC) there will be a large vapour fraction in the expansion process All this vapour
22 Some design features 15
has to be compressed back to the high pressure if the simple layout in Figure21isutilized This is illustrated with the dotted cycle in Figure29(b) The large vapouramount will not contribute to the cooling (except in the super-heating section)for a pure refrigerant and only give a minor contribution for mixed refrigerantsAn improvement is to do the expansion in two (or more) stages as illustrated inFigure29(a) The vapour from the intermediate pressure level is fed either to thecompressor as a side stream or between two compressors This configuration hastwo effects making it more desirable
bull The vapour generated in the expansion down to the intermediate pressure isnot expanded further so the necessary compressor power is reduced
bull The vapour from the intermediate pressure level is colder than the vapourthat has been compressed from the lowest pressure level so the mixing ofthetwo streams will work as inter-cooling in the compressor and the necessarycompressor power is reduced as well as the outlet temperature
Evaporator 1
Evaporator 2
1copy2copy
3copy 4copy
5copy
6copy
(a) Flowsheet
T[
C]
Position[-]
Cooling load
Evaporator 1
Evaporator 2
(b) Temperature profile
Figure 210 Cooling at two pressure levels
An interesting use of two-stage expansion occur if cooling is needed at severaltemperature levels eg a process stream that is cooled One may then havean ad-ditional evaporator at the intermediate pressure level This configuration isshownin Figure210(a) The process stream is first cooled by the intermediate pressurelevel (Evaporator 1) and then by the low pressure level (Evaporator 2) The pres-sure enthalpy diagram is still as in Figure29(b) but the amount of vapour at theintermediate pressure level is increased The temperature profile in the evaporatorsmay be illustrated by Figure210(b) Control of such cycles are discussed byWil-
16 Introduction to vapour compression cycles
son and Jones(1994) and we study degrees of freedom for this process in Chapter8
223 Internal heat exchange
Two ways of implementing internal heat exchange is illustrated in Figure211 Theconfiguration shown in Figure211(a)is common forCO2 cycles (Neksaa et al1998) but may also be used for other working fluid The positive effect is that theexpansion loss is reduced because of the extra sub-cooling before theexpansionvalve However the compressor inlet is heated which is negative for the efficiencyDepending on the working fluid and the operating point this kind of internal heatexchange may improve the efficiency (Radermacher 1989) Also if the suctionline to the compressor will heat the vapour anyhow it is better to use this coolinginternally
The internal heat exchange configuration shown in Figure211(b)is often used inLNG processes Unless the internal heat exchanger may be bypassedit does notcontribute to additional manipulated variables
(a) May improve cycle efficiency also forpure fluids often used forCO2 cycles
(b) No effect for pure refrigerants often usedin LNG processes
Figure 211 Internal heat exchange
224 Cascaded cycles
The configuration in Figure212shows two cycles in cascade both with internalheat exchange The refrigerant in the first cycle is condensed (partly of fully) by an
23 Conclusion 17
HX2
HX1
Figure 212 Cascaded cycles shown with internal heat exchange
external fluid (eg air or water) and then further cooled in HX1 This refrigerantis used as cooling in HX1 after expansion The refrigerant in the secondcycleis cooled and possibly partly condensed with the same external fluid Additionalcooling is provided in both HX1 (by heat exchange with the first refrigerant cycle)and in HX2 The second refrigerant is expanded and vaporized through HX2 toprovide the cooling at the lowest temperature This layout shown in Figure212isoften used for liquefaction of natural gas The number of cycles typicallyvariesfrom one to three and the refrigerant may be mixed or pure (then without internalheat exchange)
23 Conclusion
There are a number of design features that may be utilized in refrigeration pro-cesses based on the vapour compression cycle It is not in the scope ofthis thesisto find which of these features are part of the truly optimal design but rather to startfrom a given process design and study optimal operation However it seems thata study to quantify the energy improvements imposed by different design featuresis needed and may be an issue for further work
18 Bibliography
Bibliography
Barclay M A and Yang C C (2006) Offshore LNG The perfect startingpoint for the 2-phase expanderin lsquoOffshore Technology Conference Hous-ton Texas USArsquo
Dossat R J (2002)Principles of refrigeration Prentice Hall
Langley B C (2002)Heat pump technology Prentice Hall
Neksaa P Rekstad H Zakeri G R and Schiefloe P A (1998) lsquoCO2-heat pumpwater heater characteristics system design and experimental resultsrsquoInt JRefrigeration21 172ndash179
Radermacher R (1989) lsquoThermodynamic and heat-transfer implicationsof work-ing fluid mixtures in Rankine cyclesrsquoInt J Heat Fluid Flow10(2) 90ndash102
Wilson J A and Jones W E (1994) lsquoThe influence of plant designon refriger-ation circuit control and operationrsquoEuropean Symposium on Computer AidedProcess Engineering (ESCAPE) 4 Dublinpp 215ndash221
Chapter 3
Optimal operation of simplerefrigeration cyclesPart I Degrees of freedom andoptimality of sub-cooling
Published in Computers amp Chemical Engineering (2007) 31 pages 712-721
The paper focuses on the operation of simple refrigeration cycles Withequipment given there are from a control and operational point of viewfive steady state degrees of freedom the compressor power the heattransfer in the condenser the heat transfer in the evaporator the chokevalve opening and the active charge in the cycle Different designs foraffecting the active charge including the location of the liquid receiverare discussed With a given load (eg given cooling duty) the com-pressor power is set Furthermore it is usually optimal to maximize theheat transfer The two remaining degrees of freedom (choke valve andactive charge) may be used to set the degree of super-heatingand sub-cooling It is found that super-heating should be minimized whereassome sub-cooling is optimal For a simple ammonia cycle sub-coolinggives savings in compressor power of about 2 In this paperrefriger-ation (cooling) cycles are considered but the same principles apply toheat pumps
19
20 Simple cycles Part I
31 Introduction
Cyclic processes for heating and cooling are widely used and their powerrangesfrom less than 1kW to above 100MW In both cases vapour compression cycle isused to ldquopumprdquo energy from a low to a high temperature level
The first application in 1834 was to produce ice for storage of food which led tothe refrigerator found in most homes (Nagengast 1976) Another well-known sys-tem is the air-conditioner (AC) In colder regions a cycle operating in the oppositedirection the ldquoheat pumprdquo has recently become popular These two applicationshave also merged together to give a system able to operate in both heating andcooling mode
In Figure 31 a schematic drawing of a simple cycle is shown together with atypical pressure-enthalpy diagram for a sub-critical cycle The cycleworks asfollows
The low pressure vapour (4) is compressed by supplying workWs to give a highpressure vapour with high temperature (1) The vapour is cooled to its saturationtemperature in the first part of the condenser condensed in the middle part andpossibly sub-cooled in the last part to give the liquid (2) In the choke valve thepressure is lowered to its original value resulting in a two-phase mixture (3) Thismixture is vaporized and possibly super-heated in the evaporator (4) closing thecycle
QC
QH
12
3 4
Wsz
Ph
Pl
Evaporator
Condenser
Chokevalve
Compressor
(a)
12
3 4
P
h[Jkg-1]
Ph
Pl
∆Tsub
∆Tsup
Pmmiddot middot middot middot middot middot middot middot middot
(b)
Figure 31 (a) Simple refrigeration or heat pump cycle with (b) typical pressure-enthalpy diagram indicating both sub-cooling and super-heating
The choke valve may be replaced by an expander for improved efficiency but this
31 Introduction 21
Table 31 Structure of model equationsHeat exchangers (condenser and evaporator)Q = U middot
int∆T dA = mmiddot (houtminushin)
P = Psat(Tsat)m= ρVValvem= zmiddotCV
radic∆Pmiddotρ hout = hin
CompressorWs = m(houtminushin) = mmiddot (hsminushin)η
is not considered here The coefficient of performance for a refrigeration cycle(refrigerator AC) is defined as
COP=Qc
Ws=
m(h4minush3)
m(h1minush4)(31)
The COP is typically around 3 which indicates that 33 of the heat duty is addedas work (eg electric power)
In this paper the objective is to optimize the operation of a given cycle (Figure31)in terms of maximize the COP or specifically to minimize the compressor powerWs for a given cooling loadQc We consider only steady state operation Themodel equations are summarized in Table31 Note that pressure losses in pipingand equipment are neglected We also assume that the temperature of the hot(TH)and cold (TC) source are constant throughout the heat exchanger This assumptionholds for a cross flow heat exchanger In practice there may be some operationalconstraints for example maximum and minimum pressure constraints which arenot considered here
In industrial processes especially in cryogenic processes such as air separationand liquefaction of natural gas (LNG process) more complex refrigeration cyclesare used in order to improve the thermodynamic efficiencies These modificationslower the temperature differences in the heat exchangers and include cycles withmixed refrigerants several pressure levels and cascaded cycles Our long termobjective is to study the operation of such processes However as a start we needto understand the simple cycle in Figure31
An important result from this study is the degree of freedom analysis given inSection32 We find that the ldquoactiverdquo charge plays an important role in operation ofcyclic processes This is also directly applicable to more complex designs Unlikean open process a closed cyclic process does not have boundary conditions onpressures imposed by the flows in and out of the system Instead the pressure level
22 Simple cycles Part I
is indirectly given by the external temperatures heat exchanger sizesload andthe active chargeThe active charge is defined as the total mass accumulated inthe process equipment in the cycle mainly in the condenser and evaporator butexcluding any adjustable mass in liquid receivers (tanks)
The effect of a change in active charge on operation depends on the specific designIntuitively it seems that an increase in active charge must increase the pressureand indeed this is true in most cases For example this is the case for the mod-els used in this paper with plug-flow in the heat exchangers Then more liquidinthe condenser gives more sub-cooling which effectively reduces cooling and pres-sure increases Similarly more liquid in the evaporator gives less super-heatingeffectively increasing heat transfer and pressure increases However there may bedesigns where the effect of charge on pressure is opposite For example considera well-mixed flooded condenser where the heat transfer coefficientU to liquid islarger than to vapour An increase in charge (liquid) may then improve coolingand pressure decreases In any case the main point is that the ldquoactiverdquo charge is adegree of freedom that affects the operation of the system and this paper focuseson how to use it effectively
Although there is a vast literature on the thermodynamic analysis of refrigerationcycles there are very few authors who discuss their operation and control Somediscussions are found in text books such asStoecker(1998) Langley(2002) andDossat(2002) but these mainly deal with more practical aspectsSvensson(1994)andLarsen et al(2003) discuss operational aspects A more comprehensive recentstudy is that ofKim et al (2004) who consider the operation of trans-criticalCO2
cycles They discuss the effect of ldquoactive chargerdquo and consider alternatives forplacing the receiver
The paper also discuss super-heating and sub-cooling In the literature it is gen-erally taken for granted that there for a given cycle should be no sub-cooling andsuper-heating (∆Tsub= 0C and∆Tsup= 0C) in optimal operation For exampleStoecker(1998 page 57) states that
The refrigerant leaving industrial refrigeration condensers may be slightlysub-cooled but sub-cooling is not normally desired since it indicates thatsome of the heat transfer surface that should be be used for condensation isused for sub-cooling At the outlet of the evaporator it is crucial for protectionof the compressor that there be no liquid so to be safe it is preferable for thevapor to be slightly super-heated
In this study we confirm that super-heating is not optimal The issue of sub-cooling is less clear Of course sub-cooling in itself is always optimal as lessrefrigerant needs to be circulated The issue is whether sub-cooling is optimal
32 Degrees of freedom in simple cycles 23
for a given cold source temperature and a given condenser area because sub-cooling will reduce the temperature driving forces which must be compensatedby increasing the pressure We find contrary to popular belief that withgivenequipment sub-cooling in the condenser may give savings in energy usage (com-pressor power) in the order of 2 An ammonia case study is presented to obtainnumerical results
32 Degrees of freedom in simple cycles
321 Design versus operation
Table 32 shows typical specifications for the simple refrigeration cycle in Fig-ure 31 in design (find equipment) and in operation (given equipment) The fivedesign specifications include the load the two pressures and the degreeof sub-cooling and super-heating Based on these five design specifications external con-ditions and an assumed isentropic efficiency for the compression we may obtainthe following fourequipment parameters which can be adjusted during operationcompression work (Ws) valve opening (z) and effective heat transfer (includingUA-values) for the two heat exchangers Initially we were puzzled because wecould not identify the missing fifth equipment parameter to be adjusted during op-eration However we finally realized that we can manipulate the rdquoactive chargerdquoin the cycle which affects the operation The fact that the charge is an indepen-dent variable is unique for closed systems since there is no (external) boundarycondition for pressure which would otherwise set the active charge
Table 32 Typical specifications in design and operationGiven
Design Load (egQh) Pl Ph ∆Tsup and∆Tsub 5Operation Ws (load) choke valve opening (z)
effective heat transfer (egUA) in twoheat exchangers and active charge 5
322 Active charge and holdup tanks
For the simple cycle in Figure31we have the following overall material balance
mtot = mevap+mcon+mvalve+mcomp︸ ︷︷ ︸
mactive
+mtanks (32)
24 Simple cycles Part I
Normally the holdups in the valve and compressor are neglected and we get
mtot = mevap+mcon︸ ︷︷ ︸
mactive
+mtanks (33)
With no filling emptying or leaks the total massmtot is fixed We have not in-cluded a holdup tank in Figure31 but in practice it is common to include a tankor receiver with variable liquid mass It is assumed that a change inmtanks (egby filling or leaking) with a constant active charge (mactive) does not affect the op-eration of the cycle This implies that the tank must contain both liquid and gasin equilibrium (saturated) Then we can move mass to or from the tank withoutaffecting the pressure and thus without affecting the rest of the cycleThus theliquid tank makes operation independent of the total charge in the system
More importantly the extra tank introduces an additional degree of freedom Thiscan be seen from Equation33 With mtot constant we can by changing the mass(liquid) in the tank (mtank) change the active charge (mactive) This shows thatmtank
has an indirect steady state effect on the active charge and can therefore be usedfor control purposes of course provided that we have means of changing it
Although it is possible to introduce several tanks in a cycle we only have onematerial balance for each cycle so from Equation33this will not add any steady-state degrees of freedom with respect to the active charge
Rule 31 In each closed cycle we have one degree of freedom related to the activecharge which may be indirectly adjusted by introducing a variable liquid level(tank receiver) in the cycle
Rule 32 In each closed cycle there will be one liquid holdup that does not needto be explicitly controlled because the total mass is fixed This is usually selectedas the largest liquid volume in the closed system The remaining liquid levels(holdups) must be controlled (to avoid overfilling or emptying of tanks)
Remark 1 Note that in Rule32it says ldquodoes not needrdquo rather than ldquomust notrdquo Thus Rule32does not say that we cannot control all the liquid volumes in the system (including thelargest one) but it just states that it is not strictly necessary In fact controlling all theliquid volumes provides a way for explicitly controlling the active charge in the cycle(Rule31)
Remark 2 Introducing additional liquid tanks may be useful for operation but at leastfor pure fluids these will not introduce any additional steady-state degrees of freedombecause we can move mass from one tank to another without affecting operation Alsoto avoid that tanks fill up or empty these additional levels must be controlled (Rule32)either by self-regulation or feedback control
32 Degrees of freedom in simple cycles 25
Remark 3 In mixed refrigerantcycles two tanks may be used to indirectly change thecomposition of the circulating refrigerant In this case the two tanks have different compo-sition so moving mass from one tank to another does affect operation For more complexcycles the maximum number of degrees of freedom related to tank holdups is the numberof components in the refrigerant
Adjusting the active charge
In order to freely adjust the active charge we need to introduce a liquid tank (re-ceiver) plus an extra valveKim et al (2004) discuss alternative locations for thevariable tank holdup (liquid receiver) In Figure32 we show cycles for the twomain cases where the tank is placed (a) on the high pressure side after the con-denser and (b) on the low pressure side after the evaporator Other placements andcombinations are possible but these are only variations of these two and willnotadd any steady-state degrees of freedom for pure refrigerants
The most obvious way of introducing a means for adjusting the tank holdup is toadd an extra valve before the tank as shown in Figure32 In Figure32(a) the
z
QC
QH
Ws
Pl
Ph
Pm
(a) Liquid tank and extra valve on high pres-sure side
z
QC
QH
Ws
Pl
Ph
(b) Liquid tank and extra (non-optimal)valve on low pressure side
Figure 32 Simple cycle with variable active charge
liquid tank is located at an intermediate pressurePm after the condenser Sincethe extra valve is on the ldquosame siderdquo as the expansion valve (choke) the pressuredrop over the extra valve will not effect the efficiency of the cycle Since Pm isassumed to be the saturation pressure at the tank temperature the exit stream fromthe condenser must be sub-cooled (or super-critical but this is not considered inthis paper) Thus in Figure32(a) the pressure drop across the valve may be used
26 Simple cycles Part I
to adjust the degree of sub-cooling in the condenser To understand how the extravalve creates sub-cooling consider the pressure-enthalpy diagram inFigure31The receiver (tank) with saturated liquid operates at saturation pressure Pm andthe pressure drop for the extra valve introduces a pressure dropPhminusPm As seenfrom Figure31 the corresponding operating point 2 at the exit of the condensermust then be at a sub-cooled state
Another possibility is to place the tank after the evaporator as shown in Figure32(b) With this design the stream exiting the evaporator is not fully evapo-rated and by lowering the pressure through the extra valve the vapour exiting thevalve becomes saturated (see pressure-enthalpy diagram) Howeverin this casethe valve introduces a pressure drop that must be compensated by increasing thecompression power so a valve here is generally not optimal
A low pressure tank may not be desirable from a practical point of view since thevapour velocity will be highest at this point in the cycle and the extra equipmentwill increase the pressure drop
Extra valve removed
An extra valve is generally required to freely adjust the active charge However inmany practical cases the extra valve in Figure32(a)and32(b)is removed Whateffect does this have
bull High pressure tank without valve Without the valve we have at steady statethe same thermodynamic state at the exit of the condenser as at the exitfrom the tank Thus the exiting stream from the condenser will be satu-rated liquid The most common design is shown in Figure33 where thetank and condenser are merged together so that the saturated liquid from thecondenser drains into the receiver As we will show this is not generallyoptimal Thus in this design we have used a degree of freedom (ldquofully openvalverdquo) to set the degree of sub-cooling to zero (not optimal)
bull Low pressure tank without valve (Figure34(a)) This gives saturated vapourto the compressor Fortunately this is generally optimal for the cycle as awhole because the inlet temperature to the compressor should be as low aspossible to minimize vapour volume and save compression power Thus inthis design we have used a degree of freedom (ldquofully open valverdquo) to set thedegree of super-heating to zero (optimal) Two designs are shown in Figure34(a) one with a separate receiver and one using a flooded evaporator Thedesigns are equivalent thermodynamically but the heat transfer coefficient
32 Degrees of freedom in simple cycles 27
QH
Figure 33 Condenser with saturation at outlet giving no sub-cooling (commondesign but non-optimal)
and pressure drop will be different
In summary removing the valve gives saturation at the exit of the heat exchangerIn the case of high-pressure liquid tank we get a sub-optimal design if we removethe valve whereas for the low-pressure tank we get an optimal design if the extravalve is removed
QCTC
(a) With separate receiver
QC
(b) Flooded evaporator
Figure 34 Evaporator with saturation at outlet giving no super-heating (optimal)
323 Degrees of freedom for operation
In summary we have the following five operational or control degrees offreedomfor a simple refrigeration cycle (Figure31)
28 Simple cycles Part I
1 Compressor powerWs We assume here that it is used to set the ldquoloadrdquo forthe cycle
2 3 Effective heat transfer There are two degrees of freedom related to adjust-ing the heat transferred in the condenser and evaporator This may be donein many ways for example by introducing bypasses changing the flowratesof coolant or using a flooded condenser or evaporator to change the effec-tive UA-value However we generally find that it is optimal to maximizethe effective heat transfer in the condenser and evaporator Thereare ex-ceptions where it may not be optimal to maximize the heat transfer in thecondenser and evaporator for example because of costs related to pumpsfans or coolants but these degrees of freedom are not consideredin the fol-lowing
4 Choke valve opening (z)
5 Active charge (see Section322)
In practice we are then with a given load and maximum heat transfer left withtwosteady state degrees of freedom These are the choke valve opening (z) and theactive charge (mactive) These may be used to set the degree of super-heating anddegree of sub-cooling The pressure levels (Ph andPl ) are indirectly determined bythe given (maximum) value of the heat transfer
33 Discussion of some designs
As discussed in more detail in Section34 we find that the thermodynamic effi-ciency is optimized by having no super-heating and some sub-cooling With thisin mind we next discuss some alternative designs
331 Optimal designs
Two potentially optimal designs are shown in Figure35 The reason we say ldquopo-tentially optimalrdquo is because they will only be optimal if we use the optimal valuefor the sub-cooling and super-heating
To avoid super-heating we have in Figure35(a)and35(b)a low-pressure tank(receiver) after the evaporator This tank will give saturated vapourout of theevaporator at steady state (optimal) and also by trapping the liquid it will avoidthat we get liquid to the compressor during transient operation To avoid super-heating we must have vapour-liquid equilibrium in the tank This may be achieved
33 Discussion of some designs 29
replacements
QC
QH
Ws
Sub-coolingcontrol
zPl
Ph
(a) Optimal with 1 tank
QC
QH
WsSub-coolingcontrol
z
LC
Pl
Ph
(b) Optimal with 2 tanks
Figure 35 Two potentially optimal designs with sub-cooling and no super-heating
by letting the vapour bubble through the tank An alternative design is the floodedevaporator in Figure34(b)
At the high-pressure side we show optimal designs with both (a) no receiver and(b) a receiver and an extra valve In (a) the choke is used to control the degree ofsub-cooling (∆Tsub) Also other control policies are possible for example keep-ing the choke valve position at its optimal value or controlling the pressure butcontrolling ∆Tsub was found byJensen and Skogestad(2005) to be a good self-optimizing controlled variable
The design in Figure35(b)is thermodynamically equivalent to Figure35(a) butthe addition of the tank may prevent that we get two-phase flow with vapour ldquoblowoutrdquo through the choke We here have two adjustable holdups so from Rule 32one of them must be controlled In Figure35(b)is shown the case where the chokevalve is used to control the level in the high pressure tank but alternatively it couldcontrol the level in the low pressure tank
332 Non-optimal designs
Three non-optimal designs are shown in Figure36 Figure36(a)shows the designused in most applications except that the tank and condenser are often integrated asshown in Figure33 This common design has two errors compared to the optimalsolution 1) There is no sub-cooling in the condenser and 2) there is super-heatingin the evaporator The super-heat control is in practice accomplished witha ther-mostatic expansion valve (TEV) In theory one could get optimality by setting the
30 Simple cycles Part I
setpoint for super-heating to zero but in practice this is not possible because thiscould give liquid out of the evaporator The setpoint for super-heatingis typicallyabout 10C
In Figure36(b)we have two liquid tanks one after the evaporator and one after thecondenser This design is better since there is no super-heating in the evaporatorbut one error remains There is no sub-cooling in the condenser Note that we needto control one of the liquid levels in accordance with Rule32
Another non-optimal design is shown in Figure36(c) Here we have introducedthe possibility for sub-cooling but we have super-heating which is generally not
QC
QH
Ws
Super-heatcontrolz
Pl
Ph
(a) Non-optimal 1 This design has two er-rors 1) No sub-cooling and 2) Super-heating
QC
QH
Ws
z
LC
Pl
Ph
(b) Non-optimal 2 This design has one errorNo sub-cooling
QC
QH
Ws
Super-heatcontrol
Sub-coolingcontrol
zPl
Ph
(c) Non-optimal 3 This design has one errorSuper-heating
Figure 36 Three non-optimal designs
34 Optimality of sub-cooling 31
optimal
34 Optimality of sub-cooling
We have several times made the claim that sub-cooling may be optimal To justifythis somewhat controversial claim we start by considering a specific example
341 Ammonia case study
The objective is to cool a storage building by removing heat (QC) as illustrated inFigure37 The cycle operates between a cold medium of air inside the building(TC = Troom) and hot medium of ambient air (TH = Tamb) The steady state heatloss from the building is 20kW and the cooling loadQC is indirectly adjusted bythe temperature controller which adjusts the compressor work (Ws) to maintainTC = Ts
C
TC TsC
QH
QC
z
Ws
Ph
Pl
Figure 37 Cold warehouse with ammonia refrigeration unit
Some data for the cycle
bull Ambient temperatureTH = 25C
bull Indoor temperature setpointTsC = minus12C
bull Isentropic efficiency for compressor is 95
32 Simple cycles Part I
bull Heat transfer coefficients (U) are 1000 and 500 Wm-2K-1 for the evaporatorand condenser respectively
bull Heat exchangers with areas given in Table33
bull Thermodynamic calculations based on SRK equation of state
The equipment is given and we have 5 steady-state operational degreesof free-dom (Section32) With a given load and maximum heat transfer we have tworemaining steady state degrees of freedom which may be viewed as the degree ofsub-cooling (∆Tsub) and the degree of super-heating (∆Tsup) The performance ofthe cycle measured by the compressor powerWs was optimized with respect tothe two degrees of freedom We find as expected that super-heating is not optimalbut contrary to popular belief the results in Table33 show that sub-cooling by466C reduces the compression workWs by 174 compared to the case withsaturation out of the condenser The high pressurePh increases by 045 but thisis more than compensated by a 212 reduction in flowrate The sub-cooling in-creases the condenser chargeMcon by 501 Figure38shows the correspondingpressure enthalpy diagram for the two cases and Figure39shows the temperatureprofile in the condenser Similar results are obtained if we use other thermody-namic data if we change the compressor efficiency or if we let UA be smaller inthe sub-cooling zone
Enthalpy[Jmol-1]
Pre
ssur
e[P
a]
-7 -65 -6 -55 -5 -45times104
105
106
107
(a) Optimal operation without sub-cooling (Fig-ure36(b))
Enthalpy[Jmol-1]
Pre
ssur
e[P
a]
-7 -65 -6 -55 -5 -45times104
105
106
107
(b) Optimal operation with sub-cooling allowed(Figure35)
Figure 38 Pressure-enthalpy diagrams with and without sub-cooling
The improvement of 2 would be larger if the pressure drop in the piping andequipment was accounted for in the model because the mass flowrate is reducedwith an unchanged low pressure Thus the volumetric flowrate in the low pressure
34 Optimality of sub-cooling 33
Position[-]
Tem
pera
ture[
C]
0 02 04 06 08 130405060708090
100110120
(a) Temperature profile without sub-cooling
Position[-]
Tem
pera
ture[
C]
0 02 04 06 08 130405060708090
100110120
(b) Temperature profile with sub-cooling
Figure 39 Temperature profile in condenser
Table 33 Optimal operation with and without sub-coolingNo sub-cooling Optimal sub-cooling
Ws[W] 4648 4567QC [kW] 20 20m[kgs-1] 00177 00173Mcon
lowast [kg] 0301 0316∆Tsub[
C] 000 466∆Tsup[
C] 000 000∆Tmin con[
C] 500 0491Ph [bar] 1163 1168Pl [bar] 217 217Acon[m2] 870 870Avap[m2] 400 400
lowastEvaporator charge has no effect because of saturation (no super-heating) in the evaporator
side is reduced and this is important as pressure drop is most critical at low pres-sure The pressure losses on the high pressure side will also be slightly reduced(because of smaller flowrate and higher pressure) but this is less important for theefficiency of the cycle
34 Simple cycles Part I
342 Explanation
The irreversible isenthalpic expansion through the choke valve gives a thermody-namic loss The reason for the improvement in efficiency by sub-cooling is thatloss is reduced because less vapour is formed see Figure38 This more than com-pensates the increased irreversible loss due to larger temperature difference in thecondenser To understand this in more detail consider Figure310which shows aconceptual pressure enthalpy diagram of a typical vapour compression cycle Wehave indicated a cycle without sub-cooling (solid line) and the same cycle withsub-cooling (dotted line) Note that since we in the latter case have a higher con-denser pressure (and therefore also a higher temperature in the condensing section)we will with given equipment (UA-values) have more heat transfer whichgivesa lower outlet temperature The condenser outlet will follow the line ldquoCon outrdquowith increasing pressure The line will asymptotically approach the hot sourcetemperatureTH and we want to find the optimal operating point on this line
