Dynamic Modelling and Advanced Control of Air Conditioning and Refrigeration Systems

University of Illinois at Urbana-Champaign
Air Conditioning and Refrigeration Center A National Science Foundation/University Cooperative Research Center
Dynamic Modeling and Advanced Control

of Air Conditioning and Refrigeration Systems
B. P. Rasmussen and A. G. Alleyne
ACRC TR-244 June 2006
For additional information:
Air Conditioning and Refrigeration Center

University of Illinois
Mechanical & Industrial Engineering Dept.
1206 West Green Street Prepared as part of ACRC Project #163
Urbana, IL 61801 Advanced Dynamic Modeling, Control,
and Diagnostics for Vapor Compression Systems
(217) 333-3115 A. G. Alleyne, Principal Investigator
The Air Conditioning and Refrigeration Center was
founded in 1988 with a grant from the estate of
Richard W. Kritzer, the founder of Peerless of
America Inc. A State of Illinois Technology Challenge
Grant helped build the laboratory facilities. The
ACRC receives continuing support from the Richard
W. Kritzer Endowment and the National Science
Foundation. The following organizations have also
become sponsors of the Center.
Arçelik A. S.
Behr GmbH and Co.
Carrier Corporation
Cerro Flow Products, Inc.
Copeland Corporation
Daikin Industries, Ltd.
Danfoss A/S
Delphi Thermal and Interior
Embraco S. A.
General Motors Corporation
Hill PHOENIX
Hydro Aluminum Precision Tubing
Ingersoll-Rand/Climate Control
Lennox International, Inc.
LG Electronics, Inc.
Manitowoc Ice, Inc.
Modine Manufacturing Co.
Novelis Global Technology Centre
Parker Hannifin Corporation
Peerless of America, Inc.
Samsung Electronics Co., Ltd.
Sanden Corporation
Sanyo Electric Co., Ltd.
Tecumseh Products Company
Trane
Visteon Automotive Systems
Wieland-Werke, AG
For additional information:
Air Conditioning & Refrigeration Center

Mechanical & Industrial Engineering Dept.
University of Illinois
1206 West Green Street
Urbana, IL 61801
217 333 3115

Abstract
Over 15 billion dollars is spent on energy for residential air-conditioning alone each year, and air-
conditioning remains the largest source of peak electrical demand. Improving the efficiency of these systems has the
potential for significant economic and environmental impact, but requires not only refining individual component
designs, but increasing overall system efficiency using advanced control strategies. Transient control of vapor
compression cycles faces two significant challenges: 1) creating control-oriented models that balance simplicity with
accuracy, and capture the complex heat and mass flow dynamics, and 2) developing control strategies that can
achieve high performance over a wide range of operating conditions.
This dissertation makes contributions on both fronts and can be divided into two distinct parts. The first
portion of the dissertation presents the development, simulation, and experimental validation of a first principles
modeling framework that captures the dynamics of a variety of vapor compression cycles in a form amenable to
controller design. These models are highly nonlinear, and require a nonlinear control strategy to attain high
performance over the entire operating envelope. To this end, a gain-scheduled control approach based on local
models and local controllers is presented that uses endogenous scheduling variables. This comprises the second
portion of the dissertation, where a theoretical framework for designing gain scheduled controllers, tools for
analyzing the stability of the nonlinear closed loop system, and experimental evaluation of advanced control
strategies for vapor compression systems is presented. These results demonstrate that while linear control techniques
offer significant advantages versus traditional a/c control systems over small ranges, the gain-scheduled approach
extends these advantages over the entire operating regime.
iii
Table of Contents
Page
Abstract......................................................................................................................... iii
List of Figures ............................................................................................................ viii
List of Tables ............................................................................................................... xv
Nomenclature ............................................................................................................. xvi
Chapter 1. Introduction................................................................................................. 1
1.1 Vapor Compression Systems........................................................................................................ 1
1.2 Control-Oriented Modeling of Vapor Compression Systems .................................................... 3
1.3 Control of Vapor Compression Cycles ........................................................................................ 4
1.4 Gain Scheduling ............................................................................................................................. 5
1.5 Organization of Dissertation ......................................................................................................... 6
Chapter 2. Dynamic Modeling ...................................................................................... 7
2.1 Introduction..................................................................................................................................... 7
2.2 Modeling Paradigms ...................................................................................................................... 8
2.2.1 Lumped Parameter Models.....................................................................................................................8
2.2.2 Discretized Models.................................................................................................................................8
2.2.3 Moving Boundary Models......................................................................................................................9
2.3 Vapor Compression Systems...................................................................................................... 10
2.4 Compressors and Expansion Devices ....................................................................................... 13
2.5 Heat Exchangers .......................................................................................................................... 13
2.5.1 Moving Boundary Approach................................................................................................................13
2.5.2 Linearization.........................................................................................................................................17
2.5.3 Time Scale Separation and Model Reduction ......................................................................................18
2.5.4 Example Derivation: Single Phase Heat Exchanger.............................................................................20
2.6 Component Models ...................................................................................................................... 26
2.6.1 Compressor...........................................................................................................................................26
2.6.2 Orifice Expansion Valve ......................................................................................................................28
2.6.3 Automatic Expansion Valve.................................................................................................................29
2.6.4 Thermal Expansion Valve ....................................................................................................................30
2.6.5 Electronic Expansion Valve .................................................................................................................32
2.6.6 Single Phase Heat Exchanger...............................................................................................................33
2.6.7 Evaporator ............................................................................................................................................39
2.6.8 Condenser.............................................................................................................................................47
2.6.9 Evaporator with Receiver .....................................................................................................................58
2.6.10 Condenser with Receiver....................................................................................................................65
2.6.11 Internal Heat Exchanger .....................................................................................................................75
Chapter 3. System Simulation.................................................................................... 79
3.1 Software Introduction .................................................................................................................. 79
3.2 Library Structure .......................................................................................................................... 79
iv
3.2.1 Auxiliary Tools ....................................................................................................................................80
3.2.2 Components..........................................................................................................................................81
3.2.3 Fluid Properties ....................................................................................................................................82
3.3 Component Models ...................................................................................................................... 83
3.3.1 Component GUIs..................................................................................................................................83
3.3.2 Simulink® Block Diagram....................................................................................................................88
3.3.3 S-function .............................................................................................................................................89
3.4 System Models ............................................................................................................................. 90
3.5 Fluid Properties ............................................................................................................................ 91
3.5.1 Calculation ...........................................................................................................................................92
3.6 Library Limitations ....................................................................................................................... 92
3.7 Sample Systems ........................................................................................................................... 93
Chapter 4. Experimental System ............................................................................... 94
4.1 General System Description........................................................................................................ 94
4.2 Sensors.......................................................................................................................................... 98
4.2.1 Temperature Measurement...................................................................................................................99
4.2.2 Pressure Measurement........................................................................................................................100
4.2.3 Mass Flow Measurement....................................................................................................................100
4.2.4 Power Measurement ...........................................................................................................................101
4.3 Actuators ..................................................................................................................................... 102
4.3.1 Compressor.........................................................................................................................................102
4.3.2 Heat Exchanger Fans..........................................................................................................................106
4.3.3 Electronic Expansion Valve ...............................................................................................................108
4.3.4 Thermostatic Expansion Valve...........................................................................................................112
4.3.5 Capillary Tube Expansion ..................................................................................................................114
4.3.6 Automatic Expansion Valve...............................................................................................................115
4.3.7 Evaporator Pressure Regulating Valve...............................................................................................117
4.4 Data Acquisition ......................................................................................................................... 118
4.5 Heat Exchangers and Auxiliary Components.......................................................................... 120
4.5.1 Evaporator ..........................................................................................................................................120
4.5.2 Condenser...........................................................................................................................................122
4.5.3 Internal Heat Exchanger .....................................................................................................................124
4.5.4 Auxiliary Components .......................................................................................................................124
Chapter 5. System Identification and Model Validation ......................................... 129
5.1 System Dynamics....................................................................................................................... 129
5.1.1 Evaporator #1 .....................................................................................................................................130
5.1.2 Evaporator #2 .....................................................................................................................................134
5.2 System Identification ................................................................................................................. 142
5.2.1 Background and Identification Approach...........................................................................................142
5.2.2 System Identification Results: Evaporator #1 ....................................................................................148
5.2.3 System Identification Results: Evaporator #2 ....................................................................................153
5.3 First Principles Model Validation .............................................................................................. 157
v
5.3.1 Expansion Valve.................................................................................................................................158
5.3.2 Compressor.........................................................................................................................................159
5.3.3 Evaporator ..........................................................................................................................................160
5.3.4 Condenser with Receiver....................................................................................................................162
5.3.5 System Validation ..............................................................................................................................163
5.4 Summary ..................................................................................................................................... 167
Chapter 6. Gain-Scheduled Control......................................................................... 168
6.1 Introduction................................................................................................................................. 168
6.2 Problem Formulation ................................................................................................................. 169
6.2.1 Plant Representation...........................................................................................................................169
6.2.2 Controller Design ...............................................................................................................................171
6.2.3 Scheduling Variables and Interpolation Methods...............................................................................172
6.2.4 Closed Loop Representation...............................................................................................................174
6.3 Stability Analysis........................................................................................................................ 175
6.3.1 Stability Characterizations..................................................................................................................175
6.3.2 Stability Analysis via Linear Matrix Inequalities ...............................................................................178
6.3.3 Stability Analysis via LMIs................................................................................................................183
6.4 Youla-based Gain-Scheduling................................................................................................... 184
6.4.1 Youla Parameterization ......................................................................................................................184
6.4.2 Q-Blending .........................................................................................................................................187
6.5 Design Aspects........................................................................................................................... 192
6.6 Illustrative Example: Nonlinear Mass-Spring-Damper System.............................................. 193
6.6.1 Nonlinear Plant Model .......................................................................................................................193
6.6.2 Nonlinear Control Design ..................................................................................................................196
6.6.3 Stability Analysis ...............................................................................................................................199
6.6.4 Rate Limit Bound on Reference Changes ..........................................................................................203
6.7 Summary ..................................................................................................................................... 206
Chapter 7. Control Design and Implementation ..................................................... 207
7.1 Introduction and Background ................................................................................................... 207
7.1.1 Environmental Control .......................................................................................................................207
7.1.2 Vapor Compression Cycle Control.....................................................................................................208
7.2 Expansion Valve Control ........................................................................................................... 209
7.2.1 Superheat Regulation Step Response Tests........................................................................................209
7.2.2 FTP Automotive Drive Cycle.............................................................................................................211
7.3 Coordinated Compressor-Expansion Valve Control .............................................................. 212
7.3.1 MIMO Control Design .......................................................................................................................214
7.3.2 Gain-Scheduling.................................................................................................................................222
7.4 Summary ..................................................................................................................................... 224
Chapter 8. Conclusions and Future Work.............................................................. 226
8.1 Summary of Research Contributions....................................................................................... 226
8.1.1 Dynamic Modeling and Control of Vapor Compression Cycles ........................................................226
8.1.2 Gain Scheduling .................................................................................................................................226
vi
8.2 Future Work................................................................................................................................. 226
8.2.1 Model Validation and Extensions ......................................................................................................226
8.2.2 Control Research and Fault Detection................................................................................................226
List of References ..................................................................................................... 228
vii
List of Figures
Page
Figure 1.1 System Diagram - Ideal Subcritical Vapor Compression Cycle...................................................................2
Figure 1.2 P-h Diagram - Ideal Subcritical Vapor Compression Cycle.........................................................................2
Figure 1.3 System Diagram - Ideal Transcritical Vapor Compression Cycle................................................................3
Figure 1.4 P-h Diagram - Ideal Transcritical Vapor Compression Cycle......................................................................3
Figure 2.1 System Diagram - Ideal Subcritical Vapor Compression Cycle.................................................................11
Figure 2.2 P-h Diagram - Ideal Subcritical Vapor Compression Cycle.......................................................................11
Figure 2.3 P-h Diagram - Actual Vapor Compression Cycle ......................................................................................11
Figure 2.4 System Diagram - Subcritical Vapor Compression Cycle with Receivers................................................12
Figure 2.5 P-h Diagram - Subcritical Vapor Compression Cycle with Receivers .......................................................12
Figure 2.6 System Diagram - Subcritical Vapor Compression Cycle with Internal Heat Exchanger..........................12
Figure 2.7 P-h Diagram - Subcritical Vapor Compression Cycle with Internal Heat Exchanger................................12
Figure 2.8 Heat Exchanger with Two-Phase and Superheated Vapor Regions ...........................................................13
Figure 2.9 TEV Operation (from [1]) ..........................................................................................................................31
Figure 2.10 TEV Valve (from [59]).............................................................................................................................31
Figure 2.11 Diagram: Evaporator with two fluid regions............................................................................................34
Figure 2.12 Diagram: Evaporator with two fluid regions............................................................................................39
Figure 2.13 Diagram: Condenser with three fluid regions...........................................................................................47
Figure 2.14 Diagram: Evaporator with one fluid region and receiver .........................................................................59
Figure 2.15 Diagram: Condenser with two fluid regions and receiver ........................................................................65
Figure 2.16 Diagram: Internal Counterflow Heat Exchanger......................................................................................75
Figure 3.1 Overview of Thermosys™ Library from the Simulink® Browser.............................................................80
Figure 3.2 Overview of the Auxiliary Tools in the Thermosys™ Toolbox.................................................................81
Figure 3.3 Overview of the Components in the Thermosys™ Toolbox......................................................................82
Figure 3.4 Overview of Fluid Properties in the Thermosys™ Toolbox ......................................................................83
Figure 3.5 Evaporator GUI ..........................................................................................................................................84
Figure 3.6 GUI: Component Name Section.................................................................................................................84
Figure 3.7 GUI: Physical Parameters Section..............................................................................................................85
Figure 3.8 GUI: Operating Condition Section.............................................................................................................85
Figure 3.9 GUI: Heat Transfer Coefficients Section ...................................................................................................86
Figure 3.10 J-factor Correlation ..................................................................................................................................86
Figure 3.11 GUI: Recorded Outputs Section...............................................................................................................86
Figure 3.12 Steady State Solver Convergence Error ...................................................................................................87
Figure 3.13 GUI: Save/Load Profile Section...............................................................................................................87
Figure 3.14 Pressure and Temperature Drop Block.....................................................................................................88
Figure 3.15 Expansion Valve GUI ..............................................................................................................................88
Figure 3.16 Sample Simulink® Block Diagram..........................................................................................................89
Figure 3.17 Integration of S-function Outputs.............................................................................................................90
Figure 3.18 Sample S-function Code...........................................................................................................................90
viii
Figure 3.19 Sample System Simulation Model ...........................................................................................................91
Figure 3.20 Sample Simulink® 2-D Interpolation Table .............................................................................................92
Figure 3.21 Sample Interpolation Table: Temperature as a Function of Pressure and Enthalpy for Carbon
Dioxide ................................................................................................................................................................92
Figure 3.22 Sample Interpolation Table: Specific Heat as a Function of Pressure and Temperature for Carbon
Dioxide ................................................................................................................................................................93
Figure 4.1 Photo of original system.............................................................................................................................94
Figure 4.2 Photo of modified system...........................................................................................................................95
Figure 4.3 Schematic of Experimental System............................................................................................................96
Figure 4.4 Welded Tip Thermocouple for Air Temperature Measurement.................................................................99
Figure 4.5 Welded Tip Thermocouple for Surface Temperature Measurement ..........................................................99
Figure 4.6 Immersion Thermocouple for Refrigerant Temperature Measurement......................................................99
Figure 4.7 Pressure Transducer .................................................................................................................................100
Figure 4.8 Pressure Gauge.........................................................................................................................................100
Figure 4.9 Mass Flow Transducer .............................................................................................................................101
Figure 4.10 Mass Flow Gauge...................................................................................................................................101
Figure 4.11 Power Transducer...................................................................................................................................102
Figure 4.12 Compressor ............................................................................................................................................102
Figure 4.13 Variable Frequency Drive ......................................................................................................................103
Figure 4.14 Line Reactors..........................................................................................................................................103
Figure 4.15 Pressure Switches...................................................................................................................................104
Figure 4.16 Operating Envelope................................................................................................................................104
Figure 4.17 Range of Operating Conditions ..............................................................................................................105
Figure 4.18 Compressor Volumetric Efficiency Map................................................................................................105
Figure 4.19 Compressor Mass Flow Prediction.........................................................................................................106
Figure 4.20 Compressor Adiabatic Efficiency Map ..................................................................................................106
Figure 4.21 Evaporator Fan .......................................................................................................................................107
Figure 4.22 Condenser Fan........................................................................................................................................107
Figure 4.23 Fan Control Boards ................................................................................................................................107
Figure 4.24 Evaporator Velocity Measurement Locations ........................................................................................108
Figure 4.25 Velocity Profile for Evaporator #1 at 100% Fan Speed .........................................................................108
Figure 4.26 Evaporator and Condenser Fan Volumetric Flow Maps ........................................................................108
Figure 4.27 Photo of EEV .........................................................................................................................................109
Figure 4.28 Photo of IB Control Board .....................................................................................................................109
Figure 4.29 EEV #1 Hysteresis .................................................................................................................................109
Figure 4.30 EEV #2 Hysteresis .................................................................................................................................109
Figure 4.31 EEV #1 with Hysteresis Compensation .................................................................................................110
Figure 4.32 EEV Quantization .................................................................................................................................110
Figure 4.33 EEV #1 Effective Discharge Coefficient Map .......................................................................................110
Figure 4.34 EEV #1 Mass Flow Rate Prediction.......................................................................................................111
ix
Figure 4.35 EEV #2 Effective Discharge Coefficient Map .......................................................................................111
Figure 4.36 EEV #2 Mass Flow Rate Prediction.......................................................................................................112
Figure 4.37 Photo of TEV .........................................................................................................................................113
Figure 4.38 Diagram of TEV Operation (from [1]) ...................................................................................................113
Figure 4.39 Photo of Capillary Tube .........................................................................................................................114
Figure 4.40 Capillary Tube Discharge Coefficient Map............................................................................................114
Figure 4.41 Capillary Tube Mass Flow Rate Prediction............................................................................................115
Figure 4.42 Photo of AEV .........................................................................................................................................116
Figure 4.43 Diagram of AEV Construction (from Parker Bulletin 21-02D) .............................................................116
Figure 4.44 AEV Effective Discharge Coefficient Map............................................................................................117
Figure 4.45 AEV Mass Flow Rate Map ....................................................................................................................117
Figure 4.46 Photo of EPR..........................................................................................................................................118
Figure 4.47 Diagram of EPR Construction (from [2])...............................................................................................118
Figure 4.48 Photo of Computer and Data Acquisition...............................................................................................119
Figure 4.49 Data Acquisition Boards ........................................................................................................................119
Figure 4.50 Terminal Boards.....................................................................................................................................119
Figure 4.51 Signal Conditioning Equipment .............................................................................................................120
Figure 4.52 Diagram of the Evaporator Unit (from Larkin Bulletin RI-04) ..............................................................121
Figure 4.53 The Evaporator Unit...............................................................................................................................121
Figure 4.54 Photo of the Evaporator #1.....................................................................................................................121
Figure 4.55 Photo of the Evaporator #2.....................................................................................................................122
Figure 4.56 Schematic of the Evaporator #2 .............................................................................................................122
Figure 4.57 The Condenser Unit ...............................................................................................................................123
Figure 4.58 Schematic of the Condenser (from www.blissfield.com).......................................................................123
Figure 4.59 Diagram of the Internal Heat Exchanger (from Sherwood Product Catalog).........................................124
Figure 4.60 Internal Heat Exchanger.........................................................................................................................124
Figure 4.61 Liquid Line Receiver..............................................................................................................................125
Figure 4.62 Suction Line Accumulator......................................................................................................................125
Figure 4.63 Oil Separator ..........................................................................................................................................126
Figure 4.64 Diagram of Filter/Drier (from Sporlan Bulletin 40-10)..........................................................................126
Figure 4.65 Filter /Drier.............................................................................................................................................127
Figure 4.66 Sight Glass .............................................................................................................................................127
Figure 4.67 Manual Valve .........................................................................................................................................127
Figure 5.1 System Diagram - Subcritical Vapor Compression Cycle with High Side Receiver ...............................130
Figure 5.2 P-h Diagram - Subcritical Vapor Compression Cycle with High Side Receiver......................................130
Figure 5.3 System Dynamics: Actuator Step Changes ..............................................................................................131
Figure 5.4 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Compressor
Speed and Evaporator Superheat .......................................................................................................................132
Figure 5.5 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Compressor
x
Figure 5.6 System Dynamics: Condenser Pressure Response for Different Nominal Values of Compressor
Figure 5.7 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Air Flow Rate ....133
Figure 5.8 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Air Flow Rates .....134
Figure 5.9 System Dynamics: Condenser Pressure Response for Different Nominal Values of Air Flow Rates......134
Figure 5.10 System Dynamics: Actuator Step Changes ............................................................................................135
Figure 5.12 System Dynamics: Evaporator Air Outlet Temperature Response for Different Nominal Values of
Compressor Speed and Evaporator Superheat...................................................................................................136
Figure 5.15 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Evaporator
Air Flow Rate ....................................................................................................................................................138
Evaporator Air Flow Rate..................................................................................................................................138
Figure 5.17 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Evaporator Air
Flow Rate...........................................................................................................................................................139
Figure 5.18 System Dynamics: Condenser Pressure Response for Different Nominal Values of Evaporator Air
Flow Rate...........................................................................................................................................................139
Figure 5.19 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Condenser
Air Flow Rate ....................................................................................................................................................140
Condenser Air Flow Rate ..................................................................................................................................140
Figure 5.21 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Condenser Air
Flow Rate...........................................................................................................................................................141
Figure 5.22 System Dynamics: Condenser Pressure Response for Different Nominal Values of Condenser Air
Flow Rate...........................................................................................................................................................141
Figure 5.23 Identification Inputs ...............................................................................................................................145
Figure 5.24 Identification Results: Model Prediction................................................................................................145
Figure 5.25 Identification Results: Step Responses within 3 Standard Deviations ...................................................146
Figure 5.26 Identification Results: Bode Plot - Temperatures...................................................................................146
Figure 5.27 Identification Results: Bode Plot - Pressures .........................................................................................147
Figure 5.28 Bode Magnitude Plot: Variations in Compressor Speed (Nominal Air Flow Rate, 8º C Evaporator
Superheat)..........................................................................................................................................................150
Superheat)..........................................................................................................................................................150
xi
Figure 5.30 Bode Magnitude Plot: Variations in Evaporator Superheat (Nominal Air Flow Rate, 1200 rpm
Compressor Speed)............................................................................................................................................151
Figure 5.32 Bode Magnitude Plot: Variations in Air Flow Rates (900 rpm Compressor Speed, 16º C Evaporator
Superheat)..........................................................................................................................................................152
Figure 5.33 Bode Magnitude Plot: Variations in Air Flow Rates (1800 rpm Compressor Speed, 16º C
Evaporator Superheat) .......................................................................................................................................152
Superheat)..........................................................................................................................................................155
Superheat)..........................................................................................................................................................155
Figure 5.40 System Model Validation: Actuator Step Changes ................................................................................158
Figure 5.41 Component Model Validation: Expansion Valve Mass Flow Rate ........................................................159
Figure 5.42 Component Model Validation: Compressor Mass Flow Rate ................................................................160
Figure 5.43 Component Model Validation: Condenser Pressure and Air Outlet Temperature .................................163
Figure 5.44 System Model Validation: Mass Flow Rates .........................................................................................165
Figure 5.45 System Model Validation: System Pressures .........................................................................................166
Figure 5.46 System Model Validation: Air Temperatures.........................................................................................166
Figure 5.47 System Model Validation: Evaporator Temperatures ............................................................................167
Figure 6.1 General Nonlinear Closed Loop...............................................................................................................169
Figure 6.2 Controller Blending using a Local Controller Network ...........................................................................173
Figure 6.3 Plant Blending using a Local Model Network .........................................................................................173
Figure 6.4 Diagram of the Combined LCN/LMN .....................................................................................................174
Figure 6.5 Simplified Diagram of the LCN/LMN Interconnection ...........................................................................174
Figure 6.6 Successive Levels of Gain-Scheduled Stability .......................................................................................176
Figure 6.7 Feedback Control Loop ............................................................................................................................185
Figure 6.8 Feedback Loop Depicting the Class of All Stabilizing Controllers..........................................................185
Figure 6.9 Feedback Loop Depicting the Class of All Stabilizable Plants ................................................................185
Figure 6.10 Feedback Loop Depicting the Dual Youla Parameterization .................................................................185
Figure 6.11 Feedback Loop Depicting the Dual Youla Parameterization .................................................................187
Figure 6.12 Feedback Loop Depicting the Equivalent Stability Condition ...............................................................187
xii
Figure 6.13 Feedback Loop Depicting a Local Controller Network .........................................................................188
Figure 6.14 Feedback Loop Depicting a Local Q-Network.......................................................................................188
Figure 6.15 Feedback Loop Depicting the Local S-Network ....................................................................................189
Figure 6.16 Feedback Loop Depicting the LQN/LSN Interconnection .....................................................................189
Figure 6.17 Feedback Loop Depicting the LQN/LSN Interconnection .....................................................................191
Figure 6.18 Simplified LQN/LSN Interconnection ...................................................................................................191
Figure 6.19 Exponential Weighting Function............................................................................................................192
Figure 6.20 Quadratic Weighting Function ...............................................................................................................192
Figure 6.21 Nonlinear Mass-Spring-Damper System................................................................................................193
Figure 6.22 Root Locus for Values of y with Five Design Points .............................................................................194
Figure 6.23 Step Response Plots for Five Local Linear Models................................................................................194
Figure 6.24 Bode Plots for Five Local Linear Models ..............................................................................................195
Figure 6.25 Weighting Functions ..............................................................................................................................195
Figure 6.26 Bode Plots for Five S Models (P0 = P3) .................................................................................................196
Figure 6.27 Simulated Comparison of Nonlinear Plant, LMN, and LSN..................................................................196
Figure 6.28 Bode Plots for Five Local Linear Controllers ........................................................................................197
Figure 6.29 Closed Loop Step Response for Five Local Linear Controllers .............................................................197
Figure 6.30 Simulated Comparison of Local Controllers and LCN ..........................................................................198
Figure 6.31 Simulated Comparison of Local Controllers and LCN (Detailed View)................................................198
Figure 6.32 Controller Weighting Functions .............................................................................................................198
Figure 6.33 Bode Plots for Five Q Controllers (K0 = K3)..........................................................................................199
Figure 6.34 Simulated Comparison of LCN and LQN ..............................................................................................199
Figure 6.35 Comparison of Performance Bounds for Additional Design Points.......................................................200
Figure 6.36 Comparison of Performance Bounds for Different Choices of Nominal Plant/Controller in the
LQN/LSN Framework.......................................................................................................................................200
Figure 6.37 Comparison of H∞ Performance Bounds for LCN and LQN..................................................................201
Figure 6.38 Comparison of H2 Performance Bounds for LCN and LQN ..................................................................201
Figure 6.39 Comparison of Peak-to-Peak Performance Bounds for LCN and LQN .................................................202
Figure 6.40 Depiction of the One-Dimensional Scheduling Space with Prescribed Region of Stability ..................204
Figure 6.41 Depiction of the Two-Dimensional Scheduling Space with Prescribed Region of Stability..................204
Figure 6.42 Simulated Transition in Operating Condition using Rate Limited Reference Signal.............................205
Figure 7.1 TEV Regulation Evaporator Superheat ....................................................................................................210
Figure 7.2 EEV Regulation Evaporator Superheat ....................................................................................................210
Figure 7.3 Vehicle Speed for the SFTP-SCO3 Test ..................................................................................................211
Figure 7.4 Compressor Speed for the SFTP-SCO3 Test (Assuming the Transmission of a Small-Sized Vehicle)...211
Figure 7.5 Evaporator Superheat Regulation for SCO3 Drive Cycle ........................................................................212
Figure 7.6 Partial Load Efficiency.............................................................................................................................214
Figure 7.7 Evaporator Pressure Regulation ...............................................................................................................215
Figure 7.8 Evaporator Pressure Tracking ..................................................................................................................215
Figure 7.9 Dual SISO Control Interaction .................................................................................................................216
xiii
Figure 7.10 The General Control Problem ................................................................................................................217
Figure 7.11 Standard H∞ Control Formulation ..........................................................................................................217
Figure 7.12 Standard Formulation with Frequency Weighting .................................................................................217
Figure 7.13 Performance Frequency Weighting........................................................................................................218
Figure 7.14 Control Effort Frequency Weighting......................................................................................................219
Figure 7.15 Magnitude Response of the Plant Models ..............................................................................................220
Figure 7.16 Magnitude Response of the H∞ Controllers............................................................................................220
Figure 7.17 Singular Values of the H∞ Controllers....................................................................................................221
Figure 7.18 MIMO Control at Low Capacity ............................................................................................................221
Figure 7.19 MIMO Control at High Capacity ...........................................................................................................222
Figure 7.20 Gain Scheduled Controller Transitioning Between Design Points and Off-Design Points ....................223
Figure 7.21 Decomposition of the Controllers, J K ..................................................................................................224
Figure 7.22 Decomposition of the Controllers, Q1 ..................................................................................................224
xiv
List of Tables
Page
Table 2.1 Notation for Governing Partial Differential Equations................................................................................15
Table 2.2 Matrix Elements of Z ( x, u ) for the Gas Cooler ........................................................................................22
Table 2.3 Matrix Elements of Z ′( x ′, u ′) for the Gas Cooler ....................................................................................25
Table 2.4 Matrix Elements for Equation 2.84..............................................................................................................27
Table 2.7 Matrix Elements for Equations 2.104-2.107................................................................................................32
Table 2.8 Matrix Elements for Equation 2.113............................................................................................................33
Table 2.9 Matrix Elements of Z ( x, u ) for the Single Phase Heat Exchanger ...........................................................35
Table 2.10 Matrix Elements of Equations 2.133 – 2.136 ............................................................................................37
Table 2.11 Matrix Elements of Z ( x, u ) for the Evaporator ......................................................................................41
Table 2.13 Matrix Elements of Z ( x, u ) for the Condenser .......................................................................................49
Table 2.15 Matrix Elements of Z ( x, u ) for the Evaporator ......................................................................................60
Table 2.16 Matrix Elements of Equations 2.212-2.215 ...............................................................................................63
Table 2.17 Matrix Elements of Z ( x, u ) for the Condenser with Receiver................................................................68
Table 2.18 The elements of Equations 2.241 through 2.244. ......................................................................................72
Table 2.19 Matrix Elements for Equations 2.254 – 2.257 ...........................................................................................76
Table 4.1 Valve Designations......................................................................................................................................96
Table 4.2 Component Name, Manufacturer, Model, and URL ...................................................................................97
Table 4.3 Temperature, Pressure, and Flow Measurement Designations ....................................................................98
Table 4.4 Principal Heat Exchanger Parameters........................................................................................................124
Table 4.5 Principal Pipe Sections ..............................................................................................................................128
Table 5.1 ....................................................................................................................................................................130
Table 5.2 ....................................................................................................................................................................135
Table 5.3 Statistical Model Fits of the 6th Order Model ............................................................................................145
Table 5.4 Model Fit of the Identified Models............................................................................................................148
Table 5.5 Max Error of the Identified Models...........................................................................................................149
Table 5.6 RMS Error of the Identified Models..........................................................................................................149
Table 5.7 Model Fit of the Identified Models............................................................................................................153
Table 5.8 Max Error of the Identified Models...........................................................................................................154
Table 5.9 RMS Error of the Identified Models..........................................................................................................154
Table 5.10 Parameter Values for Model Validation ..................................................................................................164
Table 6.1 Performance Bounds for LCN/LMN Framework......................................................................................202
Table 6.2 Performance Bounds for LQN/LSN Framework.......................................................................................203
xv
Nomenclature
List of Abbreviations
AC&R – air conditioning and refrigeration
COP – coefficient of performance
LMI – Linear Matrix Inequality
MIMO – multi-input, multi-output
ODE – ordinary differential equation
PDE – partial differential equation
PRBS – pseudo-random binary sequence
R134a – refrigerant commonly used in automobile air conditioning systems
R22 – refrigerant commonly used in residential air conditioning systems
R404a – refrigerant commonly used in air conditioning systems
R744 – CO2 – carbon dioxide as a refrigerant
SISO – single-input, single-output
List of Symbols
Variable Explanation
A Area
A,B,C,D State Space Matrices
α Heat Transfer Coefficient
Cp Specific Heat
E Energy
η Efficiency
f ,g Continuous Functions
γ Void Fraction
γ Mean Void Fraction
H, h Enthalpy, Specific Enthalpy
I Identity Matrix
L Length
λ Eigenvalue
m& Mass Flow Rate
P Pressure
p Perimeter
Q Heat, Youla Parameter
Re Reynold’s Number
ρ Density
S Slip Ratio; Dual Youla Parameter
s Specific Entropy
σ Singular Value
T Temperature
t Time
U, u Internal Energy, Specific Internal Energy; Controllable Inputs
UA Lumped Heat Transfer Coefficient
V Volume; Voltage
xvi
W Work
x Quality; Dynamic States
y Outputs
z Spatial Coordinate
Z Matrix
Subscript Explanation
1,2,3 1st, 2nd, 3rd Region
a, air Air
ave Average
c Cold; Condenser; Gas Cooler
cs Cross-Sectional
d Diameter
f Liquid
g Vapor
h Hot
hx Internal Heat Exchanger
i Inner
in In
int Intermediate
k Compressor
o Outer; Steady State
out Out
r Refrigerant; Reduced Order
sys System
total Total
v Valve
w Wall
xvii
Chapter 1. Introduction
Over 15 billion dollars is spent on energy for residential air-conditioning alone each year, and air-
conditioning remains the largest source of peak electrical demand. Improving the efficiency of these systems has the
potential for significant economic and environmental impact, but requires not only refining individual component
designs, but increasing overall system efficiency using advanced control strategies. Transient control of vapor
compression cycles faces two significant challenges: 1) creating control-oriented models that balance simplicity with
accuracy, and capture the complex heat and mass flow dynamics, and 2) developing control strategies that can
achieve high performance over a wide range of operating conditions.
This dissertation makes contributions on both fronts and can be divided into two distinct parts. The first
portion of the dissertation presents the development, simulation, and experimental validation of a validated 1st
principles modeling framework that captures the dynamics of a variety of vapor compression cycles in a form
amenable to controller design. These validated models are highly nonlinear, and require a nonlinear control strategy
to attain high performance over the entire operating envelope. To this end, a gain-scheduled control approach based
on local models and local controllers is presented that uses endogenous scheduling variables. This comprises the
second portion of the dissertation, where a theoretical framework for designing gain scheduled controllers, tools for
analyzing the stability of the nonlinear closed loop system, and experimental evaluation of advanced control
strategies for vapor compression systems is presented. These results demonstrate that while linear control techniques
offer significant advantages versus traditional a/c control systems over small ranges, the gain-scheduled approach
extends these advantages over the entire operating regime.
In summary, this dissertation addresses a challenging problem in the field of thermal-fluids by applying the
theory and techniques of robust and nonlinear control. In doing so, new control techniques are developed which,
when applied, result in improvements in performance and efficiency of vapor compression cycles. The remainder of
this chapter is organized as follows. Section 1 presents background on vapor compression cycles. Section 2
summarizes the contributions to creating a control-oriented modeling framework, while an overview of control
strategies for vapor compression systems is given in Section 3. Section 4 discusses the gain-scheduling control
problem and summarizes the advances made by this dissertation to overcoming several of the analytical
shortcomings faced by industrial practitioners. Finally, Section 5 presents an outline of the dissertation.
1.1 Vapor Compression Systems