P
h[Jkg-1]
∆ws∆qC qC ws
TC-line
TH-line
Con out
Figure 310 Pressure-enthalpy diagram for a cycle with and without sub-cooling
If we consider moving from one operating point to another we require an increasein the COP for the change to be optimal
∆COP=qC +∆qC
ws+∆wsminus
qC
wsgt 0 (34)
COPmiddot∆ws lt ∆qC (35)
34 Optimality of sub-cooling 35
whereqC middot m= QC andws middot m= Ws We assume thatQC [Js-1] is given and thatm[kgs-1] andqC [Jkg-1] may vary We use∆Tsub as the independent variable andintroduce differentials The requirement for improving efficiency is then fromEquation35
(partqC
part∆Tsub
)
UAgt COPmiddot
(partws
part∆Tsub
)
UA(36)
According to Equation36 for an initial COP of 3 the increase in specific dutyin the evaporator (∆qC) should be 3 times larger than the increase in specific com-pressor power (∆ws) to give improved performance In Figure310we have that∆qC asymp ∆ws so the optimal degree of sub-cooling is clearly less than that indi-cated by this figure Note however that the ldquoCon outrdquo line is much flatter forsmaller∆qC so a small degree of sub-cooling may be optimal The optimum islocated at the degree of sub-cooling where the inequality in Equation36becomesan equality In the case study we found that the optimum outlet temperature fromthe condenser (2549C) is closer toTH (25C) than the saturation temperature(3015C)
Similar considerations on optimizing the pressurePh have been made earlier fortrans-criticalCO2-cycles (Kim et al 2004) However for sub-critical cycles likethe ammonia cycle studied above it has been assumed that the pressure is fixed bya saturation condition
343 Discussion of sub-cooling Why not found before
The above results on optimality of sub-cooling is contrary to previous claims andpopular belief Why has this result not been found before
Reason 1 Not allowed by design
The design of the condenser is often as shown in Figure33 where the saturatedliquid drains into a liquid receiver In this design it is not possible to have sub-cooling
Reason 2 Infinite area case
The optimal degree of sub-cooling becomes smaller as we increase the heattrans-fer (UA-values) In particular with an infinite heat transfer area sub-cooling is
36 Simple cycles Part I
not optimal In this case the temperature at the condenser outlet is equal to thehotsource temperatureTH Neglecting the effect of pressure on liquid enthalpy theenthalpy is also given We then find that∆qC = 0 and sub-cooling is not optimalas illustrated in Figure311
P
h[Jkg-1]
∆wsws∆qC qC
TC-line
TH-line
Figure 311 Pressure-enthalpy diagram for infinite area case where condenseroutlet is at hot source temperatureTH
In practice the enthalpy depends slightly on pressure (as indicated by thecurvedconstant temperature lines in Figure311) so ∆qC might be larger than zero butthis effect is too small to change the conclusion that sub-cooling is non-optimalwith infinite area
Reason 3 Specifying HRAT
The minimum approach temperature (∆Tmin or HRAT) is commonly used as a spec-ification for design of processes with heat exchangers The idea is to specify ∆Tmin
in order to get a reasonable balance between minimizing operating (energy)costs(favored by a small∆Tmin) and minimizing capital costs (favored by a large∆Tmin)Although specifying∆Tmin may be reasonable for obtaining initial estimates forstream data and areas it should not be used for obtaining optimal design data -and especially not stream data (temperatures) This follows because specifying∆Tmin results in an optimum with no sub-cooling This can be seen by letting theTH-line in Figure311representTH + ∆Tmin The condenser outlet temperature isthenTH +∆Tmin and similarly to the infinite area case we get∆qC = 0 (neglectingthe effect of pressure on liquid enthalpy) and sub-cooling is not optimal
The results can also be understood because specifying∆Tmin favors designs with∆T being as close as possible to∆Tmin throughout the heat exchanger and this
35 Discussion 37
clearly disfavour sub-cooling see Figure39(b)
A third way of understanding the difference is that we end up with two differentoptimization problems for design (Equation37) and operation (Equation38)
min (Ws) (37)
subject to TCminusTsC = 0
∆Ti minus∆Tmini ge 0
min (Ws) (38)
subject to TCminusTsC = 0
Amaxi minusAi ge 0
For the ammonia case study solving37with ∆Tmin = 5C gives the data for ldquoNosub-coolingrdquo in Table33 Setting the resulting areas asAmax and solving theoptimization problem38results in A=Amaxand the data for ldquoOptimal sub-coolingrdquoin Table33 We see that specifying∆Tmin gives no sub-cooling whereas fixingthe heat exchanger areas to the same value gives 466C of sub-cooling
35 Discussion
351 Super-heating by internal heat exchange
For the simple cycle in Figure31 some sub-cooling in the condenser was foundto be optimal and we here discuss whether other means of obtaining further sub-cooling in particular the use of internal heat exchange (Figure312) may be ben-eficial
Consider first the case when the vapour leaving the evaporator is saturated In thiscase the internal heat exchange in Figure312has no effect on the overall processat least for pure fluids This can be understood because there is no effect on thepressure-enthalpy diagram
Next consider the case where the vapour is super-heated which was previouslywithout internal heat exchange found to be non-optimal Depending on the prop-erties of the fluid this design may be desirable in some cases even for purere-frigerants (Radermacher 1989) In the ammonia case study presented above it isnot optimal with internal heat exchange but for a trans-criticalCO2 cycle internalheat exchange with super-heating is optimal (Neksaa et al 1998)
38 Simple cycles Part I
z
QC
QH
QA
Ws
Pl
Ph
Figure 312 Internal heat exchange
352 Selection of controlled variable
Without internal heat exchange we have found that it is generally optimalto haveno super-heating (∆Tsup= 0C) and some sub-cooling (∆Tsubgt 0C) In practiceno super-heating is easily obtained by use of a design with a low pressure tankas shown in Figure34(a)and Figure35 It is less clear how to get the rightsub-cooling In Figure35 we show a strategy where a valve is used to controlthe degree of sub-cooling∆Tsub However the optimal value of∆Tsub will varyduring operation and also∆Tsub may be difficult to measure and control so it isnot clear that this strategy is good More generally we could envisage anon-lineoptimization scheme where one continuously optimizes the operation (maximizesCOP) by adjusting the valves However such schemes are quite complex andsensitive to uncertainty so in practice one uses simpler schemes like the oneinFigure35 where the valve controls some other variable Such variables could be
bull Choke valve position setpointzs (that is the valve is left in a constant posi-tion)
bull High pressure (Ph)
bull Low pressure (Pl )
bull Temperature out of condenser (T2)
bull Degree of sub-cooling (∆Tsub= T2minusTsat(Ph))
bull Temperature out of evaporator (T4)
bull Degree of super-heating (∆Tsup= T4minusTsat(Pl ))
bull Liquid level in storage tank (to adjust charge to rest of system)
36 Conclusion 39
bull Pressure drop across the extra valve if the design in Figure35(b)is used
The objective is to achieve ldquoself-optimizingrdquo control where a constant setpoint forthe selected variable indirectly leads to near-optimal operation (Skogestad 2000)The selection of ldquoself-optimizingrdquo controlled variables for simple refrigeration cy-cles is the main topic in Part II (Jensen and Skogestad 2007)
36 Conclusion
The ldquoactive chargerdquo in a closed cycle has a steady state effect This is unlike opensystems where we have boundary conditions on pressure To adjust the degreeof freedom related to the ldquoactive chargerdquo one needs a liquid tank (receiver) in thecycle The key to make efficient use of this degree of freedom is to allow forsub-cooling in the condenser Conventional wisdom says that one should avoidsub-cooling in the condenser to maximize the efficiency However we find thatsome sub-cooling is desirable For the ammonia case study we get savings in theorder of 2 by using the design in Figure35 that allows for sub-cooling Thesavings would be even larger if we compared with the common design in Figure36(a)which in addition to having no sub-cooling also gives super-heating
Nevertheless the savings in themselves are not very large More importantly theresults show that the active charge is a degree of freedom and that the sub-coolinggives some decoupling between the high pressurePh and the hot source temper-atureTH This is similar to that found for other cycles including mixed (multicomponent) fluids and trans-criticalCO2
Frictional pressure drops in the equipment have been neglected but their inclusionwould further favor sub-cooling which has a smaller mass flow
Bibliography
Dossat R J (2002)Principles of refrigeration Prentice Hall
Jensen J B and Skogestad S (2005) Control and optimal operation of simpleheat pump cyclesin lsquoEuropean Symposium on Computer Aided Process Engi-neering (ESCAPE) 15 Barcelonarsquo
Jensen J B and Skogestad S (2007) lsquoOptimal operation of a simple refrigerationcycles Part II Selection of controlled variablesrsquoComput Chem Eng31 1590ndash1601
40 Bibliography
Kim M Pettersen J and Bullard C (2004) lsquoFundamental process and systemdesign issues in CO2 vapor compression systemsrsquoProgress in energy and com-bustion science30 119ndash174
Langley B C (2002)Heat pump technology Prentice Hall
Larsen L Thybo C Stoustrup J and Rasmussen H (2003) Control meth-ods utilizing energy optimizing schemes in refrigeration systemsin lsquoEuropeanControl Conference (ECC) Cambridge UKrsquo
Nagengast B (1976) lsquoThe revolution in small vapor compression refrigerationrsquoAmerican Society of Heating Refrigerating and Air-Conditioning Engineers(ASHRAE)18(7) 36ndash40
Neeraas B O Brendeng E Wallentinsen A and Mangersnes M (2001) A newconcept for small-scale LNG productionin lsquoAIChE Spring National Meetingrsquo
Neksaa P Rekstad H Zakeri G R and Schiefloe P A (1998) lsquoCO2-heat pumpwater heater characteristics system design and experimental resultsrsquoInt JRefrigeration21 172ndash179
Radermacher R (1989) lsquoThermodynamic and heat-transfer implicationsof work-ing fluid mixtures in Rankine cyclesrsquoInt J Heat Fluid Flow10(2) 90ndash102
Skogestad S (2000) lsquoPlantwide control the search for the self-optimizing controlstructurersquoJ Process Contr10(5) 487ndash507
Stoecker W F (1998)Industrial refrigeration handbook McGraw-Hill
Svensson M C (1994) Studies on on-line optimizing control with applicationto a heat pump PhD thesis Norwegian University of Science and TechnologyTrondheim
Chapter 4
Optimal operation of simplerefrigeration cyclesPart II Selection of controlledvariables
Published in Computers amp Chemical Engineering (2007) 31 pages 1590-1601
The paper focuses on operation of simple refrigeration cycles and con-siders the selection of controlled variables for two different cycles Oneis a conventional sub-critical ammonia refrigeration cycle and the otheris a trans-criticalCO2 refrigeration cycle There is no fundamental dif-ference between the two cycles in terms of degrees of freedomand op-eration However in practical operation there are differences For theammonia cycle there are several simple control structuresthat give self-optimizing control that is which achieve in practice close-to-optimaloperation with a constant setpoint policy For theCO2 cycle on theother hand a combination of measurements is necessary to achieve self-optimizing control
41 Introduction
Refrigeration and heat pump cycles are used both in homes cars and in industryThe load and complexity varies from small simple cycles like a refrigerator or air-conditioner to large complex industrial cycles like the ones used in liquefactionof natural gas
41
42 Simple cycles Part II
TCTC (building)
TH (ambient)
TH (ambient)
T1T2
T3
T4Ts
C
QH
z
Ws
QC
Qloss
Ph
Pl
(chokevalve)
(compressor)
(evaporator)
(condenser)
Vl
Figure 41 Simple refrigeration cycle studied in this paper (shown for the ammo-nia case)
The simple refrigeration process illustrated in Figure41is studied in this paper InPart I (Jensen and Skogestad 2007b) we showed that the cycle has five steady-statedegrees of freedom the compressor power the heat transfer in the condenser theheat transfer in the evaporator the choke valve opening and the ldquoactive chargerdquoDifferent designs for affecting the active charge including the locationof the liquidstorage were discussed in Part I
It was found in Part I that there are normally three optimally active constraintsmaximum heat transfer in condenser maximum heat transfer in evaporatorandminimum (zero) super-heating The cycle in Figure41obtains the latter by havinga liquid receiver before the compressor which gives saturated vapourentering thecompressor In addition we assume that the load (eg cooling duty) is specifiedThere is then one remaining unconstrained steady-state degree of freedom relatedto the outlet temperature of the condenser which should be used to optimize theoperation The main theme of the paper is to select a ldquoself-optimizingrdquo controlledvariable for this degree of freedom such that a constant setpoint policy(indirectly)achieves near-optimal operation
We consider two systems
bull a conventional sub-critical ammonia cycle for cold storage (TC = minus10C)
bull a trans-criticalCO2 cycle for cooling a home (TC = 20C)
TheCO2 cycle is included since it always has an unconstrained degree of freedomthat must be used for control This is because there is no saturation condition on the
42 Selection of controlled variable 43
high pressure side which is usually said to introduce one extra degree offreedomto the cycle (Kim et al 2004) However as shown in Part I (Jensen and Skogestad2007b) this ldquoextrardquo degree of freedom is also available in a conventional sub-critical cycle if we allow for sub-cooling in the condenser The sub-cooling willto some extent decouple the outlet temperature and the saturation pressure inthecondenser More importantly some sub-cooling is actually positive in terms ofthermodynamic efficiency (Jensen and Skogestad 2007b) The ammonia cycle isincluded to show that there are no fundamental differences between a sub-criticaland a trans-critical cycle There is some confusion in the literature on this
Although there is a vast literature on the thermodynamic analysis of closed refrig-eration cycles there are few authors who discuss the operation and control of suchcycles Some discussions are found in text books such asStoecker(1998) Lang-ley (2002) andDossat(2002) but these mainly deal with more practical aspectsSvensson(1994) and Larsen et al(2003) discuss operational aspects A morecomprehensive recent study is that ofKim et al(2004) who consider the operationof trans-criticalCO2 cycles
This paper considers steady-state operation and the objective is to find which con-trolled variables to fix The compressor power is used as the objective function(costJ = Ws) for evaluating optimal operation
42 Selection of controlled variable
We consider here the simple cycle in Figure41 where the liquid receiver on thelow pressure side ensures that the vapour entering the compressor is saturatedNote that there is no liquid receiver after the condenser and thus no assumptionof having saturated liquid at the condenser outlet Furthermore it is assumed thatthe heat transfer in both the condenser and evaporator are maximized Finally atemperature controller on the stream to be cooled (here the building temperatureTC) is used to adjust the compressor power
There then remains one unconstrained degree of freedom (choke valve positionz)which should be used to optimize the operation for all disturbances and operatingpoints We could envisage an real-time dynamic optimization scheme where onecontinuously optimizes the operation (minimize compressor power) by adjustingzHowever such schemes may be quite complex and sensitive to uncertainty Theseproblems can be reduced by selecting a good control variable and ideallyone geta simple constant setpoint scheme with no need for real-time optimization Whatshould be controlled (and fixed at least on the short time scale) Some candidates
44 Simple cycles Part II
are
bull Valve positionz(ie an open-loop policy where the valve is left in a constantposition)
bull High pressure (Ph)
bull Low pressure (Pl )
bull Temperature out of compressor (T1)
bull Temperature before valve (T2)
bull Degree of sub-cooling in the condenserlowast (∆Tsub= T2minusTsat(Ph))
bull Temperature approach in hot source heat exchanger (T2minusTH)
bull Temperature out of evaporator (T4)
bull Degree of super-heating in the evaporatordagger (∆Tsup= T4minusTsat(Pl ))
bull Liquid level in the receiver (Vl ) to adjust the active charge in the rest of thesystem
bull Liquid level in the condenser (Vl con) or in the evaporator (Vl vap)
bull Pressure drop across the ldquoextrardquo valve in Figure411dagger
The objective is to achieve ldquoself-optimizingrdquo control where a constant setpoint forthe selected variable indirectly leads to near-optimal operation (Skogestad 2000)Note that the selection of a good controlled variable is equally important in an ldquoad-vancedrdquo control scheme like MPC which also is based on keeping the controlledvariables close to given setpoints
The selection of controlled variables is a challenging task especially if one con-siders in detail all possible measurements so we will first use a simple screeningprocess based on a linear model
421 Linear analysis
To find promising controlled variables the ldquomaximum gainrdquo rule (Halvorsen et al2003) will be used For the scalar case considered in this paper the rule isPrefer controlled variables with a large scaled gain|Gprime| from the input (degree of
lowastNot relevant in theCO2 cycle because of super-critical high pressuredaggerNot relevant for our design (Figure41)
42 Selection of controlled variable 45
freedom) to the output (controlled variable)Procedure scalar case
1 Make a small perturbation in each disturbancesdi and re-optimize the oper-ation to find the optimal disturbance sensitivitypart∆yoptpartdi Let ∆di denotethe expected magnitude of each disturbance and compute from this the over-all optimal variation (here we choose the 2-norm)
∆yopt =
radic
sumi
(part∆yopt
partdimiddot∆di
)2
2 Identify the expected implementation errorn for each candidate controlledvariabley (measurement)
3 Make a perturbation in the independent variablesu (in our caseu is the chokevalve positionz) to find the (unscaled) gainG = ∆y∆u
4 Scale the gain with the optimal span (spanyequiv ∆yopt+n) to obtain for eachcandidate output variabley the scaled gain|Gprime| = |G|
spany
The worst-case lossL = J(ud)minusJopt(ud) (the difference between the cost with aconstant setpoint and re-optimized operation) is then for the scalar case (Skogestadand Postlethwaite 2005 page 394)
L =|Juu|
21
|Gprime|2(41)
whereJuu = part 2Jpartu2 is the Hessian of the cost functionJ In our caseJ = Ws
(compressor work) Note thatJuu is the same for all candidate controlled variablesy
The most promising controlled variables should then be tested on the non-linearmodel using realistic disturbances to check for non-linear effects including feasi-bility problems
422 Combination of measurements
If the losses with a fixed single measurement are large as for theCO2 case studythen one may consider combinations of measurements as controlled variablesThesimple null space method (Alstad and Skogestad 2007) gives a linear combinationwith zero local loss for the considered disturbances
c = h1 middoty1 +h2 middoty2 + (42)
46 Simple cycles Part II
The minimum number of measurementsy to be included in the combination isny = nu+nd In our casenu = 1 and if we want to consider combinations ofny = 2measurements then onlynd = 1 disturbance can be accounted exactly for With theldquoexact local methodrdquo (Halvorsen et al 2003) or the ldquoextended null space methodrdquo(Alstad and Skogestad 2007) it is possible to consider additional disturbancesThe local loss is then not zero and we will minimize the 2-norm of the effect ofdisturbances on the loss
43 Ammonia case study
The cycle operates between air inside a building (TC = Troom = minus10C) and am-bient air (TH = Tamb = 20C) This could be used in a cold storage building asillustrated in Figure41 The heat loss from the building is
Qloss= UAloss(TH minusTC) (43)
The nominal heat loss is 15kW The temperature controller shown in Figure41maintainsTC = minus10C and will indirectly giveQC = Qloss at steady-state
431 Modelling
The structure of the model equations are given in Table41and the data are givenin Table42 The heat exchangers are modelled assuming ldquocross flowrdquo with con-stant temperature on the air side (TH = 20C andTC = minus10C) The isentropicefficiency for the compressor is assumed constant The SRK equation of state isused for the thermodynamic calculations The gPROMS model is available on theinternet (Jensen and Skogestad 2007a)
Table 41 Structure of model equationsHeat exchangers (condenser and evaporator)Q = U middot
int∆T dA = mmiddot (houtminushin)
P = Psat(Tsat)m= ρVValvem= zmiddotCV
radic∆Pmiddotρ hout = hin
CompressorWs = m(houtminushin) = mmiddot (hsminushin)η
43 Ammonia case study 47
Table 42 Data for the ammonia case studyTH = 20CTC = Ts
C = minus10CCondenser (UA)C = 2500WK-1
Evaporator (UA)E = 3000WK-1
Compressor isentropic efficiencyη = 095Choke valveCV = 00017m2
Building UAloss= 500WK-1
432 Optimal steady-state operation
At nominal conditions the compressor power was minimized with respect to thedegree of freedom (z) The optimal results are given in Table43 and the corre-sponding pressure enthalpy diagram and temperature profile in the condenser areshown in Figure42 Note that the optimal sub-cooling out of the condenser is58C This saves about 20 in compressor power (Ws) compared to the conven-tional design with saturation
h [Jmol-1]
P[P
a]
-7 -65 -6 -55 -5 -45times104
105
106
107
2
3 4
1
(a) Pressure enthalpy diagram
Position[-]
T[
C]
0 02 04 06 08 12030405060708090
100110120
T1
T2
Tsat(Ph)
(b) Temperature profile in condenser
Figure 42 Optimal operation for the ammonia case study
433 Selection of controlled variables
There is one unconstrained degree of freedom (choke valve openingz) whichshould be adjusted to give optimal sub-cooling in the condenser We want tofind a good controlled variable (see Section42 for candidates) to fix such that
48 Simple cycles Part II
Table 43 Optimal steady-state for ammonia case studyWs[kW] 2975
z[-] 0372Ph [bar] 1070Pl [bar] 235
QH [kW] 1796m[kgs-1] 00127
∆Tsub= T2minusTsat(Ph) [C] 580T1 [C] 1026T2 [C] 209T3 [C] -150T4 [C] -150
we achieve close-to-optimal operation in spite of disturbances and implementationerror (ldquoself-optimizing controlrdquo)
Linear analysis of alternative controlled variables
The following disturbance perturbations are used to calculate the optimal variationin the measurementsylowast
d1 ∆TH = plusmn10C
d2 ∆TsC = plusmn5C
d3 ∆UAloss= plusmn100WK-1
The assumed implementation error (n) for each variable is given in Table44whichalso summarizes the linear analysis and gives the resulting scaled gains in orderfrom low gain (poor) to high gain (promising)
Some notes about Table44
bull Pl andT4 have zero gains and cannot be controlled The reason for the zerogains are that they both are indirectly determined byQloss
Qloss= QC = (UA)C (T4minusTC) and Pl = Psat(T4) (44)
lowastIn order to remain in the linear region the optimal variations were computedfor a disturbanceof magnitude 1100 of this and the resulting optimal variations were then multiplied by 100 to get∆yopt(di)
43 Ammonia case study 49
Table 44 Linear ldquomaximum gainrdquo analysis of candidate controlled variablesy forammonia case study
|∆yopt(di )|Variable (y) Nom G d1 (TH ) d2 (TC) d3 (UAloss) |∆yopt| n span y |Gprime|Pl [bar] 235 000 0169 0591 0101 0623 0300 0923 000T4 [C] -150 000 0017 0058 0010 0061 100 106 000∆Tsup[
C] 000 000 000 000 000 000 100 100 000T1 [C] 1026 -14374 38 173 62 422 100 432 333Ph [bar] 1071 -1739 412 041 0460 417 100 517 337z[-] 0372 1 00517 00429 00632 0092 005 0142 703T2 [C] 209 28795 104 020 0300 104 100 114 253Vl [m
3] 100 51455 9e-03 0011 12e-03 00143 005 0064 801∆Tsub[
C] 580 -34078 213 108 108 262 150 412 828Vl con[m3] 067 -57 58e-03 24e-03 14e-03 00064 005 0056 1010T2minusTH [C] 089 -28795 0375 0174 0333 0531 150 203 1418
bull The degree of super-heating∆Tsup can obviously not be controlled in ourcase because it is fixed at 0C (by design of the cycle)
bull The loss is proportional to the inverse of squared scaled gain (see Equation41) This implies for example that a constant condenser pressure (Ph)which has a scaled gain of 337 would result in a loss in compressor powerJ = Ws that is(828337)2 = 603 times larger than a constant sub-cooling(∆Tsub) which has a scaled gain of 828
bull The simple policies with a constant pressure (Ph) or constant valve position(z) are not promising with scaled gains of 337 and 703 respectively
bull A constant level in the liquid receiver (Vl ) is a good choice with a scaledgain of 801 However according to the linear analysis the liquid level inthe condenser (Vl con) is even better with a scaled gain of 1010
bull Controlling the degree of sub-cooling in the condenser (∆Tsub= T2minusTsat(Ph))is also promising with a scaled gain of 828 but the most promising is thetemperature approach at the condenser outlet (T2minusTH) with a scaled gain of1418
bull The ratio between the implementation errorn and the optimal variation∆yopt
tells whether the implementation error or the effect of the disturbance is mostimportant for a given control policy For the most promising policies we seefrom Table44 that the contribution from the implementation error is mostimportant
Nonlinear analysis
The nonlinear model was subjected to the ldquofullrdquo disturbances to test more rigor-ously the effect of fixing alternative controlled variables The main reason for
50 Simple cycles Part IIW
s[k
W]
TH [C]10 15 20 25 30
1
2
3
4
5mdash y = Phmdash y = z- - No sub-cooling
mdash y = ∆Tsubmdash y = Vlmdash y = Vconlmdash y = T2minusTHmdash Opt
(a) Disturbance inTH (d1)
Loss
[kW
]
TH [C]10 15 20 25 30
01
02
03
04
05
Ph
∆Tsub
z
zNo sub-cooling
Vl
Vconl
T2minusTH
(b) Disturbance inTH (d1)
Ws[k
W]
TC [C]-15 -10 -5
1
2
3
4
5
(c) Disturbance inTC (d2)
Loss
[kW
]
TC [C]-15 -10 -5
01
02
03
04
05
Phz
z
No sub-cooling
∆Tsub
Vl
Vconl
T2minusTH
AAAAAU
(d) Disturbance inTC (d2)
Ws[k
W]
UA[W C-1]400 500 600
1
2
3
4
5
(e) Disturbance inUAloss (d3)
Loss
[kW
]
UA[W C-1]400 500 600
01
02
03
04
05
Ph z
z
No sub-cooling
∆Tsub
Vl
Vconl
T2minusTH
(f) Disturbance inUAloss (d3)
Figure 43 Ammonia case Compressor power (left) and loss (right) for differentdisturbances and controlled variables A line that ends corresponds to infeasibleoperation
43 Ammonia case study 51
considering the full disturbances is to check for non-linear effects in particularpossible infeasible operation which cannot be detected from the linear analysisFigure43shows the compressor powerWs (left) and lossL =WsminusWsopt (right) fordisturbances inTH (d1) TC (d2) andUAloss (d3) Wsopt is obtained by re-optimizingthe operation for the given disturbances As predicted from the linear analysiscontrol ofPh or z should be avoided as it results in a large loss and even infeasi-bility (a line that ends corresponds to infeasible operation) Controlling the degreeof sub-cooling∆Tsub gives small losses for most disturbances but gives infeasibleoperation whenTH is low Controlling the liquid level either in the receiver orin the condenser gives small losses in all cases Another good policy is tomain-tain a constant temperature approach out of the condenser (T2minusTH) This controlpolicy was also the best in the linear analysis and has as far as we know notbeensuggested in the literature for ammonia cycles
A common design for refrigeration cycles also discussed in Part I is to have nosub-cooling in the condenser In practice this might be realized with the designin Figure41 by adding a liquid receiver after the condenser and using the chokevalve to control this liquid level or using the design in Figure411with the ldquoextrardquovalve between the condenser and tank removed The performance of thisdesign(ldquono sub-coolingrdquo) is shown with the dashed line in Figure43 The loss (rightgraphs) for this design is always nonzero as it even at the nominal point has a lossof 006kW and the loss increases with the cooling duty of the cycle Neverthelesswe note that the loss with this design is low (less than about 02kW or 35) for allconsidered disturbances This may be acceptable although it is much higherthanthe best controlled variables (Vl Vconl andT2minusTH) where the maximum losses areless than 0005kW
Figure44shows the sensitivity to implementation error for the four best controlledvariables Controlling a temperature difference at the condenser exit (either T2minusTH or ∆Tsub) has a small sensitivity to implementation error On the other handcontrolling either of the two liquid levels (Vl or Vl con) might lead to infeasibleoperation for relatively small implementation errors In both cases the infeasibilityis caused by vapour at the condenser exit In practice this vapour ldquoblow outrdquo maybe ldquofeasiblerdquo but certainly not desirable
A third important issue is the sensitivity to the total charge of the system whichis relevant for the case where we control the liquid level in the receiver (y = Vl )There is probably some uncertainty in the initial charge of the system and maybemore importantly there might be a small leak that will reduce the total charge overtime Optimally the total charge has no steady-state effect (it will only affect theliquid level in the receiver) However controlling the liquid level in the receiver(y = Vl) will make the operation depend on the total charge and we have lost one
52 Simple cycles Part IILo
ss[k
W]
Vl [m3]095 1 105
01
02
03
04
05
y = Vl
(a)
Loss
[kW
]
Vconl [m3]062 064 066 068 07 072
01
02
03
04
05
y = Vl con
(b)
Loss
[kW
]
T2minusTH [C]-15 -1 -05 0
01
02
03
04
05
y = T2minusTH
(c)
Loss
[kW
]
∆Tsub[C]
5 6 7
01
02
03
04
05
y = ∆Tsub
(d)
Figure 44 Ammonia case Loss as function of implementation error
of the positive effects of having the liquid receiver The other control structureswill not be affected by varying total charge
Conclusion ammonia case study
For the ammonia case controlling the temperature approach at the condenser exit(T2minusTH) seems to be the best choice as the losses caused by implementation error(Figure44) and disturbances (Figure43) are very small This control implemen-tation is shown in Figure45where we also have introduced an inner ldquostabilizingrdquoloop for pressure However thesetpointfor the pressure is used as a degree offreedom so this loop does not affect the results of this study which are based onsteady-state Although not optimal even nominally another acceptable policyis
44CO2 case study 53
to use the conventional design with no sub-cooling (Figure411 with the ldquoextravalverdquo removed and minimum super-heating)
PSfrag
TC
TC
PC
TC (building)
minusT1
T2
T3 T4
TsC
QH
z
Ws
QC
Qloss
Ph
Pl
Psh
TH (ambient)
(T2minusTH)s
Figure 45 Proposed control structure for the ammonia cycle
44 CO2 case study
Neksaa(2002) shows thatCO2 cycles are attractive for several applications bothfrom an efficiency point of view and from an environmental perspective Skaugen(2002) gives a detailed analysis of the parameters that affect the performance of aCO2 cycle and discusses pressure control in these systems
The simple cycle studied in this paper see Figure46(a) operates between airinside a room (TC = 20C) and ambient air (TH = 30C) This could be an air-conditioner for a home as illustrated in Figure46(a) The heat loss out of thebuilding is given by Equation43 and the temperature controller shown in Figure46(a)indirectly givesQC = Qloss The nominal heat loss is 40kW
We consider a cycle with an internal heat exchanger see Figure46(a) This heatexchanger gives further cooling before the choke valve by super-heating the sat-urated vapour from the evaporator outlet This has the advantage of reducing theexpansion loss through the valve although super-heating increases thecompres-sor power For theCO2 cycle it has been found that the internal heat exchangerimproves efficiency for some operating points (Domanski et al 1994) For the
54 Simple cycles Part II
CO2 cycle we find that the internal heat exchanger gives a nominal reduction of99 in Ws For the ammonia cycle the effect of internal heat exchange to givesuper-heating is always negative in terms of efficiency
441 Modelling
Table41shows the structure of the model equations and the data are given in Table45 Constant air temperature is assumed in the evaporator (TC) The gas coolerand internal heat exchanger are modelled as counter-current heat exchangers with6 control volumes each The Span-Wagner equation of state (1996) is used forthe thermodynamic calculations The MATLAB model is available on the internet(Jensen and Skogestad 2007a)
z
TC
TH
TH
Qihx
Qloss
Ws
QC
QHPh
Pl
TC
T1T2
T3 T4
TsC
Vl
T prime2
T prime4
(Gascooler)
(Internalhex)
(a) TheCO2 cycle
h [Jmol-1]
Pre
ssur
e[P
a]
0C
-41 -405 -4 -395times105
106
107 2
3 4
1
(b) Pressure enthalpy diagram
Figure 46 TheCO2 cycle operates trans-critical and is designed with an internalheat exchanger
442 Optimal operation
Some key parameters for optimal operation of theCO2 cycle are summarized inTable46 and the pressure enthalpy diagram is given in Figure46(b) Figure47shows the optimal temperature profiles in the gas cooler and in the internal heatexchanger
44CO2 case study 55
Table 45 Conditions for theCO2 case studyEvaporator(UA)vap = 798WC-1
Gas cooler(UA)gco = 795WC-1
Internal heat exchanger(UA)ihx = 153WC-1
Compressor isentropic efficiencyη = 075Ambient TH = 30CAir flow gas coolermcp = 250JC-1s-1
RoomTC = TsC = 20C
RoomUAloss= 400WC-1
Choke valveCV = 121middot10minus6m2
Table 46 Optimal operation forCO2 caseWs[W] 958
z[-] 034Ph [bar] 9761Pl [bar] 5083QH [W] -4958
Qihx [W] 889m[kgs-1] 0025
T1 [C] 896T2 [C] 255T3 [C] 150T4 [C] 312
Note that when the ambient air goes below approximatelyTH = 25C the opti-mal pressure in the gas cooler is sub-critical We will only consider trans-criticaloperation so we assume that the air-conditioner is not used below 25C
443 Selection of controlled variable
We want to find what the valve should control In addition to the variables listedinSection42 we also consider internal temperature measurements in the gas coolerand internal heat exchanger Note that the ldquono sub-coolingrdquo policy is not possiblefor theCO2 cycle because it operates trans-critical
As discussed in more detail below there are no obvious single measurementstocontrol for this application One exception is the holdupm on the high pressureside of the cycle However measuring the holdup of a super-critical fluidis not
56 Simple cycles Part II
Positionφ [-]
T[
C]
0 02 04 06 08 130
40
50
60
70
80
90T1
T prime2
(a) Gas cooler
Positionφ [-]
T[
C]
0 02 04 06 08 110
15
20
25
30
35
40
T2
T prime2
T4
T prime4
(b) Internal heat exchanger
Figure 47 CO2 case Temperature profile in gas cooler and internal heat ex-changer
easy (one might use some kind of scale but this will be to expensive in most ap-plications) Thus we will consider measurement combinations First we will tryto combine two measurements and if this is not acceptable for all disturbanceswe may try more measurements Any two measurements can be combined andwe choose here to combinePh and T2 The reason is thatPh is normally con-trolled anyway for dynamic reasons andT2 is simple to measure and is promisingfrom the linear analysis Also temperature corrected setpoint for pressure hasbeen proposed before (Kim et al 2004) We use the ldquoexact local methodrdquo (Al-stad and Skogestad 2007) and minimize the 2-norm ofMd = HFWd whereF =partyoptpartdi is the optimal sensitivity ofyprime = [Ph T2] with respect to disturbancesdprime = [TH TC (UA)loss] The magnitude of the disturbances are given inWd Wefind that the linear combinationc= h1 middotPh+h2 middotT2 with k= h2h1 =minus853barC-1
minimizes the 2-norm of the three disturbances on the loss This can be imple-mented in practice by controlling the combined pressure and temperature
Phcombine= Ph +k middot(T2minusT2opt
)(45)
whereT2opt = 255C andk = minus853barC-1 An alternative is to use a morephysically-based combination For an ideal gas we havem = PVmiddotMW
RT and sincethe gas cooler holdupmgco seems to be a good variable to control we will includePT in the gas cooler as a candidate controlled variable
44CO2 case study 57
Table 47 Linear ldquomaximum gainrdquo analysis of controlled variables forCO2 case|∆yopt(di)|
Variable (y) Nom G d1 (TH ) d2 (TC) d3 (UAloss) |∆yopt| n spany |Gprime|
PhT prime2 [barC-1] 032 -0291 0140 -0047 0093 0174 00033 0177 025
Ph [bar] 9761 -7885 483 -155 310 594 10 604 131T prime
2 [C] 355 367 1627 -293 764 1821 1 192 191T prime
2 minusTH [C] 362 24 410 -192 500 675 15 825 291z[-] 034 1 015 -004 018 024 005 029 345Vl [m3] 007 003 -002 0005 -003 0006 0001 0007 477T2 [C] 255 6014 837 090 318 900 1 100 602Phcombine[bar] 9761 -5920 -231 -231 391 330 953 425 139mgco[kg] 483 -1118 0151 -0136 0119 0235 044 0675 1655
Linear method
We first use the linear ldquomaximum gainrdquo method to find promising controlled vari-ables The following disturbanceslowast are considered
d1 ∆TH = plusmn10C
d2 ∆TC = plusmn5C
d3 ∆UAloss from minus100 to +40WC-1
The linear results are summarized in Table47 Some controlled variables (Pl T prime4
∆Tsub and∆Tsup) are not considered because they as discussed earlier can not befixed or are not relevant for this cycle The ratioPhT prime
2 in the gas cooler is notfavourable with a small scaled gain This is probably because the fluid in thegascooler is far from ideal gas soPhT prime
2 is not a good estimate of the holdupmgcoFrom Table47 the most promising controlled variables are the holdup in the gascooler (mgco) and the linear combination (Phcombine) Fixing the valve openingzs
(no control) or the liquid level in the receiver (Vl ) are also quite good
Non-linear analysis
Figure48 shows the compressor power (left) and loss (right) for some selectedcontrolled variables We see that the two most important disturbances are thetem-peraturesTH andTC which gives larger losses than disturbance in the heat loss outof the building Controlling the pressurePh gives infeasible operation for small
lowastIn order to remain in the linear region the optimal variations were computedfor a disturbanceof magnitude 1100 of this and the resulting optimal variations were then multiplied by 100 to get∆yopt(di)
58 Simple cycles Part IIW
s[k
W]
TH [C]25 30 35 400
1
2
3
4
5mdash y = Phmdash y = T2mdash y = z- - y = Vl vap- - y = mgco y = Phcombinemdash Opt
(a) Disturbances inTH (d1)
TH [C]
Loss
[kW
]
25 30 35 400
01
02
03
04
05Ph
Ph
VlT2
z
z mgco
(b) Disturbance inTH (d1)
Ws[k
W]
TC [C]15 20 250
1
2
3
4
5
(c) Disturbance inTC (d2)
TC [C]
Loss
[kW
]
15 20 250
01
02
03
04
05Ph
Vl
(d) Disturbance inTC (d2)
Ws[k
W]
UA[W C-1]240 280 320 360 400 4400
1
2
3
4
5
(e) Disturbance inUAloss (d3)
UA[W C-1]
Loss
[kW
]
240 280 320 360 400 4400
01
02
03
04
05
Ph
(f) Disturbance inUAloss (d4)
Figure 48CO2 case Compressor power (left) and loss (right) for different dis-turbances and controlled variables A line that ends corresponds to infeasible op-eration
44CO2 case study 59
disturbances in the ambient air temperature (TH) The nonlinear results confirmthe linear gain analysis with small losses forPhcombineandmgco
Vl [m3]
Loss
[kW
]
y = Vl
0070075 008
01
02
03
04
05
(a)
z[-]
Loss
[kW
]
y = z
031 033 035 037
01
02
03
04
05
(b)
Phcombine[bar]
Loss
[kW
]
c = Phcombine
90 95 100 105
01
02
03
04
05
(c)
mgco[kg]
Loss
[kW
]
y = mgco
45 5
01
02
03
04
05
(d)
Figure 49CO2 case Loss as function of implementation error
Another important issue is the sensitivity to implementation error From Figure49we see that the sensitivity to implementation error is very large fory = Vl Thethree best controlled variables are constant valve opening (z) constant holdup inthe gas cooler (mgco) and the linear combination (Phcombine)
ConclusionCO2 case study
For thisCO2 refrigeration cycle we find that fixing the holdup in the gas coolermgco gives close to optimal operation However since the fluid is super-critical
60 Simple cycles Part II
holdup is not easily measured Thus in practice the bestsinglemeasurement is aconstant valve openingz (ldquono controlrdquo) A better alternative is to usecombinationsof measurements We obtained the combinationPhcombine= Ph + k middot (T2minusT2opt)using the ldquoexact local methodrdquo This implementation is shown in Figure410 Thedisturbance loss compared with single measurements is significantly reduced andthe sensitivity to implementation error is very small
z
TC
TH
TH
Qloss
Ws
QC
QH
Pl
TC
T1
T3 T4
TsC
k
PC
Ph
T2
Pshcombine
Vl
T prime4
Figure 410 Proposed control structure for theCO2 cycle
45 Discussion
451 Super-heating
An important practical requirement is that the material entering the compressormust be vapour (either saturated or super-heated) Saturation can be achieved byhaving a liquid receiver before the compressor as shown in Figure41 Howeverin many designs the receiver is located at the high pressure side and super-heatingmay be controlled with the choke valve (eg thermostatic expansion valve TEV)as shown in Figure411 A minimum degree of super-heating is required to handledisturbances and measurement errors Since super-heating is not thermodynami-cally efficient (except for some cases with internal heat exchange) this minimaldegree of super-heating becomes an active constraint With the configuration inFigure 411 the ldquoextrardquo valve is the unconstrained degree of freedom (u) that
45 Discussion 61
should be adjusted to achieve optimal operation Otherwise the results from thestudy hold both for the ammonia andCO2 cycle
TC
TCTC (building)
minus
T1T2
T3 T4
TsC
QH
z
Ws
QCQloss
Ph
Pl
ldquoExtrardquovalve
Figure 411 Alternative refrigeration cycle with liquid receiver on high pressureside and control of super-heating
452 Heat transfer coefficients
We have assumed constant heat transfer coefficients in the heat exchangers Nor-mally the heat transfer coefficient will depend on several variables such as phasefraction velocity of the fluid and heat transfer rate However a sensitivity analysis(not included) shows that changing the heat transfer coefficients does not affectthe conclusions in this paper For theCO2 cycle we did some simulations using aconstant air temperature in the gas cooler which may represent a cross flow heatexchanger and is an indirect way of changing the effective UA value We found thatthe losses for a constant liquid level control policy (y = Vl ) was slightly smallerbut the analysis presented here is still valid and the conclusion that a combinationof measurements is necessary to give acceptable performance remains the same
453 Pressure control
This paper has only considered steady-state operation For dynamic reasons inorder to ldquostabilizerdquo the operation a degree of freedom is often used to control onepressure (Pl or Ph) However thesetpointfor the pressure may be used as a degreeof freedom at steady-state so this will not change the results of this study An
62 Bibliography
example of a practical implementation using cascade control is shown in Figure45 where the temperature difference at the condenser outlet is controlled whichwas found to be the best policy for the ammonia case study The load in the cycleis controlled by adjusting the setpoint to the pressure controller that stabilize thelow pressure (Pl )
46 Conclusion
For a simple cycle there is one unconstrained degree of freedom that should beused to optimize the operation For the sub-critical ammonia refrigeration cycleagood policy is to have no sub-cooling Further savings at about 2 are obtainedwith some sub-cooling where a good control strategy is to fix the temperatureapproach at the condenser exit (T2minusTH) see Figure45 One may argue that 2savings is very little for all the effort but larger savings are expected for caseswith smaller heat exchanger areas (Jensen and Skogestad 2007b) and allowingfor sub-cooling shows that there is no fundamental difference with theCO2 case
For the trans-criticalCO2 cycle the only single ldquoself-optimizingrdquo measurementseems to be the holdup in the super-critical gas cooler (mgco) However sincethis holdup is difficult to measure a combination of measurements is needed Wepropose to fix a linear combination of pressure and temperaturePhcombine= Ph +k middot (T2minusT2opt) see Figure410 This is a ldquoself-optimizingrdquo control structure withsmall losses for expected disturbances and implementation errors
Acknowledgment
The contributions of Tore Haug-Warberg and Ingrid Kristine Wold on implement-ing the thermodynamic models are gratefully acknowledged
Bibliography
Alstad V and Skogestad S (2007) lsquoThe null space method for selecting optimalmeasurement combinations as controlled variablesrsquoInd Eng Chem Res
Domanski P A Didion D A and Doyle J P (1994) lsquoEvaluation of suction-lineliquid-line heat exchange in the refrigeration cyclersquoInt J Refrig17 487ndash493
Bibliography 63
Dossat R J (2002)Principles of refrigeration Prentice Hall
Halvorsen I J Skogestad S Morud J C and Alstad V (2003)lsquoOptimal selec-tion of controlled variablesrsquoInd Eng Chem Res42 3273ndash3284
Jensen J B and Skogestad S (2007a) gPROMS and MATLAB model code forammonia andCO2 cycles See additional material for paper at homepage of SSkogestad
Jensen J B and Skogestad S (2007b) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Kim M Pettersen J and Bullard C (2004) lsquoFundamental process and systemdesign issues in CO2 vapor compression systemsrsquoProgress in energy and com-bustion science30 119ndash174
Langley B C (2002)Heat pump technology Prentice Hall
Larsen L Thybo C Stoustrup J and Rasmussen H (2003) Control meth-ods utilizing energy optimizing schemes in refrigeration systemsin lsquoEuropeanControl Conference (ECC) Cambridge UKrsquo
Neksaa P (2002) lsquoCO2 heat pump systemsrsquoInt J Refrigeration25 421ndash427
Skaugen G (2002) Investigation of TranscriticalCO2 Vapour Compression Sys-tems by Simulation and Laboratory Experiments PhD thesis Norwegian Uni-versity of Science and Technology
Skogestad S (2000) lsquoPlantwide control the search for the self-optimizing controlstructurersquoJ Process Contr10(5) 487ndash507
Skogestad S and Postlethwaite I (2005)Multivariable feedback control Secondedn John Wiley amp Sons
Span R and Wagner W (1996) lsquoA new equation of state for carbondioxidecovering the fluid region from the triple-point temperature to 1100 K at pressuresup to 800 MParsquoJ Phys Chem Ref Data25(6) 1509ndash1596
Stoecker W F (1998)Industrial refrigeration handbook McGraw-Hill
Svensson M C (1994) Studies on on-line optimizing control with applicationto a heat pump PhD thesis Norwegian University of Science and TechnologyTrondheim
64 Bibliography
Chapter 5
Problems with specifying∆Tmin indesign of processes with heatexchangers
Accepted for publication in Industrial amp Engineering Chemistry Research
We show in this paper that the common method of specifying∆Tmin
for individual heat exchangers may lead to wrong decisions and shouldbe used with care when designing heat exchanger systems In particu-lar design with constraints on∆Tmin may result in operation conditionswhich are not optimal when the resulting areas are installed In additiondifferentU-values for the heat exchangers are not easily handled Wepropose an alternative method (simplified TAC) to avoid these problemsand compare it with the∆Tmin-method on three vapour compression (re-frigeration) cycle case studies
51 Introduction
In process design one seeks to optimize the future income of the plant Thismight be realized by minimizing the total annualized cost (TAC)JTAC = Joperation+Jcapital[$year-1] see Problem51 below However findingJTAC requires detailedequipment and cost data which is not available at an early design stage
An alternative simple and common approach for design of processes with heat ex-changers especially at an early design stage is to specify the exchanger minimumapproach temperature (EMAT= ∆Tmin) in each heat exchanger see Problem52
65
66 Problems with the minimum approach temperature
below The idea is that this specification should give a reasonable balancebetweenminimizing operating costsJoperation (favored by a small∆Tmin) and minimizingcapital costsJcapital (favored by a large∆Tmin)
As an example Figure51 shows a hot stream (T2) transferring heat to a coldstream with constant temperature (T1) Stream 2 may be hot exhaust gas which iscooled to recover its energy and this is done by vaporizing water in stream 1 Asmall value of∆Tmin means that a lot of the energy is recovered but it requiresa large heat exchanger On the other hand a larger value of∆Tmin requires lessarea but the outlet temperatureT2 will be higher and less energy is recoveredThere exists many rules of thumb for the value of∆Tmin For exampleTurton et al(1998 page 250) recommends 10C for fluids and 5C for refrigerants
0 1 Position[-]
T[
C]
T1 = constant
T2(Small area)
T2(Large area)
∆Tmin(Large area)
∆Tmin(Small area)
Figure 51 The effect of different values for∆Tmin
Note that what we call∆Tmin in this paper is the individual exchanger minimumapproach temperature (EMAT) This should not be confused with the ldquopinch tem-peraturerdquo (heat recovery approach temperature HRAT) used in design of heat ex-changer networks
The use of∆Tmin as a constraint is only used fordesign as it is reasonably wellknown that we should never specify∆Tmin during operation (here the areas shouldbe used as constraints rather than∆Tmin) However even when it comes to designspecifying∆Tmin (and maybe varying it in an outer loop) does not always resultin a good design except for simple cases To understand why specifying ∆Tmin indesign may not be correct consider the following three problems
Problem 51 Detailed optimal designbased on minimizing TAC[$year-1] (egBiegler et al 1997)
minu1
(Joperation+Jcapital
)(51)
51 Introduction 67
subject to model equations and operational constraints (this applies to all problemsin this paper) The degrees of freedomu1 include all the equipment data (sizes)and operating variables The annualized capital cost is often obtained as
Jcapital= sumiisinUnits
(Cfixedi+CvariableimiddotSnii )T (52)
HereSi is the characteristic size for the unit (area in m2 for heat exchangers) andthe cost factors (Cfixedi andCvariablei) and cost scaling factorni are constants foreach unit (eg heat exchangers)T is the capital depreciation time egT =10years The operating costJoperationare given by the prices of feeds products andutilities (energy) plus other fixed and variable operating costs eg see Equation58below
Problem 52 Simplified optimal designwith specified∆Tmin
minu2
(Joperation) (53)
subject to ∆Ti minus∆Tmin ge 0
Here the degrees of freedomu2 include the heat transfer in the heat exchanger(Qi) and the operating variables (flows works splits etc) After solving this prob-lem one can calculate the heat exchanger areasAi from the resulting temperatures(usingQi =
intUi∆Ti dAi) Note that Problem52will favor designs where the tem-
perature difference∆T is close to∆Tmin throughout the heat exchangers becausethis improves energy efficiency but does not cost anything Specifying∆Tmin willtherefore tend to give designs with large heat exchanger areas In addition dif-ferentU-values can not be handled easily as they are not part of the optimizationproblem in Equation53 An indirect approach is too use different∆Tmin i for eachheat exchanger in Equation53
Let us now consider steady-state operation where the equipment data includingheat exchanger areas are given and the degrees of freedomu3 include only theoperating variables For each heat exchanger there is at steady-state one operat-ing variable which may be chosen as theeffectiveareaAi (in practiceAi may bechanged using a bypass) However we must requireAi le Amax
i whereAmaxi is the
ldquoinstalled areardquo eg found from Problem51or 52 We then have
Problem 53 Optimal operationwith given heat exchanger areas
minu3
(Joperation) (54)
subject to Ai minusAmaxi le 0
68 Problems with the minimum approach temperature
Note that in many cases including the examples in this paper it is optimal to haveAi = Amaxi
The solution to Problem53in terms of optimal stream data (temperatures) will bethe same as to Problem51 butnot generally the same as to Problem52 see themotivating example below To understand this note that in Problem53 with theareas given there is no particular incentive to make the temperature difference∆Tldquoevenrdquo (due to∆Tmin) throughout the heat exchangers Provided there are degreesof freedom we will therefore find that∆T from Problem53varies more throughthe heat exchangers than∆T from Problem52 In particular the∆Tmin obtainedfrom Problem53 is often smaller than that specified in design (Problem52)see the introductory example below Thus the optimal nominal operating point(solution to Problem53) is not the same as the nominal simplified design point(solution to Problem52) From this it is clear that specifying∆Tmin in design isnot a good approach
The objective of this paper is to study the∆Tmin-method (Problem52) in moredetail and suggest an alternative simple design method (called the simplified TAC-method) for heat exchanger systems The optimization problems are solved usingthe gPROMS software
TCTC (building)
TH (ambient)
TH (ambient)
T1T2
T3
T4Ts
C
QH
(condenser)
z(chokevalve)
(compressor)Ws
QC (evaporator)
Qloss
Ph
Pl
Figure 52 An ammonia refrigeration system
52 Motivating example Ammonia refrigeration cycle
The ammonia refrigeration cycle for cold storage presented inJensen and Skoges-tad(2007a) is shown in Figure52 We use the following conditions
52 Motivating example Ammonia refrigeration cycle 69
Table 51 Structure of model equations for the ammonia case studyHeat exchangers (condenser and evaporator)Q =
intU∆T dA = mmiddot (houtminushin)
P = Psat(Tsat)m= ρVValvem= zmiddotCV
radic∆Pmiddotρ hout = hin
CompressorWs = m(houtminushin) = mmiddot (hsminushin)η
bull Qloss= 20kW
bull Ambient temperatureTH = 25C
bull Cold storage (indoor) temperature set pointTsC = minus12C
bull Heat transfer coefficient for the evaporator and condenserU = 500Wm-2 C-1
bull The ammonia leaving the evaporator is saturated vapour (which is alwaysoptimal for this cycleJensen and Skogestad(2007b))
The temperature controller is assumed to adjust the compressor power to maintainTC = Ts
C which indirectly sets the loadQC = Qloss The main model equations aregiven in Table51
521 ∆Tmin design-method
The operational cost is given by the compressor power (Joperation=Ws) so with the∆Tmin-method the optimal design problem see Problem52 becomes
minu2
(Ws)
subject to ∆Tvapminus∆Tmin vapge 0 (55)
∆Tconminus∆Tminconge 0
where the degrees of freedomu2 include the heat transfered in the two heat ex-changers (Qi) (but note thatQC = 20kW) plus three other operating variableslowast
(eg two pressures and refrigerant flow) We choose∆Tmin = 10C in both the
lowastThe simple cycle in Figure52has five operating variables (Jensen and Skogestad 2007b) buttwo of these have been specified (given load and saturated vapour)
70 Problems with the minimum approach temperature
evaporator and the condenser (Shelton and Grossmann 1986) The resulting heatexchanger areasA are then obtained fromQ =
int(U middot∆T)dA
As noted in the introduction the solution to Equation55 does not generally giveoptimal operation with the resulting areas The re-optimized operation problemwith given areas see Problem53 becomes
minu3
(Ws)
subject to AvapminusAmaxvap le 0 (56)
AconminusAmaxcon le 0
whereAmaxvap andAmax
con are the result of the∆Tmin-method design problem (Equation55) The degrees of freedomu3 include Avap and Acon (which in this case areoptimally equal to their maximum design values) plus the three other operatingvariables (eg two pressures and refrigerant flow)
The results for the two problems are summarized in the two left columns of Table52 and we note that re-optimization reduces the operating cost (Ws) by 32Figure53shows the corresponding temperature profiles in the condenser
Positionφ [-]
T[
C]
0 02 04 06 08 130405060708090
100110120
(a) Optimal design with specified∆Tmin
Positionφ [-]
T[
C]
0 02 04 06 08 130405060708090
100110120
(b) Re-optimized operation with specifiedA
Figure 53 Temperature profile in the condenser for the∆Tmin-method
bull In the design case (55) (with fixed∆Tmin) there is no sub-cooling of ammo-nia in the condenser In the re-optimized operation (56) however there is asub-cooling of 89C The optimality of sub-cooling in simple refrigerationcycles is discussed in detail inJensen and Skogestad(2007b)
bull The high pressurePh is increased by 07 in the re-optimized case but thisis more than compensated for by a 37 reduction in flowrate
53 Proposed simplified TAC method 71
Table 52 Ammonia case study∆Tmin-method Simplified TAC (Eq59)
Design (55) Operation (56) C0 = 818lowast C0 = 8250dagger
∆Tmin = 10C Re-optimized∆Tvap
min [C] 100 100 913 267∆Tcon
min [C] 100 153 184 100Acon[m2] 450 450 412 147Avap[m2] 400 400 438 150Atot [m2] 850 850 850 297HX cost[-] 100 100 100 051Pl [bar] 174 174 181 077Ph [bar] 135 136 140 284∆Tsub[
C] 00 89 96 285m[mols-1] 107 103 103 110Ws[W] 6019 5824 5803 12479COP[-] 332 343 345 160
Given data is shown in boldfacelowastC0 adjusted to get same total heat exchanger area as the∆Tmin-methoddaggerC0 adjusted to get same∆Tmin as used in the∆Tmin-method
In summary we find that the∆Tmin design-method does not result in optimal oper-ating variables for the given example A better design method is therefore needed
53 Proposed simplified TAC method
The original cost function for Problem51 requires quite detailed cost data plusa lot of other information which is not available at an early design stage Thesimplified Problem52on the other hand is easy to formulate and solve but herewe cannot easily handle differentU-values for heat exchangers and the optimaldesign point is not generally the same as the optimal operating point (even nom-inally) Therefore the objective is to find a better simplified formulation Thestarting point is to replace the equipment cost (Equation52) in Problem51witha simplified expression First we assume that the structure of the design is givensuch that we need not consider the fixed cost terms (ie we setCfixedi = 0) Sec-ond we only consider heat exchanger costs For a vapour compression cycle thisis justified if the capital cost for the compressor is proportional to the compressorpowerWs which corresponds to assumingni = 1 for the cost exponent for the com-pressor We can then include the capital cost for the compressor in the operatingcost of the compressor Third we assume that all heat exchangers have the same
72 Problems with the minimum approach temperature
cost factors (Cvariablei= C0 andni = n)
The resulting ldquosimplified TACrdquo optimal design problem becomes
Problem 54 Simplified optimal designwith simplified TAC
minu4
(
Joperation+C0 middotsumi
Ani
)
(57)
subject to model equations and operating constraints and whereu4 includes theheat exchanger areasAi plus the operating variables In the general case
Joperation= sum pFi Fi minussum pPj Pj +sum pQkQk +sum pWslWsl [$year-1] (58)
whereFi are feedsPj are productsQk are utilities (energy)Wsl are the mechan-ical work the prsquos are respective prices For a heat exchanger network problemthis can be reduced toJoperation= sum pQi Qi [$year-1] which in many cases simpli-fies to Joperation= QH [$year-1] whereQH is the supplied heat (Gundersen andNaess 1988) For the refrigeration cycles considered in this paperJoperation=Ws[$year-1]
In the examples we choosen = 065 for the heat exchangers and useC0 as thesingle adjustable parameter (to replace∆Tmin) There are several benefits comparedwith the∆Tmin-method
bull The heat exchanger temperatures depend onUiAi for each heat exchangerThus differentU-values for each exchanger are easily included
bull The optimal design (Equation57) and the optimal operation (Problem53)have the same solution in terms of optimal stream data This follows sincethe termC0 sumi A
ni is constant in operation
bull The assumption of using the sameC0 for all heat exchangers is generally amuch better than assuming the same∆Tmin
On the other hand compared to the∆Tmin design method the simplified TAC de-sign method requires calculation of∆T inside all exchangers during the optimiza-tion and the optimization problem is also a bit more difficult to solve Howeverthe proposed method does not require any additional data compared with the∆Tmin
method
54 Other case studies 73
531 Revisit of ammonia case study
The optimization problem (Equation57) for the proposed simplified TAC-methodbecomeslowast
min(Ws+C0 middot (Ancon+An
vap)) (59)
The right two columns of Table52shows the optimal design withn= 065 and twodifferent values ofC0 FirstC0 = 818 gives the same total heat exchanger area andalmost the same capital cost as the∆Tmin-method but the area is better distributedbetween the evaporator and condenser This results in a 360 reduction in op-erating cost (Ws) compared with the∆Tmin-method (036 after re-optimizing theoperating point for the∆Tmin-method) SecondC0 = 8250 gives∆Tmin = 100Cas specified in the∆Tmin method The compressor work is increased with 107(114) but the heat exchanger area is reduced by 60 and this is theonly designthat truly satisfies the∆Tmin we selected initially
The simplified TAC method confirms that sub-cooling is optimal and we see thatthe degree of sub-cooling increases with decreasing heat transfer area (increasedC0)
Note that the heat transfer coefficientsUi were assumed to be equal but the sim-plified TAC method will automatically distribute the heat transfer area optimallyalso if the heat exchangers have different heat transfer coefficients For examplewith Uvap = 2Ucon the energy savings (for the same heat exchangers cost) are evenlarger (6) using the simplified TAC method compared with the∆Tmin-method
54 Other case studies
We here briefly present results from two other case studies
541 CO2 air-conditioner
CO2 as a working fluid in air-conditioners and heat-pumps is gaining increasedpopularity because of its low environmental impact (Lorentzen 1995 Neksaa2002) We consider a trans-criticalCO2 air-condition unit with the following data
lowastIn a more realistic design one may also consider additional constraints such as maximumcompressor suction volumes and pressure ratio but this is not discussed here
74 Problems with the minimum approach temperature
Table 53CO2 air-conditioner∆Tmin-method Simplified TAC
Design Operation C0 = 253lowast C0 = 185dagger C0 = 877Dagger
∆Tmin = 5C Re-optimized∆Tgco
min [C] 500 356 241 207 500∆Tvap
min [C] 500 500 578 501 115∆T ihx
min [C] 500 475 - - -Agco[m2] 131 131 176 202 092Avap[m2] 160 160 138 160 070Aihx [m2] 023 023 0 0 0Atot [m2] 314 314 314 362 161HX cost[-] 100 100 091 100 059Ph [bar] 878 916 928 910 1070Pl [bar] 508 508 499 508 433m[mols-1] 065 059 069 070 067Ws[W] 892 859 871 814 1328COP[-] 449 465 459 492 301
Given data is shown in boldfacelowastC0 adjusted to get same total area as∆Tmin-methoddaggerC0 adjusted to get same heat exchanger cost as the∆Tmin-methodDaggerC0 adjusted to get same∆Tmin as used in the∆Tmin-method
bull Heat transfer coefficientU = 500Wm-2K-1 for the evaporator condenserand internal heat exchanger