A majority of household, automotive and industrial air conditioning and refrigeration devices operate using
a vapor compression cycle. These systems are a type of thermodynamic machinery, which utilize a compressible
fluid to transfer heat. While these devices can be used for heat generation, this dissertation deals specifically with
cooling applications. For ease of explanation, we categorize these systems into two groups: subcritical and
transcritical cycles. These titles refer to fluid’s condition throughout the cycle; whether the fluid is always below its
critical point (subcritical) or whether the fluid operates both below and above the critical point (transcritical).
Subcritical systems form the bulk of the systems used today, whether in homes, automobiles, or industry.
A diagram of the system is shown with the accompanying P-h diagram in Figures 1.1-1.2. The simplest cycle
operates with four components: compressor, condenser, expansion valve, and evaporator. The fluid enters the
1
compressor as a superheated vapor at a low pressure. The fluid is compressed to a high pressure by the compressor
and then enters the condenser. At this higher pressure, the fluid has a higher temperature than the ambient
conditions, and as a fan blows air across the condenser, heat is transferred to the air, and the fluid condenses. The
fluid exits the condenser as a subcooled liquid at a high pressure. The fluid then passes through an expansion
device. At the exit of the expansion valve the fluid is generally two-phase, and at a low pressure. The fluid then
enters the evaporator. At this lower pressure the fluid has a lower temperature than ambient conditions, and as a fan
blows air across the evaporator, heat is transferred to the fluid, and the fluid evaporates. The fluid exits the
evaporator as a superheated vapor and enters the compressor.
Pressure
3 2
4 1
Liquid Two-Phase Vapor
Enthalpy
Figure 1.1 System Diagram - Ideal Subcritical Figure 1.2 P-h Diagram - Ideal Subcritical Vapor
Vapor Compression Cycle Compression Cycle
A slight variation to this system is the addition of receivers. When a receiver is placed at the exit of the
evaporator, the fluid enters and exits the evaporator as a two-phase fluid. The two-phase fluid enters the receiver,
and the compressor draws fluid from the top of the receiver. In this manner the fluid entering the compressor is
always a saturated vapor so as to avoid damaging the compressor with liquid flow. Similarly, when a receiver is
placed at the exit of the condenser, the fluid does not exit as a subcooled liquid, but as a two-phase fluid. The
entrance to the expansion device is from the bottom of the receiver. Thus the fluid entering the expansion device is
always a saturated liquid, thus avoiding choked flow conditions.
The most common transcritical systems use carbon dioxide (CO2 or R744) as the working fluid. The main
disadvantage of this system is the high operating pressures necessary. Recently the lower environmental impact of
this refrigerant has led to an increase in its appeal. A diagram of the system is shown with the accompanying P-h
diagram in Figures 1.3-1.4.
2
Evaporator Fan
4
Evaporator
Expansion Valve
Supercritical
3 2
Filter/Dryer
Pressure
4 1
1
Internal Heat Exchanger
Compressor

Oil Seperator
Gas Cooler Enthalpy

2
Condenser Fan
Figure 1.3 System Diagram - Ideal Figure 1.4 P-h Diagram - Ideal Transcritical Vapor
Transcritical Vapor Compression Cycle Compression Cycle
This system resembles the subcritical system with a few alterations. As the fluid leaves the compressor, the
fluid has an extremely high pressure and is above the critical point. This supercritical fluid behaves differently than
both of the liquid and gas phases. At this state the fluid is compressible, which must be taken into account when
modeling the heat exchanger. Because the fluid does not condense as it flows through the heat exchanger, the term
“gas cooler” is used to describe the heat exchanger. Additionally, there is often a third heat exchanger in the system.
This is a counterflow heat exchanger and placed in between the gas cooler and expansion device, and between the
evaporator and compressor. This is generally referred to as the internal heat exchanger or suction line heat
exchanger. With this device the high temperature fluid leaving the gas cooler heats the fluid leaving the evaporator
before it enters the compressor. This ensures that the fluid entering the compressor is superheated vapor, as well as
providing some efficiency benefits that are more pronounced for CO2 than for most subcritical refrigerants. The
thermodynamic assumptions and performance considerations for this system are the same as those for the subcritical
system.
1.2 Control-Oriented Modeling of Vapor Compression Systems

To model these systems a few standard assumptions are made. First, the compression of the fluid is
assumed to be adiabatic with an isentropic efficiency. Second, isobaric conditions in the condenser and evaporator
are assumed. Third, expansion through the valve is assumed to be isenthalpic. The expansion valve and the
compressor are modeled with static, semi-empirical relationships, while the dynamic heat exchanger models are
derived using a moving boundary, lumped parameter approach. These nonlinear models are linearized about an
3
operating condition, and reduced in order to include only the most important system dynamics. The full order,
nonlinear models and the reduced order, linear models are validated using data taken on experimental systems
available at the University of Illinois at Urbana-Champaign.
In reality, an AC&R system is a highly nonlinear, infinite-dimensional system. Many advanced control
design techniques generally yield high order controllers when based on high order models. Some of the difficulties
in developing low order controllers for high order systems are discussed by Anderson in [7]. In practice a low order
controller is achieved by: 1) reducing the model order prior to controller design, or 2) designing a high order
controller, and then numerically reducing the order of the controller. Although both approaches could be applied,
the first approach is desired because of the added benefits of identifying the dominant system dynamics. A key
contribution of this research is the identification of the dominant dynamic modes in terms of physical parameters.
This permits the derivation of a minimal representation of the system dynamics which is essential for model based
control design, as well as providing invaluable insight to the underlying dynamics of the system. The resulting
models are used to create the Thermosys™ Toolbox for MATLAB, a suite of simulation tools specifically suited for
real-time simulation, control design and evaluation, and fault detection.
1.3 Control of Vapor Compression Cycles

Clearly, an increase of efficiency in AC&R systems would have a notable effect on the nation’s economy
as well as the average individual’s budget. Additionally, increasing the efficiency of these systems would have a
much larger societal and environmental impact because of the reduction in fossil fuels required to provide this
energy. Significant progress has been made in recent years to improve component efficiency in AC&R systems.
With the increasing availability of inexpensive computing power, a veritable leap in increased efficiency is possible
using advanced control techniques.
Residential and industrial AC&R systems have extremely large start-up times. Automotive AC systems
rarely operate at steady state conditions, and are constantly attempting to compensate for changing setpoints and
external conditions. For each type of system, traditional control strategies have included single-input single-output
(SISO) control or simple on/off (“bang-bang”) control. On/off control schemes limit the system’s overall efficiency
due to large power requirements and significant thermal inertia during start-up transients. Furthermore, multiple
SISO control techniques are less efficient because of the extensive cross-coupling of the system dynamics.
Multivariable control strategies could improve efficiency while simultaneously benefiting from the coupled system
dynamics. Not only would a model-based MIMO control strategy eliminate the problems associated with current
industry practice, but it would also allow several simultaneous control objectives.
To maximize efficiency of this system for a cooling application, the portion of the evaporator with two-
phase flow needs to be maximized. Because the heat transfer coefficient between the liquid and the evaporator walls
is much higher than the heat transfer coefficient between the vapor and the evaporator walls, the two-phase portion
of the evaporator provides virtually all of the cooling capacity of the system. However, to ensure safe and reliable
operation of the compressor, the fluid entering the compressor must be completely vapor. Systems with a receiver at
the evaporator exit can ensure safe operation of the compressor, while maximizing the evaporator’s performance.
4
For systems without a receiver, a generally acceptable compromise is for the fluid to be 5° C above the saturation
temperature. This is generally referred to as 5° C of superheat.
Superheat control is generally achieved using a mechanical feedback for the control of an expansion valve,
but often results in a phenomenon known in the industry as “valve hunting.” This condition is characterized by
oscillations in the length of two-phase flow in the evaporator, and thus oscillations in the amount of superheated
vapor at the evaporator exit. This condition was first qualitatively documented in 1963 by Zahn in [116]. In 1966
Wedekind and Stoecker published data demonstrating this phenomenon in [110]. Later Broersen and van der Jagt in
[18] used a lumped parameter model to show that this phenomenon is caused by the interaction of the controller with
the system dynamics. They made various suggestions how to avoid valve hunting, but all have noted disadvantages
and require the system to function at less efficient conditions. This condition has been widely observed in industry
and is usually solved by adjusting the valve parameters resulting in decreased performance. By using multivariable
control schemes this phenomenon could be avoided while allowing the system to function at more efficient
operating conditions.
While achieving multiple performance objectives with AC&R systems is desirable, it is generally
impractical. More sophisticated control techniques traditionally have been used only in large industrial systems,
where the economy of scale made these techniques cost effective. These techniques required extensive tuning using
empirical data and lumped parameters.
Part of the solution to the problem again lies in model-based multivariable control strategies. By
developing a general lumped parameter model, a large number of AC&R systems could be included within the
framework. Multivariable control schemes developed using a form of this model would allow the control scheme to
be adapted based on physical parameters of the individual systems. Additionally, these control schemes could be
designed to meet multiple performance objectives such as maximizing COP, exchanging efficiency for capacity
when needed, controlling not only temperature, but also humidity, and so on. This dissertation demonstrates the
effectiveness in advanced model based control strategies in both simulation and experiment.
1.4 Gain Scheduling

The field of gain scheduling has been said to be “another example of the lamented theory/application gap”
except that the application is ahead of the theory [86]. Indeed, the foremost challenge facing gain scheduling is
guaranteeing stability and performance, despite the extensive experience demonstrating its effectiveness in practice.
Many physical systems exhibit dynamics that vary appreciably over the typical operating regime, therefore
a single linear controller may fail to achieve acceptable performance throughout the envelope of conditions. One of
the most popular solutions to this nonlinear control design problem is the use of gain-scheduling, in which a
nonlinear controller is constructed by interpolating a family of local controllers. Thus the nonlinear control design
problem can be divided into several control problems where linear design tools are generally employed. This
“divide-and-conquer” [58] methodology has found success in a variety of applications. The success of the gain-
scheduled approach seems directly attributable to its simplicity in design and implementation, as well as the maturity
of the linear control design tools and their computer-aided control system design (CACSD) implementations.
5
Despite the overwhelming success in industrial practice, the principal difficulty of using gain-scheduled
control is guaranteeing stability. By blending several linear controllers, the resulting global controller is nonlinear,
and results in the addition of “hidden coupling terms” [85] or additional dynamics due to the interpolation functions.
Thus, although the gain-scheduled controller may be stable at every fixed value of the scheduling variable, the true
global closed loop may not be stable.
Stability of the closed loop system is generally assessed either by extensive simulations or experiments, or
by proving that the gain-scheduled controller is stable for all fixed values of the scheduling variable, and then
inferring overall stability by assuming slowly varying scheduling parameters. The former can be time or cost
prohibitive, and the latter may not be appropriate for many gain-scheduled systems.
This dissertation presents a theoretical framework for designing gain scheduled controllers, and tools for
analyzing the practical stability of the nonlinear closed loop system using Linear Matrix Inequalities. Additionally,
an alternative framework for scheduling dynamic controllers is presented that is based on the dual Youla
parameterization. This framework is shown to provide greater stability while offering an element of design
freedom.
1.5 Organization of Dissertation

The remainder of this dissertation is organized as follows. Chapter 2 describes a 1st principles modeling
framework that captures the dynamics of a variety of vapor compression cycles in a form amenable to controller
design. The modeling procedure for each of the components is presented, along with analytical representations of
both nonlinear and linearized models. Chapter 3 presents the simulation environment developed for simulation,
model validation, and controller design. Chapter 4 details the experimental setup, while model validation of the first
principles modeling framework is given in Chapter 5. These validated models are highly nonlinear, and require a
nonlinear control strategy to attain high performance over the entire operating envelope. To this end, a gain-
scheduled control approach based on local models and local controllers is presented that uses endogenous
scheduling variables. Chapter 6 presents a theoretical framework for designing gain scheduled controllers, and tools
for analyzing the stability of the nonlinear closed loop system. Chapter 7 presents the experimental evaluation of
advanced control strategies for vapor compression systems, demonstrating that while linear control techniques offer
significant advantages over small ranges, the gain-scheduled approach extends these advantages over the entire
operating regime. Conclusions and recommendations for future work are given in Chapter 8.
6
Chapter 2. Dynamic Modeling
This chapter discusses the modeling approach employed to capture the salient dynamic characteristics of
vapor compression cycles, while yielding a model that is still sufficiently simple to be a useful tool for multivariable
control design. First, the need for control-oriented models is argued, followed by a discussion of different modeling
paradigms with particular attention given to the moving boundary approach for modeling two-phase flows. After
presenting these introductory topics, dynamic component models are presented, both in linear and nonlinear form.
The models for compressor and expansion devices are presented first, followed by the presentation of the heat
exchanger models. To illustrate the modeling approach, assumptions, and resulting physical insight, a complete
derivation is given for a generic single-phase heat exchanger. This is followed by the presentation of nonlinear and
linear models for various types of two-phase heat exchangers. Although the complete derivation for the two-phase
heat exchangers is omitted, sufficient detail is given for the reader to replicate the modeling process.
2.1 Introduction
Controls engineers are often required to develop dynamic models for a class of complex physical systems
with little prior knowledge of the system dynamics. Adding to the challenge of this task is the desire to gain physical
insight into the underlying dynamics. This task can sometimes span years or decades before the dynamics of the
given class of systems are well understood and characterized with simple control-oriented models (e.g. [10]).
The foremost difficulty in this process is balancing accuracy with simplicity. Although many processes can
be accurately modeled with high order ordinary or partial differential equations, the practicing control engineer often
needs practical low order controllers [7]. This means that significant time must be spent either 1) developing
simple, accurate, control-oriented models prior to the control design phase, or 2) reducing high order controllers
after the design phase, but prior to implementation. The distinct advantage of developing simple models is the
physical insight gained. This insight is invaluable as a tool in developing general control strategies (i.e. selecting
sensors, actuators) and in system design (e.g. sizing components).
Because of these advantages, engineers expend considerable effort to develop low order physical models.
For example, efforts to model the two-phase fluid dynamics associated with drum boilers spanned several decades
before finding an appropriate balance of simplicity and accuracy [10]. However, once an accurate low order model
has been identified, the development of control strategies follows quickly. As an example, this was evident with the
publications that followed the Moore-Greitzer compressor model [66],[38].
This dissertation is seeks to develop a control-oriented modeling framework for vapor compression
systems. The modeling approach should be sufficiently accurate to capture the essential dynamic behavior, while
remaining simple enough to provide insight into the dynamics and adequately tractable to be useful as a control
design tool. An essential consideration before beginning the task of developing vapor compression models is the
intended use for the model. Dynamic models that capture the slow dynamics (hours/days) are most useful for
analyzing issues of comfort and energy use, and would focus on modeling the heating or cooling zone affected by
the vapor compression system. From this perspective, the refrigerant circuit dynamics are virtually instantaneous,
and are unnecessary to consider when developing the dynamic model. Indeed, the bulk of the modeling task lies in
modeling the surroundings, not the system. In contrast, the dynamic models that capture the fast dynamics
7
(seconds/minutes) are generally used to design and evaluate system control strategies (e.g. superheat regulation,
pressure regulation, etc.). In this case, the ambient conditions are generally viewed as static. There is a litany of
publications regarding the building zone control. However, focus of this work is on the dynamics of the refrigerant
circuit, and the potential benefits of advanced control strategies in achieving higher operating efficiencies and better
performance.
2.2 Modeling Paradigms