bull Ambient temperatureTH = 30C
bull Set point for room temperatureTC = 20C
bull Heat loss into the roomQloss= 40kW
The details about the model are found inJensen and Skogestad(2007a) In the op-timization we have included an internal heat exchanger (with areaAihx) that trans-fers heat from before the compressor to before the valve Otherwise the flowsheetis as for the ammonia cycle shown in Figure52
For solving Problem52 we use a design∆Tmin = 50C in all heat exchangersAgain we find that re-optimizing for operation (Problem53) gives a better oper-ating point with 370 less compressor power The results given in Table53aresimilar to the ammonia cooling cycle although there is no sub-cooling sincePh isabove the critical pressure
Interestingly with the simplified TAC method we obtainAihx = 00m2 whichmeans that it is not optimal from an economical point of view to pay for the area for
54 Other case studies 75
the internal heat exchanger (although the internal heat exchanger would of coursebe used if it were available free of charge) This is a bit surprising sincewe havenot included the fixed cost of installing a heat exchanger which would make iteven less desirable to invest in an internal heat exchanger On the otherhand ifwe require a lot of super-heating before the compressor then it might be better toachieve this super-heating in an internal heat exchanger but this is notdiscussedhere
With C0 = 253 we get the same total heat transfer area as for the∆Tmin-method butthe shaft work is reduced by 426 (058 compared to re-optimized)C0 = 185gives the same cost of heat exchanger area (without even considering the savingsof completely removing a heat exchanger) andWs is reduced by 1222 (885)With C0 = 877 we get the only design with∆Tmin = 50C The heat exchangercost is reduced by 41 and the compressor power is increased by 49 (55)compared with the∆Tmin-method
542 PRICO LNG process
The PRICO LNG process (Price and Mortko 1996) is a simple configuration uti-lizing mixed refrigerants Details about the model is presented elsewhere (Jensen2008) Note that we are not considering constraints on compressor suction volumeand pressure ratio for the compressor This will be important in an actual designbut we have tried to keep the case study simple to illustrate the effect of specifying∆Tmin
A design∆Tmin of 20C is used for the∆Tmin method From Table54 we seethat re-optimizing reduces the energy usage (Ws) by 48 This is achieved byincreasing the pressure ratio (by 255) and reducing the refrigerant flowrate (by167) The composition of the refrigerant is also slightly changed but this isnot shown in Table54 We were quite surprised by the rather large improvementobtained by re-optimizing with fixed heat transfer areas considering the relativelylow value for the initial∆Tmin
With the simplified TAC method we get a 41 reduction (02 increase com-pared to re-optimized) inWs for the same total heat transfer area (C0 = 2135) Thesmall increase inWs compared with the re-optimized∆Tmin design is because thesimplified TAC method minimizes the heat exchanger cost and not the total areaWith the same cost (C0 = 2090) the TAC-method gives a reduction in compressorpower of 43 (01) The saving compared with the re-optimized case is smallbecause of the small∆Tmin resulting in very large heat exchangers A more rea-sonable design is achieved withC0 = 7350 which gives a design with a true∆Tmin
76 Problems with the minimum approach temperature
Table 54 PRICO LNG process∆Tmin-method Simplified TAC
Design Operation C0 = C0 = C0 =∆Tmin = 2C Re-optimized 2135lowast 2090dagger 7350Dagger
∆TminHOT [C] 200 089 090 086 200∆TminNG[C] 200 098 109 108 222AHOT middot10minus3 [m2] 982 982 1012 1027 431ANG middot10minus3 [m2] 299 299 269 272 145ATot middot10minus3 [m2] 1281 1281 1281 1299 577HX cost[-] 100 100 099 100 060Ph [bar] 201 271 270 268 378Pl [bar] 27 29 29 29 191m[kmols-1] 30 25 25 25 23Ws[MW] 1894 1814 1817 1812 2216
Given data is shown in boldfacelowastC0 adjusted to get same total area as∆Tmin-methoddaggerC0 adjusted to get same heat exchanger cost as the∆Tmin-methodDaggerC0adjusted to get same∆Tmin as used in the∆Tmin-method
of 20C The heat exchanger capital cost is reduced by 40 but the compressorpower is increased by 170 (222)
55 Discussion
551 ∆Tmin-method
There are some main points that are important to note from this analysis of the∆Tmin-method
1 ∆Tmin is treated as an important parameter in heat exchanger design butthe theoretical basis seems weak as ldquoviolatingrdquo∆Tmin in operation may givelower operating cost
2 The∆Tmin-method will not always give the optimal operating point so sub-optimal setpoints might be implemented
3 The size distribution between the heat exchanger will not be optimal al-though this may be partly corrected for by individually adjusting the valueof ∆Tmin for each heat exchanger
4 More seriously the results might lead to wrong structural decisions andthis
55 Discussion 77
can not be changed by iterating on the∆Tmin-values In the ammonia casestudy one would incorrectly conclude that sub-cooling is not optimal andthus implement a liquid receiver after the condenser This would duringoperation achieve no sub-cooling From the true optimum however we seethat some sub-cooling is optimal
5 One potential advantage with the∆Tmin-method is that it only requires anoverall energy balance for the heat exchangers However for more com-plex cases a more detailed model of the heat exchangers is needed so in thegeneral cases this advantage is lost
In summary the∆Tmin-method is not satisfactory for realistic design problems
552 Other approaches
The question is whether there are other ways of specifying temperature differencesIn terms of area rather than specifying∆Tmin it would be better to specify themean temperature difference∆T = 1
A
int∆TdA since∆T is directly linked to the
heat transfer area by (assuming constant heat transfer coefficient)
Q = UA∆T (510)
However this method has several drawbacks that makes it unattractive touseFirst it might be cumbersome to calculate the integral of the temperature and sec-ond and more importantly there are no general rules in selecting values for∆TWe therefore propose to use the simplified TAC-method
553 Heat exchanger network design
The results in this paper show that one should notuse a constraint on∆Tmin (EMAT)for the design of individual heat exchangers What about the standard heat ex-changer network (HEN) design problem (egGundersen and Naess 1988) wherea constraint on the heat recovery approach temperature∆Tmin (HRAT) for the net-work is used Our results do notinvalidate this approach First the stream data(inlet and outlet temperatures and flows) are fixed for the standard HEN prob-lem It then follows that specifying∆Tmin (HRAT) is equivalent to specifying theheat recovery or equivalently the required hot utility (QH) The solution to thisparticular design problem will therefore result in optimal operating data if wein-stall the resulting areas (and remove the∆Tmin specification) However note thatthe simplified TAC-formulation in Equation57with Joperation= sumi pQi Qi is much
78 Bibliography
more general than the standard heat exchanger network design problem where thestream data are specified
56 Conclusion
We have shown that the method of specifying∆Tmin for design of heat exchangers(min J subject to∆T ge ∆Tmin) may fail to give an optimal operating point In theammonia refrigeration case study the∆Tmin-method fails to find that sub-coolingin the condenser is optimal As a simple alternative we propose the simplified totalannualized cost (TAC) method (min(J+C0 sumi A
ni )) whereC0 replaces∆Tmin as the
adjustable parameter A high value ofC0 corresponds to increasing the investment(capital) costs relative to the operating (energy) costs and favors small areas and alarger∆Tmin ThusC0 can be adjusted to get a desired value for∆Tmin or the totalarea or it can be obtained from cost data With the alternative method differentheat transfer coefficientsUi can also be accounted for
Another important conclusion is related to the temperature difference profilein theheat exchanger According to exergy or entropy minimization rules of thumb(egSauer et al 1996) it is optimal to have even driving forces which suggests that∆Tshould be constant in heat exchangers The results presented here however suggestthat this is not true The∆Tmin approach (Problem52) favors a more constant∆Tprofile (see Figure53(a)) but in optimal operation (Problem53) we find that thetemperature difference is small in one end (see Figure53(b))
Bibliography
Biegler L T Grossmann I E and Westerberg A W (1997)Systematic Methodsof Chemical Process Design Prentice Hall
Gundersen T and Naess L (1988) lsquoThe synthesis of cost optimalheat exchang-ers networksrsquoComput Chem Eng12(6) 503ndash530
Jensen J B (2008) Optimal operation of cyclic processes With application toLNG processes PhD thesis Norwegian University of Science and Technology
Jensen J B and Skogestad S (2007a) lsquoOptimal operation of a simple refrig-eration cycles Part II Selection of controlled variablesrsquoComput Chem Eng31 1590ndash1601
Bibliography 79
Jensen J B and Skogestad S (2007b) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Lorentzen G (1995) lsquoThe use of natural refrigerants a complete solution to theCFCHCFC predicamentrsquoInt Journal of Refrigeration18 190ndash197
Neksaa P (2002) lsquoCO2 heat pump systemsrsquoInt J Refrigeration25 421ndash427
Price B C and Mortko R A (1996) PRICO - a simple flexible provenapproachto natural gas liquefactionin lsquoGASTECH LNG Natural Gas LPG interna-tional conference Viennarsquo
Sauer E Ratkje S K and Lien K M (1996) lsquoEquipartition of forces A newprinciple for process design and optimizationrsquoInd Eng Chem Res35 4147ndash4153
Shelton M R and Grossmann I E (1986) lsquoOptimal synthesis of integrated re-frigeration systemsmdashi Mixed-integer programming modelrsquoComput ChemEng10(5) 445ndash459
Turton R Bailie R C Whiting W B and Shaeiwitz J A (1998)Analysissynthesis and design of chemical processes Pearson Education Ltd
80 Bibliography
Chapter 6
Optimal design of a simple LNGprocess
In this chapter we discuss the design optimization of a single cyclemixed fluid LNG process the PRICO process A simple objective func-tion is stated and used on nine different cases with varying constraintsImportant constraints are discussed and the results are compared withthe commercial PRICO process and with other publications
61 Introduction
Large amounts of natural gas are found at locations that makes it infeasible or noteconomical to transport it in gaseous state to the customers The most economicway of transporting natural gas over long distances is to first produce liquefiednatural gas (LNG) and then transport the LNG by ships LNG has approximately600 times the density of gaseous natural gas
At atmospheric pressure LNG has at saturated conditions a temperature of ap-proximatelyminus162C so the process requires large amounts of energy to generatethe necessary cooling Several different process designs are used and they can begrouped roughly as follows
1 Pure fluid cascade process The refrigerants are pure fluids but several cy-cles are used to improve the efficiency
2 Mixed fluid refrigerant The refrigerant composition is adjusted to matchthe cooling curve of the natural gas
81
82 Optimal design of a simple LNG process
3 Mixed fluid cascade process Combination of the two where energy effi-ciency is further improved by using several mixed refrigerant cycles
The PRICO LNG process considered in this paper see Figure61 belongs to thesecond group and has a single cycle with a mixed refrigerant The mixed refrig-erant gives a good temperature match throughout the heat exchanger comparedwith using pure component cycles (group 1) However compared with multiplemixed refrigerant processes the thermodynamic efficiency is lower so theprocessis mainly used for smaller plants (eg peak shaving plants) up to 2MTPA (milliontons per annum)
Stebbing and OrsquoBrien(1975) reports on the performance of the first commercialPRICO plants in operationPrice and Mortko(1996) from the Black amp Veatchcompany discuss the process and give some key values for several oftheir plantsWith respect to academic workLee et al(2002) used the PRICO process as oneof their case studies for testing their approach to design optimization The samegroup later published some updated results (Del Nogal et al 2005)
LNG Flash gas
Fuel gasPre-treatedNG SW
Fuel HXNG HX
Refrigerantcompressor
Fuelcompressor
Condenser Ph
Pl
PmTout
Figure 61 Simplified flowsheet of the PRICO process also showing flash gasheating and recompression (not considered here)
61 Introduction 83
611 Optimal design
Mathematically optimal design may be expressed as the solution to the followingoptimization problem
min JTAC = Joperation+Jcapital (61)
subject to cle 0
whereJcapital[$year-1] is the annualized cost of the equipment andJoperation[$year-1]is the annual operating cost (Joperation= Jutility +Jfeeds+Jproducts) The total annual-ized cost (TAC) is minimized with respect to the design variables which includesboth the structure (integer variables) and the design parametersc is a set of designconstraints (ie maximum design pressure non-negative flows and feedcomposi-tion)
The truly optimal design is a complex task involving mixed integer non-linearprogramming In our case however we start from a given process structure so wemay use a simpler approach with no integer variables
For heat exchanger design a common approach is to minimize only the operatingcost but subject to a minimum approach temperature in all heat exchangers(Leeet al 2002 Del Nogal et al 2005) that is∆Ti ge ∆Tmin is introduced as a designconstraint The idea is that the heat exchanger minimum approach temperaturelowast∆Tmin gives a balance between low operating cost (favored by low∆Tmin) and lowcapital cost (favored by high∆Tmin) This approach will however not generallyresult in a true optimal design even when iterating on∆Tmin Specifically re-optimizing to find the optimal operation given the resulting equipment will usuallyresult in a lower optimal value of∆Tmin (Chapter5) This also implies that a designapproach based on specifying∆Tmin and altering it in an outer loop will not work
Therefore we choose to use the simplified total annual cost (sTAC) methodpre-sented in Chapter5 The capital cost is often expressed as a fixed and a variablecontributionJcapital= sumi (Cfixedi+CvariableimiddotS
nii )T whereSi is the characteristic
size for the unit (area in m2 for heat exchangers) The cost factors (Cfixedi andCvariablei) and the scaling factorni are assumed constants for each unitT is thecapital depreciation time egT = 10years First if the structure of the network isfixed the size independent parameter does not matter (so we may useCfixedi = 0)Second we here consider only the compressor and heat exchanger costs (mainequipment) Third we assume that the exponentn = 1 for the compressors Wecan then add the operation and capital cost for the compressor into a singleterm
lowast∆Tmin in this paper refers to the individual exchanger approach temperature(ERAT) and notthe heat recovery approach temperature (HRAT) or ldquopinchrdquo temperature of the entire network
84 Optimal design of a simple LNG process
see Equation62 below Fourth we assume that the size dependent parameterand the exponent are equal for all heat exchanger (Cvariablei= C0 andni = n) Wechoose to fixn= 065 for the heat exchanger areas and useC0 as a single adjustableparameter
The resulting ldquosimplified TACrdquo design problem with the cost factorsC0 andk asparameters becomes
min
k middotWs+ pfeedmiddot mfeedminus pLNG middot mLNG minus pfuel middot mfuel︸ ︷︷ ︸
Jflows
+C0 middotsumi
Ani
[$year]
(62)
This minimization is subject to some design constraints see Section612 Ai isthe heat transfer area for the two hot streams in the main heat exchanger (naturalgas and warm refrigerant) and for the sea water heat exchanger Note thatk is notonly the price of the compressor energy as it also includes the annualized capitalcost for the compressor One may viewk[$MWyear] as the price of ldquorentingrdquo acompressor
With no capacity limits this problem (Equation62) is actually ill-posed as it isoptimal to have infinite feed (capacity) In practice one either designs fora givenfeed or imposes constraints that limit the capacity In this paper we assume thatwe wish to find the optimal design with a given maximum compressor shaft powerWmax
s The factork then does not matter as the termkWs is fixed We also donot include the condenser in the design and instead specify the outlet temperatureof the condenser A mass balance gives ˙mfuel = mfeedminus mLNG and we may writeJflows = mfeed(pfeedminus pfuel)minusmLNG (pLNG minus pfuel) =minusmLNG (pLNG minus pfuel) wherewe have assumedpfeed= pfuel This is reasonable since feed may be used as a fuelSinceWs is constrained the optimal design problem (Equation62) simplifies to
min(minusmLNG +C0
(A065
HOT +A065NG
))(63)
subject to Ws leWmaxs
mfuel le mmaxfuel
cle 0
The minimization is with respect to design parameters (AHOT andANG) operatingparameters (flows pressures splits etc) We may adjust the parameterC0 in Equa-tion 63to obtain a reasonable∆Tmin This will unlike the ldquo∆Tmin-methodrdquo (wherewe maximizemLNG with ∆Ti ge ∆Tmin as constraints) give the same operating pointif one re-optimizes the operation with the areas fixed at the optimal design valuesThe setc is constraints that may vary for the different cases
61 Introduction 85
612 Design constraints
In design it is necessary to impose some constraints for the optimization to assurethat a feasible solution is found
Pressure There are typically constraints on pressures in the process These con-straints may be related to equipment manufacturing limits For example alarge conventional centrifugal compressor has a maximum pressure in theorder of 30 to 40bar while a vertical split centrifugal compressor may haveoutlet pressure up to about 80bar (General Electric Oil and Gas January2007) Constraints may also arise from the fact that higher pressures are notdesirable from a safety point of view (eg explosion danger) One may alsoconstrain certain pressures to simplify the objective function by removingthe pressure dependence (higher pressure requires more expensive pipingheat exchangers etc)
Temperatures There may be constraints on temperature gradients (eg in heatexchangers) or on absolute temperatures Both of these may be related tomechanical issues such as stressThis has not been considered here
Compressor suction volume (Vsuc) In operationthe compressor suction volumeand compressor head are linked to the speed of rotation through the charac-teristic curve In designone may have to limit the compressor size to keepit within physical limits (eg by limiting the compressor suction volume)The current maximum limit for a single flow centrifugal compressor seemsto be 380000m3h-1 (General Electric Oil and Gas January 2007)
Compressor head A simple correlation for the maximum head (or specific en-thalpy rise) per compressor wheel is for a centrifugal compressor (see Equa-tion 173 on page 37 inLudtke 2004)
Head= ∆h = smiddotu2 [kJkg-1] (64)
wheresasymp 057minus066 is the work input factor andu[ms-1] is the velocity atthe wheel tip Adding the numbers for each wheel gives the total head forthe compressor For example one compressor wheel with 17m diameterand a rotational speedω = 3600RPM gives the following head
u =3600min-1
60smin-1middotπ middot17m= 320ms-1 (65)
∆h = (320ms-1)2 middotsasymp 58minus68kJkg-1 (66)
Table61gives the head for different rotational speeds and wheel diameters(s= 057) For large compressors the number of wheels is typically six orless (up to eight for the barrel type)
86 Optimal design of a simple LNG process
Table 61 Maximum head∆h[kJkg-1] per compressor wheel for different wheeldiameters and rotational speeds (Equation64 with s = 057) The last columnshows the total head for a compressor with one of each of the wheels in the table
Wheel diameterRotational speed 17m 16m 15m 14m 13m 12m Sum 6 wheels3600RPM 596 528 464 404 348 297 26363000RPM 414 366 322 280 242 206 1830
Compressor pressure ratio (Pr) The maximum compressor pressure ratio is analternative simple way of limiting the compressor head (egPrice andMortko (1996) report the value of 55 for their compressor) but this ap-proach is less exact since the head also depends on the mass flowrate Theuse of Equation64 is therefore preferred
Compressor shaft work (Wmaxs ) A maximum value for the compressor shaft work
may be imposed for example due to physical limitations in the driver to thecompressor or the compressor itself or in the available power supply
Mach number The gas velocity must be below the sonic velocity (Mach numberless than one) High Mach numbers are typically a problem at low pressuresso this is an issue at the compressor inlet However the velocity is onlyknown at the compressor outlet (the tip speed) so it is common to reportthe machine Mach number as the tip speed over the sonic velocity at theinlet The machine Mach number may be higher than one (as high as 125according toLudtke(2004)) without having an actual Mach number higherthan oneWe have not considered this constraint
LNG outlet conditions There may be constraints on the LNG heating value andcompositionThis is not considered here We have however assumed stor-age as saturated liquid atP = 11bar
Fuel specifications There may be constraints on the fuel heating value and com-positionThis is not considered here We have limited the maximum amountof fuel to a value somewhat larger than the energy needed by the refrigerantcompressor (by combustion in a gas turbine)
62 Process description
Figure61shows a simplified flowsheet of the PRICO processNatural gasis fedto the main heat exchanger (NG HX) after some pretreatment (removal of water
62 Process description 87
CO2 etc) which is not included in this paper The natural gas is cooled liquefiedand sub-cooled by heat exchange in NG HX with the cold refrigerant To furtherlower the temperature the sub-cooled liquid is then expanded and the liquid isseparated from the vapour (flash gas) and sent to storage as LNG The vapour(flash gas) is heated re-compressed and used as fuel but this part of the plant hasnot been included in this work Instead the flash gas is considered as a productwith a given fuel prize The further simplified process considered in this work isshown in Figure62
LNG Flash gas
Pre-treated NG SW
NG HX
Refrigerantcompressor
Condenser
Ph
Pl
PmTout
Figure 62 Further simplified flowsheet of the PRICO process used in this work
The refrigerant is partially condensed in the sea water (SW) cooler (condenser)The main part (all in our case see Figure62) of the refrigerant is fed to the NGHX and is cooled together with the natural gas stream The refrigerant is a sub-cooled liquid at the outlet of NG HX and is expanded to the low pressure (Pl ) Theresulting two-phase mixture provides the cooling in NG HX by vaporization Theoutlet from the heat exchanger (NG HX) is slightly super-heated partly toavoiddamage to the compressor We have assumed∆Tsup= 10C in most cases
We have made some simplifications compared to the process described inPrice andMortko (1996) First we do not consider removal of heavy components from thenatural gas feed This removal may be done upstream or integrated in the refriger-ation process (as shown byPrice and Mortko(1996)) A more detailed discussion
88 Optimal design of a simple LNG process
on this is given belowSecond we have not included the refrigerant separator andrefrigerant pump indicated byPrice and Mortko(1996) This however will notchange the thermodynamics if the pump work is neglectedThird as already men-tioned we have not considered the flash gas heating (Fuel HX) and compressionThe resulting flowsheet is shown in Figure62
Some of the data reported byPrice and Mortko(1996) which we have used asdesign constraints is shown in Table62 In addition we use the following
bull Composition of natural gas (mole-) 897 methane 55 ethane 18propane 01 n-butane and 28 nitrogen Note that this composition ismore methane rich than that reported inPrice and Mortko(1996) (897versus 832) The mole weight isMW = 00176kgmolminus1
bull The temperature after the refrigerant condenser and the temperature in thenatural gas feed are both 30C Price and Mortko(1996) report 29C and32C respectively
bull Pressure drops
ndash 5bar on natural gas side in main heat exchanger
ndash 01bar in SW cooler
ndash 4bar for hot refrigerant in main heat exchanger
ndash 1bar for cold refrigerant in main heat exchanger
ndash 03bar from main heat exchanger to compressor
bull Constant heat transfer coefficients
bull The refrigerant is a mix of nitrogen (N2) methane (C1) ethane (C2) propane(C3) and n-butane (nC4) and the composition is found by optimizationPriceand Mortko(1996) use hydrocarbons ranging fromC1 to iC5
bull The SRK equation of state is used for the thermodynamic calculations
63 Results for optimal design
The numerical results from the optimization for nine cases are reported in Table63 Boldface numbers indicate specifications or active contraints We have ad-justedC0 in Equation63to obtain∆Tmin asymp 20C in the main heat exchanger (NGHX) for all cases We have assumed 10C super-heating at the compressor inlet
63 Results for optimal design 89
Table 62 Design constraints based on data fromPrice and Mortko(1996)Pfeed[bar] 40Ws[MW] (max) 775Vsuc[m3s-1] (max) 88lowast
Ph [bar] (max) 22Pr [-] (max) 55
lowastThis was reported as 317m3s-1 but this is probably a misprint and should be 317000m3 h-1
(∆Tsup= 10C) except in Cases63and64 Note from the results that it is optimalin all cases to have noC3H8 in the refrigerant
Table 63 Optimal design results for nine different casesCase 61 62 63 64 65 66 67 68 69
mfeed[kgs-1] 522 450 453 448 496 496 514 761 808mLNG [kgs-1] 446 417 420 415 463 463 481 711 758mfuel [kgs-1] 77 333 333 333 333 333 333 50 50
mREF [kgs-1] 478 475 472 443 251 298 320 611 617Tout[
C] -144 -156 -156 -156 -157 -157 -157 -157 -156∆Tsup[
C] 100 100 116 257 100 100 100 100 100∆Tmin [C] 196 197 195 203 197 194 194 200 204
η [] 828 828 828 828 828 828 828 828 828Ws[MW] 775 775 775 775 775 775 775 120 120
Ph [bar] 220 220 220 220 504 300 370 300 300Pressure ratio[-] 55 55 55 55 225 166 117 73 72
Pl [bar] 40 40 40 40 224 181 317 411 416Head[kJkg-1] 134 135 136 145 256 216 200 162 161
Vsuc[m3s-1] 843 833 840 839 751 106 70 106 106UAHOT [MW C-1] 384 409 413 398 187 229 268 518 522UANG [MW C-1] 48 44 44 46 57 55 58 80 82UAtot [MW C-1] 431 453 457 444 244 284 326 598 604
C0 middot10minus3 [kgs-1m-13]lowast 110 120 130 107 37 51 54 3000 3000Refrigerant composition
xCH4 [mole-] 333 323 323 325 311 292 319 325 332xC2H6 [mole-] 353 332 334 347 323 329 327 329 335xC3H8 [mole-] 00 00 00 00 00 00 00 00 00
xnminusC4H10 [mole-] 250 246 243 228 267 303 252 234 235xN2 [mole-] 64 99 100 100 99 76 102 112 98
lowastC0 adjusted to obtain∆Tmin asymp 20C
Case 61Nominal design using data fromPrice and Mortko(1996) in Table62
90 Optimal design of a simple LNG process
We specify the LNG temperature at the exit of NG HX (Tout) to minus144C (Priceand Mortko 1996) The LNG production ( ˙m = 252kmoles-1 = 446kgs-1) isslightly larger (37) than that reported byPrice and Mortko(1996) (mLNG =243kmols-1lowast) but note that the feed composition is different and that we haveneglected the removal of heavy components On the other hand we have notincluded the heating of flash gas (in Fuel HX) before re-compression to turbinefuel which would have further increased the LNG production by providing somecooling for free
The resulting flash gas is 77kgs-1 and will produce about 230MW of energy bycombustion in a gas turbine (assuming 60 efficiency and 50MJkg-1) This is toohigh considering that the refrigerant compressor consumes less than 80MW In theremaining cases we have therefore limited the amount of flash gas to 333kgs-1 toachieve about 100MW equivalents of fuel which replaces the specification onTout
Case 62Constraint on the amount of flash gas after expansion (333kgs-1) suchthat it gives about100MWworth of energy in a gas turbine
Note that the temperature out of the heat exchanger (Tout) is reduced fromminus144to minus156C to reduce the amount of flash gas This results in a 60 reduction inproduction compared with Case61 (mLNG drops from 446kgs-1 to 417kgs-1)This is because we are unable to cool as much natural gas and this is not compen-sated for by the increased liquid fraction after expansion The effect of the outlettemperature (Tout) is further discussed below
Case 63Optimized super-heating
We find by removing the constraint on super-heating that∆Tsup increases from100C to the optimal value of 116C This gives an 08 increase in LNGproduction compared with Case62 This illustrates as discussed byJensen andSkogestad(2007) that the optimal super-heating is not zero for a system withinternal heat exchange
Case 64Higher degree of super-heating
In this case we specify a higher degree of super-heating (257C compared to theoptimal of 116C) This gives only a 13 reduction in LNG production com-pared to Case63 which shows that the optimum is ldquoflatrdquo in terms of super-heatingWith 022C super-heating we get a reduction of 23 in LNG production com-pared with Case63
lowastCalculated from 471MNm3day-1
63 Results for optimal design 91
In reality we expect that the heat transfer coefficient is lower in the super-heatingsection than in the vaporization section This suggests that the optimal degreeofsuper-heating will be lower than what we find with constant heat transfercoeffi-cients
Until now we have fixed the two refrigerant pressures Specifically thedischargepressurePh is fixed at 22bar and the pressure ratioPr = PhPl at 55 (Table63)However some authors have published optimization results with discharge pres-sure much higher than 220bar (Lee et al 2002 Del Nogal et al 2005) Theyalso claim that the refrigerant flowrate should be about 3-4 times the flowrate ofnatural gas on mole basis For the cases up till now we have obtained a ratio ofabout 6 which is about 50 higher These two observations are closelyrelated asthe amount of refrigerant depends on the pressure ratio (Pr)
Case 65No pressure constraints
Here we optimize the process without the constraint on discharge pressure andpressure ratio We see that the production is increased from 417kgs-1 to 463kgs-1
(11) while the refrigerant amount is reduced from 147kmoles-1 to 764kmoles-1
(which gives a ratio 29 between refrigerant and LNG flowrate) To achieve thisthe high pressure is increased toPh = 504bar and the pressure ratio isPr = 22
Some other interesting results to note are
bull the compressor suction volume decreases
bull the necessary heat transfer area for the warm refrigerant stream isless thanhalfUA is 188MWC compared to 409MWC for Case62
Both these effects are related to the fact that much less refrigerant is needed buthow can this be explained The cooling duty per kg of refrigerant is closely re-lated to the compressor head[kJkg-1] which again is closely related to the pres-sure ratio So increasing the compressor head (and pressure ratio) willincreasethe cooling duty per kg of refrigerant and thus decrease the required amount ofrefrigerant
There is a potential problem with this design A high pressure ratio usually requiresmore compressor stages (casings) and this may not be desirable althoughsome ofthe extra capital cost related to the extra compressor casing and higher pressurewill be offset by the reduction in heat transfer area
We wish to limit the PRICO process to one compressor casing which may not befeasible with the high pressure ratio of 22 in Case65 To get a realistic design weuse performance specifications for the MCL1800 series compressor from GeneralElectric Oil and Gas(January 2007)
92 Optimal design of a simple LNG process
Case 66MCL1800 series compressor
MCL1800 is a centrifugal compressor with casing diameter of 1800mm The re-ported maximum suction volume is 380000m3h-1 or about 106m3s-1 the maxi-mum discharge pressure is 30bar and the maximum shaft work is 120MW (Gen-eral Electric Oil and Gas January 2007) In this case we keep 775MW as themaximum compressor shaft work to compare with the other cases and specifya maximum compressor suction volume of 106m3s-1 and maximum pressure of30bar
Interestingly the results show that we are able to almost match the production andpressure ratios obtained in Case65with realistic specifications and one compres-sor casing The total head in the compressor may be achieved with one compressorcasing with 5 wheels and a rotational speed of 3600RPM see Table61
Note that the suction volumeVsuc is an active constraint for the three last cases inTable63where actual compressor data are utilized
The MCL1800 compressor seems a bit large for the specified duty so let us try asmaller compressor
Case 67MCL1400 series compressor
MCL1400 is a centrifugal compressor with a casing diameter of 1400mm Thereported maximum suction volume is 250000m3h-1 or about 70m3s-1 the maxi-mum discharge pressure is 37bar and the maximum shaft work is 75MW (GeneralElectric Oil and Gas January 2007) We have assumed that it is possible to use775MW power
We get a slightly higher production (36) compared with Case66 This illus-trates that the increase in outlet pressure (from 30bar to 37bar) more than compen-sates for the reduction in compressor suction volume (from 106m3s-1 to 70m3s-1)The reqired head is slightly reduced because of a higher suction pressure How-ever the maximum head is reduced compared with the MCL1800 compressorbecause of a reduced tip speed The total achievable head in the compressor isstrongly affected by the reduction in wheel diameter throughout the compressorcasing Using a 70cm reduction from one wheel to the next and an intial wheeldiameter of 135m gives an estimated maximum total head of 172kJkg-1 with 6compressor wheels This is significantly less than the required amount and we be-lieve that the design in Case67 is infeasible with one compressor casing but adetailed compressor design is necessary to verify this conclusion
Finally we would like to find the maximum train capacity limit for the PRICOprocess with a single compressor casing
63 Results for optimal design 93
Case 68Again we utilize the larger MCL1800 but we allow for more shaft powernamely120MW
We find that we may produce 711kgs-1 LNG in a single PRICO train with onecompressor casing using realistic design data Note that the required compressorhead is reduced from 216kJkg-1 to 162kJkg-1 compared to Case66 so for thiscase we may use a slower driver (see Table61) for example a Frame 9 gas turbinewith rotational speed of 3000RPM (General Electric Oil and Gas January 2007)
Note that the cost factorC0 = 3000 is increased tenfolds compared to the othercases This was necessary to achieve∆Tmin = 20C There seems to be no prac-tical reasons for this large increase in the cost factor and an alternativedesignoptimization is given below
1copy 2copy
Figure 63 A flowsheet of the PRICO process illustrating the use of liquid turbineand valve as expansion device
The LNG expansion valve and the refrigerant expansion valve shown inFigure61may be exchanged with a combination of liquid turbine and valve see Figure63Ideally one would do the entire expansion in a turbine but two-phase turbines areto our knowledge not in use so it is necessary to use the combination of a liquidturbine and a valve (Barclay and Yang 2006) The liquid turbine will then take thepressure down to slightly above the saturation pressure and the expansion valvewill take care of the two-phase expansion The main advantage with this solutionis not that power is generated by the liquid turbine (this is actually quite small)but rather the extra cooling provided by the isentropic expansion in the turbinecompared to isenthalpic expansion in the valve This reduces i) the amount of
94 Optimal design of a simple LNG process
flash gas that does not contribute to the product for LNG expansion andii) theamount of flash gas that does not contribute significantly to the cooling butstillneeds to be compressed by the compressor for the refrigerant expansion
Case 69A liquid turbine is included in the expansion of the natural gas1copy andin the expansion of the refrigerant2copy
The production is further increased by 66 compared with Case68 The totalheat transfer area is increased by 10 Note that also for this case we had tospecify a highC0 to obtain∆Tmin asymp 20C
We have assumed an isentropic efficiency of 80 in the liquid turbines We haveused a back-off from saturation at the turbine outlet of 20bar A turbine will beeven more advantageous for cases with higher pressures both in the refrigerantcycle and in the natural gas stream
In summarywe see that some design constraints strongly affect the optimal solu-tion These constraints are related to the compressor performance maximumsuc-tion volume maximum discharge pressure maximum head and maximum shaftwork The changes given by these constraints are illustrated in cases65 to 68Other constraints have less influence on the optimal solution these are the degreeof super-heating and the amount of flash gas (determined by the temperature aftercooling or the amount of flash gas) The changes given by these constraints areillustrated by the first four cases
631 An alternative design optimization
Above we adjusted the cost factorC0 to achieve∆Tmin asymp 20C For Case68 thisresulted in a cost factor unrealistically much larger than for the other cases Wesuspect that this is due to the non-linear behaviour of∆Tmin A better approachmay be to fixC0 HereC0 = 110middot103kgs-1
(m-2)065
is used for all cases
Table 64 shows key results for all nine cases with the same specifications forsuper-heating maximum work and pressures as before Note that for Case68weget an increase by 86 in the LNG production compared to the correspondingcase in Table63 whereC0 = 3000middot 103kgs-1
(m-2)065
This is achieved by in-creasing the total heat transfer area by 247 and∆Tmin is reduced from 20C to15C A similar increased production and heat transfer areas is achieved for Case69 For cases65to 67the production is reduced compared to the results in Table63
The results in Table64 are obtained more easily than the results in Table63
64 Discussion 95
Table 64 Optimal design withC0 = 110middot103kgs-1(m-2)065
Alternative Case 61 62 63 64 65 66 67 68 69mLNG [kgs-1] 446 454 458 443 365 426 451 772 814
∆Tmin [C] 196 196 186 204 289 251 237 148 165UAHOT [MW C-1] 384 417 423 397 80 184 225 642 628UANG [MW C-1] 48 46 47 45 29 41 46 103 104UAtot [MW C-1] 431 464 469 442 109 225 271 746 732
Change compared with Table63∆mLNG [] 00 89 90 -11 -212 -80 -62 86 74∆UAtot [] 00 24 26 -045 -555 -210 -169 247 212
where it was necessary to adjustC0 until we got∆Tmin asymp 20C
In terms of optimality it seems better to fix the cost factorC0 rather than the∆Tmin
In Cases62 and63 the relative (percent) increase in production ( ˙mLNG) is morethan three times the relative increase in heat transfer area (UAtot) even though thework Ws is constant This indicates strongly that the cost factorC0 is too high Analternative approach would therefore be to adjustC0 such that the relative increasein mLNG andUAtot are similar
64 Discussion
641 Compressor
The number of impellers or wheels in the compressor will affect the achievablehead Ludtke (2004) reports that centrifugal compressor may be manufacturedwith up to ten impellers in one casing but that more than five impellers usuallygives a cost in terms of reduced efficiencyGeneral Electric Oil and Gas(January2007) indicates that their compressors may be delivered with up to eight impellersBased on the numbers for maximum head per impeller in Table61 we have nodifficulty achieving the necessary head for the cases presented with the MCL1800compressor see Table63 However there is some uncertainty
bull The wheel diameter is decreasing through the compressor and finding theoptimal design requires a detailed analysis of the compressor
bull The number of wheels may affect the efficiency which we have assumedconstant at 828
96 Optimal design of a simple LNG process
Price and Mortko(1996) report that the PRICO process uses an axial compressorThe efficiency may be slightly higher for an axial compressor than for a centrifugalcompressor
642 Heavy extraction
We have neglected the process for removing heavy components from the naturalgas stream The paper byPrice and Mortko(1996) shows that this process is in-tegrated in the refrigeration process as shown in Figure64(a) The natural gas istaken out of the heat exchanger after being cooled somewhat The heavy compo-nents are then in the liquid phase and is extracted (methane and ethane in the liquidis recovered and mixed with the LNG product) The vapour stream is sent back tothe heat exchanger for further cooling
Alternatively one may have the removal of heavy components before the refriger-ation process This is shown in Figure64(b)
The choice between upstream and integrated NGL recovery depends onseveralfactors and will not be treated here
The fraction of NGL in the natural gas feed is quite small for the case we haveconsidered so we believe inclusion of an integrated NGL extraction will nothavea large impact on our results
(a) Integrated NGL recovery (b) Upstream NGL recovery
Figure 64 Alternative locations of the NGL recovery
64 Discussion 97
643 Feed pressure
Higher feed pressure gives higher maximum production because the main part ofthe cooling will be at a higher temperature (since the condensation temperaturerange is higher with increased pressure) Figure65shows the maximum produc-tion as function of feed pressure However this effect is not for free since a feedcompressor is needed For example compressing the feed from 40bar to60barrequires a compressor with about 53MW power for the natural gas flowrate inCase68 To find the optimal feed pressure it is necessary to also consider thefeed compressor and the extra capital cost related to higher pressure inthe heatexchanger and piping
The feed pressure increase is limited if the NGL extraction is integrated (becausethe separation in NGL extraction is harder at higher pressure eg propane has acritical pressure of about 42bar) so very high feed pressures areonly feasible forplants with NGL extraction prior to refrigeration
P[bar]
mLN
G[k
gs-1
]
40 45 50 55 6075
80
85
90
95
100
Figure 65 Maximum LNG production as function of feed pressure for Case69
644 Large PRICO plants
Larger production rates requires more than one compressor so one then has thechoice of using more PRICO trains in parallel or to use more cycles in cascade withone compressor for each cycle An obvious advantage with choosing more parallelPRICO trains is that one may operate on part load if one compressor shuts downBut important issues are thermodynamic efficiency and capital cost compared withcascaded cycles
Based on Case69 in Table 63 we get a maximum LNG production of about
98 Optimal design of a simple LNG process
76kgs-1 see Figure65 with P = 40bar This is about 22MTPA based on 335operating days per year so two trains in parallel gives about 44MTPA while threetrains in parallel gives about 66MTPA However if a large plant is constructedone may wish to have the NGL extraction up front and install a feed compressorprior to the refrigeration process From Figure65 we see that a feed pressure of60bar gives a maximum production of about 97kgs-1 or 28MTPA per train Thisresults in a maximum production of 56MTPA and 84MTPA for two and threeparallel trains respectively Even higher feed pressures may be feasible
The specific work is 439kWht-1 for 40bar feed pressure and 344kwht-1 for 60barfeed pressure for the process with liquid turbine for expansion of LNG and refrig-erant This is without considering the shaft work generated by the turbines andrequired by the feed and fuel compressor
Roberts et al(2004) presents different driver configurations for the AP-X processfrom the Air Products company The configuration with the highest production(10MTPA) is with three Frame 9 gas turbines and and additional 10MW fromthe helperstarter motor giving a total of 400MW We have assumed 130MWfor each Frame 9 gas turbine this the ISO rating (General Electric Oil and GasJanuary 2007) With the same available shaft work we would achieve
28MTPA120MW
middot400MW= 93MTPA
for the PRICO process with 60bar feed pressure and liquid turbine for expansionof both LNG and refrigerant This is 7 lower for the same shaft work but theactual numbers will depend heavily on several factors such as feed pressure feedtemperature feed composition and actual compressor performance
65 Conclusion
An objective function for design optimization has been derived for a PRICO pro-cess The process is optimized for several different constraints and compared withthe commercial process and other publications Using compressor specificationsfound online we are able to increase the LNG production compared to the com-mercial process Important constraints especially concerning the compressor fea-sibility are discussed Finally we found that the PRICO process has about 7less production than the AP-X process with the same available shaft power
Bibliography 99
Bibliography
Barclay M A and Yang C C (2006) Offshore LNG The perfect startingpoint for the 2-phase expanderin lsquoOffshore Technology Conference Hous-ton Texas USArsquo
Del Nogal F Kim J Smith R and Perry S J (2005) lsquoImproved design ofmixed refrigerant cycles using mathematical programmingrsquoGas Processors As-sociation (GPA) Europe Meeting Amsterdam
General Electric Oil and Gas (January 2007) lsquoLiquefied natural gas En-hanced solutions for LNG plantsrsquowwwgepowercombusinessesge_oilandgasendownloadsliquified_natural_gaspdf
Jensen J B and Skogestad S (2007) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Lee G C Smith R and Zhu X X (2002) lsquoOptimal synthesis of mixed-refrigerant systems for low-temperature processesrsquoInd Eng Chem Res41(20) 5016ndash5028
Ludtke K H (2004)Process centrifugal compressors Springer-Verlag BerlinHeidelberg
Price B C and Mortko R A (1996) PRICO - a simple flexible provenapproachto natural gas liquefactionin lsquoGASTECH LNG Natural Gas LPG interna-tional conference Viennarsquo
Roberts M J Liu Y Bronfenbrenner J C and Petrowski J M (2004) Reduc-ing LNG capital cost in todayrsquos competitive environmentin lsquoLNG 14 confer-ence Doha Quatarrsquo
Stebbing R and OrsquoBrien J (1975) An updated report on the PRICO(TM) pro-cess for LNG plantsin lsquoGASTECH LNG Natural Gas LPG international con-ference Parisrsquo
100 Bibliography
Chapter 7
Optimal operation of a simpleLNG process
Considering the large amount of work that goes into the design of LNGprocesses there is surprisingly little attention to theirsubsequent oper-ation This partly comes from the assumption that optimal design andoptimal operation are the same but this is not generally true In this pa-per we study the optimal operation of a relatively simple LNGprocessnamely the PRICO process We find that the process has four opera-tional degrees of freedom (neglecting the degrees of freedom related torefrigerant composition) We then study the two modes of operation i)given production and ii) maximum production
71 Introduction
The process considered in this paper is a single mixed refrigerant process namelythe PRICO process (Stebbing and OrsquoBrien 1975 Price and Mortko 1996) Thisis the simplest configuration used commercially for liquefaction of natural gasandit has been optimized in several publications (Lee et al 2002 Del Nogal et al2005 Chapter6) but only with respect to designSingh and Hovd(2006) studythe controllability of the process but they do not consider optimal operation whichis the theme in this paper
An important issue in plantwide control is to find the degrees of freedom thatmaybe used for online optimization (Skogestad 2002) In our case these are the sameas the steady-state operational degrees of freedom and this number is important forseveral reasons First it determines the degrees of freedom availablefor solving
101
102 Optimal operation of a simple LNG process
the optimization problem However more importantly in terms of operation it de-termines the number of steady-state controlled variables that need to be selectedOptimal operation is normally implemented by keeping the selected variables atconstant setpoints Note that the selection of controlled variables is equally im-portant if we use a model-based control structure such as model predictive control(MPC)
There are two main modes of operation for a LNG process
1 With a given LNG production (load) minimize the compressor shaft workWs (optimize efficiency)
2 Maximize the LNG production rate subject to given constraint (maximumshaft workWmax
s )
In general to implement optimal operation we first need to control active con-straints Second we need to find controlled variables for the unconstrained degreesof freedom We here use the self-optimizing control approach ldquoSelf-optimizingcontrol is when we can achieve acceptable loss with constant setpoint values forthe controlled variables (without the need to re-optimize when disturbances oc-cur)rdquo (Skogestad 2000)
72 Process description
Figure71shows a simplified flowsheet of the PRICO process The PRICO processworks as follows After compression to pressurePh the mixed refrigerant is cooledto 30C in a sea water (SW) cooler before it is further cooled together with thenatural gas in the main heat exchanger The high pressure sub-cooledliquid is thensent through a liquid turbine and a choke valve to give a low-temperature saturatedliquid at pressurePm in the receiver The liquid is further expanded to low pressurePl to give a two-phase mixture which is vaporized in the main heat exchanger toprovide the necessary cooling duty The vapour is slightly super-heatedbeforeit is compressed back to the high pressure The PRICO process is discussed inmore detail in Chapter6 We here consider Case69 from Chapter6 where aliquid turbine is included both in the expansion of natural gas and expansionof therefrigerant
Note that the refrigerant is only partially condensed at pressurePh in the sea water(SW) cooler so both liquid and vapour are fed to the main heat exchangerWe haveplaced the liquid receiver at an intermediate pressure (Pm) before the choke valve
72 Process description 103
The extra choke valve between the liquid turbine and receiver is to give a safetymargin to saturation at the liquid turbine outlet
NG30C40bar
Ph
Pl
Pm
N
minus157C35bar Choke valveTurbine
Main heat exchanger
Refrigerantcompressor
SW
30C
Figure 71 A simplified flowsheet of the PRICO process
Note that we have not included any extraction of heavy components from the nat-ural gas feed
721 Nominal conditions
bull The natural gas enters after pretreatment with a pressure of 40bar and atemperature of 30C
bull Composition of natural gas (mole-) 897 methane 55 ethane 18propane 01 n-butane and 28 nitrogen
bull Pressure drops
ndash 5bar in natural gas stream
ndash 01bar in SW cooler
ndash 4bar for hot refrigerant in main heat exchanger
104 Optimal operation of a simple LNG process
ndash 1bar for cold refrigerant in main heat exchanger
ndash 03bar for the compressor suction
bull Constant heat transfer coefficients (UA)
bull The refrigerant is a mix of nitrogen (N2) methane (C1) ethane (C2) propane(C3) and n-butane (nC4) and the composition is used as a degree of freedomin the optimization
bull Cooling of refrigerant to 30C in SW cooler
bull Maximum shaft work in the compressorWmaxs = 120MW
bull We have not considered the turbine characteristic and we have assumed aconstant pressure drop of 20bar across the valve after the turbine Thisvalve is necessary to guarantee that no vapour is formed in the liquid turbine
722 Manipulated inputs
There are 9 manipulated inputs with a steady-state effect (potential controldegreesof freedom for controlu0)
1 Compressor rotational speedN
2 Choke valve valve openingz
3 Turbine rotational speed
4 Cooler flow of sea water (SW) in cooler
5 Load Feed flow of natural gas (can also be considered a disturbance)
6-9 Composition of refrigerant (5 components give 4 independent compositions)These degrees of freedom are not considered for control but need to be op-timized at the design stage
Assuming maximum cooling of refrigerant in the SW cooler or rather fixingT =30C after the SW cooling which consumes 1 degree of freedom this leaves 8degrees of freedom To find the nominal optimal steady-state operating point wewill use all 8 degrees of freedom but during operation we will assume constantrefrigerant composition so there are only 4 steady-state control degrees of freedom
72 Process description 105
723 Operational constraints
In general there are many constraints (represented by the equationc le 0 in theoptimization problems) that must be satisfied during operation
bull Super-heating (∆Tsup) The stream entering the compressor must not containliquid (but note that this is not necessary an active constraint in this casebecause there is internal heat exchange so it is actually optimal with somesuper-heating see Chapter6)
bull ToutLNG Natural gas temperature out of the main heat exchanger should be
within certain bounds This temperature sets the amount of flash gas andaffects the composition of flash gas and LNG
bull Refrigerant pressure Maximum bound (not considered in this paper)
bull Compressor outlet temperature Maximum bound (not considered in thispaper)
bull Compressor power (Ws) We assume maximum at 120MW see Chapter6
bull Compressor rotational speed (N) We assume maximum at 100
bull Compressor surge The compressor may in theory be operated in the surgeregion using active surge control (Gravdahl and Egeland 1999) but nor-mally one would like to operate with a certain margin to surge In this paperwe use the peak of the compressor characteristic curve as the limit (see Fig-ure 73) We define the variable∆msurge to be the distance from the peakof the characteristic curve and use the value∆msurgege 00kgs-1 (ie noback-off)
bull All flows must be non-negative and also have upper bounds
In particular the cooling water flow has a maximum value and it is clear fromphysical insight that maximum cooling is optimal (active constraint) Assumingthat we have a large area in this heat exchanger we will in the following replacethis constraint by the following
bull Maximum cooling Assume refrigerant hasT = 30C after SW cooler
With the assumption ofT = 30C after the SW cooler (flow of sea water at maxi-mum) we are left with 8 steady-state degrees of freedom (4 for control)
106 Optimal operation of a simple LNG process
73 Model
When switching from design simulations to operation it is necessary to reformu-late parts of the models because the equipment is fixed Our goal is to have simplemodels that capture the most important operational effects We will here discussthe features that we have included in the operational models and also brieflymen-tion the effects that we have not considered
The process is modelled using the gPROMS software with the accompanying Mul-tiflash package for thermodynamic calculations The SRK equation of state isused for thermodynamic calculations both for the natural gas and the refrigerantThe main heat exchanger is a distributed system which for modelling purposeshas been discretized into 100 cells using forward and backward finite differencemethod for the cold and hot streams respectively
The pressure drop at the compressor inlet (suction) is modelled as
∆Psuc= ∆Psuc0
(Vsuc
Vsuc0
)2
(71)
The remaining pressure drops are assumed constant The structure of the modelequations are summarized in Table73
731 Compressor characteristic
In operation one normally uses compressor characteristic curves to model thecompressor behaviour These curves relating flow efficiency pressure increaseand rotational speed are normally supplied by the compressor vendor ofthe in-stalled compressor but since we do not have a specific design and vendor we needa more general approach
Non-dimensional groups
Compressor characteristics are easiest represented using non-dimensional groupsFor exampleSaravanamuttoo et al(2001) assume the following functional depen-dencelowast
f (DNmP1P2T1T2 R) = 0 (72)
lowastThey actually use the two groupsRT1 andRT2 instead of our three termsT1 T2 andR
73 Model 107
where the characteristic lengthD [m] is usually taken as the compressor wheeldiameterN [s-1] is the rotational speed ˙m[kgs-1] is the mass flowrateP[kgms-2]is the pressureR = RMW [Jkg-1K-1] is the specific gas constant andT [K] isthe temperature Subscripts 1 and 2 represent the compressor inlet and outletrespectively Using the Pi theorem of dimensional analysisSaravanamuttoo et al(2001) reduce these 7 function dependencies to 4 independent non-dimensionalgroups
Pr =P2
P1 Tr =
T2
T1 mr =
mradic
RT1
D2P1and Nr =
NDradic
RT1
(73)
(pressure ratio temperature ratio reduced flow and reduced speed) From thesefour groups it is possible to express one group in terms of the remaining threebut Saravanamuttoo et al(2001) claims that the groupsP2
P1and T2
T1may be plotted
against the last two groups The justification for this (which we could not find inSaravanamuttoo et al(2001)) is probably that the outlet conditions of the compres-sor (T2 andP2) should depend on the inlet conditions only which are expressed bymr andNr
It is common to report the isentropic efficiencyη instead of the temperature riseT2T1
so the following dependencies are used to quantify the steady-state operation ofthe compressor
Pr = f (mrNr) (74)
η = f (mrNr) (75)
Characteristic compressor curves
The dependencies in74 and75 are normally given graphically as ldquocurvesrdquo butwe are here looking for simple algebraic relationships
We use the method ofMoore and Greitzer(1986) cited inGravdahl and Egeland(1999) with some adjustmentsMoore and Greitzer(1986) proposed to use a cubicequation to predict the characteristic curve of a compressor The equation for thepressure ratio is (using our own nomenclature)
Pr = Pr0 +H
(
1+32
(mrW
minus1
)
minus12
(mrW
minus1
)3)
(76)
wherePr = P2P1
is the pressure ratio over the compressor (the first non-dimensionalgroup in Equation73) Pr0 is the shut-off value for the compressor (the pressure
108 Optimal operation of a simple LNG process
ratio delivered at zero flow)H andW are called the semi-height and semi-widthof the characteristic curve respectively and ˙mr is the reduced mass flowrate (thesecond non-dimensional group in Equation73)
The cubic equation has three parameters (Pr0 H andW) which physical signifi-cance are indicated in Figure72 The curve is for a given reduced rotational speedNr (the third non-dimensional group in Equation73)
2H
2W
Pr0
Pr [-]
mr[-]
Figure 72 A cubic compressor characteristic curve for a constant reduced com-pressor speed
The surge point where dynamic instability occurs is somewhere near the peak valueof the pressure ratio in Figure72 Operation to the left of this point is unstablewithout active (feedback) control This is discussed extensively byGravdahl andEgeland(1999) Note that the surge point is normally close to the ldquooptimalrdquo oper-ating point with peak pressure ratio and peak efficiency
To get the entire compressor map for all values of the reduced speedNr we pro-pose the following dependency onNr for the parameterH andW
H = H0minus12
(
H0 +Pr0
2minus1
)
middot (1minusNr) (77)
W = W0(Nr)13 (78)
This is by no means an exact approach but we get compressor characteristic curvesthat are similar to typical example curves shown in textbooks (egSaravanamuttooet al 2001)
For the isentropic efficiency we propose to use the following function
η = η0
((
1minus
(H minusH0
H0
)2)
minus1000(mrminus2W)2
)
(79)
73 Model 109
This captures the two most important effects the efficiency has a peak value fora given reduced rotational speed and the peak value is slightly affected by therotational speed
A sample compressor map is shown in Figure73 To compute ˙mr we have useda compressor with wheel diameterD = 17m and a working fluid with molecu-lar weightMW = 0032kgmol-1 and the following values for the parameters inEquation76to Equation79
bull Pr0 =minus29 Note that we have used a negative value for the pressure ratio atzero flowrateGravdahl and Egeland(1999) state thatPr0 gt 0 but we areonly interested in the operating regime to the right of the peak value so thisparameter does not have any physical meaning in our model We have useda high negative value to get a steeper characteristic curve
bull H0 = 18125
bull W0 = 00698
bull η0 = 822
Pr[
-]
mr[-]
η[-]
mr[-]0 002 004 006 008 01 012 014 016
0 002 004 006 008 01 012 014 016
0
05
1
0
2
4
6
8
10
N = 100
1
10 N = 100-
Figure 73 Compressor map for the refrigerant compressor using Equations76-79with N in the range 10 to 100 and nominal inlet temperature The red dotsindicates the peak for the pressure ratio curve
110 Optimal operation of a simple LNG process
74 Objective function
The objective function for optimal operation is simpler than for optimal designdiscussed in Chapter6 because the investments are already made and the capi-tal cost does not need to be considered The cost function to be minimized thenbecomes
Joperation=pWs middotWsminus pWsturbine middotWsturbine+ pSWmiddotQC (710)
minus pLNG middot mLNG + pfeedmiddot mfeedminus pfuel middot mfuel
We make the following assumptions
bull Same price for fuel and feed Thenpfeedmiddotmfeedminuspfuel middotmfuelminuspLNG middotmLNG =(pfeedminus pLNG) middot mLNG = pLNG middot mLNG
bull Neglect income from turbine workpWsturbine = 0
bull Neglect cost of coolingpSW = 0
The optimization problem then becomes
minu
Wsminus pLNG middot mLNG (711)
subject to cle 0
Herecle 0 represent the mathematical formulation of the operational constraintsand the model equations
Depending on product prize and other external factors there are two main operatingmodes
Mode I Given throughput With a given feed flowrate or given LNG productionthe optimization problem simplifies to
minu
Ws (712)
subject to mfeed= given (or mLNG = given)
cle 0
Mode I will result in the same optimal operation as the nominal optimal de-sign provided the optimal design is done correctly (eg using the simplifiedTAC method in Chapter6) andmfeed is kept at the nominal feedrate
Mode II Maximum throughput If the LNG prize ( ˆpLNG) is sufficiently high andthere is no active constraint related to available feed or product distribution
75 Nominal optimum given production case (Mode I) 111
Table 71 The nominal operating point for Mode I - Given production Mode II -Maximum production
Mode I Mode IIWs[kW] Compressor work 106 120Ph [bar] Cycle high pressure 268 300Pl [bar] Cycle low pressure 367 414N [] Compressor rotational speed 100 100η [] Compressor efficiency 828 828
mLNG [kgs-1] LNG flowrate 698 767mREF[kgs-1] Refrigerant flowrate 549 614
∆msurge[kgs-1] Surge margin 0000 0000∆Tsup[
C] Super-heating before compressor 129 113Tout[
C] NG temperature after cooling -157 -157xCH4 [mole-] Methane in refrigerant 319 327
xC2H6 [mole-] Ethane in refrigerant 352 343xC3H8 [mole-] Propane in refrigerant 00 00
xnminusC4H10 [mole-] nButane in refrigerant 247 233xN2 [mole-] Nitrogen in refrigerant 82 97
Boldface Specifications and active constraints
(sommaxLNG is not an active constraint) then it will be optimal to maximize the
production of LNG and the objective function may be simplified
minu
minus mLNG (713)
subject to cle 0
Note that the operation in this mode may be quite different from the ldquonomi-nalrdquo optimum found for mode I
75 Nominal optimum given production case (Mode I)
Since the production rate (or feed rate) is fixed there are 7 steady-state degrees offreedom including the 4 refrigerant compositions