The literature is replete with attempts to model vapor compression system dynamics. Lebrun [56] and
Bendapudi [13] both provide literature reviews of notable research in this area. Bendapudi notes that there is
considerable interest in deriving models that are simple mathematically but without losing relevant detail.
Bendapudi also reports that the “largest task in modeling a refrigeration system was, usually, the modeling of the
heat exchangers.” These heat exchanger models can be generally classified into three groups: lumped parameter
models, discretized models, and moving boundary models.
In general, the discretized or finite difference approach results in models that are fairly accurate, but of high
dynamic order. Thus these models are appropriate for simulation, but of limited use for developing controller
strategies. Although a pure lumped parameter approach results in fewer equations, it frequently ignores important
dynamics due to the complex heat exchanger behavior. Specifically, a two-phase fluid heat exchanger is often
modeled as a single lumped subsystem, which ignores the important dynamics associated with the moving boundary
between the two-phase flow region and the single-phase flow regions. Hence, a lumped parameter model with a
moving interface boundary was proposed as the solution to capturing these important dynamics, while preserving
the simplicity of lumped parameter models.
2.2.1 Lumped Parameter Models

Dynamic and control studies that model heat exchangers or vapor compression systems with pure lumped
parameter models are too replete to attempt explicit mention. Often these studies have as a primary focus something
other than the heat exchanger dynamics, and so the model development is treated superficially (often assuming
simply a first order time constant and/or delay [99],[72]. However, a particularly notable effort to develop nontrivial
lumped parameter models that consider multiple fluid regions in the heat exchanger, albeit temporally invariant, is
the work by Chi and Didion (1982) [22]. However, like most publications in this area, limited experimental
validation is presented.
2.2.2 Discretized Models

Finite difference or distributed approaches to dynamic modeling of these dynamics exist in various stages
of development and are available as commercial software packages (e.g. e-Thermal [6], Modelica [105], EASY5 [4],
or SINDA/FLUENT [24]). The first notable model using this approach was reported by Gruhle and Isermann in
1984 [39], and since has been followed by many researchers. The attractiveness of this approach is the ability to
model the fluid behavior in a detailed manner. This, reportedly, lends itself to higher accuracy prediction than the
lumped parameter approaches that rely on effective parameters and only describe the dominant physical mechanisms
for heat transfer, fluid flow, etc.
8
Gruhle and Isermann’s work used a discretized model of the evaporator for model based design of PI valve
control strategy (no validation of the model is presented) [39]. Shoureshi (1984) presented a model of a condenser
using bond graphs [93]; limited validation of the method was given. MacArthur (1984) presented a discretized
model [61], and later (1989) gave extensive experimental validation of the model [62]. Sami (1987) presented a
distributed model of a heat pump system with no validation [88]. Recent research has also employed this approach,
such as Mithraratne (2000) who presented a discretized model of a TEV controlled evaporator with validation [65],
and [12] who presented a distributed model of a liquid chiller system with validation. This approach has also been
used for transient modeling of transcritical systems [78].
2.2.3 Moving Boundary Models

The moving boundary approach seeks to capture the dynamics of multiple fluid phase heat exchangers
while preserving the simplicity of lumped parameter models. At the focal point of this modeling approach is the
ability to predict the effective position of phase change in a heat exchanger. Wedekind was among the first to study
the transient behavior along these lines. In 1966 he and W. Stoecker published research regarding the transient
response of the effective dry out point in evaporating flows [110],[108]. In 1978 he proposed using a mean void
fraction to develop a transient model for the evaporating and condensing flows of a vapor compression system
[109],[11]. Wedekind’s research plays a critical role in the moving boundary method of modeling vapor
compression cycles. The mean void fraction assumption is applied almost universally by other researchers
developing moving boundary models, allowing the two-phase region to be modeled in lumped form.
Void fraction is defined as the ratio of vapor volume to total volume, and many experimental correlations
have been proposed for predicting void fraction for various conditions and fluids. Excellent reviews of well known
correlations can be found in [84] and [112]. In general, these can be categorized into four divisions, listed in order
of increasing complexity: Homogeneous, Slip Ratio, Lockhart-Martinelli, and Mass Flux Dependent. The latter two
types are notably complex, whereas the first two are remarkably simple. Independent of the correlation used, most
authors assume that the value of mean void fraction is time-invariant for small transients. A discussion of validity of
this assumption will be treated later in this chapter with the derivation of heat exchanger model.
In his paper [109] Wedekind presented comparisons of model and experimental data sampled at
approximately 4Hz, showing good agreement between model and data. Unfortunately, in the moving boundary
publications since that time, few have presented complete validation results, and often at sampling frequencies so
low as to question if significant transient behavior is being lost.
Dhar and Soedel (1979) developed a dynamic model of an entire air conditioning system using a lumped
parameter, moving boundary approach, but presented only simulation results [26]. Broersen and van der Jagt (1980)
employ a moving boundary model of an evaporator to analyze hunting of thermostatic expansion valves [18]. No
validation results are presented. Kapadia (1984) presented a moving boundary model for a condenser in Rankine
cycle, with extremely limited validation results [48]. In 1992 Grald and MacArthur followed their work on
distributed heat exchanger models with the presentation of a moving boundary model of an evaporator [35]. For
model validation the evaporator was incorporated into a overall heat pump model (with a discretized condenser
model). Validation of the model was extremely limited. Pressure predictions from the original discretized system
9
model is compared to data for compressor start up and shutdown (sampled at approximately 0.1 Hz). The
discretized system model is then compared to the system model with a moving boundary evaporator for only
evaporator cooling capacity and compressor power.
X.D. He was first author to suggest the use of moving boundary models for multivariable control design
(1997) [41],[40]. He presented linearized moving boundary models for the evaporator and condenser, and provided
adequate validation. Although reduced order models were used for MIMO control design, these reduced order
models were never validated, so no conclusion can be drawn regarding the validity of the simplifying assumptions.
The experimental control results, however, demonstrated improved performance for coordinated control strategies.
Following these publications, several authors have presented identical models [46] or extensions of the
modeling framework for heat exchangers operating under different conditions [111], including the presentation of a
switching strategy for transitioning between moving boundary models as fluid regions disappear or reappear [77].
However, none of these publications present experimental validation of the models. Leducq (2003) presents a
reduced order model similar to that suggested by He et al, but does include a comparison of superheat and outlet
water temperature for a step change in compressor speed, with a sampling frequency of 0.1 Hz [57]. In contrast with
all the previous research applied to subcritical vapor compression systems, Rasmussen (2004) derives a system
model with moving boundary models for the heat exchangers for a transcritical vapor compression cycle with
comparison between nonlinear, linearized, reduced order models and data [82].
2.3 Vapor Compression Systems

The vapor compression cycle is used in refrigeration, air-conditioning, and heat pump applications. The
ideal vapor compression cycle consists of four processes [20]: 1) isentropic compression in a compressor, 2) isobaric
heat rejection in a condenser, 3) isenthalpic expansion in a expansion valve, and 4) isobaric heat absorption in an
evaporator. This process and the associated four components are shown in Figures 2.1-2.2. A majority of vapor
compression systems operate below the critical point of the internal fluid, thus being termed subcritical systems.
Alternatively, transcritical vapor compression systems reject heat above the critical point while the internal fluid is
in a supercritical state, leading to the term transcritical systems. This would include CO2 systems which have
attracted attention as a possible replacement of traditional subcritical refrigerants because CO2 is a natural fluid, and
does not have the negative environmental impacts of traditional refrigerants. Although the work here can be applied
to both subcritical and transcritical vapor compression systems, the experimental results will focus exclusively on
subcritical systems. The dynamics of transcritical systems was treated in a previous work [81], and the interested
reader is referred to [82] for more information.
10
Pressure
3 2
4 1
Enthalpy
Figure 2.1 System Diagram - Ideal Subcritical Figure 2.2 P-h Diagram - Ideal Subcritical Vapor
Vapor Compression Cycle Compression Cycle
Actual vapor compression systems deviate from the ideal in several respects. First, momentum losses as
the fluid travels through the heat exchanger leads to pressure drops as the fluid evaporates and condenses. Second,
the compression of the fluid is not isentropic. Third, for practical reasons, the fluid leaving the evaporator and
condenser is not saturated, but superheated vapor and subcooled liquid respectively. The superheated vapor ensures
that two-phase fluid does not damage the compressor, and subcooled liquid at the valve inlet ensures proper
metering of the fluid flow.
Pressure
3 2
4 1
Enthalpy
Figure 2.3 P-h Diagram - Actual Vapor Compression Cycle
The efficiency of the actual vapor compression system is characterized by a Coefficient of Performance
(COP). For a Carnot refrigerator (the theoretical limit) the COP is given as: COPCarnot =
Tlow
Thigh −Tlow . For an actual
vapor compression system, the COP is given as the ratio of the amount of heat absorbed to the amount of work
h1 − h4
required, or simply COP = =
Qlow
Win h2 − h1 .
Actual systems may have several additional components to facilitate operation, including oil separators,
filter/driers, etc. (Fig. 2.4-2.5). The addition of a low side accumulator prevents two-phase fluid from entering the
compressor, while allowing the fluid flow through the evaporator to be entirely two-phase. Because two-phase fluid
11
generally has a higher heat transfer coefficient that superheated vapor, this is attractive from an efficiency standpoint
as well. The addition of a high side receiver provides a location for storing excess refrigerant, and ensures liquid
flow to the expansion valve. The addition of an internal heat exchanger (Fig. 2.6-2.7) serves to superheat the fluid
entering the compressor and subcool the liquid leaving the condenser, thus increasing the cooling capacity of the
system. Other possible additions include flood tanks, flash gas bypass, etc. These are not in widespread use, and are
not treated in this work.
Evaporator Fan
Accumulator
4 1
Evaporator
Pressure
Expansion Valve Compressor
3 2
Filter/Dryer Oil Seperator
Condenser
4 1
3 2
Receiver Liquid Two-Phase Vapor
Enthalpy
Condenser Fan
Figure 2.4 System Diagram - Subcritical Figure 2.5 P-h Diagram - Subcritical Vapor Compression
Vapor Compression Cycle with Receivers Cycle with Receivers
Evaporator Fan
4
Evaporator
Expansion Valve
Filter/Dryer
Pressure
3
3 2
1
Internal Heat Exchanger
Compressor 4 1
Oil Seperator
Condenser
2 Enthalpy
Condenser Fan
Figure 2.6 System Diagram - Subcritical Vapor

Figure 2.7 P-h Diagram - Subcritical Vapor Compression
Compression Cycle with Internal Heat
Cycle with Internal Heat Exchanger
Exchanger
12
2.4 Compressors and Expansion Devices
As mentioned previously, the issue of time scale is of critical importance when deriving dynamic models of
vapor compression systems. This work is focused on refrigerant circuit dynamics that evolve on the order of
seconds or minutes. Because the dynamics of compressors or expansion devices are generally an order of
magnitude faster than those of the heat exchangers, these mass flow devices are generally modeled with static
relationships. Two principle relationships are necessary for modeling these components: first, the prediction of mass
flow rate, and second, the prediction of outlet enthalpy.
Mass flow rate through an ideal reciprocating compressor is given using the volumetric capacity of the
&k =
compressor multiplied by the speed of rotation, m ρV k ω kη k . Compressing inefficiencies are captured using
an empirically determined volumetric efficiency, η k . The prediction of outlet enthalpy is given by assuming an
isentropic efficiency. These efficiencies are often assumed to vary as a function of the pressure ratio.
Mass flow through the expansion valves is modeled with variations on the orifice equation
m& v = C d Av ρ (Pin − Pout ) , with the discharge coefficient is determined empirically as a function of operating
condition. The valves are assumed to be isenthalpic.
2.5 Heat Exchangers

The dynamics of a vapor compression system are assumed to be dominated by the dynamics of the heat
exchangers. As mentioned in the previous section, the dynamics of the actuating components (compressor, valve,
etc.) are considered to be fast relative to the dynamics of the heat exchangers. Thus, while the actuating components
are modeled with static (algebraic) relationships, the modeling of the heat exchangers is significantly more complex.
2.5.1 Moving Boundary Approach

The approach most applicable to the objectives of this research is known as the lumped parameter, moving
boundary approach. This approach assumes a time-varying boundary between regions of different fluid state (i.e.
subcooled liquid, two-phase, or superheated vapor). Separate control volumes are considered for each of the fluid
regions, and the necessary distributed parameters are “lumped” for each of these regions.
x=1
Twall,1(t) Twall,2(t)
m& in hin
P(t) & outhout (t)
m
xin > 0
Two-Phase Superheat
L1(t) L2(t)
LTotal
Figure 2.8 Heat Exchanger with Two-Phase and Superheated Vapor Regions
13
To derive governing differential equations suitable for simulation or analysis, the most common
method[111],[46],[41] is to begin with the governing partial differential equations (PDEs) for fluid flow in a tube.
After applying a few simplifying assumptions, given below, these PDEs can be integrated along the length of the
heat exchanger to remove the spatial dependence and yield several ordinary differential equations (ODEs). The
number of ODEs depends on the number of fluid regions assumed. An alternative method uses the unsteady state
form of the conservation of mass and energy equations. By assuming control volumes associated with each of the
fluid regions, these equations can be expanded, yielding an identical set of ODEs. Although, this method will be
shown to be completely equivalent to the first method, it has some distinct advantages including derivation
simplicity, conceptual simplicity, and freedom in choosing the dynamic state variables. This will be essential for
applying the physical-based model reduction approaches presented.
2.5.1.1 Modeling Assumptions

The modeling methods to be presented require several assumptions about the fluid flow in the heat
exchangers. These assumptions were commonly used in past modeling efforts and are as follows:
• The heat exchanger is assumed to be a long, thin, horizontal tube.
• The refrigerant flowing through the heat exchanger tube can be modeled as a one-dimensional fluid flow.
• Axial conduction of refrigerant is negligible.
• Pressure drop along the heat exchanger tube due to momentum change in refrigerant and viscous friction is
negligible (refrigerant pressure along the entire heat exchanger tube can be assumed to be uniform). Thus
the equation for conservation of momentum is not needed.
2.5.1.2 Governing Partial Differential Equations

As mentioned in the introduction, the method for deriving ordinary differential equations using the lumped
parameter, moving boundary approach is to integrate the governing partial differential equations along the length of
the heat exchanger tube to remove spatial dependence. An explanation of the partial differential equations for
conservation of refrigerant mass and energy can be found in [35], and are replicated in Equations 2.1 – 2.2
r r
respectively, where u is the fluid velocity vector, f is the body force vector and σ is the stress tensor. (The
conservation of momentum is neglected and not included here.) By applying the assumptions outlined in the
previous section, it is possible to simplify these equations to one-dimensional PDEs. A detailed explanation of these
steps can be found in [35]. (The derivation presented in this thesis differs from that presented in the cited source
only by not neglecting the rate change of pressure with respect to time in the conservation of energy equation.) The
resulting equations for conservation of refrigerant mass and energy in the heat exchanger tube are given in Equations
2.3 – 2.4. Additionally an equation for the conservation of heat exchanger wall energy is given in Equation 2.5.
The necessary notation is described in Table 2.1.
∂ρ r
+ ∇ ⋅ ( ρu ) = 0 ( 2.1 )
∂t
r
∂ ( ρu ) rr r
+ ∇ ⋅ ( ρu u ) = ρf + ∇ ⋅ σ ( 2.2 )
∂t
14
∂( ρAcs ) ∂(m& )
+ =0 ( 2.3 )
∂t ∂z
∂( ρAcs h − Acs P ) ∂(m& h )
+ = piα i (Tw − Tr ) ( 2.4 )
∂t ∂z
(C p ρA)w ∂(Tw ) = piα i (Tr − Tw ) + poα o (Ta − Tw ) ( 2.5 )