With a given production rate ˙mLNG = 698kgs-1 the nominal optimum is found bysolving the optimization problem in Equation712 The results are summarized inthe left column of Table71 The work in712was minimized with respect to the7 degrees of freedom including the 4 refrigerant compositions Note thatwe have
112 Optimal operation of a simple LNG process
assumed that the refrigerant is 30C after the SW cooling The optimal operationof the compressor was found subject to compressor maps (Figure73) We findthat the following constraints are active at the nominal optimum
1 Temperature of natural gas after cooling at maximum (Tout = minus157C)
2 Surge margin at minimum (∆msurge= 0)
3 Compressor speed at maximum (N = 100)
Thus at the nominal optimum the only unconstrained degrees of freedom are therefrigerant compositions
76 Nominal optimum maximum production case (ModeII)
Here we consider mode II where maximum production is the objective see Equa-tion 713 Since the production rate (or feed rate) is free there are 8 steady-statedegrees of freedom (with 30C after SW cooling) For numerical reasons opti-mal operation in mode II was found by solving mode I for increasing values ofmLNG = given until no feasible solution was found
h[Jmol-1]
P[P
a]
-11 -105 -10 -95 -9 -85 -8 -75times104
105
106
107
Figure 74 Optimal (nominal) pressure enthalpy diagram for the maximum pro-duction case (mode II)
76 Nominal optimum maximum production case (Mode II) 113
Pr[
-]
mr[-]
η[-]
mr[-]0 002 004 006 008 01 012 014 016
0 002 004 006 008 01 012 014 016
0
05
1
0
2
4
6
8
Figure 75 The circle shows the optimal (nominal) compressor operating point forthe maximum production case (mode II)
The nominal optimum including the optimal composition of the refrigerant issummarized in the right column of Table71 The cycle is illustrated in a pressureenthalpy diagram in Figure74 and the optimal temperature and temperature dif-ference profiles in the main heat exchanger are shown in Figure76(a)and76(b)respectively
We find that the following constraints are active at the nominal optimum
1 Compressor work at maximum (Ws = 120MW)
2 Surge margin at minimum (∆msurge= 0)
3 Temperature of natural gas after cooling at maximum (Tout = minus157C)
4 Compressor rotational speed at maximum (N = 100)
Note that there are two ldquocapacityrdquo constraints that are active (1 and 4) Again theonly unconstrained degrees of freedom are related to the refrigerantcomposition
We have now identified the nominal optimum for the two cases but how should
114 Optimal operation of a simple LNG process
Position[-]
T[
C]
0 20 40 60 80 100-200
-150
-100
-50
0
50
(a) Temperature profile
Position[-]
T[
C]
0 20 40 60 80 1000
5
10
15
20
25
30
(b) Temperature difference profile
Figure 76 Optimal temperatures in the heat exchanger for the maximum produc-tion case (mode II)
we control the process to maintain close to optimal operation when the processisexposed to disturbances This is discussed next
77 Optimum with disturbances
Table 72 Nominal minimum and maximum values for the disturbances Thenumbers in parentheses are for mode I
Nominal Min Max NameWmax
s [MW] 120 110 130 d1lowast
Pfeed[bar] 40 35 45 d2
Tin [C]dagger 30 25 35 d3
xCH4 [] 327 (319) 294 (287) 360 (351)d4
xC2H6 [] 343 (352) 309 (317) 377 (387)d5
xC4H10 [] 233 (247) 210 (222) 256 (272)d6
xN2 [] 97 (82) 87 (74) 107 (90) d7
mLNG [kgs-1] 698 665 731 d8Dagger
lowastOnly used for mode IIdaggerThe temperature of natural gas and refrigerant at the inlet to the main heat exchangerDaggerOnly used for mode I
The next step is to consider optimal operation with disturbances and we considerthe eight disturbance variables given in Table72 We here fix the refrigerant com-
77 Optimum with disturbances 115
position because it is assumed to be unavailable as a degree of freedom duringnormal operation (instead the composition is treated as a disturbance see Table72) With a fixed temperature (30C) after the SW cooler there are then 4 remain-ing degrees of freedom During operation it is always optimal to cool the naturalgas tominus157C (to avoid product give-away) and one degree of freedom is spentto set the load
Mode I The production rate is given
Mode II In the maximum production case it is always optimal (for all distur-bances) to operate the compressor at its maximum (Ws = Wmax
s )
Thus two constraints are always active and this leaves for both modes only 2steady-state operational degrees of freedom and the resulting optimal operationconditions for various disturbances are summarized in Table75 The results arealso shown graphically as dots in Figure77 and Figure78 for mode I and modeII respectively
Recall that the surge margin constraint (∆msurge= 0) and compressor maximumspeed constraint (N = Nmax) were active in the nominal point and we find as onewould expect that these remain active for most of the disturbances but not allAlso note that some disturbances are not feasible in mode I probably because ofthe fixed refrigerant composition
To obtain optimal operation we should always implement the active constraintsand then find rdquoself-optimizingrdquo variables for the remaining unconstrained degreesof freedom in each region Strictly speaking to be truly optimal we then needtoconsider four regions
1 N = Nmax and∆msurge= 0 is optimal (two active constraints implementationis obvious)
2 ∆msurge= 0 andN lt Nmax is optimal (unconstrained optimum ie need tofind an associated controlled variable forN)
3 N = Nmax and∆msurgegt 0 is optimal (unconstrained optimum ie need tofind an associated controlled variable for∆msurge)
4 N lt Nmax and∆msurgegt 0 is optimal (unconstrained optimum ie need tofind an associated controlled variable forN and∆msurge)
All this cases can occur as seen in Table75 This becomes rather complicatedFirst a large effort is required to find the best self-optimizing variables in each ofthe three last regions Second even if we can find the self-optimizing variables ineach region it is not clear when to switch between the regions (that it it is easy to
116 Optimal operation of a simple LNG process
identify when to switch when we encounter a constraint eg going from region 2to 1 but it is more difficult to determine when to leave a constraint region)
For practical implementation we would therefore prefer to have the same con-trolled variables in all regions and in our case the obvious policy is to considerkeeping the variables at constraint (∆msurge= 0 andN = 100) in all regions Ob-viously this is not optimal but the loss is rather small as discussed next except forsome cases in mode I where it seems operation is not feasible (without changingthe composition or feed rate)
771 Selection of controlled variables
A preliminary screening was performed by using the maximum scaled gain method(Halvorsen et al 2003) Some variables where discarded based on these results(eg the degree of sub-cooling cycle high pressure) Also we found that thesurge margin (∆msurge) is a much more promising controlled variable than any ofthe alternatives we tested Thus we choose to fix∆msurge= 0 and this gives onlyminor losses as seen below
Figure77 shows compressor shaft work as a function of 6 of the 7 disturbances(d3 minus d8) considered for mode I∆msurge= 0 and the following controlled vari-ables are tested for the remaining degree of freedomN = 100∆Tsup= 129CPl = 367barTout
com = 126C andmref = 549kgs-1 The dots shows re-optimizedoperation where bothN and∆msurgehas been optimized Note that the composi-tion of the refrigerant is still the same as for the nominal operating point The plotsare obtained by ldquobruteforce evaluationrdquo which involves fixing the variables andcomputing the resulting operating point for varying disturbances The disturbanceis plotted on the X-axis with the nominal value at the center and the Y-axis showsthe corresponding compressor shaft work in MW
For the four last disturbances (d5minusd8) we are not able to find a feasible solutionfor some values of the disturbances This means that we are not able to satisfythe specified production rate with the given refrigerant composition The best con-trolled variable to maintain constant is the compressor rotational speedN Forsome disturbances is it the only feasible controlled variable and it remains close tooptimal for all operating points
Figure78shows LNG production as a function of 6 of the 7 disturbances (d1d2d4minusd6) considered for mode II∆msurge= 0 and the following alternative controlledvariables are tested for the remaining degree of freedomN = 100 ∆Tsup =113C Pl = 414barTout
com = 124C andmref = 614kgs-1 The dots shows re-
77 Optimum with disturbances 117
TC [C]
Ws[M
W]
25 30 35100
105
110
115
120
Pl
mref
AAAU
N
Toutcom
∆Tsup
N
(a) Disturbanced3 = Tin
XCH4 []
Ws[M
W]
29 30 31 32 33 34 35100
105
110
115
120
∆Tsup
Toutcom
RNN
(b) Disturbanced4 = xCH4
XC2H6 []
Ws[M
W]
32 33 34 35 36 37 38100
105
110
115
120
N
N
∆Tsup
Toutcom
(c) Disturbanced5 = xC2H6
XC4H10 []
Ws[M
W]
2324 25 26 27100
105
110
115
120
NN
Toutcom
∆Tsup
(d) Disturbanced6 = xC4H10
XN2 []
Ws[M
W]
75 8 85 9100
105
110
115
120
N Nmref
R
∆Tsup
Toutcom
(e) Disturbanced7 = xN2
mLNG [kgs-1]
Ws[M
W]
67 68 69 70 71 72 73100
105
110
115
120
N
N
mref
Pl
∆Tsup
Toutcom
(f) Disturbanced8 = mLNG
Figure 77 Shaft work for different control structures as function of disturbancesfor mode I Dots are re-optimized operation with constant composition
118 Optimal operation of a simple LNG process
d1 = Ws[MW]
LNG
[kg
s-1]
110 115 120 125 13065
70
75
80
85
Pl
Toutcom
∆Tsup
mre fN
(a) Disturbanced1 = Wsmax
d2 = Tin [C]
LNG
[kg
s-1]
25 30 3565
70
75
80
85
Pl
N∆Tsup
Toutcom
mre f
(b) Disturbanced3 = Tin
d4 = xCH4 []
LNG
[kg
s-1]
30 31 32 33 34 35 3665
70
75
80
85
N
PlToutcom
∆Tsup mref
(c) Disturbanced4 = XCH4
d5 = xC2H6 []
LNG
[kg
s-1]
31 32 33 34 35 36 3765
70
75
80
85NPl
Toutcom
R∆Tsup
mref
(d) Disturbanced5 = XC2H6
XC4H10 []
LNG
[kg
s-1]
2122 23 24 2565
70
75
80
85
Pl
Nmref Tout
com
∆Tsup
(e) Disturbanced6 = XC4H10
XN2 []
LNG
[kg
s-1]
9 95 10 10565
70
75
80
85
Toutcom
mref
Pl
N ∆Tsup
N
(f) Disturbanced7 = XN2
Figure 78 LNG production for different control structures as function of distur-bances for mode II Dots are re-optimized operation with constant composition
77 Optimum with disturbances 119
Ws[MW]
∆Tsu
p[C
]
110 115 120 125 1300
5
10
15
20
25
30
Figure 79∆Tsup as function of disturbance inWmaxs for the constantN control
strategy
optimized operation where bothN and∆msurge has been optimized The distur-bance is plotted on the X-axis with the nominal value at the center and the Y-axisshows the corresponding production rate in kgs-1 Note that the composition ofthe refrigerant is still the same as for the nominal operating point
N is actually the only feasible control structure for some disturbance directionsThis may be seen from Figure78(a) where all lines except the constantN lineends at the nominal point (from left to right) The reason for this is that theothercontrol structures would require a non-feasible rotational speed (higher than themaximum of 100) to maintain the controlled variable constant
From Figure78 we see that controllingN seems to be a good self-optimizingcontrol strategy that gives close to optimal operation over the entire disturbancerange This means that the implementation is quite simple sinceN = Nmax isoptimal in one region and close to optimal in the other regions
Note that we have notconsidered the constraint on∆Tsup in Figure 77 and inFigure78 The nominal value for∆Tsup is 129C (mode I) and 113C (mode II)which should be more than sufficient for normal operation For disturbances thatleads the process into operation in the region whereN = Nmax is optimal (eg tothe right in Figure78(a)) the super-heating is gradually reduced see Figure79but still remains positive At some point it will be necessary to discard one ofthe optimally active constraints (eg maximum shaft workWs = Wmax
s ) to satisfya minimum degree of super-heating One solution to this could be to select arefrigerant composition that would assure sufficient super-heating for all operatingpoints Another solution is to implement some logic that will switch controlledvariables depending on the operating point
120 Optimal operation of a simple LNG process
N []
LNG
[kg
s-1]
80 85 90 95100105110 115 12065
70
75
80
85
Figure 710 LNG production as function ofN
In summaryFor both mode I and mode II fixing∆msurge= 0 andN = 100is optimal for the nominal operating point and in some of the disturbance regionsSince maintaining constantN and∆msurgeis also close to optimal for the remainingdisturbance regions we propose to use this control structure
78 Discussion
781 Moving temperature profile
An interesting result to note from Figure77and Figure78is the ldquokinkrdquo that seemsto be at the nominal point for all disturbances This was at first puzzling tous sincewe are not changing the set of constraints that are active
Let us consider the nominal point withN as degree of freedom (not worrying aboutthe constraintNmax which can be exceeded for a short time) Looking at the re-sulting graphs see Figure710 we note that ˙mLNG as a function ofN has a discon-tinuity at the nominal point (100) and one at approximately 105 We foundthat the reason behind this rather strange behaviour is the shape of the obtainedtemperature profile in the heat exchanger Figure76(b)shows that the tempera-ture difference profile in the heat exchanger has two clear peaks oneat the heatexchanger inlet and one about 70 through the heat exchanger Theldquokinkrdquo thatwe observe in the objective function ( ˙mLNG) occurs when the peak at the heat ex-changer inlet leaves the heat exchanger which causes an abrupt change in the heattransfer at the inlet to the heat exchanger Figure79 shows the effect in super-heating out of the heat exchanger
78 Discussion 121
782 Compressor characteristic
The compressor characteristics used to link the flowrate pressure ratioefficiencyand the rotational speed of the compressor is not fitted to the data of a real com-pressor so it is uncertain if an installed compressor has the behaviour ourmodelpredicts An issue for further research could be to investigate if our conclusion onthe proposed control structure (∆msurge= 0kgs-1 andN = 100) is affected bythe compressor characteristic
783 Refrigerant composition
The refrigerant composition has been optimized for the nominal operating pointonly Since it is not adjusted in operation it would have been better to optimizeit with respect to the expected range of operating points (given by the distur-bances) This strategy ldquorobust optimumrdquo is discussed inGovatsmark and Sko-gestad(2005)
784 Additional considerations
The optimization problems for mode I and mode II presented above are simplifiedfor example
bull There might be a given schedule of ships arriving to transport the LNG tomarkets elsewhere Because of boil-off from the storage tanks it may bedesirable to minimize the storage but on the other hand there should besufficient storage to minimize the loading time of the ships This kind ofthinking is discussed byZaim(2002)
bull If there are more parallel production trains one needs to decide on howmany trains that should operate and how large production each train shouldhave This is also discussed byZaim(2002)
bull If the ambient temperature is changing (for example from night to day orseasonal changes) it may be optimal to produce more LNG during night(or in colder periods) when the gas turbine has higher efficiency and thecondenser temperature is lower Related topics are discussed for simplerrefrigeration systems inCai et al(2007) (also included as AppendixB)
122 Bibliography
Bibliography
Cai J Jensen J Skogestad S and Stoustrup J (2007) Balancing energy con-sumption and food quality loss in supermarket refrigeration systemin lsquoIEEErsquo
Del Nogal F Kim J Smith R and Perry S J (2005) lsquoImproved design ofmixed refrigerant cycles using mathematical programmingrsquoGas Processors As-sociation (GPA) Europe Meeting Amsterdam
Govatsmark M and Skogestad S (2005) lsquoSelection of controlled variables androbust setpointsrsquoIndEngChemRes44(7) 2207ndash2217
Gravdahl J T and Egeland O (1999)Compressor Surge and Rotating StallModeling and Control London Springer Verlag
Halvorsen I J Skogestad S Morud J C and Alstad V (2003)lsquoOptimal selec-tion of controlled variablesrsquoInd Eng Chem Res42 3273ndash3284
Lee G C Smith R and Zhu X X (2002) lsquoOptimal synthesis of mixed-refrigerant systems for low-temperature processesrsquoInd Eng Chem Res41(20) 5016ndash5028
Moore F K and Greitzer E M (1986) lsquoA theory of post-stall transients in a axialcompressor systems Part I - Development of equationsrsquoJournal of Engineeringfor Gas Turbines and Power108 68ndash76
Price B C and Mortko R A (1996) PRICO - a simple flexible provenapproachto natural gas liquefactionin lsquoGASTECH LNG Natural Gas LPG interna-tional conference Viennarsquo
Saravanamuttoo H I H Rogers G F C and Cohen H (2001)Gas turbinetheory 5th edn Pearson Education Limited Harlow England
Singh A and Hovd M (2006) Dynamic modeling and control of the PRICOLNG processin lsquoAmerican Institute of Chemical Engineers (AIChE) annualmeetingrsquo
Skogestad S (2000) lsquoPlantwide control the search for the self-optimizing controlstructurersquoJ Process Contr10(5) 487ndash507
Skogestad S (2002) Plantwide control towards a systematic procedure in lsquoEu-ropean Symposium on Computer Aided Process Engineering (ESCAPE)- 12The Hague The Netherlandsrsquo
Stebbing R and OrsquoBrien J (1975) An updated report on the PRICO(TM) pro-
Bibliography 123
cess for LNG plantsin lsquoGASTECH LNG Natural Gas LPG international con-ference Parisrsquo
Zaim A (2002) Dynamic optimization of an LNG plant Case study GL2Z LNGplant in Arzew Algeria PhD thesis RWTH Aachen
Table 73 Structure of model equations
Unit Equations
Compressor Ws = mmiddot (houtminushin) Ws = m(hsminushin)ηPoutPin
= f (mr Nr) η = f (mr Nr)lowast
Turbines Ws = mmiddot (houtminushin) Ws = m(hsminushin)η
Valves hout = hin m= zmiddotCv middotradic
∆Pmiddotρ
SW cooler Q = mmiddot (houtminushin)
Heat exchanger Qi(z) = Ui middot∆Ti(z)δAi mmiddot parthi(z)partz = Qi(z)dagger
partPi(z)partz = constant
lowastConsult Section782for details regarding the compressor characteristicdaggeri is the stream either natural gas hot refrigerant or cold refrigerant
Table 74 Data for the PRICO processUA for natural gas in main heat exchanger 845MWC-1
UA for warm refrigerant in main heat exchanger 532MWC-1
UA for cold refrigerant in main heat exchanger 616MWC-1
Nominal compressor isentropic efficiency 822 lowast
Isentropic efficiency for the liquid turbines 800
lowastAt nominal conditions andN = 100 see Section782for further details
124 Bibliography
Table 75 Optimal operation with disturbances (nominal refrigerant composition)Mode I Mode II
Ws Ph Pl N ∆msurge mLNG Ph Pl N ∆msurge
[MW] [bar] [bar] [] [kgs-1] [kgs-1] [bar] [bar] [] [kgs-1]d1(minus1) 695 266 375 999 0489d1(minus05) 727 281 394 100 000d1(0) Not relevant for mode I 767 300 414 100 000d1(05) 785 316 434 100 000d1(1) 798 332 453 100 000d2(minus1) 115 297 402 100 0000 736 301 415 100 0000d2(minus05) 110 280 382 100 0000 752 301 415 999 0000d2(0) 106 268 367 100 0000 767 300 414 100 0000d2(05) 105 263 375 960 1125 778 299 416 994 0000d2(1) 104 258 359 100 0000 786 298 421 980 0000d3(minus1) 100 256 347 100 0000 798 309 418 100 0000d3(minus05) 103 261 357 100 0000 782 304 416 999 0000d3(0) 106 268 367 100 0000 767 300 414 100 0000d3(05) 111 279 388 100 0024 733 292 411 100 0000d3(1) 117 294 410 991 0000 708 287 409 100 0602d4(minus1) 110 293 391 100 0000 753 311 420 100 0000d4(minus05) 107 277 375 100 0000 762 306 417 100 0000d4(0) 106 268 367 100 0000 767 300 414 100 0000d4(05) 109 270 388 956 1832 749 290 409 100 1002d4(1) 111 270 380 100 0 738 285 407 100 0471d5(minus05) 106 270 369 100 0000 758 298 421 979 0000d5(0) 106 268 367 100 0000 767 300 414 100 0000d5(05) 108 271 372 100 0000 764 301 415 100 0000d5(1) Infeasible 757 303 415 100 0000d6(minus1) 112 267 379 100 0000 720 280 410 980 0000d6(minus05) 110 268 376 100 0000 743 288 416 976 0000d6(0) 106 268 367 100 0000 767 300 414 100 0000d6(05) 108 280 377 100 0000 757 308 426 979 0000d6(1) Infeasible 755 320 431 100 0000d7(minus1) Infeasible 757 302 416 100 0503d7(minus05) 108 273 374 100 0000 763 300 414 100 1971d7(0) 106 268 367 100 0000 767 300 414 100 0000d7(05) 106 267 372 986 0367 764 299 418 987 0000d7(1) 107 268 376 978 0657 756 297 425 963 0000d8(minus1) 103 254 353 100 0000d8(minus05) 104 260 370 962 2493d8(0) 106 268 367 100 0000 Not relevant for mode IId8(05) 111 282 385 100 0000d8(1) Infeasible
di( j) i is the disturbance number (see below) andj is the fraction of the full disturbance (egj = minus1 is the minimum value of the disturbance andj = 05 is in the middle of the nominal valueand the maximum value of the disturbance)
d1 = Wmaxs d5 = xC2H6
d2 = Pfeed d6 = xC4H10
d3 = Tin d7 = xN2
d4 = xCH4 d8 = mLNG
Chapter 8
Degrees of freedom forrefrigeration cycles
Submitted for publication in Industrial amp Engineering Chemistry Research
An important issue for optimal operation and plantwide control is tofind the degrees of freedom available for optimization A previouslypublished systematic approach to determine the steady state degrees offreedom is expanded to take into account the active charge asa possibledegree of freedom in cyclic processes Additional degrees of freedomrelated to composition of the circulating refrigerant are also discussedTwo LNG processes of current interest the C3MR LNG process fromAir Products and the MFC process developed by Statoil-LindeLNGTechnology Alliance are studied with respect to operational degrees offreedom
81 Introduction
This paper considers degrees of freedom for available for optimization of refriger-ation processesSkogestad(2000) points out that it is normally steady-state thateffects the plant economics so we will consider only steady-state operation Weare then interested in the steady-state degrees of freedom that affects the plant eco-nomics (objective functionJ) Nopt = NMV minusN0 whereNMV is the manipulatedvariables (valves etc) andN0 is the variables that does not affect the economics(eg liquid level in an open process)Skogestad(2002 2004) The number of ma-nipulated variables are usually quite easily obtain by counting the valves pumpsand other inputs to the process TheN0 variables that does not affect the plant
125
126 Degrees of freedom for refrigeration cycles
economics however requires detailed process overview and understanding A listof the potential degrees of freedom for some typical process units is given inSko-gestad(2002) with an updated version inAraujo et al(2007) We wish to extendthis list to also include degrees of freedom that are special for closed cycles
Glemmestad et al(1999) discuss degrees of freedom for heat exchanger networks(HEN) The review paper on plantwide control byLarsson and Skogestad(2000)discuss degrees of freedom as this is an important step in plantwide controlAmore recent study on degrees of freedom is that ofKonda et al(2006)
Processes for liquefaction of natural gas are very cost intensive and requires largeamounts of energy in operation It is therefore important that the plants arebothwell designed and later operated close to optimum also for changing conditionsThe optimal design of LNG processes has been studied extensively by severalcompanies such as Air Products Shell Phillips and Statoil-Linde LNG Technol-ogy Alliance It seems however that the subsequent operation of LNGplants hasbeen less studied at least in the open literature This is a bit surprising consid-ering the large throughputs which makes even small improvements economicallyattractive There are some publications regarding control of LNG plants (Mandler2000 Singh and Hovd 2006) but they consider the dynamic performance andcontrollability rather than the optimal steady-state operationZaim (2002) lookedinto dynamic optimization of a plant with several trains in parallel
Degrees of freedom for refrigeration processes are covered in thenext section andin Section83we apply the findings on some case studies including the two LNGprocesses
bull The propane pre-cooled mixed refrigerant (C3MR) process from AirProd-ucts
bull The mixed fluid cascade (MFC) process from Statoil-Linde LNG Technol-ogy Alliance
82 Degrees of freedom
An important issue in plantwide control is to find the degrees of freedom thatmay be used for optimization (Skogestad 2000) which in our case is equal to thenumber of steady-state degrees of freedomNss This is an important number forseveral reasons First it determines the degrees of freedom availablefor solvingthe optimization problem However more importantly in terms of operation itdetermines the number of steady-state controlled variables (Nss) that need to be
82 Degrees of freedom 127
Table 81 Potential operational degrees of freedom (Nmaxss ) for some typical pro-
cess unitsProcess unit Potential DOFFeed 1 (feedrate)Splitter number of exit streams - 1Mixer 0Compressor turbine pump 1 (work)Adiabatic flash tank 0lowast
Liquid phase reactor 1Gas phase reactor 0lowast
Heat exchanger 1 (bypass or flow)Column (excluding heat exchangers) 0lowast + number of side streams Valve 0lowast
Choke valve 1Each closed cycleActive charge (holdup fluid) 1dagger
Composition of fluid NCminus1Dagger
lowastPressure is normally assumed to be given by the surrounding processand is then not a degree offreedom However on must add one degree of freedom for each extra pressure that is independentlyset and which has a steady-effect (need a manipulated input not already counted eg a valve)
daggerThe active charge in the equipment is a potential degree of freedom but it may not be availablein some designs
DaggerNC is the number of components in the working fluid (refrigerant)
selected Optimal operation is normally implemented by keeping those variablesat constant setpoints
Rule (actual degrees of freedom)The numberNss of steady-state degrees of free-dom may be obtained by counting the number of manipulated variablesNMV andsubtracting the followingN0 variables (Skogestad 2002 2004)
bull manipulated variables with no effect on the costJ eg extra bypasses ofheat exchangers (only used to improve dynamic performance)
bull variables with no steady state effect that need to be controlled eg liquidholdups with no steady state effect
ThusNss= NMV minusN0
128 Degrees of freedom for refrigeration cycles
Potential (maximum) degrees of freedom
Based on this ruleSkogestad(2002) derived the potential number of degrees offreedomNmax
ss for some typical process units and an updated version was publishedby Araujo et al(2007) In Table81 the list is further updated to include also thepotential degrees of freedom for cyclic processes The additions areshown belowthe dotted line in the table First a valve should normally not be counted unlessitaffects a pressure that has a steady-state effect An example is a chokevalve whichis installed to lower the pressure and it has therefore been added explicitlyas adegree of freedom in the table In addition we potentially haveNCminus1 degrees offreedom related to the fluid composition in the cycle Finally the active chargeinthe cycle is a potential degree of freedom For example it may change the pressurelevel This is explained in more detail below
Many designs will have fewer actual degrees of freedom than given inTable81For example a heat exchanger will not have any degrees of freedomif all flowsare given and there is no bypass or other means of affecting the heat transfer egthe main exchanger in a LNG process Similarly one may not be able to adjust thetotal active charge or fluid composition in practice
Potential degrees of freedom1 Compressor2 Heat exchangers1 Choke valve
+ 1 ldquoActive chargerdquoNmax
ss = 5 degrees of freedomQC
QH
Wsz
Ph
Pl
Evaporator
Condenser
Figure 81 A simple refrigeration cycle withNmaxss = 5 potential degrees of free-
dom However if the drawing shows the actual process there will only beNss= 4degrees of freedom because there is no means of changing the active charge byfilling or removing refrigerant
ExampleFrom Table81 the simple cooling cycle shown in Figure81with purefluid (NC = 1) has five potential degrees of freedom (Nmax
ss ) related to one com-pressor two heat exchangers one choke valve and the active charge Exactly how
82 Degrees of freedom 129
these degrees of freedom may be changed depends on the case For example fora compressor the actual manipulated variable (MV) may be the rotational speed orthe fraction of time the compressor is on (home or automotive installations) Theactive charge in Figure81 may be changed by filling or removing refrigerant tothe cycle but since this is not included the actual process does not have a degreeof freedom related to the active charge and thus has only 4 degrees of freedom(Nss)
Table 82 Actual degrees of freedom for refrigeration cyclesNss= NMV minusN0
Process unit Actual DOFEach MV (Valve heat exchanger compressor turbine etc) 1For each cycle subtract variables with no steady-state effect (N0)Pure fluidLiquid receivers exceeding the firstlowast -1Multi componentLiquid receiver exceeding theNC firstdagger -1
lowastThe first receiver is not subtracted in a closed cycle as this has a steadystate effectdaggerAssumes composition different in each of theNC first receivers otherwise the number of de-
grees of freedom is less
821 Remark on active charge or ldquofeedrdquo for closed cycles
The degree of freedom related to the total active charge is not obvious so we herediscuss this ldquoextrardquo degree of freedom that may occur in closed cyclesConsiderthe process shown in Figure82 where we have included an external tank with avalve for filling refrigerant into the closed cycle A temperature controller adjuststhe compressor speed to assure that the cooling load is constant The choke valveis a thermostatic expansion valve (TEV) that controls the degree of super-heatingat the evaporator outlet With the fan speeds for the two heat exchangersfixed (egat maximum) there are then one potential remaining degree of freedom related tothe active charge
First assume that the process is operated with just enough charge (refrigerant in theclosed cycle) to obtain saturation out of the condenser see Figure82(a) We thenopen the external valve for some time and then close it again to fill more refrigerantinto the cycle We then get to the operating point shown in Figure82(b)where theadded amount of refrigerant has accumulated in the condenser This follows sincethe evaporator will not change its holdup (significantly) due to the thermostaticexpansion valve that indirectly sets the area available for super-heating so the only
130 Degrees of freedom for refrigeration cycles
TC
Super-heatcontrol
(a) Saturation at condenser outlet
TC
Super-heatcontrol
(b) Sub-cooling at condenser outlet
Figure 82 A simple (not closed) refrigeration cycle The external fillingemptyingsystem illustrates the degree of freedom related to the active charge
place the extra refrigerant can go is to the condenser The increased charge inthe condenser will lead to a higher pressurePh which again leads to larger meantemperature difference and thus more heat transfer and the liquid at the outlet fromthe condenser will be sub-cooled as indicated in Figure82(b) It has been shownthat this may reduce the necessary compressor power in some cases (Jensen andSkogestad 2007b) but the main point here is to note that the charge in the systemhas an effect on the steady state operation
Remark 1 Note that Table82discussed below does not apply because the cycle in Figure82 is not closed
Remark 2 If we add a liquid receiver to the cycle in Figure82then we loose one degreeof freedom (as we have a level with no steady-state effect that needs to be controlled) Toregain this degree of freedom we (at least) would need to add another valve
822 Actual degrees of freedom for refrigerant cycles
For a closedcycle in order to adjust the active charge during operation we needa liquid receiver (variable holdup) in the cycle However for a pure refrigerantadding additional (two or more) receivers will not increase the number ofdegreesof freedom as the holdup has no steady-state effect and needs to be controlled(actually it may reduce the number unless we also add a valve for this purpose)In general for a multicomponent (mixed) refrigerant the holdup of theNC firsttanks dohave a steady-state effect provided the tanks have different compositions
82 Degrees of freedom 131
z
QC
QH
Ws
Pl
Ph
(a) No extra choke valveNMV = 4 andNss= 4
z
QC
QH
Ws
Pl
Ph
Pm
(b) An extra choke valveNMV = 5 andNss= 5
z
QC
QH
Ws
Pl
Ph
Pm
(c) Extra choke valve and liquid receiver on lowpressure sideNMV = 5 butNss= 4
Figure 83 Simple cycle with liquid receiver on the high pressure side
132 Degrees of freedom for refrigeration cycles
as they provide an indirect means of adjusting the fluid composition Only liquidreceivers exceeding theNC first will have no steady-state effect These argumentsare the basis for Table82which gives the actual(rather than the potentialin Table81) degrees of freedom for vapour compression cycles
Let us apply Table82 to the process in Figure81 which hasNMV = 4 (two heatexchangers one valve and one compressor) Since there are no liquidreceiversthere is no variables that need to be subtracted(N0 = 0) ThusNss= 4 degrees offreedom which confirms our earlier findings
Note that adding a liquid receiver somewhere in this cycle will not change thenumber of steady-state degrees of freedom In order to do this we also need to adda valve for example upstream of the receiver The addition of a liquid receiver tothe cycle is shown in Figure83 We have from Table82the following degrees offreedom for the three cases in Figure83
Figure 83(a)NMV = 4-N0 = 0Nss = 4
Figure 83(b)NMV = 5-N0 = 0Nss = 5
Figure 83(c)NMV = 5-N0 = 1Nss = 4
For the two first cases the liquid level does not need to be subtracted as ithas asteady-state effect (see Figure82) and also does not need to be controlled in aclosed cycle
The design in Figure83(a)with no extra valve does not allow for adjusting theactive charge The design in Figure83(b)with an extra choke valve has an ad-ditional degree of freedom This may be more optimal (Jensen and Skogestad2007b) since it allows for the condenser outlet to be sub-cooled This is becausethe pressure in the receiver is equal to the saturation pressure and the extra valvegivesP gt Psat (sub-cooling) at the condenser outlet
In Figure83(c) we have added also a liquid receiver on the low pressure sideThus we have lost one degree of freedom compared to Figure83(b) because oneof the liquid levels need to be controlled Another way of understanding whythereis one less degree of freedom is that we now always have saturated vapour at theinlet to the compressor whereas it before could be super-heated
Remark Note that a loss of a degree of freedom does not mean that the process is lessoptimal In fact in this case it is opposite because for a simple cycle (without internalheat exchange) it is optimal to have saturation (no super-heating) before the compressorThus Figure83(c)is optimal by design whereas in Figure83(b)one needs to adjust oneof the degrees of freedom to get optimality and this may be difficult to achieve in practice
83 Case studies 133
83 Case studies
We will here present some more complex case studies First we will look at twoprocesses not directly related to LNG plants a two-pressure level refrigeration cy-cle cooling a process stream and a heat integrated distillation column (two differentconfigurations) Then we will look at three LNG case studies a small scaleLNGprocess with a single mixed refrigerant the propane pre-cooled mixed refrigerant(C3MR) process from Air Products and the mixed fluid cascade (MFC) processfrom Statoil-Linde LNG Technology Alliance
831 Two-pressure level refrigeration
Figure84(a)shows a refrigeration system with two-stage expansion using a purerefrigerant (NC = 1) The process stream is first cooled by the intermediate-pressure refrigerant (Evaporator 1) and then by the low-pressurerefrigerant (Evap-orator 2) The evaporators are kettle type boilers so there is no super-heating of thevapour The temperature profile in the evaporators is illustrated in Figure84(b)Control of such cycles is discussed byWilson and Jones(1994) The two com-pressors are usually driven with a common driver so there is only one manipulatedvariable for the two compressors1copy There are three valves2copy 3copy and 4copy shownin Figure84(a) The valve between Evaporator 1 and the second compressor4copyis present to limit the amount of cooling in Evaporator 1 if necessary In additionwe may manipulate the flow of coolant in the condenser5copy and the process stream6copy Thus this two-pressure level cycle hasNMV = 6 manipulated variables
The two evaporators in Figure84(a)will function as liquid receivers so there are intotal three variable liquid levels Two of the liquid levels typically the evaporatorsneed to be controlled for stabilization and since the refrigerant is pure thesetpointsfor these (levels) has no steady-state effect (see also Table82) Thus we end upwith Nss= NMV minusN0 = 6minus2 = 4 degrees of freedom To operate the system weneed to decide on four controlled variables In general these should beselectedas the possible active constraints (eg max cooling5copy fully open) plus ldquoself-optimizingrdquo variables
It is interesting to compare the actualdegrees of freedom with the potentialdegreesof freedom according to Table81 We have for Figure84(a)
134 Degrees of freedom for refrigeration cycles
Evaporator 1
Evaporator 2
1copy2copy
3copy 4copy
5copy
6copy
Ph
Pl
T
Processin
Processout
(a) FlowsheetT
[C
]
Position[-]
Process stream
Evaporator 1
Evaporator 2
(b) Temperature profile
Figure 84 Cooling at two pressure levelsNss= 4
Actual DOFNMV = 6minusN0 = 2Nss = 4
Potential DOF1 Feed1 Compressors3 Heat exchangers3 Valve
+ 1 Active chargeNmax
ss = 9 degrees of freedom
The 5 ldquolostrdquo degrees of freedom are related to
1 No sub-cooling in the condenser (not optimal)
23 No super-heating in the two evaporators (optimal)
45 No bypass for the two evaporators (optimal)
Thus four of the five lost degrees of freedom are ldquooptimal by designrdquoin this caseThe only possible loss is related to not allowing for sub-cooling of the streamleaving the condenser To fix this would require an additional valve between thecondenser and the liquid receiver
832 Heat integrated distillation
To reduce the energy consumption in distillation one may use a heat pump be-tween the condenser and reboiler (Salim et al 1991) Two possible designs are
83 Case studies 135
1copy
2copy
3copy
4copy
5copy6copy
7copy
(a) External working fluid
1copy
2copy
3copy
4copy
5copy6copy
7copy
(b) Vapour overhead as working fluid
Figure 85 Two ways of using a heat pump to integrate the reboiler with the con-denser in a distillation column
shown in Figure85
a) With external working fluid Control of such columns have been studiedbyJoslashrgenses and coworkers (Hallager et al 1990 Nielsen et al 1987 1988Li et al 2003) who also have an experimental setup
b) With the vapour overhead as working fluid (not a closed cycle)
There are in total 7 manipulated variables for both systems (see Figure85) Feedflowrate 1copy Bottom product flowrate2copy Top product flowrate3copy Reflux flowrate4copy Compressor5copy Choke valve6copy Cooling water flowrate7copy
On the column side there are two liquid levels with no steady-state effect that mustbe controlled
bull Condenser liquid level (eg may be controlled by the reflux flowrate4copy)
bull Reboiler liquid level (eg may be controlled by the bottom product flowrate2copy)
Using the method in Table82 combined with the general rule we get the actualdegrees of freedom
136 Degrees of freedom for refrigeration cycles
Figure 85(a)NMV = 7minusN0 = 3lowast
Nss = 4
lowastOne of the two levels in the heat pumpcycle must be controlled and has no steady-state effect
Figure 85(b)NMV = 7minusN0 = 3lowast
Nss = 4
lowastThe heat pump cycle is no longer closedso the single liquid level (receiver) must becontrolled
Thus in both cases there are 4 steady-state degrees of freedom whenthe feedrateand column pressure are included This is the same as for an ldquoordinaryrdquo distillationcolumn
Govatsmark(2003) studied the process with external working fluid (Figure85(a))and found that the top composition and the column pressure are at their constraintsfor the case where the top product is the valuable product Assuming a given feedflowrate there is then one unconstrained degree of freedom left This unconstraineddegree of freedom could then be used to control a temperature in the bottomsectionof the column which is found to be a good self-optimizing controlled variable(Govatsmark 2003)
It is interesting to compare the actualdegrees of freedom with the potential(max-imum) degrees of freedom according to Table81
Figure 85(a)1 Feed0 Column1 Pressure (in column)1 Compressor3 Heat exchanger1 Choke valve
+ 1 Active chargeNmax
ss = 8 degrees of freedom
Figure 85(b)1 Feed0 Column1 Pressure (in column)1 Compressor2 Heat exchanger1 Choke valve
+ 1 Pressure (in condenser)Nmax
ss = 7 degrees of freedom
In Figure85(a)the four ldquolostrdquo degrees of freedom are related to
12 No bypass of two heat exchangerslowast (optimal)
3 Saturation before compressor (optimal)
4 Saturation at condenser outlet (not optimal)
For Figure85(b)there are three ldquolostrdquo degree of freedom related to
1 No bypass of the column reboiler (optimal)
lowastNot the cooler because the flow of coolant is an actual degree of freedom
83 Case studies 137
2 Saturation before compressor (optimal)
3 Saturation at condenser outlet (not optimal)
The only non-optimal ldquolostrdquo degree of freedom is for both cases related tosatu-ration out of the condenser in the heat pump cycle This may be close to optimal(Jensen and Skogestad 2007a) However to gain this degree of freedom it is nec-essary to have a valve between the cooler and the liquid receiver This valve willthen give sub-cooling out of the cooler
833 Small scale LNG process
From Table 82NMV = 6minusN0 = 0lowast
Nss = 6
lowastNo subtraction of liquid receivers be-cause of multicomponent working fluid withNCge 2
1copy
2copy 3copy
4copy
5copy
6copy NG
LNG
Figure 86 A small scale LNG concept to illustrate the degrees of freedom relatedto changing the composition via liquid levels
Consider the small-scale LNG process in Figure86 with a mixed refrigerantNCge 2 (Neeraas and Brendeng 2001) Note that the original process design hasadditional degrees of freedom related to individual heat exchangersfor refriger-ant cooling and natural gas cooling Our simplified flowsheet assumes thatit is
138 Degrees of freedom for refrigeration cycles
optimal to cool the refrigerant and the natural gas to the same temperature Wefind (using Table82) that the process has six manipulated variablesNMV = 6 seeFigure86 Since there are two liquid levels we need to control at least one forstabilization Thus it is tempting to remove one degree of freedom Howeverthelevel setpoint has a steady-state effect because the composition in the two tanksare different and we may change the composition of the circulating refrigerant byshifting mass from one tank to the other ThusNss= NMV = 6
Using Table81we get the potential degrees of freedom if we assumeNC = 3
Figure 861 Feed1 Compressor3 Choke valve4 Heat exchanger1 Active charge2 Compositions
Nmaxss = 12 degrees of freedom
We have the following 6 lost degrees of freedom
1-3 No bypass of process heat exchangers (optimal)
45 Pressure in the two flash drums (optimal discussed below)
6 Composition of the refrigerant (not optimal)
The ldquolostrdquo degree of freedom related to the pressure in the two flash drumsare notobvious Adding a valve before the second flash drum will give a lower tempera-ture in the flash drum and thus also of the vapour that is sent through the secondheat exchanger This is not optimal since the vapour will then be colder than thenatural gas which it is cooled together with The same is true for the first flashdrum if the natural gas feed is cooled with the same coolant as the refrigerant aftercompression
834 Propane pre-cooled mixed refrigerant (C3MR)
The C3MR process developed by the Air Products company has a large marketshare of the existing liquefaction plants worldwide A flowsheet of the C3MRprocess is given in Figure87 The first cycle is with a pure refrigerant usuallypropane The second cycle is with a mixed refrigerant We identify the followingmanipulated variables
bull Natural gas feed1copy
83 Case studies 139
1copy
2copy 3copy 4copy
5copy 6copy 7copy
8copy
9copy
10copy
11copy
12copy13copy
Figure 87 Flowsheet of the C3MR processPropane (C3) Mixed refrigerant(MR) Natural gas (NG)
bull 6 choke valves for propane pre-cooling (one for each pressure level for natu-ral gas cooling and one for each pressure level for mixed refrigerant cooling)2copy 3copy 4copy 5copy 6copy and 7copy
bull Propane compressor one speed8copy
bull Flow of cooling water or air in propane condenser9copy
bull Two choke valves for mixed refrigerant cycle10copy and11copy
bull Mixed refrigerant compressor12copy
bull Flow of cooling water or air in mixed refrigerant cooler13copy
For the propane cycle we need to control 6 of the 7 liquid levels (eg the heatexchanger levels) and sinceNC = 1 none of these level setpoints will have a steady-state effect AssumingNC = 3 for the mixed refrigerant cycle we get the followingactual and potential degrees of freedom
140 Degrees of freedom for refrigeration cycles
Actual DOFNMV = 13minusN0 = minus6Nss = 7
Potential DOF1 Feed2 Compressor10 Heat exchanger8 Choke valve2 Active charge
+ 2 Compositions (MR)Nmax
ss = 25 degrees of freedom
The 18 ldquolostrdquo degrees of freedom are related to
1-8 No bypass of process heat exchangers (optimal)
9-14 Saturation at propane compressor inlet (no super-heating optimal)
15 Saturation out of propane condenser (no sub-cooling not optimal)
16 Pressure in the flash drum in the MR cycle (optimal)
1718 Composition of the mixed refrigerant (not optimal)
The only loss in efficiency due to the ldquolostrdquo degrees of freedom are caused by hav-ing no sub-cooling in the propane condenser and by not being able to change thecomposition in the mixed refrigerant cycle To get sub-cooling in the condenser itis necessary to have a valve between the condenser and the liquid receiver Adjust-ing the composition is discussed further in Section84
It is optimal to have the flash drum in the mixed refrigerant cycle at the samepressure as the outlet from the last propane cooler Otherwise the warmrefrigerantwould be colder than the natural gas in the first mixed refrigerant heat exchangercausing a non-optimal temperature profile
Optimal operationLet us consider two different operating strategies
Case 81Maximum production given available shaft work
Case 82Minimum shaft work given feed flowrate
For both cases the following is true
It is not economical to cool more than necessary so the natural gas outlettemper-ature is at its maximum constraintlowast this may be controlled by the last choke valve
lowastThis temperature will implicitly set the amount of flash gas see Chapter6 and the compositionof both the flash gas and the LNG
83 Case studies 141
in the mixed refrigerant cyclelowast removing one degree of freedom Cooling wateris usually cheap so it is usually wise to maximize the flow of cooling water whichremoves two additional degrees of freedom We are then left with 4 degrees offreedom to optimize the operation
For Case81 Two degrees of freedom are used to maximize compressor shaftwork one for each cycle (active constraints) This leaves us with two uncon-strained degrees of freedom so two setpoints must be specified for this case (egPh in the mixed refrigerant cycle andPl in the propane cycle)
For Case82 The feed flowrate is given so we loose one degree of freedom Thisleaves us with three unconstrained degrees of freedom so three setpoints must bespecified for this case (egPl andPh in the mixed refrigerant cycle andPl in thepropane cycle)
Remark If the cooling water is too cold it may be necessary to limit the flow of coolingwater to avoid violating constraints on the compressor suction pressure (minimum con-straint) This however does not change the analysis sincethe number of active constraintsare the same since we exchange the maximum cooling constraint with the minimum pres-sure constraint
835 Mixed fluid cascade (MFC)
The Statoil-Linde LNG Technology Alliance has developed a mixed fluid cascade(MFC) process (Bach 2002 Forg et al 1999) A flowsheet of the process is givenin Figure88 It consists of three refrigeration cycles i) Pre-cooling cycle (PR) ii)liquefaction cycle (LC) and iii) sub-cooling cycle (SC) All three refrigerant cyclesuse mixed refrigerants
There are in total 13 manipulated variables (NMV = 13)
bull Natural gas feed1copy
bull Flow of cooling waterair in PC cycle2copy
bull Choke valve intermediate pressure level PC cycle3copy
bull Choke valve low pressure level PC cycle4copy
bull Compressor PC cycle5copy
bull Flow of cooling waterair in LC cycle6copy
lowastAnother solution is to use the natural gas feed flowrate by adjusting the LNG expansion Thisexpansion device (valve or turbine) is not shown here as we have indicated the feed manipulator atthe inlet of the stream instead
142 Degrees of freedom for refrigeration cycles
1copy2copy
3copy
4copy
5copy
6copy
7copy8copy
9copy
10copy
11copy12copy
13copy
NGLNG
Figure 88 Flowsheet of the MFC processPre-cooling cycle (PC) Liquefactioncycle (LC) Sub-cooling cycle (SC)andNatural gas (NG)
bull Extra valve LC cycle7copy
bull Choke valve LC cycle8copy
bull Compressor LC cycle9copy
bull Flow of cooling waterair in SC cycle10copy
bull Extra valve SC cycle11copy
bull Choke valve SC cycle12copy
bull Compressor SC cycle13copy
There are no liquid levels that must be controlled (N0 = 0) AssumingNC = 3 foreach cycle we get the following actual and potential degrees of freedom
84 Discussion 143
Actual DOFNMV = 13minusN0 = 0Nss = 13
Potential DOF1 Feed3 Compressor7 Heat exchangers4 Choke valve3 Active charge
+ 6 CompositionNmax
ss = 24 degrees of freedom
The 11 ldquolostrdquo degrees of freedom are related to
1-4 No bypass of process heat exchangers (optimal)
5 Saturation at PC condenser outlet (not optimal)
6-11 Two compositions for each of the three cycles (not optimal)
Also for this process it is the saturation specification out of the condenserfor thefirst cycle and the fixed compositions that will give the losses
Optimal operation Let us again consider optimal operation for two cases TheLNG outlet temperature must be controlled and the amount of cooling water inthree sea water coolers are maximized giving 9 unconstrained degrees of freedom
For Case81(maximum feed)We use three degrees of freedom to maximize thecompressor shaft work in each cycle We are then left with 6 unconstrained degreesof freedom so 6 setpoints must be specified
For Case82(given feed)The feed is given so there are then 8 unconstrained de-grees of freedom For this case we need 8 setpoints
Note that there may be active constraints on the temperature after PCHX1 andorthe temperature after PCHX2 if the NGL extraction is integrated For each activeconstraint we will have one less unconstrained degree of freedom thatwe have tofind a controlled variable for
84 Discussion
841 Degrees of freedom
Table 81 is not straightforward to use in practice This is mainly because anintermediate pressure may or may not have a steady-state effect This is noteasilycaptured by simply counting the degrees of freedom for each unit operation but
144 Degrees of freedom for refrigeration cycles
requires a process understanding The method shown in Table81 is only one ofseveral alternatives that will lead to the same result
842 Refrigerant composition
For mixed refrigerants there areNC minus 1 potential degrees of freedom related tothe composition of the refrigerant However all of these are usually not realizedas actual degrees of freedom We have claimed above that liquid level setpoints(provided different composition) may be used to effectively utilize the degreesof freedom related to refrigerant composition see the small scale LNG processpresented above This is not a practical solution with several compositionsas it isnecessary with equally many liquid tanks (with sufficiently different compositions)and control elements (valves) So in practice one will instead rely on a constantcomposition that may be changed on a larger timescale by utilizing the make-upsystem
843 Saturation in condenser
Another potential degree of freedom that is sometimes ldquolostrdquo is related to the con-denser pressure By having a liquid receiver after the condenser it will not be possi-ble to have a condenser pressure different from the saturation pressure (Pcon= Psat)By having a valve in between the condenser and the liquid receiver it is possibleto have sub-cooling in the condenser (Pconge Psat) This is discussed in more detailin Jensen and Skogestad(2007b) However it may also be necessary with a dif-ferent condenser design if the design does not allow for sub-coolingThis is thecase if the liquid formed in the condenser leaves the heat transfer zone (eg due togravity)
85 Conclusion
The degrees of freedom available for optimization (Nss) is an important numberfor several reasons It determines the number of free variables available to solvethe optimization problem However more importantly it determines how manysteady-state controlled variables that must be selected to operate the process
This paper extends an earlier published simple approach to determine the potentialdegrees of freedom (Nmax
ss ) based on unit operations to also cover vapour compres-sion cycles A simple method to determine the actualdegrees of freedom (Nss)
Bibliography 145
for vapour compression cycles is also presented Both methods are illustrated onfour case studies where the difference between the potentialand actualdegrees offreedom are explained and related to the process layout
For the two LNG case studies (C3MR and MFC) we also illustrate the effect ofoperating strategy (maximum production and given production) on the number ofunconstrained degrees of freedom
Bibliography
Araujo A Govatsmark M and Skogestad S (2007) lsquoApplication of plantwidecontrol to the HDA process I - steady-state optimization and self-optimizingcontrolrsquo Control engineering practice15 1222ndash1237
Bach W A (2002) lsquoDevelopments in the mixed fluid cascade process (MFCP)for LNG baseload plantsrsquoReports on science and technology Linde63
Forg W Bach W Stockmann R Heiersted R S Paurola P and FredheimA O (1999) lsquoA new LNG baseload process and manufacturing of the mainheat exchangerrsquoReports on science and technology Linde61
Glemmestad B Skogestad S and Gundersen T (1999) lsquoOptimal operation ofheat exchanger networksrsquoComput Chem Eng23 509ndash522
Govatsmark M S (2003) Integrated Optimization and Control PhD thesisNor-wegian University of Science and Technology (NTNU) Trondheim Norway
Hallager L Jensen N and Joslash rgensen S B (1990) Control system configura-tion for a heat pump operating as energy source for a distillation columninlsquoPresented at Nordic CACE symposium Lynby Denmarkrsquo
Jensen J B and Skogestad S (2007a) lsquoOptimal operation of a simple refrig-eration cycles Part II Selection of controlled variablesrsquoComput Chem Eng31 1590ndash1601
Jensen J B and Skogestad S (2007b) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Konda N Rangaiah G and Krishnaswamy P (2006) lsquoA simple and effectiveprocedure for control degrees of freedomrsquoChem Engng Sci61 1184ndash1194
Larsson T and Skogestad S (2000) lsquoPlantwide control - a reviewand a newdesign procedurersquoModeling Identification and Control21 209ndash240
146 Bibliography
Li H Gani R and Joslash rgensen S B (2003) Integration of design and controlfor energy integrated distillationin lsquoEuropean Symposium on Computer AidedProcess Engineering (ESCAPE) 13 Lappeenranta Finlandrsquo
Mandler J A (2000) lsquoModelling for control analysis and design in complexindustrial separation and liquefaction processesrsquoJournal of Process Control10 167ndash175
Neeraas B O and Brendeng E (2001) lsquoMethod and device for small scale liq-uefaction of a product gasrsquo US Patent number 6751984
Nielsen C S Andersen H W Brabrand H and Joslashrgensen S B(1988) Adap-tive dual composition control of a binary distillation column with a heat pumpin lsquoPreprints IFAC Symp on Adaptive Control of Chem Processesrsquo
Nielsen C S Andersen H W Brabrand H Toftegard B and Joslashrgensen S B(1987) Dynamics and identification of distillation pilot plant with heat pumpin lsquoPresented at CHISArsquo87 Praguersquo pp 1ndash10
Salim M A Sadasivam M and Balakrishnan A R (1991) lsquoTransient analysisof heat pump assisted distillation system 1 the heat pumprsquoInt J of EnergyResearch15 123ndash135
Singh A and Hovd M (2006) Dynamic modeling and control of the PRICOLNG processin lsquoAmerican Institute of Chemical Engineers (AIChE) annualmeetingrsquo
Skogestad S (2000) lsquoPlantwide control the search for the self-optimizing controlstructurersquoJ Process Contr10(5) 487ndash507
Skogestad S (2002) Plantwide control towards a systematic procedure in lsquoEu-ropean Symposium on Computer Aided Process Engineering (ESCAPE)- 12The Hague The Netherlandsrsquo
Skogestad S (2004) lsquoNear-optimal operation by self-optimizing control Fromprocess control to marathon running and business systemsrsquoComput Chem Eng29 127ndash137
Wilson J A and Jones W E (1994) lsquoThe influence of plant designon refriger-ation circuit control and operationrsquoEuropean Symposium on Computer AidedProcess Engineering (ESCAPE) 4 Dublinpp 215ndash221
Zaim A (2002) Dynamic optimization of an LNG plant Case study GL2Z LNGplant in Arzew Algeria PhD thesis RWTH Aachen
Chapter 9
Conclusion
The charge (holdup) does not affect the steady-state for anopenprocess (eg liquidlevel in a buffer tank) because of the boundary conditions on pressure It has beenshown that ldquoactive chargerdquo inclosedcycles has a steady-state effect One wayof affecting the ldquoactive chargerdquo is by having a liquid receiver in the cycle Thisdegree of freedom is often lost by designing the cycle without sub-cooling in thecondenser We find that some sub-cooling is desirable and for the ammonia casestudy the compressor shaft work is reduced by about 2 by allowing forsub-cooling The savings are not very large but more importantly the results showthat the active charge is a degree of freedom and that the sub-cooling gives somedecoupling between the high pressurePh and the hot source temperatureTH Thisshows that there are no fundamental differences between the typical sub-criticalcycles and the trans-criticalCO2 cycles
In terms of practical operation there are differences between the sub-critical am-monia cycle and the trans-criticalCO2 cycle For the ammonia cycle several simplecontrol structures gives acceptable performance The best controlstructure foundis to control the temperature approach at the condenser exit For theCO2 cycle wehad to use a linear combination of measurements to get acceptable performance
It is common to do the early design of refrigeration cycles by specifying a mini-mum approach temperature in the heat exchangers (minJ subject to∆T ge ∆Tmin)This method fails to give the optimal operating point and also fails to find that sub-cooling is optimal As a simple alternative we propose the simplified TAC method(min(J+C0 sumAn
i )) whereC0 replaces∆Tmin as the adjustable parameter
A PRICO LNG process has been designed by using realistic compressor specifi-cations found online We are able to increase the LNG production comparedwith
147
148 Conclusion
the commercial PRICO process Compared with the AP-X LNG process we findthat the PRICO process has about 7 less production for the same available shaftpower
Operation of the PRICO process is studied for two modes of operation i) minimumshaft work (for given production) and ii) maximum production Both modeshas 2controlled variables that must be selected (after controlling constraints thatare al-ways optimally) We find that maintaining a minimum distance to surge (∆msurge=00kgs-1) and maximum rotational speed of the compressor (N = 100) gives op-timal operation for the nominal operating point and for some of the disturbances(for both modes) Since this control structure also gives close to optimal opera-tion for the remaining disturbances we propose to control∆msurge= 00kgs-1 andN = 100
The degrees of freedom available for optimization is an important number forsev-eral reasons First it determines the number of free variables to solve theoptimiza-tion problem Second it determines how many steady-state controlled variablesthat must be selected to operate the process Finding the degrees of freedom isnot straightforward and requires a detailed process understanding An earlier pub-lished simple approach to determine the potential degrees of freedom basedon unitoperations is extended to also cover vapour compression cycles A simple methodto determine the actual degrees of freedom for vapour compression cycles is alsopresented Both the methods are illustrated on four case studies where the differ-ence between potential and actual degrees of freedom are explained and related tothe process layout
Appendix A
Optimal operation of a mixedfluid cascade LNG plant
Studies on the operation of complex vapour compression cycles likethe one used for the production of liquefied natural gas (LNG) are notwidely reported in the open literature This is a bit surprising consider-ing the large amount of work that has been put into optimizingthe designof such processes It is important that the process is operated close tooptimum to fully achieve the maximum performance in practice Thereare possibilities for savings both due to (a) identifying the optimal pointof operation and (b) selecting the controlled variables such that the op-timal operation depends weakly on disturbancesIn this paper we study the mixed fluid cascade (MFC) LNG process de-veloped byThe Statoil Linde Technology Alliance We study the degreesof freedom and how to adjust these to achieve optimal steady-state op-eration
A1 Introduction
Large amounts of natural gas (NG) are found at locations that makes it infeasible ornot economical to transport it in gaseous state (in pipelines or as compressed NG)to the customers The most economic way of transporting NG over long distancesis to first produce liquefied natural gas (LNG) and then transport the LNG by shipsAt atmospheric pressure LNG has approximately 600 times the density of gaseousNG
At atmospheric pressure LNG has a temperature of approximatelyminus162C so
149
150 Optimal operation of a mixed fluid cascade LNG plant
the process of cooling and condensing the NG requires large amounts of energySeveral different process designs are used and they can be grouped roughly asfollows
bull Mixed refrigerant The refrigerant composition is adjusted to match thecooling curve of NG Some are designed with a separate pre-cooling cycle
bull Cascade process (pure fluid) Several refrigerant cycles are used to limit themean temperature difference in the heat exchange
bull Mixed fluid cascade process Energy efficiency is further improved byusingseveral mixed refrigerant cycles
The process considered in this paper is the Mixed Fluid Cascade (MFC) processdeveloped byThe Statoil Linde Technology Alliance(Bach 2002) The MFC pro-cess has three different cycles all with mixed refrigerant and the firstcycle withtwo pressure levels
The steady-state model for this plant is implemented in gPROMS (nd) result-ing in approximately 14000 equations Optimizing the plant takes in the orderof 2 hours on a Pentium 4 computer with 28 GHz and 512 MB RAM runningGNULinux