∂t
Table 2.1 Notation for Governing Partial Differential Equations
ρ density of refrigerant
P pressure of refrigerant
h enthalpy of refrigerant
pi inner perimeter (interior surface area per unit length)
po outer perimeter (exterior surface area per unit length)
Tr temperature of refrigerant
Tw tube wall temperature
α i
heat transfer coefficient between tube wall and internal
fluid
αo heat transfer coefficient between tube wall and external
fluid
Acs cross-sectional area of the inside of tube
m& mass flow rate of refrigerant flowing along the tubes
(C p ρA)w thermal capacitance of tube wall per unit length
To perform the necessary integrations, an integration rule commonly known as Leibniz’s equation is used
(Equation 2.6), with z being the spatial coordinate. Thus the limits of integration depend on the how the regions are
defined for each heat exchanger.
z2 (t ) z (t )
∂f ( z , t ) d ⎡2 ⎤ d ( z 2 (t )) d ( z1 (t ))
∫ ∂ t
dz = ⎢ ∫ f ( z , t )dz ⎥ − f ( z 2 (t ), t )
dt ⎢ ⎥ dt
+ f ( z1 (t ), t )
dt
( 2.6 )
z1 ( t ) ⎣ z1 ( t ) ⎦
The resulting ordinary differential equations can be combined, simplified, and organized into matrix form.
This form is generally referred to in controls applications as “state space form.” The general form for a linear, time-
invariant system is given in Equation 2.7. In this form, u is the vector of inputs, y the vector of outputs, and x the
vector of states. {A, B, C , D} are constant matrices. Because the systems resulting from the outlined modeling
approach are nonlinear models, an alternate state space form is used (Equation 2.8). The resulting set of equations
can be solved numerically given appropriate initial conditions. The dynamic order, or number of state variables for
each component, reflects the relative complexity of the component dynamics.
15
x& = Ax + Bu
( 2.7 )
y = Cx + Du
x& = f ( x, u )
( 2.8 )
y = g ( x, u )
Combining the final results of the integrated PDEs into a matrix form results in an equation of the form
Z ( x, u ) ⋅ x& = f ( x, u ) . The state variables typically include pressures, outlet enthalpy, length of the fluid regions,
and wall temperatures.
2.5.1.3 Dynamic Conservation Equations

Perhaps a simpler and more intuitive method for deriving the governing equations is by writing the
unsteady state conservation equations for refrigerant mass, energy and wall energy. For each region the
conservation of refrigerant energy is given in Equation 2.9 where U& is the rate of change of the total internal
energy of the refrigerant in the region considered, H& in = m& in hin is the rate of energy entering the region by means
of refrigerant mass, H& out = m& out hout is the rate of energy leaving the region by means of refrigerant mass,
Q& w = α i Ai (Tr − Tw ) is the rate of energy leaving the region through heat transfer to the heat exchanger wall, and
W& is the rate of moving boundary work being performed because of a change in the boundary between the regions.
Special care regarding the proper sign convention of this term is essential. In general
d
W& = ( PV ) = + P& V − PV& where the sign of each of these two terms depends on the definition of work. For
dt
this thesis, the work added by an increase in pressure is positive, + P& V , and the work done by an increasing the
volume by a change in the moving boundary is negative, − PV& .

U& = H& in − H& out − Q& w − W& ( 2.9 )
Similarly for each region, the conservation of wall energy is given in Equation 2.10 where E& w is the rate
of change of the total energy of the heat exchanger wall in the region considered, Q& a = α o Ao (Tw − Ta ) is the
rate of energy leaving the heat exchanger wall through heat transfer to the external fluid, and E& int is the rate of
energy being transferred to another region of the heat exchanger wall by a change in the boundary between the
regions.
E& w = Q& w − Q& a − E& int ( 2.10 )

The conservation of mass for the entire heat exchanger is given in Equation 2.11 where the rate of change
of the total refrigerant mass in the heat exchanger is equal to the difference between mass entering and leaving the
gas cooler.
m& = m& in − m& out ( 2.11 )
16
These equations together form a complete set for characterization of the dynamics, although it is more
convenient to expand the time derivative terms U& and E& w into more palatable forms. In doing so, there is
complete freedom in selecting the system states. By selecting the same variables as used in the first derivation
method, the equivalence of the two approaches can be verified. For a more detailed discussion of this exercise, see
[81] or [82].
2.5.2 Linearization
The moving boundary models developed for the heat exchangers are highly nonlinear. For analysis and
control design purposes, a linear model is desired. The standard linearization procedure, where the partial
derivatives of the nonlinear functions with respect to the states and inputs are calculated neglecting the 2nd and
higher order terms, will be used [52]. This is straightforward for the algebraic models of the mass flow devices.
However, the heat exchanger models have a unique form, and the linearization procedure for these components is as
follows.
The heat exchanger models are of the form of Equation 2.12. Assuming Z ( x, u ) is full rank for all x
and u , this can be rearranged as Equation . The assumption that Z ( x, u ) is full rank is true if the original modeling
assumptions are true. Specifically, as long as the length of any of the assumed regions is greater than zero, Z ( x, u )
will be invertible.
Z ( x, u ) ⋅ x& = f ( x, u ) ( 2.12 )
x& = Z ( x, u ) −1 f ( x, u )
( 2.13 )
= g ( x, u )
Using the assumption x = x o + δx , a local linearization of this, neglecting higher order terms, would be
Equation 2.14, and with the substitution δx = x − x o , becomes Equation 2.15.
⎡ ∂g ⎤ ⎡ ∂g ⎤
δx& = ⎢ ⎥δx + ⎢ ⎥δu ( 2.14 )
⎢⎣ ∂x x0 ,u0 ⎦⎥ ⎣⎢ ∂u ⎥
x0 , u 0 ⎦
⎡ ∂g ⎤ ⎡ ∂g ⎤
x& = ⎢ ⎥ (x − xo ) + ⎢ ⎥ (u − u o ) ( 2.15 )
⎣⎢ ∂x ⎥
x0 , u 0 ⎦ ⎣⎢ ∂u ⎥
x0 , u 0 ⎦
Expanding the first term of Equation 2.15 results in Equation 2.16. Likewise, expanding the second term
results in Equation 2.17. This is of the familiar form x& = Ax + Bu (Equation 2.18). This form will be denoted as
Equation 2.19, or in the standard form as Equation 2.20 using the substitutions in Equation 2.21.
−1
⎡ ∂g
⎢
⎤
[
⎥= Z ]
−1 ⎡ ∂f
⎢
⎤
⎥− Z[ ]
−2 ⎡ ∂Z
⎢
⎤
[ ]
⎥ f x0 , u 0
⎣⎢ ∂x ⎣⎢ ∂x ⎣⎢ ∂x
x0 , u 0 x0 , u 0
⎥
x0 , u 0 ⎦ ⎥
x0 , u 0 ⎦ ⎥ 123
x0 , u 0 ⎦
0 ( 2.16 )
= Z[ ]
−1 ⎡ ∂f
⎢
⎤
⎥
⎢⎣ ∂x
x0 , u 0
x0 , u 0 ⎥
⎦
17
⎡ ∂g
⎢
⎤
⎥= Z [ ]
−1 ⎡ ∂f
⎢
⎤
⎥ ( 2.17 )
⎣⎢ ∂u ⎣⎢ ∂u
x0 , u 0
⎥
x0 , u 0 ⎦ ⎥
x0 , u 0 ⎦
x& = Z [ x0 , u 0
]
−1 ⎡ ∂f
⎢
∂
⎤
⎥ (x − xo ) + Z [ x0 , u 0
]
−1 ⎡ ∂f
⎢
∂
⎤
⎥ (u − u o ) ( 2.18 )
⎣⎢ 42
1
x
4 3
⎥
x0 , u 0 ⎦ ⎣⎢ 42
1
u ⎥
x0 , u 0 ⎦
4 3
Fx Fu
x& = Z −1 Fx δx + Z −1 Fu δu ( 2.19 )
x& = Aδx + Bδu ( 2.20 )
−1
A = Z Fx
( 2.21 )
B = Z −1 Fu
The nonlinear output equations are denoted as Equation 2.22. The linearized version is then given as
Equation 2.23, or in the standard form as Equation 2.24, using the substitutions in Equation 2.25.
y = g ( x, u ) ( 2.22 )
δy = G x δx + Gu δu ( 2.23 )
δy = Cδx + Dδu ( 2.24 )
C = Gx
( 2.25 )
D = Gu
2.5.3 Time Scale Separation and Model Reduction
Although the derivation procedures presented above yield low order models of the heat exchangers, an
analysis of the eigenvalues for any of the single or two-phase heat exchangers reveal that the dynamics are
singularly perturbed [53]. Specifically, a single zero eigenvalue is present that corresponds to the conservation of
refrigerant mass (pure integration), while the remaining eigenvalues are separated by more than an order of
magnitude. As an example, the eigenvalues of a single and a two-phase heat exchanger are given from [82] in
Equations 2.26 and 2.27 respectively. Both confirm the presence of a zero eigenvalue, as well as indicating
multiple-time scale behavior or a singularly perturbed system. This behavior is observed at the component level of
the modeling procedure, but is a characteristic that remains after combining the component models to form the
overall system model.
⎡− 49.9 ⎤
λ ( Ac ) = ⎢⎢ − 0.123⎥⎥ ( 2.26 )
⎢⎣ 0 ⎥⎦
⎡− 53.4 ⎤
⎢ − 13.7 ⎥
⎢ ⎥
λ ( Ae ) = ⎢ − 0.411⎥ ( 2.27 )
⎢ ⎥
⎢ − 0.132⎥
⎢⎣ 0 ⎥⎦
18
For the purposes of control-oriented modeling, the fast dynamic modes evolve almost instantaneously
relative to the dominant dynamic modes. This presents an opportunity for model reduction, as these dynamic modes
could be replaced with their algebraic equivalents with minimal loss in accuracy. The nonessential nature of these
modes can be verified with analysis of the controllability and observability of the system model. The Hankel
singular values reveal that there are several weakly controllable/observable modes at the system level. Prior
research has shown the fast dynamics modes of a singularly perturbed system correspond to weakly
controllable/observable modes [30]. Finally, data driven model construction confirms that lower order models are
sufficient for capturing the essential system dynamics (see Chapter 5 or [81]).
Because the above derivation procedures result in a singularly perturbed representation of the system
dynamics, model reduction techniques can be employed to obtain a minimal representation of the system dynamics.
The objective is not only to derive reduced order component models, but also gain physical insight into which
physical phenomena are relatively fast/slow. Numerical methods of model reduction using balanced realizations
transform the state variables and thus eliminate their physical meaning. Conversely, singular perturbation techniques
provide means for justifying the elimination of fast dynamic states without losing the physical nature of the states.
Recall that the redundant nature of thermodynamic properties results in freedom in choosing the physical
representation of the dynamic states. With the proper choice of system states, the fast and slow dynamic modes are
virtually decoupled, and a reduced order model can be calculated by residualizing the fast states [95]. The state
space form of the full order model is reordered into the form of Eq. 2.28, and the reduced order model is calculated
using Eq. 2.29.
⎡ x& ⎤ ⎡ A11 A12 ⎤ ⎡ x ⎤ ⎡ B1 ⎤

⎢ z& ⎥ = ⎢ A +
A22 ⎥⎦ ⎢⎣ z ⎥⎦ ⎢⎣ B2 ⎥⎦
u
⎣ ⎦ ⎣ 21
( 2.28 )
⎡ x⎤
y = [C1 C 2 ]⎢ ⎥ + Du
⎣z⎦
−1
Ar = A11 − A12 A22 A21
−1
Br = B1 − A12 A22 B2
( 2.29 )
−1
C r = C1 − C 2 A22 A21
−1
Dr = D − C 2 A22 B2
For a more complete explanation of the approach and a demonstration of the efficacy, the interested reader
is referred to [82]. Reduced order models are calculated for the gas cooler, evaporator, and internal heat exchanger.
After combining these reduced order component models, the resulting system model approximates well the
eigenvalues of the full order system model. Furthermore, one additional state can be removed because of a
modeling redundancy. There is a conservation of mass equation for each of the heat exchangers. However, because
the net loss in refrigerant mass for one of these components equals the net mass gain of the other, these two dynamic
equations are redundant and one can be removed. The final reduced order model approximates the full order mode
with negligible error. More importantly, residualization of the alternative models yields the insight that the dynamic
19
modes associated with the refrigerant energy states (pressure, moving boundary length, etc.) are fast compared to the
wall temperature dynamics and the location of refrigerant mass, and can be replaced with algebraic equations.
2.5.4 Example Derivation: Single Phase Heat Exchanger

For pedagogical purposes, the entire approach for deriving the governing equations will be demonstrated in
this section for a single phase heat exchanger (e.g. gas cooler for transcritical vapor compression systems).
2.5.4.1 Conservation of Refrigerant Mass

The PDE for conservation of refrigerant mass is given in Equation 2.30. Each term of this equation is
integrated from z = 0 to z = Ltotal . Integrating the first term and assuming a constant cross-sectional area results
in Equation 2.31. Applying Leibniz’s equation results in Equation 2.32. Assuming an average density in the gas
cooler, ρ c , and performing the integration results in Equation 2.33. Taking the time derivative results in Equation
2.34. Selecting pressure and enthalpy as the independent variables for calculating thermodynamic properties, and
assuming an average enthalpy, hc , Equation 2.34 can be rewritten as Equation 2.35. Integrating the second term of
the PDE results in Equation 2.36. Combining the results of the integration results in Equation 2.37 for the
conservation of refrigerant mass.
∂( ρAcs ) ∂(m& )
+ =0 ( 2.30 )
∂t ∂z
∂ ( ρAcs ) ⎡ Ltotal ∂( ρ ) ⎤
Ltotal
∫
0
∂t
dz = Acs ⎢ ∫
⎣ 0 ∂t
dz ⎥
⎦
( 2.31 )
∂ ( ρAcs ) ⎡ d Ltotal ⎤
Ltotal
∫
0
∂t
dz = Acs ⎢
⎣ dt 0
∫ ρdz ⎥⎦ ( 2.32 )
∂( ρAcs )
Ltotal
dz = Acs ⎡⎢ ( ρ c Ltotal )⎤⎥

d
∫
0
∂t ⎣ dt ⎦
( 2.33 )
∂( ρAcs )
Ltotal
∫ dz = Acs Ltotal [ ρ& c ] ( 2.34 )

0
∂t
∂( ρAcs ) ⎡⎛ ∂ρ ⎞ ⎛ ⎞ ⎤
⎟ P&c + ⎜ ∂ρ
Ltotal
∫ dz = Acs Ltotal ⎢⎜ ⎟h&c ⎥ ( 2.35 )

∂t ⎜ ⎟ ⎜ ∂h ⎟ ⎥
0 ⎣⎢⎝ ∂Pc hc ⎠ ⎝ c Pc ⎠ ⎦
∂(m& )
Ltotal
∫
0
∂z
dz = m& out − m& in ( 2.36 )
⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ⎞ ⎤
⎢⎜⎜ ⎟ Acs Ltotal ⎥ P&c + ⎢⎜ ∂ρ ⎟ Acs Ltotal ⎥ h&c + m& out − m& in = 0 (2.37 )
⎢⎣⎝ ∂Pc ⎟ ⎥⎦ ⎢⎣⎜⎝ ∂hc ⎟ ⎥⎦
hc ⎠ Pc ⎠
2.5.4.2 Conservation of Refrigerant Energy

The PDE for conservation of refrigerant energy is given in Equation 2.38. Each term of this equation is
integrated from z = 0 to z = Ltotal . Integrating the first term and assuming a constant cross-sectional area results
in Equation 2.39. Applying Leibniz’s equation results in Equation 2.40. Assuming an average density in the gas
20
cooler, ρ c , and an average enthalpy, hc , results in Equation 2.41. Taking the time derivative results in Equation
2.42. Selecting pressure and enthalpy as the independent variables for calculating thermodynamic properties
Equation 2.42 can be rewritten as Equation 2.43, and then rearranged into Equation 2.44. Integrating the second
term of the PDE results in Equation 2.45. Taking the time derivative and performing the integration results in
Equations 2.46 and 2.47 respectively. Integrating the third term of the PDE results in Equation 2.48. Integrating the
right side of the equation and rearranging results in Equations 2.49 and 2.50 respectively. Combining the results of
the integration results in Equation 2.51 for the conservation of refrigerant energy.
∂( ρAcs h − Acs P ) ∂(m& h )

+ = piα i (Tw − Tr ) ( 2.38 )
∂t ∂z
∂ ( ρAcs h ) ⎡ Ltotal ∂( ρh ) ⎤
Ltotal
∫0 ∂t dz = Acs ⎢⎣ ∫0 ∂t dz ⎥⎦ ( 2.39 )
∂ ( ρAcs h ) ⎡ d Ltotal ⎤
Ltotal
∫
0
∂t
dz = Acs ⎢
⎣ dt 0
∫ ρhdz ⎥
⎦
( 2.40 )
∂ ( ρAcs h )
Ltotal
dz = Acs ⎡⎢ ( ρ c hc Ltotal )⎤⎥

d
∫
0
∂t ⎣ dt ⎦
( 2.41 )
∂( ρAcs h )
Ltotal
∫ dz = Acs Ltotal [ρ& c hc + ρ c h&c ] ( 2.42 )

0
∂t
∂( ρAcs h ) ⎡ ⎡⎛ ∂ρ ⎞ ⎛ ⎞ ⎤ ⎤
⎟ P&c + ⎜ ∂ρ
Ltotal
∫ dz = Acs Ltotal ⎢ ⎢⎜ ⎟h&c ⎥ hc + ρ c h&c ⎥ ( 2.43 )

0
∂t ⎢⎣ ⎢⎣⎜⎝ ∂Pc ⎟
hc ⎠
⎜ ∂h
⎝ c
⎟ ⎥
Pc ⎠ ⎦ ⎥⎦
∂( ρAcs h ) ⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ⎞ ⎤
⎟hc ⎥ Acs Ltotal P&c + ⎢⎜ ∂ρ
Ltotal
∫ dz = ⎢⎜ ⎟hc + ρ c ⎥ Acs Ltotal h&c ( 2.44 )

∂t ⎜ ⎟ ⎥ ⎢⎣⎜⎝ ∂hc ⎟
0 ⎣⎢⎝ ∂Pc hc ⎠ ⎦ Pc ⎠ ⎦⎥
∂ ( Acs P ) ∂( P )
Ltotal Ltotal
∫
0
∂t
dz = Acs ∫
0
∂t
dz ( 2.45 )
∂( Acs P )
Ltotal Ltotal
∫
0
∂t
dz = Acs ∫ P& dz
0
c ( 2.46 )
∂( Acs P )
Ltotal
∫
0
∂t
dz = Acs Ltotal P&c ( 2.47 )
∂(m& h )
Ltotal
∫
0
∂z
dz = m& out hout − m& in hin ( 2.48 )
Ltotal
∫ p α (T
0
i i w − Tr )dz = p i Ltotalα i (Tw − Tr ) ( 2.49 )
Ltotal
∫ p α (T
0
i i w − Tr )dz = α i Ai (Tw − Tr ) ( 2.50 )
21
⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ⎞ ⎤
⎢⎜⎜ ⎟hc + 1⎥ Acs Ltotal P&c + ⎢⎜ ∂ρ ⎟hc + ρ c ⎥ Acs Ltotal h&c
⎢⎣⎝ ∂Pc hc ⎟⎠ ⎥⎦ ⎢⎣⎜⎝ ∂hc ⎟
Pc ⎠ ⎥⎦ ( 2.51 )
+ m& out hout − m& in hin = α i Ai (Tw − Tr )

2.5.4.3 Conservation of Tube Wall Energy
The PDE for the conservation of tube wall energy is given in Equation 2.52. Integrating each side of this
equation from z = 0 to z = Ltotal results in Equation 2.53, and can be simplified to Equation 2.54.

∂t
(C p ρA)w Ltotal ∂(Tw ) = pi Ltotalα i (Tr − Tw ) + po Ltotalα o (Ta − Tw ) ( 2.53 )

∂t
(C p ρV )w ∂(Tw ) = α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ( 2.54 )

∂t
2.5.4.4 Governing Ordinary Differential Equations
Combining the final results of the integrated PDEs into a matrix form results in Equation 2.55, which is of
the Z ( x, u ) ⋅ x& = f ( x, u ) form, with states x = P [ hc Tw ] , and where the elements of the Z ( x, u ) matrix
T
are given in Table 2.2.
⎡ z11 z12 0 ⎤ ⎡ P&c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤

⎢z ⎢ ⎥
z 22 0 ⎥ ⎢ h&c ⎥ = ⎢ m& in − m& out ⎥ ( 2.55 )
⎢ 21 ⎥ ⎢ ⎥
⎣⎢ 0 0 z 33 ⎦⎥ ⎢⎣T&w ⎥⎦ ⎣⎢ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎦⎥
Table 2.2 Matrix Elements of Z ( x, u ) for the Gas Cooler
⎡⎛ ∂ρ c ⎞ ⎤
z11 ⎢⎜⎜ ⎟hc − 1⎥ Acs Ltotal
⎟
⎢⎣⎝ ∂Pc hc ⎠ ⎥⎦
⎡⎛ ∂ρ c ⎞ ⎤
z12 ⎢⎜⎜ ⎟hc + ρ c ⎥ Acs Ltotal
⎟
⎢⎣⎝ ∂hc Pc ⎠ ⎦⎥
⎛ ∂ρ c ⎞
z 21 ⎜ ⎟ Acs Ltotal
⎜ ∂P ⎟
⎝ c hc ⎠
⎛ ∂ρ c ⎞
⎜ ∂h ⎟
⎝ c Pc ⎠
z 33 (C p ρV )w
2.5.4.5 Dynamic Conservation Equations
The conservation of refrigerant energy for each region is given in Equation 2.56 where U& is the rate of
change of the total internal energy of the refrigerant in the region considered, H& in is the rate of energy entering the
22
region by means of refrigerant mass, H& out is the rate of energy leaving the region by means of refrigerant mass,
Q& w is the rate of energy leaving the region through heat transfer to the heat exchanger wall, and W& is the rate of
moving boundary work being performed because of a change in the boundary between the regions. Special care
d
regarding the proper sign convention of this term is essential. In general W& = ( PV ) = + P& V − PV& where the
dt
sign of each of these two terms depends on the definition of work. For this dissertation, the work added by an
increase in pressure is positive, + P& V , and the work done by an increasing the volume by a change in the moving
boundary is negative, − PV& .
U& = H& in − H& out − Q& w − W& ( 2.56 )
The conservation of wall energy for each region is given in Equation 2.57 where E& w is the rate of change
of the total energy of the heat exchanger wall in the region considered, Q& a is the rate of energy leaving the heat
exchanger wall through heat transfer to the external fluid, and E& int is the rate of energy being transferred to another
region of the heat exchanger wall by a change in the boundary between the regions.
E& w = Q& w − Q& a − E& int ( 2.57 )

The conservation of mass for the entire heat exchanger is given in Equation 2.58 where the rate of change
of the total refrigerant mass in the heat exchanger is equal to the difference between mass entering and leaving the
gas cooler.
m& = m& in − m& out ( 2.58 )

Applying the unsteady state form of the conservation of mass, refrigerant energy, and wall energy to the
single region of the gas cooler, and arranging in matrix form results in Equation 2.59.
⎡U& c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤

⎢ ⎥ ⎢ ⎥
⎢ m& c ⎥ = ⎢ m& in − m& out ⎥ ( 2.59 )
⎢ E& w ⎥ ⎢⎣ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎥⎦
⎣ ⎦
2.5.4.6 Equivalence
By applying the assumptions regarding operating condition, the time derivative terms in the above
equations can be expanded in terms of other variables. This allows some freedom in choosing the dynamic state
variables. The results of this approach can be shown to be equivalent to the results of the PDE approach by selecting
the same state variables for the energy approach as those given by the PDE approach.
Defining the total internal energy in terms of the total refrigerant mass and an average specific internal
energy U c = mc uc , the time derivative of this term can be expanded as in Equation 2.60. Defining the refrigerant
mass in terms of an average density, ρ c , and the internal volume, Vc = Acs Ltotal , yields Equation 2.61. Since
density and internal energy can be given as functions of pressure, Pc , and enthalpy, hc , the time derivative of these
23
variables can be written in terms of the desired state (Equation 2.62). Simplifying this expression and substituting
∂u 1
the formal definition of enthalpy h = u + Pv or h = u + P ρ , and its partial derivatives of =− and
∂P h ρ
∂u P ⎛ ∂ρ ⎞
= 1+ ⎜ ⎟ results in Equations 2.63 and 2.64 respectively. Likewise expanding the time derivative
∂h P ρ 2 ⎝ ∂h P ⎠
of total mass inventory results in Equations 2.65 and 2.66. Finally, defining the total wall energy as the product of
thermal capacitance and temperature Ew = C p ρV ( )T

w w
, the time derivative of this term can be written as in
Equation 2.67. Comparing Equations 2.64, 2.66, and 2.67 to Equation 2.55 reveals that they are equivalent.
U& c = m& c u c + mc u& c ( 2.60 )
U& c = ( ρ& c u c + ρ c u& c ) Acs Ltotal ( 2.61 )
⎡⎛ ⎛ ∂ρ ⎞ ⎛ ⎞ ⎞ ⎛⎛ ⎞ ⎛ ⎞ ⎞ ⎤
U& c = ⎢⎜ ⎜ c ⎟ P&c + ⎜ ∂ρ c ⎟h&c ⎟u c + ⎜ ⎜ ∂u c ⎟ P&c + ⎜ ∂u c ⎟h&c ⎟ ρ c ⎥ Acs Ltotal ( 2.62 )
⎢⎣⎜⎝ ⎜⎝ ∂Pc ⎟
hc ⎠
⎜ ∂h
⎝ c
⎟ ⎟
Pc ⎠ ⎠
⎜ ⎜ ∂P
⎝⎝ c
⎟
hc ⎠
⎜ ∂h
⎝ c
⎟ ⎟ ⎥
Pc ⎠ ⎠ ⎦
⎡⎛ ∂ρ ⎞ ⎛ ⎞ ⎤ ⎡⎛ ⎞ ⎛ ⎞ ⎤
U& c = ⎢⎜ c ⎟u c + ⎜ ∂u c ⎟ ρ c ⎥ Acs Ltotal P&c + ⎢⎜ ∂ρ c ⎟u c + ⎜ ∂u c ⎟ ρ c ⎥ Acs Ltotal h&c ( 2.63 )
⎢⎣⎜⎝ ∂Pc ⎟
hc ⎠
⎜ ∂P
⎝ c
⎟ ⎥
hc ⎠ ⎦ ⎢⎣⎜⎝ ∂hc ⎟
Pc ⎠
⎜ ∂h
⎝ c
⎟ ⎥
Pc ⎠ ⎦
⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ∂ρ ⎞ ⎤
U& c = ⎢⎜ c ⎟hc − 1⎥ Acs Ltotal P&c + ⎢⎜ c ⎟hc + ρ c ⎥ Acs Ltotal h&c ( 2.64 )
⎢⎣⎜⎝ ∂Pc hc ⎟⎠ ⎥⎦ ⎢⎣⎜⎝ ∂hc ⎟
Pc ⎠ ⎥⎦
m& c = ρ& c Acs Ltotal ( 2.65 )
⎡⎛ ∂ρ ⎞ ⎛ ∂ρ ⎞ ⎤
m& c = ⎢⎜ c ⎟ P&c + ⎜ c ⎟h&c ⎥ Acs Ltotal ( 2.66 )
⎜ ⎟ ⎜ ∂h ⎟ ⎥
⎣⎢⎝ ∂Pc hc ⎠ ⎝ c Pc ⎠ ⎦
E& = (C ρV ) T&
w p w w ( 2.67 )
2.5.4.7 Alternative State Representations

An alternative state representation can be derived by simply expanding the time derivative of U c = mc uc
in terms of pressure, Pc , and density, ρc (Equations 2.68 – 2.71). This results in what will be referred to as the
second representation, and will be denoted with the prime notation (i.e. Z ′( x ′, u ′) ⋅ x& ′ = f ′( x ′, u ′) ). When the
resulting equations are arranged in matrix form, the resulting model is of the same form as Equation 2.55, but with
alternative state variables of x′ = [P mc Tw ] . The resulting model is given in Equation 2.72 where the
T
elements of the Z ′( x ′, u ′) matrix are given in Table 2.3.
U& c = m& c u c + mc u& c ( 2.68 )
⎛ ⎛ ∂u ⎞ ⎛ ⎞ ⎞
U& c = m& c u c + mc ⎜ ⎜ c ⎟ P&c + ⎜ ∂u c ⎟ ρ& c ⎟ ( 2.69 )
⎜ ⎜ ∂P ⎟ ⎜ ∂ρ ⎟ ⎟
⎝⎝ c ρc ⎠ ⎝ c Pc ⎠ ⎠
24
⎛ ⎛ ∂u ⎞ ⎛ ⎞ m& c ⎞
U& c = m& c u c + mc ⎜ ⎜ c ⎟ P&c + ⎜ ∂u c ⎟ ⎟ ( 2.70 )
⎜ ⎜ ∂P ⎟ ⎜ ∂ρ ⎟V ⎟
⎝⎝ c ρc ⎠ ⎝ c Pc ⎠ c ⎠
⎡ ⎛ ∂u ⎞m ⎤ ⎡ ⎛ ∂u ⎞⎤
U& c = ⎢u c + ⎜ c ⎟ c ⎥ m& c + ⎢mc ⎜ c ⎟⎥ P&c ( 2.71 )
⎜ ∂ρ ⎟ ⎜ ⎟
⎣⎢ ⎝ c Pc ⎠ Vc ⎦⎥ ⎣⎢ ⎝ ∂Pc ρ c ⎠⎦⎥
′ z12
⎡ z11 ′ 0 ⎤ ⎡ P&c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤
⎢0 ⎢ ⎥
1 0 ⎥ ⎢m& c ⎥ = ⎢ m& in − m& out ⎥ ( 2.72 )
⎢ ⎥ ⎢ ⎥
⎣⎢ 0 0 z 33′ ⎦⎥ ⎢⎣T&w ⎥⎦ ⎣⎢ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎦⎥
Table 2.3 Matrix Elements of Z ′( x ′, u ′) for the Gas Cooler
⎛ ∂u ⎞
′
z11 mc ⎜ c ⎟
⎜ ∂P ⎟
⎝ c ρc ⎠
⎛ ∂u ⎞
′
z12 uc + ⎜ c ⎟ ρ c
⎜ ∂ρ ⎟
⎝ c Pc ⎠
′
z 33 (C p ρV )w
A useful result that stems from the equivalence of the various representations is that the matrix Z ′( x ′, u ′)
can be calculated directly from the elements of Z ( x, u ) . By observing that f ( x, u ) = f ′( x ′, u ′) , and thus
Z ( x, u ) ⋅ x& = Z ′( x ′, u ′) ⋅ x& ′ , algebraic manipulation results in the relationships in Equations 2.73 – 2.75 for
defining the elements of Z ′( x ′, u ′) . Evaluation of these elements in terms of their thermodynamic functions
confirms these relationships.
z11 z 22 − z12 z 21
′ =
z11 ( 2.73 )
z 22
z12
′ =
z12 ( 2.74 )
z 22
′ = z 33
z 33 ( 2.75 )
Finally, the original derivation result of the energy approach is defined as the third representation, and
denoted with the double prime notation (i.e. Z ′′( x ′′, u ′′) ⋅ x& ′′ = f ′′( x ′′, u ′′) ). This model is also of the same form
as Equation 2.55, but with alternative state variables of x′′ = U c [ mc Ew ] . The resulting model is given in
T
Equation 2.76 where the Z ′′( x ′′, u ′′) matrix is defined simply as Z ′′( x ′′, u ′′) = I
3x3
.
25
⎡U& c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤
⎢ ⎥ ⎢ ⎥
⎢ m& c ⎥ = ⎢ m& in − m& out ⎥ ( 2.76 )
⎢ E& w ⎥ ⎢⎣ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎥⎦
⎣ ⎦
2.5.4.8 Summary
The gas cooler is the simplest case for all the possible heat exchangers. However, the general methods
presented here can be extended to each of the other types of heat exchangers. More importantly, the equivalence of
model representations independent of modeling approach or choice of dynamic state variables extends to all other
heat exchangers as well. Furthermore, the advantages of the energy approach become more apparent and more
important when applied to the heat exchangers with phase changes. Specifically, the simplicity of the approach and
the ability to choose the state variables becomes of great interest when evaluating possibilities for model reduction.
2.6 Component Models

This section provides the detailed derivation for the basic components of a vapor compression cycle. This
includes compressors, expansion valves, and a heat exchangers under various conditions.
2.6.1 Compressor
2.6.1.1 Nonlinear Model
Two algebraic relationships are used to model the compressor. Mass flow rate is calculated in Equation
2.77 where ρ k = ρ (Pk ,in , hk ,in ) , and a volumetric efficiency, η vol , is assumed. Additionally, compression is
assumed to be an adiabatic process with an isentropic efficiency, and therefore the relationship between the entrance
and exit enthalpies is given in Equation 2.78, where hout ,isentropic = h(Pout , s k ) and s k = s (Pin , hin ) . For
implementation, this is rearranged to give Equation 2.79. Both the volumetric and isentropic efficiencies are
assumed to change with operating condition, and are given by semi-empirical maps (Equations 2.80 and 2.81). For
simulation purposes, the change of compressor speed is rate limited to reflect the limitations of a real compressor.
m& k = ω k Vk ρ kη vol ( 2.77 )
hout ,isentropic − hin