A2 Process description
A simplified flowsheet is given in FigureA1 For more details about the processconsultBach(2002) andForg et al(1999)
Nominal conditions
bull Feed NG enters withP = 615bar andT = 11C after pretreatment Thecomposition is 888 methane 57 ethane 275 propane and 275nitrogen Nominal flow rate is 1kmols-1
bull Product LNG is atP = 551bar andT = minus155C
bull The refrigerants are a mix of nitrogen (N2) methane (C1) ethane (C2) andpropane (C3) and the compositions are used in optimization
bull The refrigerant vapour to the compressors are super-heated 10C
bull The refrigerants are cooled to 11C in all sea water (SW) coolers (assumedmaximum cooling)
A3 Degree of freedom analysis 151
NG
LC
SC
PC1 PC2
LNG
PCHX1 PCHX2 LCHX SCHX
Ph
Ph
Ph
Pl
Pl
Pl
Pm
Pm
SWSW
SW
SW
Figure A1 Simplified flowsheet of the MFC process SC - sub-cooling cycleLC - liquefaction cycle PC - pre-cooling cycle (two stages 1 and 2) Degrees offreedom associated with variable active charge in each cycle are not shown
bull Pressure drops are 05bar in SW coolers 05bar for hot flows in main heatexchangers and 02bar for cold refrigerant in main heat exchangers
The SRK equation of state is used both for NG and the refrigerants The heatexchangers are distributed models with constant heat transfer coefficients Thecompressors are isentropic with 90 constant efficiencies
A3 Degree of freedom analysis
In this section we present a detailed degree of freedom analysis which is an impor-tant result of this work
In a single simple vapour compression cycle (eg a home refrigerator) there aretwo obvious manipulated inputs namely the compressor and the valvelowast In addi-tion there is a less obvious manipulated variable This is the ldquoactive chargerdquoin thecycle which may be modified by introducing a unit (tank) with variable holdup(Jensen and Skogestad 2007) The active charge may be changed by placing tanks
lowastIn addition one might control flow of hot and cold fluid but this is outside thecycle so let usoverlook that for now
152 Optimal operation of a mixed fluid cascade LNG plant
at many different locations but from a simple mass balance it may be verifiedthatfor each cycle one may have only one independent variable (tank) associated withthe active charge Thus for the cycles the number of manipulated variablesare thenumber of compressors and valves plus one active charge for each cycle
Let us now look at the MFC process
A31 Manipulated variables (MVrsquos)
From the discussion above we find that there are in total 26 manipulated variables(degrees of freedom)
bull 5 Compressor powersWsi
bull 4 Choke valve openingszi
bull 4 SW flows in coolers
bull 1 NG flow (can also be considered a disturbance)
bull 9 Composition of three refrigerants
bull 3 active charges (one for each cycle)
A32 Constraints during operation
There are some constraints that must be satisfied during operation
bull Super-heating The vapour entering the compressors must bege 10C super-heated
bull ToutLNG NG Temperature out of SCHX must beleminus155C or colder
bull Pressure 2barge Ple 60bar
bull NG temperature after PCHX1 and PCHX2 (not considered in this paper)
bull Compressor outlet temperature (not considered in this paper)
A33 Active constraints
We are able to identify some constraints that will be active at optimum In totalthere are 11 active constraints
A4 Optimization results 153
bull Excess cooling is costly soToutLNG = minus155C
bull Optimal with low pressure in cycles soPl = 2bar (for all 3 cycles)
bull Maximum cooling AssumeT = 11C at 4 locations
A34 Unconstrained degrees of freedom
After using 11 of the 26 manipulated inputs to satisfy active constraints we areleft with 15 unconstrained degrees of freedom In this work we considerthe NGflow given from elsewhere (disturbance to the process) In addition weassumethat the degree of super-heating is controlled at∆Tsup= 10C so we are left with13 degrees of freedom in optimization For a steady state analysis the pairingofinputs and outputs is insignificant so say we are left with the following subsetofthe MVrsquos
bull 3 NG temperatures (after PCHX1 PCHX2 and LCHX)
bull Pm in SC
bull 9 Refrigerant compositions
In this paper we will not consider manipulating refrigerant composition in opera-tion (only in the optimization) so of the 13 unconstrained degrees of freedom weare left with 4 during operation
A4 Optimization results
In this section we are optimizing on the 13 degrees of freedom given abovetolocate the optimal operation of a given MFC LNG plant The resulting temperatureprofiles for the four main heat exchangers are given in FigureA2 Some key valuesof the refrigerant cycles are given in TableA1 where the nomenclature is given inFigureA1
Some remarks
bull The total shaft work is 10896MW
bull The optimal NG temperature out of PCHX1 PCHX2 and LCHX isminus173Cminus515C andminus771C respectively
154 Optimal operation of a mixed fluid cascade LNG plant
middot - middot LIQ- - - PREmdash- NG SUBmdash- PREcold
Tem
pera
ture
[K]
0 50 100 150250
255
260
265
270
275
280
285
(a) NG1A
middot - middot LIQ- - - PREmdash NG SUBmdash PREcold
0 50 100 150215220225230235240245250255260
(b) NG1B
mdash NGmiddot - middot LIQ SUBmdash LIQcold
Position
Tem
pera
ture
[K]
0 50 100 150185
190
195
200
205
210
215
220
225
(c) NG2
mdash NG SUBmdash SUBcold
Position0 50 100 150
120130140150160170180190200
(d) NG3
Figure A2 Temperature profiles
bull In the true design there will separators at the high pressure side of the cycleswhich has not been considered here Further work will include an analysisof the effect of this sub-optimal design
bull In the SC cycle the pressure ratios over the two compressor stages are farfrom equal (which is a rule of thumb for compression ratios) This is becausethe inlet temperature to the first stage (approximatelyminus80C) is much lowerthan inlet temperature to the second stage (11C)
bull Nitrogen is present in SC only to satisfy the minimum pressure of 2bar
A5 Control structure design 155
Table A1 Optimal operation of a MFC processPC1 PC2 LC SC
Pl [bar] 645 200 200 200Pm[bar] 645 - 2838Ph [bar] 1503 1503 2058 5699C1 [] 000 000 402 5299C2 [] 3770 3770 8296 4245C3 [] 6230 6230 1302 000N2 [] 000 000 000 455n[mols-1] 464 685 390 627Ws[MW] 12565 + 2644 2128 3780+1086
A5 Control structure design
NG LNG
PCHX1 PCHX2 LCHX SCHX
SWSW
SW
SW
TC
TCTCTC
PC
SHSHSH
SH
Figure A3 Suggested control structure for the MFC process SH is degree ofsuper-heating controllers PC and TC are pressure and temperature controllers re-spectively Not shown Three pressure controllers on the low pressure side usingthe active charge in each cycle
In the section above we where able to identify the optimum for the process buthow should this optimum be implemented in practice First we need to control theactive constraints
156 Optimal operation of a mixed fluid cascade LNG plant
bull Pl is for each of the 3 cycles For this we may use ldquoactive chargerdquo (seediscussion above)
bull Maximum cooling in 4 SW coolers SW flow at maximum
bull LNG outlet temperature atminus155C May use first compressor stage in SUB
In addition we choose to control
bull Degree of super-heating (4 locations) For this we may use the correspond-ing choke valve opening
The four remaining degrees of freedom should be used to control variables whichhave good self optimizing properties
ldquoSelf optimizing control is when we can achieve acceptable loss with constantsetpoint values for the controlled variables (without the need to re-optimize whendisturbances occur)rdquo (Skogestad 2000)
To evaluate the loss one needs to consider the effect of disturbances and implemen-tation errors A steady-state analysis is usually sufficient because the economicsare primarily determined by the steady-state
Based on physical insight the following four variables may been suggested
bull ToutNG1A
bull ToutNG1B
bull ToutNG2
bull Pm
A possible control structure with these four variables and the active constraintscontrolled is shown in FigureA3 However note that the ldquopairingsrdquo of controlledand manipulated inputs are included primarily to illustrate that we have availabledegrees of freedom as this does not matter for evaluating self-optimizing controlat steady-state It will be the subject of future work to compare this choiceofcontrolled variables with one that follows from a systematic procedure
A6 Conclusion
We have shown that the degrees of freedom in vapour compression cycles are equalto the number of compressors and valves plus one The extra degree of freedom is
Bibliography 157
related to the ldquoactive chargerdquo in the system and a tank with variable holdup shouldbe included to gain this degree of freedom
A detailed degree of freedom analysis for the MFC process reveals thatthere arefour unconstrained degrees of freedom in operation (not considering manipulatingrefrigerant compositions) To fully achieve the potentially high thermodynamicefficiency of the MFC process it is important that these four unconstrained degreesof freedom are utilized optimally
Bibliography
(nd)httpwwwpsenterprisecomproducts_gpromshtml
Bach W A (2002) lsquoDevelopments in the mixed fluid cascade process (MFCP)for LNG baseload plantsrsquoReports on science and technology Linde63
Forg W Bach W Stockmann R Heiersted R S Paurola P and FredheimA O (1999) lsquoA new LNG baseload process and manufacturing of the mainheat exchangerrsquoReports on science and technology Linde61
Jensen J B and Skogestad S (2007) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Skogestad S (2000) lsquoPlantwide control the search for the self-optimizing controlstructurersquoJ Process Contr10(5) 487ndash507
158 Bibliography
Appendix B
On the Trade-off between EnergyConsumption and Food QualityLoss in SupermarketRefrigeration Systems
J Cailowast J B Jensendagger S Skogestaddagger J Stoustruplowast
Submitted for publication in the proceedings of the 2008 American Control Con-ference Seattle USA
This paper studies the trade-off between energy consumption and foodquality loss at varying ambient conditions in supermarket refrigerationsystems Compared with the traditional operation with pressure controla large potential for energy savings without extra loss of food quality isdemonstrated We also show that by utilizing the relativelyslow dynam-ics of the food temperature compared with the air temperature we areable to further lower both the energy consumption and the peak valueof power requirement The Pareto optimal curve is found by off-lineoptimization
lowastAutomation and Control Department of Electronic Systems Aalborg University (AAU) 9220Aalborg Denmark (e-mail jcesaaudk jakobesaaudk)
daggerDepartment of Chemical Engineering Norwegian University of Science and Tech-nology (NTNU) 7491 Trondheim Norway (e-mail jorgenbachemengntnunoskogechemengntnuno)
159
160 Trade-off between Energy Consumption and Food Quality Loss
B1 Introduction
Increasing energy costs and consumer awareness on food productssafety and qual-ity aspects impose a big challenge to food industries and especially to supermar-kets which have direct contacts with consumers A well-designed optimal controlscheme continuously maintaining a commercial refrigeration system at its opti-mum operation condition despite changing environmental conditions will achievean important performance improvement both on energy efficiency and food qual-ity reliability
Many efforts on optimization of cooling systems have been focused on optimizingobjective functions such as overall energy consumption system efficiency capac-ity or wear of the individual components seeJakobsen and Rasmussen(1998)Jakobsen et al(2001) Larsen and Thybo(2004) Leducqa et al(2006) Swens-son(1994) They have proved significant improvements of system performanceunder disturbances while there has been little emphasis on the quality aspectoffoodstuffs inside display cabinets
This paper discusses a dynamic optimization of commercial refrigeration systemsfeaturing a balanced system energy consumption and food quality loss A formerdeveloped quality model of food provides a tool for monitoring and controlling thequality loss during the whole process seeCai et al(2006)
The paper is organized as follows Operation and modelling of a refrigerationsystems is presented in SectionB2 In SectionB3 we introduce the problem for-mulation used for optimization Different optimization schemes and results arepresented in SectionB4 Finally some discussions and conclusions follow in Sec-tion B5 and SectionB6
B2 Process description
A simplified sketch of the process is shown in FigureB1 In the evaporator thereis heat exchange between the air inside the display cabinet and the cold refrigerantgiving a slightly super-heated vapor to the compressor After compression the hotvapor is cooled condensed and slightly sub-cooled in the condenser This slightlysub-cooled liquid is then expanded through the expansion valve giving a cold two-phase mixture
The display cabinet is located inside a store and we assume that the store hasaconstant temperature This is relative true for stores with air-conditioningThe
B2 Process description 161
Eva
pora
tor13
Compressor13
Expansion Valve13
Con
dens
er13
Fan13Fan13
Display cabinet13Outdoors13
amb13T13
C13P13
CF13N13
C13N13E13P13
OD13
EF13N13
cabin13T13
e13Q13 13EF13W13 13
C13W13 13
CF13W13 13
Store13
Food13
s13c13Q13213 13
c13g13Q13213 13
food13T13
Figure B1 Sketch of a simplified supermarket refrigeration system studiedin thispaper
condenser and fans are located at the roof of the store Condensationis achievedby heat exchange with ambient air
B21 Degree of freedom analysis
There are 5 degrees of freedom (input) in a general simple refrigeration systemseeJensen and Skogestad(2007) Four of these can be recognized in FigureB1as the compressor speed (NC) condenser fan speed (NCF) evaporator fan speed(NEF) and opening degree of the expansion valve (OD) The fifth one is related tothe active charge in the system
Two of the inputs are already used for control or are otherwise constrained
bull Constant super-heating (∆Tsup = 3C) This is controlled by adjusting theopening degree (OD) of the expansion valve
bull Constant sub-cooling (∆Tsub = 2C) We assume that the condenser is de-signed to give a constant degree of sub-cooling which by design consumesthe degree of freedom related to active charge seeJensen and Skogestad(2007)
So we are left with three unconstrained degrees of freedom that shouldbe used tooptimize the operation These are
1 Compressor speedNC
2 Condenser fan speedNCF
3 Evaporator fan speedNEF
These inputs are controlling three variables
1 Evaporating pressurePE
162 Trade-off between Energy Consumption and Food Quality Loss
2 Condensing pressurePC
3 Cabinet temperatureTcabin
However the setpoints for these three variables may be used as manipulatedinputsin our study so the number of degrees of freedom is still three
B22 Mathematical model
The model equations are given in TableB1 please seeLarsen(2005) for the mod-elling of refrigeration systems We assume that the refrigerator has fast dynamicscompared with the display cabinet and food so for the condenser evaporator valveand compressor we have assumed steady-state For the display cabinet and foodwe use a dynamic model as this is where the slow and important (for economics)dynamics will be The food is lumped into one mass and the air inside the cabinettogether with walls are lumped into one mass The main point is that there are twoheat capacities in series For the case with constant display cabinet temperaturewe will also have constant food temperature There are then no dynamics and wemay use steady-state optimization
Some data for the simulations are given in TableB2 please seeLarsen(2005) forfurther data
B23 Influence of setpoints on energy consumption
As stated above this system has three setpoints that may be manipulatedPC PE
andTcabin In FigureB2 surface shows that under 2 different cabinet tempera-tures the variation of energy consumption with varyingPC andPE PointA is theoptimum for cabinet temperatureTcabin1 and pointB is the optimum forTcabin2Tcabin1 is lower thanTcabin2 so the energy consumption is higher in pointA than inpointB
B24 Influence of setpoint on food quality
Food quality decay is determined by its composition factors and many environ-mental factors such as temperature relative humidity light etc Of all the envi-ronmental factors temperature is the most important since it not only stronglyaffects reaction rates but is also directly imposed to the food externally Theotherfactors are at least to some extent controlled by food packaging
B2 Process description 163
Table B1 Model equationsCompressor
WC =mre f middot(his(PePc)minushoe(Pe))
ηis
hic =1minus fq
ηismiddot (his(PePc)minushoe(Pe))+hoe(Pe)
mre f = NC middotVd middotηvol middotρre f (Pe)CondenserWCF = K1CF middot (NCF)3
mairC = K2CF middotNCF
Taoc = Tc +(TambminusTc) middotexp(
minus(αC middot mmCairC)(mairC middotCpair)
)
0 = mre f middot (hic(PePc)minushoc(Pc))minus mairC middotCpair middot (TaocminusTamb)EvaporatorWEF = K1EF middot (NEF)3
mairE = K2EF middotNEF
Taoe= Te+(TcabinminusTe) middotexp(
minus(αE middot mmEairE)(mairE middotCpair)
)
0 = Qeminus mairE middotCpair middot (TcabinminusTaoe)Display cabinetQc2 f = UAc2 f middot (TcabinminusTf ood)Qs2c = UAs2c middot (TstoreminusTcabin)dTf ood
dt = (mCpf ood)minus1 middot Qc2 f
dTcabindt = (mCpcabin)
minus1 middot (minusQc2 f minus QE + Qs2c)
Qf oodloss=int t ft0 100middotDTre f exp(Tf oodminusTre f
z )dt
Table B2 Some data used in the simulationDisplay cabinetlowast
heat transfer areaUAs2c= 160WK-1
heat capacitymCpcabin= 10kJK-1
Foodheat transfer areaUAc2f = 200WKminus1
heat capacitymCpfood = 756kJK-1
quality parameterDTre f = 02day-1quality parameterTre f = 0Cquality parameterZ = 10C
164 Trade-off between Energy Consumption and Food Quality Loss
15
2
25
3
35
456789101
15
2
25
PE (bar)
PC
(bar)
Wto
t (K
W)
A
B
Tcabin 1
Tcabin 2
Figure B2 Energy consumption under different setpoints
Here we focus on the temperature influence to food qualityQfood The only set-point directly influencing food temperature (and thus food quality) isTcabin FigB3 shows the daily quality loss for chilled cod product under 4 casesTfood of2 1C andTsin Tsin1 andTsin2 are the sinusoidal function with mean value of1C amplitude of 1C and 3C respectively period is 24h Note that the qualityloss is higher with higher temperature but there is only minor extra loss over 24hby using a sinusoidal temperature with small amplitude A sinusoidal with largeamplitude has a larger influence on quality due to the non-linearity of the qualityfunction it will not be considered here
0 6 12 18 240
5
10
15
20
25
30
35
Time [hr]
Qua
lity
loss
[]
Tfood
=2 C
Tfood
=Tsin2
Tfood
=Tsin1
Tfood
=1 C
Figure B3 Fresh fish quality loss when stored at different temperatures
B3 Problem formulation
We here consider at a time horizon of three days ambient temperature (Tamb) fol-lows a sinusoidal function with a mean value of 20C period of 24 hours andamplitude of 6C This is a normal temperature profile in Denmark during sum-mer seeDanmarks Meteorologiske Institut(2007)
B3 Problem formulation 165
The objective is to minimize the energy consumption subject to maintaining afixed quality loss by using those 3 DOF This can be formulated mathematicallyas
min(NC(t)NCF(t)NEF(t))
J (B1)
where J =int t f
t0(WC(t)+WCF(t)+WEF(t)︸ ︷︷ ︸
Wtot(t)
)dt (B2)
The quality loss of the food could be included in the objective function directlybut we choose to limit it by using constraints The optimization is also subjected toother constraints such as maximum speed of fans and compressor minimum andmaximum value of evaporator and condenser pressure etc
In this paper the food is a fresh cod product Danish food authorities requireit to be kept at a maximum of 2C The control engineer will normally set thetemperature setpoint a little lower for example at 1C
Case 1Traditional operation with constant pressures (PE) (PC) and constanttemperatures (Tcabin= Tfood = 1C)
There are usually large variations in the ambient temperature during the yearso intraditional operation it is necessary to be conservative when choosing the setpointfor condenser pressure To reduce this conservativeness it is common to use onevalue for summer and one for winter We will here assume that the summer settingis used
To get a fair comparison with traditional control which operates at 1C we willillustrate our optimization by considering the following cases
Case 2Tcabin andTfood constant at 1CTwo remaining unconstrained degrees of freedom as functions of time areused for minimizing the energy consumption inB1
Case 3Tfood = 1t fminust0
int t ft0 Tfood(t)dt = 1C
Three remaining unconstrained degrees of freedom as functions of time areused for minimizing the energy consumption inB1
Case 4Qfood loss(t f ) le 755Three remaining unconstrained degrees of freedom as functions of time areused for minimizing the energy consumption inB1 755 is the qualityloss at constant temperature of 1C obtained in cases 1 and 2
166 Trade-off between Energy Consumption and Food Quality Loss
B4 Optimization
B41 Optimization
The model is implemented ingPROMSr and the optimization is done by dynamicoptimization (except for Case 1) For the Case 2 we have used piecewiselinearmanipulated variables with a discretisation every hour For the cases with varyingcabinet temperature (Case 3 and 4) we have used sinusoidal functionsu= u0+Amiddotsin(π middot t24+ φ) whereu0 is the nominal inputA is the amplitude of the inputtis the time andφ is the phase shift of the input
Using a sinusoidal function has several advantages
bull There are much fewer variables to optimize on only 3 for each input com-pared with 3 parameters for each time interval for discrete dynamic opti-mization
bull There are no end-effects
In all cases we find that the phase shift is very small
B42 Optimization results
TableB3 compares the four cases in terms of the overall costJ end quality lossmaximum total power (Wtotmax) and maximum compressor power (WCmax) Thetwo latter variables might be important if there are restrictions on the maximumcompressor power or on the total electric power consumption
Some key variables including speed and energy consumption for compressor andfans as well as temperatures are plotted for each case in FigureB5 through FigureB8
Table B3 Traditional operation and optimal operation for three differentcon-straints
Case 1lowast Case 2dagger Case 3Dagger Case 4sect
J [MJ] 2737 2428 2407 2414Qf oodloss(t f ) [] 755 755 761 755WCmax[W] 955 1022 836 879Wtotmax[W] 1233 1136 946 981
For Case 1 (traditional operation) the total energy consumption over threedays is2737MJ Note that the condenser temperature (and pressure) is not changing withtime
B4 Optimization 167
If we keepTcabin = Tfood constant at 1C but allow the pressures (and temper-atures) in the condenser and evaporator to change with time (Case 2) we mayreduce the total energy consumption by 113 to 2428MJ Fig B6 shows thatthe evaporator temperature is constant because we still control the cabinet tem-perature while the condenser temperature varies with ambient temperatureThequality is the same as in Case 1 because of the constant cabinet temperatureThepower variations are larger but nevertheless the maximum total power (Wtotmax)is reduced by 79 to 1136W
Next we also allow the cabinet temperature to vary but add a constraint ontheaverage food temperaturesT food = 10C (Case 3) This reduces the total energyconsumption with another 09 while the food quality loss is slightly higherNote from FigureB7 that the evaporator cabinet and food temperature is varyinga lot
Finally in Case 4 we do not care about the average food temperature but insteadrestrict the quality loss WithQfoodloss(t f )le 755 which is the same end qualitywe obtained for Case 1 we save 118 energy compared with Case 1 but useslightly more than for Case 3 (029) Note from FigureB8 that the amplitudefor food cabinet and evaporator temperature are slightly reduced compared to Case3
An important conclusion is that most of the benefit in terms of energy savingsisobtained by letting the setpoint forPE andPC vary (Case 2) The extra savings bychanging also the cabinet temperatureTcabin (Case 3 and 4) are small Howeverthe peak value for compressor power and total system power is significantly de-creased for Case 3 and 4 This is also very important because a lower compressorcapacity means a lower investment cost and a lower peak value of total powerconsumption will further reduce the bill for supermarket owner according to thefollowing formula
Cop =int year
month(Pel(t) middotEel(t)+max(Pel(t)) middotEeldem(t))dt (B3)
whereCop is the operating costEel is the electricity ratePel is the electric powerEeldemis the electricity demand charge max(Pel(t)) is the maximum electric powerduring one month
B43 Trade-off between energy consumption and food qualityloss
Fig B4 plots the Pareto optimal curve between food quality loss and energyconsumption It shows that reducing quality loss and saving energy is a conflicting
168 Trade-off between Energy Consumption and Food Quality Loss
objective to a system An acceptable tradeoff between these two goals can beselected by picking a point somewhere along the line It also shows that Case 1 isfar away from optimization Case 4 is one optimal point while Case 2 and 3 arenear optimal solutions
75 80 85 90 95 100 105 110 11514
16
18
20
22
24
26
28
Daily Energy Consumption [MJ]
Dai
ly Q
ualit
y Lo
ss [
]
Case 1
Case 4
Case 3 Case 2
Pareto Optimal Curve
Figure B4 Optimization between food quality loss and energy consumption
B5 Discussion
Having oscillations in the pressures will impose stress and cause wear on theequip-ment This might not be desirable in many cases but in this study the oscillationsare with a period of one day so this should not be an issue
Experiments on the influence of fluctuating temperatures on food quality werereviewed byUlrich (1981) where marginal reduction in final quality due to fluc-tuations was reported In our case food temperature is only slowly varying andwith an amplitude of less than 1C Thus this will not pose any negative influenceon food quality
B6 Conclusion
We have shown that traditional operation where the pressures are constant givesexcessive energy consumption Allowing for varying pressure in the evaporatorand condenser reduces the total energy consumption by about 11 Varying foodtemperature gives only minor extra improvements in terms of energy consumptionbut the peak value of the total power consumption is reduced with an additional14 for the same food quality loss
Reducing quality loss and saving energy is a conflicting objective Our optimiza-tion result will help the engineer to select an acceptable tradeoff between these two
Bibliography 169
NC
NCF
NEF
Time [h]
N[r
ads-1
]
12 24 36 48 60 720
10
20
30
40
50
60
(a) Speed
Time [h]
W[W
]
Wtot
WC
WEFWCF12 24 36 48 60 720
200400600800
100012001400
(b) Power
Time [h]
T[
C]
Tcon
Tamb
12 24 36 48 60 7210
20
30
40
(c) High temperatures
Time [h]T
[C
]
Tevap
Tfood
Tcabin
12 24 36 48 60 72-8
-6
-4
-2
0
2
4
(d) Low temperatures
Figure B5 Traditional operation withTcabin= 1C PE = 24bar andPC = 80bar(Case 1)
goals by picking a point somewhere along the Pareto front line
This paper investigates the potential of finding a balancing point between qualityand energy consumption by open-loop dynamic optimizations It uses the sinusoidambience temperature as one example In real life weather patterns are not exactlya sinusoidal function but real weather conditions can be easily obtainedin advancefrom forecast Practical implementation including selecting controlled variablesand using closed-loop feedback control will be the theme of future research
Bibliography
Cai J Risum J and Thybo C (2006) Quality model of foodstuff in a refrig-erated display cabinet In proc International refrigeration conference PurdueUSA
170 Bibliography
NC
NCFNEF
Time [h]
N[r
ads-1
]
12 24 36 48 60 720
10
20
30
40
50
60
(a) Inputs
Time [h]
W[W
] Wtot
WC
WEFWCF
12 24 36 48 60 720200400600800
100012001400
(b) Power
Time [h]
T[
C] Tcon
Tamb
12 24 36 48 60 7210
20
30
40
(c) High temperatures
Time [h]
T[
C]
Tevap
Tfood
Tcabin
12 24 36 48 60 72-8
-6
-4
-2
0
2
4
(d) Low temperatures
Figure B6 Optimal operation forTcabin= 1C (Case 2)
Danmarks Meteorologiske Institut(2007)URL wwwdmidk
Jakobsen A and Rasmussen B (1998) Energy-optimal speed control of fans andcompressors in a refrigeration system Eurotherm 62 Nancy Francepp pp317ndash323
Jakobsen A Rasmussen B Skovrup M J and Fredsted J (2001) Develope-ment of energy optimal capacity control in refrigeration systems In procIn-ternational refrigeration conference Purdue USA
Jensen J B and Skogestad S (2007) lsquoOptimal operation of simple refrigera-tion cycles Part I Degrees of freedom and optimality of sub-coolingrsquoComputChem Eng31 712ndash721
Larsen L F S (2005) Model based control of refrigeration system PhD thesisDepartment of Control Engineering Aalborg University Fredrik Bajers Vej 7CDK 9220 Aalborg Denmark
Bibliography 171
NCNCFNEF
Time [h]
N[r
ads-1
]
12 24 36 48 60 720
10
20
30
40
50
60
(a) Speed
Time [h]
W[W
]
Wtot
WC
WEFWCF
12 24 36 48 60 720200400600800
100012001400
(b) Power
Time [h]
T[
C] Tcon
Tamb
12 24 36 48 60 7210
20
30
40
(c) High temperatures
Time [h]T
[C
]Tevap
Tfood
Tcabin
12 24 36 48 60 72-8
-6
-4
-2
0
2
4
(d) Low temperatures
Figure B7 Optimal operation forT food = 1C (Case 3)
Larsen L F S and Thybo C (2004) Potential energy saving in refrigerationsystem using optimal setpoint In proc Conference on control applicationsTaipei Taiwan
Leducqa D Guilparta J and Trystramb G (2006) lsquoNon-linear predictive controlof a vapour compression cyclersquoInternational Journal of Refrigeration29 761ndash772
Swensson M C (1994) studies on on-line optimizing control with application toa heat pump PhD thesis Department of refrigeration Engineering NTNU
Ulrich R (1981) lsquoVariations de temperature et qualite des produits surgelesrsquoRe-view Generale due Froid71 371 V389
172 Bibliography
NCNCFNEF
Time [h]
N[r
ads-1
]
12 24 36 48 60 720
10
20
30
40
50
60
(a) Speed
Time [h]
W[W
]
Wtot
WC
WEFWCF
12 24 36 48 60 720200400600800
100012001400
(b) Power
Time [h]
T[
C] Tcon
Tamb
12 24 36 48 60 7210
20
30
40
(c) High temperatures
Time [h]
T[
C]
Tevap
Tfood
Tcabin
12 24 36 48 60 72-8
-6
-4
-2
0
2
4
(d) Low temperatures
Figure B8 Optimal operation forQfood le 755 (Case 4)
- Contents
- List of Figures
- List of Tables
- 1 Introduction
-
- 11 Motivation
- 12 Thesis overview
- 13 Publication list
-
- 2 Introduction to vapour compression cycles
-
- 21 Introduction
-
- 211 Fundamentals
- 212 Expansion device
- 213 Condenser
- 214 Evaporator
- 215 Active charge
- 216 Operation of simple cycles
-
- 22 Some design features
-
- 221 Constant versus varying temperature loads
- 222 Two-stage expansion and two pressure levels
- 223 Internal heat exchange
- 224 Cascaded cycles
-
- 23 Conclusion
- Bibliography
-
- 3 Simple cycles Part I
-
- 31 Introduction
- 32 Degrees of freedom in simple cycles
-
- 321 Design versus operation
- 322 Active charge and holdup tanks
- 323 Degrees of freedom for operation
-
- 33 Discussion of some designs
-
- 331 Optimal designs
- 332 Non-optimal designs
-
- 34 Optimality of sub-cooling
-
- 341 Ammonia case study
- 342 Explanation
- 343 Discussion of sub-cooling Why not found before
-
- 35 Discussion
-
- 351 Super-heating by internal heat exchange
- 352 Selection of controlled variable
-
- 36 Conclusion
- Bibliography
-
- 4 Simple cycles Part II
-
- 41 Introduction
- 42 Selection of controlled variable
-
- 421 Linear analysis
- 422 Combination of measurements
-
- 43 Ammonia case study
-
- 431 Modelling
- 432 Optimal steady-state operation
- 433 Selection of controlled variables
-
- 44 CO2 case study
-
- 441 Modelling
- 442 Optimal operation
- 443 Selection of controlled variable
-
- 45 Discussion
-
- 451 Super-heating
- 452 Heat transfer coefficients
- 453 Pressure control
-
- 46 Conclusion
- Bibliography
-