= ηk ( 2.78 )
hout − hin
hout =
ηk
1
[h out , isentropic ]
+ hin (η k − 1) ( 2.79 )
η vol = f 1 (Pratio , ω k ) ( 2.80 )

η k = f 2 (Pratio , ω k ) ( 2.81 )
2.6.1.2 Linearized Model
The nonlinear equations are linearized assuming inputs and outputs defined by Equations 2.82 and 2.83.
Thus y = f (u ) . A local linearization is given as δy = Dδu , where D is defined in Equation 2.84, and the
26
∂η k
matrix elements are listed in Table 2.4, where the gradients of the semi-empirical map are denoted as and
∂ω k
∂η k
.
∂Pratio
u = [ω k hin ]
T
Pin Pout ( 2.82 )
y = [m& k Tout ]
T
hout ( 2.83 )
⎡ d11 d12 d13 d14 ⎤

∂f
= D = ⎢⎢d 21 d 22 d 23 d 24 ⎥⎥ ( 2.84 )
∂u
⎢⎣d 31 d 32 d 33 d 34 ⎥⎦
Table 2.4 Matrix Elements for Equation 2.84
∂η vol
d11 Vk ρ kη vol + Vk ρ k ω k
∂ω k
⎛ ∂ρ ⎞
d12 ω k Vk ⎜ k ⎟η − ω V ρ ∂η vol ⎛ Pratio
⎜⎜
⎞
⎟⎟
⎜ ∂Pin ⎟ vol k k k
∂Pratio ⎝ Pin ⎠
⎝ hin ⎠
∂η vol ⎛ 1 ⎞
d13 ω k Vk ρ k ⎜⎜ ⎟⎟
∂Pratio ⎝ Pin ⎠
⎛ ∂ρ ⎞
d14 ω k Vk ⎜ k⎟η
⎜ ∂hin ⎟ vol
⎝ Pin ⎠
⎛ hin − hout , s ⎞ ∂η k
d 21 ⎜ ⎟
⎜ η 2 ⎟ ∂ω
⎝ k ⎠ k
⎛ hin − hout , s ⎞⎛ ∂η k ⎞⎛ − Pratio ⎞ ⎛ 1 ⎞⎛⎜ ∂hout , s ⎞⎛ ∂s
⎟⎜ k
⎞
⎟
d 22 ⎜ ⎟⎜ ⎟⎟⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟
⎜ η 2 ⎟⎜ ∂P
⎝ k ⎠⎝ ratio ⎠⎝ Pin ⎠ ⎝ηk ⎠⎜⎝ ∂s k Pout
⎟⎜ ∂Pin
⎠⎝ hin
⎟
⎠
⎛ hin − hout , s ⎞ ⎛ ∂η k ⎞⎛ 1 ⎞ ⎛ 1 ⎞⎛⎜ ∂hout , s ⎞
⎟
d 23 ⎜ ⎟⎜ ⎟⎟⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟
⎜ η 2 ⎟ ⎜ ∂P
⎝ k ⎠⎝ ratio ⎠⎝ Pin ⎠ ⎝ηk ⎠⎜⎝ ∂Pout sk
⎟
⎠
⎛ 1 ⎞⎛⎜ ∂hout , s ⎞⎛ ∂s
⎟⎜ k
⎞ ⎛η −1⎞
⎟+⎜ k
d 24 ⎜⎜ ⎟⎟ ⎟
⎝ηk ⎠⎜⎝ ∂s k ⎟⎜ ∂hin ⎟ ⎜⎝ η k ⎟⎠
Pout ⎠⎝ Pin ⎠
⎛ ∂T ⎞ ⎡⎛ hin − hout , s ⎞ ∂η k ⎤
⎜ out ⎟ ⎢⎜ ⎟
d 31
⎜ ∂hout ⎟ ⎢⎜ η 2 ⎟ ∂ω ⎥
⎝ Pout ⎠ ⎣⎝ k ⎠ k ⎦⎥
27
⎛ ∂T ⎞ ⎡⎛ hin − hout , s ⎞⎛ ∂η k ⎞⎛ − Pratio ⎞ ⎛ 1 ⎞⎛⎜ ∂hout , s ⎞⎛ ∂s ⎞⎤
d 32 ⎜ out ⎟ ⎢⎜ ⎟⎜ ⎟⎟⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ ⎟⎜ k ⎟⎥
⎜ ∂hout ⎟ ⎢⎜ η 2 ⎟⎜ ∂P
⎝ Pout ⎠ ⎣⎝ k ⎠⎝ ratio ⎠⎝ Pin ⎠ ⎝ η k ⎠⎜⎝ ∂s k Pout
⎟⎜ ∂Pin
⎠⎝
⎟⎥
hin ⎠ ⎦
⎛ ∂T ⎞ ⎛ ∂T ⎞ ⎡⎛ hin − hout , s ⎞⎛ ∂η k ⎞⎛ 1 ⎞ ⎛ 1 ⎞⎛⎜ ∂hout ,s ⎞⎤

d 33 ⎜ out ⎟ + ⎜ out ⎟ ⎢⎜ ⎟⎜ ⎟⎜ ⎟⎟ + ⎜⎜ ⎟⎟ ⎟⎥
⎜ ∂Pout ⎟ ⎜ ∂hout ⎟ ⎢⎜ η 2 ⎟⎜ ∂P ⎟⎜ P
⎝ hout ⎠ ⎝ Pout ⎠ ⎣⎝ k ⎠⎝ ratio ⎠⎝ in ⎠ ⎝ηk ⎠⎜⎝ ∂Pout sk
⎟⎥
⎠⎦
⎛ ∂T ⎞ ⎡⎛ 1 ⎞⎛⎜ ∂hout , s ⎞⎛ ∂s ⎞ ⎛ η − 1 ⎞⎤
d 34 ⎜ out ⎟ ⎢⎜ ⎟⎟ ⎟⎜ k ⎟+⎜ k ⎟⎥
⎜ ∂hout ⎟ ⎢⎜⎝ η k ⎠⎜⎝ ∂s k ⎟⎜ ∂hin ⎟ ⎜⎝ η k ⎟⎠⎥
⎝ Pout ⎠⎣ Pout ⎠⎝ Pin ⎠ ⎦
2.6.2 Orifice Expansion Valve
Two algebraic relationships are used to model the orifice valve. Mass flow rate is calculated assuming
standard orifice flow (Equation 2.85), and using a semi-empirical map for the discharge coefficient (Equation 2.86).
The discharge coefficient is assumed to be a function simply of pressure differential, ΔP = (Pin − Pout ) .
Additionally, expansion is assumed to be an isenthalpic process (Equation 2.87).
m& v = C d ρ (Pin − Pout ) ( 2.85 )
C d = f (ΔP ) ( 2.86 )
hv ,in = hv ,out ( 2.87 )

∂C d
matrix elements are listed in Table 2.5, where the slope of the semi-empirical map is denoted as . The fluid
∂ΔP
exiting the expansion valve is assumed to be a two-phase mixture.
u = [Pin hin ]
T
Pout ( 2.88 )
y = [m& v Tout ]
T
hout ( 2.89 )
⎡d11 d12 d13 ⎤

∂f
= D = ⎢⎢ 0 0 d 23 ⎥⎥ ( 2.90 )
∂u
⎢⎣ 0 d 32 0 ⎥⎦
28
⎛ ∂C d ⎞ ⎡ ⎛ ∂ρ ⎞⎤
⎟[ρ (Pin − Pout )]2 + 12 C d [ρ (Pin − Pout )] 2 ⎢ ρ + ΔP⎜ ⎟⎥
1
−1
d11 ⎜
⎝ ∂ΔP ⎠ ⎢ ⎜ ∂Pin ⎟⎥
⎣ ⎝ hin ⎠⎦
⎛ ∂C ⎞
− ⎜ d ⎟[ρ (Pin − Pout )]2 − 12 ρC d [ρ (Pin − Pout )] 2
1
−1
d12
⎝ ∂ΔP ⎠
⎛ ∂ρ ⎞
C d ΔP[ρ (Pin − Pout )] 2 ⎜ ⎟
1 −1
d13 2
⎜ ∂hin ⎟
⎝ Pin ⎠
d 23 1
⎛ dTout ⎞
d 32 ⎜⎜ ⎟⎟
⎝ dPsat ⎠
2.6.3 Automatic Expansion Valve
Two algebraic relationships are used to model the automatic expansion valve (AEV). Mass flow rate is
calculated assuming standard orifice flow (Equation 2.91), and using a semi-empirical map for the discharge
coefficient (Equation 2.92). The discharge coefficient is assumed to be a function of inlet pressure, Pin , and
pressure differential, ΔP = (Pin − Pout ) . Additionally, expansion is assumed to be an isenthalpic process
(Equation 2.93).
m& v = C d ρ (Pin − Pout ) ( 2.91 )
C d = f (Pin , ΔP ) ( 2.92 )
hv ,in = hv ,out ( 2.93 )
∂C d
matrix elements are listed in Table 2.6, where the gradient of the semi-empirical map is denoted as and
∂Pin
∂C d
. The fluid exiting the expansion valve is assumed to be a two-phase mixture.
∂ΔP
u = [Pin hin ]
T
Pout ( 2.94 )
y = [m& v Tout ]
T
hout ( 2.95 )
29
⎡d11 d12 d13 ⎤
∂f
= D = ⎢⎢ 0 0 d 23 ⎥⎥ ( 2.96 )
∂u
⎢⎣ 0 d 32 0 ⎥⎦
⎛ ∂C d ∂C d ⎞ ⎡ ⎛ ∂ρ ⎞⎤
⎟⎟[ρ (Pin − Pout )]2 + 12 C d [ρ (Pin − Pout )]− 2 ⎢ ρ + ΔP⎜ ⎟⎥
1 1
d11 ⎜⎜ +
⎝ ∂Pin ∂ΔP ⎠ ⎢ ⎜ ∂Pin ⎟⎥
⎣ ⎝ hin ⎠ ⎦
⎛ ∂C ⎞
− ⎜ d ⎟[ρ (Pin − Pout )]2 − 12 ρC d [ρ (Pin − Pout )] 2
1
−1
d12
⎝ ∂ΔP ⎠
⎛ ∂ρ ⎞
1 −1
d13 2
⎜ ∂hin ⎟
⎝ Pin ⎠
d 23 1
⎛ dTout ⎞
d 32 ⎜⎜ ⎟⎟
⎝ dPsat ⎠
2.6.4 Thermal Expansion Valve
This component is assumed to have a simple first order dynamic that is associated with the mechanical
feedback mechanism. The sensing bulb pressure, Pbulb , is assumed to respond with a first order time constant, τ ,
with respect to changes in the outlet temperature at the evaporator. This first order dynamic is due to the heat
transfer through the pipe and sensing bulb P&bulb = τ1 [Psat (Tero ) − Pbulb ], and the time constant is selected by the
user. Mass flow rate is calculated using Equation 2.97, where the discharge coefficient, C d , is given by a semi-
empirical map (Equation 2.98), which is a function of the pressure differential, ΔP = (Pin − Pout ) , and the area of
the valve opening, Av , which is a function of the bulb pressure, Pbulb , the evaporator pressure, Pevap , and the TEV
setting, θ. Additionally, expansion is assumed to be an isenthalpic process (Equation 2.100).
m& v = Av C d ρ (Pin − Pout ) ( 2.97 )
C d = f 1 (ΔP ) ( 2.98 )
Av = f 2 (Pbulb , Pevap , θ ) ( 2.99 )
hv ,in = hv ,out ( 2.100 )
30
Fbulb
Fevap
do
ks
Fs
Figure 2.9 TEV Operation (from [1]) Figure 2.10 TEV Valve (from [59])
The area of the valve opening of a thermostatic expansion valve is governed by a force balance which
operates on a diaphragm within the valve, Fbulb = Fevap + Fspring , where individual forces are given
as Fbulb = Pbulb Ad , Fevap = Pe Ad , and Fs = k s δ . Solving for the spring displacement gives
Ad
δ = (Pbulb − Pe ) . The diameter of the opening is given as a function of the spring displacement and the tip
ks
⎛ ⎛ α ⎞⎞
angle d i = 2(δ 0 − δ )⎜⎜ tan⎜ ⎟ ⎟⎟ , where the initial spring displacement is assumed to be a linear function of the
⎝ ⎝ 2 ⎠⎠
TEV adjustment screw position δ 0 = a 0 + a1θ . The area of the valve opening is then given as
π
Av =
4
(d 2
o − d i2 ) . Thus Av = b0 + [b1 + b2θ + b3 (Pbulb − Pe )] .
2

The nonlinear equations are linearized assuming states, inputs, and outputs defined by Equations 2.101,
δx& = Aδx + Bδu
2.102, and 2.103. A local linearization is given as , where A, B, C , D are defined in
δy = Cδx + Dδu
Equations 2.104-2.107, and the matrix elements are listed in Table 2.7, where the gradients of the semi-empirical
∂Av ∂Av ∂C d
maps are denoted as , and . The fluid exiting the expansion valve is assumed to be a two-phase
∂Pbulb ∂Pout ∂ΔP
mixture.
x = [Pbulb ]
T
( 2.101 )
u = [Tero hin ]
T
Pin Pout ( 2.102 )
y = [m& v Tout ]
T
hout ( 2.103 )
A = [a11 ] ( 2.104 )
31
B = [b11 0 0 0] ( 2.105 )
⎡c11 ⎤
C = ⎢⎢ 0 ⎥⎥ ( 2.106 )
⎢⎣ 0 ⎥⎦
⎡0 d12 d13 d14 ⎤
D = ⎢⎢0 0 0 d 24 ⎥⎥ ( 2.107 )
⎢⎣0 0 d 33 0 ⎥⎦
Table 2.7 Matrix Elements for Equations 2.104-2.107
1
a11 −
τ
1 ⎛ dPsat ⎞
b11 ⎜ ⎟⎟
τ ⎜⎝ dTero ⎠
⎛ ∂Av ⎞
c11 ⎜⎜ ⎟⎟C d ρ (Pin − Pout )
⎝ ∂Pbulb ⎠
⎛ ∂C d ⎞ ⎡ ⎛ ∂ρ ⎞⎤
⎟ Av [ρ (Pin − Pout )]2 + 12 Av C d [ρ (Pin − Pout )] 2 ⎢ ρ + ΔP⎜ ⎟⎥
1
−1
d12 ⎜
⎝ ∂ΔP ⎠ ⎢ ⎜ ∂Pin ⎟⎥
⎣ ⎝ hin ⎠ ⎦
⎛ ∂Av ⎞ ⎛ ∂C ⎞
⎟⎟C d [ρ (Pin − Pout )]2 − ⎜ d ⎟ Av [ρ (Pin − Pout )]2 − 12 ρAv C d [ρ (Pin − Pout )]− 2
1 1 1
d13 ⎜⎜
⎝ ∂Pout ⎠ ⎝ ∂ΔP ⎠
⎛ ∂ρ ⎞
1 −1
d14 2
⎜ ∂hin ⎟
⎝ Pin ⎠
d 24 1
⎛ dTout ⎞
d 33 ⎜⎜ ⎟⎟
⎝ dPsat ⎠
2.6.5 Electronic Expansion Valve
Two algebraic relationships are used to model the electronic expansion valve (EEV). Mass flow rate is
calculated assuming standard orifice flow (Equation 2.108), and using a semi-empirical map for the discharge
coefficient (Equation 2.109). The discharge coefficient is assumed to be a function of valve input, u v , and pressure
differential, ΔP = (Pin − Pout ) . Additionally, expansion is assumed to be an isenthalpic process (Equation 2.110).
For simulation purposes, the change in electronic input is rate limited to reflect the limitations of a real expansion
valve.
32
m& v = C d ρ (Pin − Pout ) ( 2.108 )
C d = f 1 (u v , ΔP ) ( 2.109 )
hv ,in = hv ,out ( 2.110 )
∂C d
matrix elements are listed in Table 2.8, where the gradients of the semi-empirical map are denoted as and
∂u v
∂C d
. The fluid exiting the expansion valve is assumed to be a two-phase mixture.
∂ΔP
u = [u v hin ]
T
Pin Pout ( 2.111 )
y = [m& v Tout ]
T
hout ( 2.112 )
⎡d11 d12 d13 d14 ⎤

∂f
= D = ⎢⎢ 0 0 0 d 24 ⎥⎥ ( 2.113 )
∂u
⎢⎣ 0 0 d 33 0 ⎥⎦
⎛ ∂C d ⎞
d11 ⎜⎜ ⎟⎟ ρ (Pin − Pout )
⎝ ∂u v ⎠
⎛ ∂C d ⎞ ⎡ Cd ⎤⎡ ⎛ ∂ρ ⎞⎤
d12 ⎜ ⎟ ρ (Pin − Pout ) + ⎢ ⎥ ⎢ ρ + (Pin − Pout )⎜ ⎟⎥
⎝ ∂ΔP ⎠ ⎢⎣ 2 ρ (Pin − Pout ) ⎥⎦ ⎢⎣ ⎜ ∂Pin
⎝
⎟⎥
hin ⎠ ⎦
⎛ ∂C ⎞ ⎡ Cd ρ ⎤
d13 − ⎜ d ⎟ ρ (Pin − Pout ) − ⎢ ⎥
⎝ ∂ΔP ⎠ ⎣⎢ 2 ρ (Pin − Pout ) ⎦⎥
⎡ Cd ⎤⎡ ⎛ ∂ρ ⎞⎤
d14 ⎢ ⎥ ⎢(Pin − Pout )⎜ ⎟⎥
⎢⎣ 2 ρ (Pin − Pout ) ⎥⎦ ⎢⎣ ⎜ ∂hin
⎝
⎟⎥
Pin ⎠ ⎦
d 24 1
⎛ dTout ⎞
d 33 ⎜⎜ ⎟⎟
⎝ dPsat ⎠
2.6.6 Single Phase Heat Exchanger

A single phase heat exchanger, such as the gas cooler found in a transcritical vapor compression cycle, is
modeled with one regions. There is no moving interface boundary and thus this is a purely lumped parameter
model. The governing ordinary differential equations (ODEs) are obtained by integrating the governing partial
33
differential equations (PDEs) (Eqs. 2.114-2.116) along the length of the assumed heat exchanger and assuming
lumped parameters.
∂( ρAcs ) ∂(m& )
+ =0 ( 2.114 )
∂t ∂z
∂( ρAcs h − Acs P ) ∂(m& h )
+ = piα i (Tw − Tr ) ( 2.115 )
∂t ∂z

∂t
Twall(t)
& inhin
m P(t) & outhout (t )
m
Single Phase
LTotal
Figure 2.11 Diagram: Evaporator with two fluid regions

Several assumptions are made regarding the lumped parameters. For heat transfer a weighted average of the
air temperature across the evaporator is assumed Ta = Ta ,in (μ ) + Ta ,out (1 − μ ) . In the single-phase region,
hin + hout
average properties are assumed, i.e. hc = , Tr = T (Pc , hc ) , and ρ c = ρ (Pc , hc ) .
2
The governing equations for the conservation of refrigerant mass, refrigerant energy, and heat exchanger
wall energy are given as follows:
2.6.6.1.1 Conservation of Refrigerant Mass
⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ⎞ ⎤
⎢⎜⎜ ⎟ Acs Ltotal ⎥ P&c + ⎢⎜ ∂ρ ⎟ Acs Ltotal ⎥ h&c + m& out − m& in = 0 ( 2.117 )
⎢⎣⎝ ∂Pc ⎟ ⎥⎦ ⎢⎣⎜⎝ ∂hc ⎟ ⎥⎦
hc ⎠ Pc ⎠
2.6.6.1.2 Conservation of Refrigerant Energy
⎡⎛ ∂ρ ⎞ ⎤ ⎡⎛ ⎞ ⎤
⎢⎜⎜ ⎟hc + 1⎥ Acs Ltotal P&c + ⎢⎜ ∂ρ ⎟hc + ρ c ⎥ Acs Ltotal h&c
⎢⎣⎝ ∂Pc hc ⎟⎠ ⎥⎦ ⎢⎣⎜⎝ ∂hc ⎟
Pc ⎠ ⎥⎦ ( 2.118 )
+ m& out hout − m& in hin = α i Ai (Tw − Tr )
2.6.6.1.3 Conservation of Tube Wall Energy
(C p ρV )w ∂(Tw ) = α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ( 2.119 )

∂t
2.6.6.1.4 Algebraic Combination and Simplification
34
Combining the final results of the integrated PDEs into a matrix form results in Equation 2.120, which is of
Z ( x, u ) ⋅ x& = f ( x, u ) form, with states x = [P hc Tw ] , and where the elements of the Z ( x, u ) matrix
T
the
are given in Table 2.9.
⎡ z11 z12 0 ⎤ ⎡ P&c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤

⎢z ⎢ ⎥
z 22 0 ⎥ ⎢ h&c ⎥ = ⎢ m& in − m& out ⎥ ( 2.120 )
⎢ 21 ⎥ ⎢ ⎥
⎢⎣ 0 0 z 33 ⎥⎦ ⎢⎣T&w ⎥⎦ ⎢⎣ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎥⎦
Table 2.9 Matrix Elements of Z ( x, u ) for the Single Phase Heat Exchanger
⎡⎛ ∂ρ c ⎞ ⎤
z11 ⎢⎜⎜ ⎟hc − 1⎥ Acs Ltotal
⎟
⎢⎣⎝ ∂Pc hc ⎠ ⎥⎦
⎡⎛ ∂ρ c ⎞ ⎤
z12 ⎢⎜⎜ ⎟hc + ρ c ⎥ Acs Ltotal
⎟
⎢⎣⎝ ∂hc Pc ⎠ ⎦⎥
⎛ ∂ρ c ⎞
⎜ ∂P ⎟
⎝ c hc ⎠
⎛ ∂ρ c ⎞
⎜ ∂h ⎟
⎝ c Pc ⎠
z 33 (C p ρV )w

Recall that the single phase heat exchanger could be modeled as Z ( x, u )x = f ( x, u ) with states given as
Equation 2.123 and where the function f ( x, u ) is defined in Equation 2.121. The model outputs are given as
nonlinear functions of the states and inputs, y = g ( x, u ) . Let the inputs and outputs be defined by Equations 2.124
and 2.125.
⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤

f ( x, u ) = ⎢⎢ m& in − m& out ⎥
⎥ ( 2.121 )
⎢⎣ α i Ai (Tr − Tw ) − α o Ao (Tw − Ta ) ⎥⎦
⎡ Pc ⎤ ⎡ Pc ⎤
⎢h ⎥ ⎢ hout ⎥
⎢ out ⎥ ⎢ ⎥
⎢ Tw ⎥ ⎢ Tw ⎥
g ( x, u ) = ⎢ ⎥=⎢ 1
( ) μ
( ) ⎥
⎢Ta ,out ⎥ ⎢ 1− μ Ta − 1− μ Ta ,in ⎥
( 2.122 )
⎢Tr ,out ⎥ ⎢ T (Pc , hout ) ⎥

⎢ ⎥ ⎢ ⎥
⎢⎣ mc ⎥⎦ ⎢⎣ ρ c Acs LTotal ⎥⎦
35
x = [Pc Tw ]
T
hc ( 2.123 )
u = [m& in m& a ]
T
m& out hin Ta ,in ( 2.124 )
y = [Pc mc ]
T
hout Tw Ta ,out Tr ,out ( 2.125 )
The energy balance for the air given a heat exchanger with n regions is given in Equation 2.126. Solving
for Ta and assuming two regions results in Equation 2.127.
⎡ n L ⎤
m& air C p ,air (Ta ,in − Ta ,out ) = α o Ao ⎢∑ i (Ta − Tw,i )⎥ ( 2.126 )
⎣ i =1 LTotal ⎦
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟Ta ,in + Tw1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
Ta = ( 2.127 )
⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
For the linearization, the partial derivatives of the average air temperature are required. First recall that the
air-side heat transfer coefficient is a function of mass flow rate of air. Specifically, we assume that the heat transfer
coefficient scales with Reynold’s number (where the prime denotes initial values) as given in Equation 2.128. Thus
the partial derivative of heat transfer coefficient with respect to mass flow rate of air can be written as Equation
2.129. The partial derivatives of air temperature are then given in Equations 2.130 – 2.132.
α o ⎛ Re l ⎞
m m
⎛ m& ⎞
=⎜ ⎟ = ⎜ air ⎟ ( 2.128 )
α o′ ⎝ Re′l ⎠ ⎝ m& ′air ⎠
∂α o ⎛ m& ⎞ ⎛ α ′ ⎞
m
= m⎜ air ⎟ ⎜ o ⎟ ( 2.129 )
∂m& air ⎝ m& ′air ⎠ ⎝ m& air ⎠
∂Ta 1
= ( 2.130 )
∂Tw,1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟
∂Ta ⎝ 1 − μ ⎠⎝ α o Ao ⎠
= ( 2.131 )
∂Ta ,in ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛1+ μ ⎞ ⎛ 1 ⎞⎛ n Li Tw,i ⎞
⎜⎜ ⎟⎟Ta ,in − ⎜⎜ ⎟⎟⎜⎜ ∑ ⎟⎟
∂Ta ⎝1− μ ⎠ ⎝ 1 − μ ⎠⎝ i =1 LTotal ⎠ ⎛⎜ C p ,air ⎞⎛ m& air ∂α o
⎟⎟⎜⎜1 −
⎞
⎟⎟
= ⎜α A ( 2.132 )
∂m& air ⎛ ⎛ 1 ⎞⎛ m& air C p ,air ⎞ ⎞
2
⎝ o o ⎠⎝ α o ∂m& air ⎠
⎜⎜ ⎟⎜ ⎟⎟ + 1⎟
⎜ ⎜⎝ 1 − μ ⎟⎠⎜ α A ⎟
⎝ ⎝ o o ⎠ ⎠
36
∂Tr ⎛⎜ ∂T ⎞ ∂T
⎟, 1 ⎛ ∂T ⎞
The following partial derivatives are also used: = r
= ⎜ ⎟,
∂Pc ⎜ ∂Pc ⎟ ∂hin 2 ⎜ ∂hc ⎟
⎝ hc ⎠ ⎝ Pc ⎠
∂Tr 1 ⎛ ∂T ⎞ ∂ρ c ⎛⎜ ∂ρ c ⎞ ∂ρ
⎟, 1 ⎛ ∂ρ ⎞ ⎛
⎟ , and ∂ρ c = 1 ⎜ ∂ρ c
⎞
= ⎜ ⎟, = c
= ⎜ c ⎟.
∂hout 2 ⎜⎝ ∂hc ⎟ ∂P ⎜ ∂P
Pc ⎠
⎟ ∂hin 2 ⎜ ∂hc ⎟ ∂hout 2 ⎜ ∂hc ⎟
c
⎝ c hc ⎠ ⎝ Pc ⎠ ⎝ Pc ⎠
The partial derivatives of the functions f ( x, u ) and g ( x, u ) with respect to the states and inputs are
defined in Equations 2.133 – 2.136, with the matrix elements listed in Table 2.10.
⎡ f x ,11 f x ,12 f x ,13 ⎤

∂f ⎢ ⎥
= Fx = ⎢ 0 0 0 ⎥ ( 2.133 )
∂x
⎢⎣ f x ,31 f x ,32 f x ,33 ⎥⎦
⎡ f u ,11 f u ,12 f u ,13 0 0 ⎤
∂f ⎢ ⎥
= Fu = ⎢ f u , 21 f u , 22 0 0 0 ⎥ ( 2.134 )
∂u
⎢⎣ 0 0 0 f u ,34 f u ,35 ⎥⎦
⎡ 1 0 0 ⎤
⎢ 0 g x , 22 0 ⎥⎥
⎢
∂g ⎢ 0 0 1 ⎥
= Gx = ⎢ ⎥ ( 2.135 )
∂x ⎢ 0 0 g x , 43 ⎥
⎢ g x ,51 g x ,52 0 ⎥
⎢ ⎥
⎢⎣ g x ,61 g x ,62 0 ⎥⎦
⎡0 0 0 0 0 ⎤
⎢0 0 g u , 23 0 0 ⎥
⎢ ⎥
∂g ⎢0 0 0 0 0 ⎥
= Gu = ⎢ ⎥ ( 2.136 )
∂u ⎢0 0 0 g u , 44 g u , 45 ⎥
⎢0 0 g u ,53 0 0 ⎥
⎢ ⎥
⎣⎢0 0 0 0 0 ⎦⎥
Table 2.10 Matrix Elements of Equations 2.133 – 2.136
⎛ ∂T ⎞
f x ,11 − α i Ai ⎜ r ⎟
⎜ ∂Pc ⎟
⎝ hc ⎠
⎛ ∂T ⎞
f x ,12 − 2mo − α i Ai ⎜ r ⎟
⎜ ∂hc ⎟
⎝ Pc ⎠
f x ,13 α i Ai
37
⎛ ∂T ⎞
f x ,31 α i Ai ⎜ ⎟r
⎜ ∂Pc ⎟
⎝ hc ⎠
⎛ ∂T ⎞
f x ,32 α i Ai ⎜ ⎟r
⎜ ∂hc ⎟
⎝ Pc ⎠
⎛ ∂T ⎞
f x ,33 − α i Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw ⎠
f u ,11 hin
f u ,12 − hout
f u ,13 min + mout
f u , 21 1
f u , 22 −1
⎛ ∂Ta ⎞
f u ,34 α o Ao ⎜⎜ ⎟
⎟
⎝ ∂Ta ,in ⎠
⎛ ∂α o ⎞ ⎛ ∂Ta ⎞
f u ,35 ⎜⎜ ⎟⎟ Ao (Ta − Tw ) + α o Ao ⎜⎜ ⎟⎟
⎝ ∂m& air ⎠ ⎝ ∂m& air ⎠
g x , 22 2
⎛ ∂T ⎞
g x , 43 2⎜⎜ a ⎟⎟
⎝ ∂Tw ⎠
⎛ ∂Tr ,out ⎞
g x ,51 ⎜ ⎟
⎜ ∂P ⎟
⎝ c hout ⎠
⎛ ∂Tr ,out ⎞
g x ,52 2⎜ ⎟
⎜ ∂h ⎟
⎝ out Pc ⎠
⎛ ∂ρ ⎞
g x , 61 ⎜ c ⎟V
⎜ ∂Pc ⎟ c
⎝ hc ⎠
⎛ ∂ρ ⎞
g x ,62 ⎜ c ⎟V
⎜ ∂hc ⎟ c
⎝ Pc ⎠
g u , 23 −1
⎛ ∂Ta ⎞
g u , 44 2⎜⎜ ⎟ −1
⎟
⎝ ∂Ta ,in ⎠
38
⎛ ∂Ta ⎞
g u , 45 2⎜⎜ ⎟⎟
⎝ ∂m& air ⎠
⎛ ∂Tr ,out ⎞
g u ,53 −⎜ ⎟
⎜ ∂h ⎟
⎝ out Pc ⎠
2.6.7 Evaporator
The evaporator is modeled with two regions: a two-phase region, and a superheat region. The moving
interface boundary between these regions is allowed to be a dynamic variable. The governing ordinary differential
equations (ODEs) are obtained by integrating the governing partial differential equations (PDEs) (Eqs. 2.114-2.116)
along the length of the assumed heat exchanger and assuming lumped parameters associated with each fluid region.
x=1
m& in hin
P(t) & outhout (t)
m
xin > 0
Two-Phase Superheat
L1(t) L2(t)
LTotal
Figure 2.12 Diagram: Evaporator with two fluid regions

air temperature across the evaporator is assumed Ta = Ta ,in (μ ) + Ta ,out (1 − μ ) . In the first region, the fluid
properties are determined by assuming a mean void fraction. Thus ρ 1 = ρ f (1 − γ ) + ρ g (γ ) , and so forth. In
hg + hout
the second region, average properties are assumed, i.e. h2 = , Tr 2 = T (Pe , h2 ) , and ρ 2 = ρ ( Pe , h2 ) .
2
For this derivation, the time derivative of the mean void fraction is neglected. This assumption is valid not only
because the change in mean void fraction tends to be small during transients considered, but also because its time
dependence is related to dynamic modes that are much faster than the dominant system dynamics. Thus any mean
void fraction dynamics can be replaced with their instantaneous, algebraic equivalents.
39
2.6.7.1.1 Conservation of Refrigerant Mass (Two-Phase and Superheat Regions)
⎛ dρ f dρ ⎞
⎜⎜ (1 − γ ) + g (γ )⎟⎟ Acs L1 P&e + (ρ f − ρ g )(1 − γ )Acs L&1 = m& in − m& int ( 2.137 )
⎝ dPe dPe ⎠
⎡⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞ ⎞⎤ 1 ⎛ ∂ρ ⎞
⎢⎜ 2 ⎟ + ⎜ 2 ⎟⎛⎜ dhg ⎟⎟⎥ AL2 P&e + ⎜ 2 ⎟ AL h&
⎢⎜⎝ ∂Pe ⎟ 2 ⎜ ∂h2 ⎟⎜⎝ dPe ⎠⎥⎦ 2 ⎜ ∂h2 ⎟ 2 out ( 2.138 )
⎣ h2 ⎠ ⎝ Pe ⎠ ⎝ Pe ⎠
+ (ρ g − ρ 2 )AL&1 = m& int − m& out
2.6.7.1.2 Conservation of Refrigerant Energy (Two-Phase and Superheat Regions)
⎛ d (ρ f h f ) d (ρ h ) ⎞
⎜⎜ (1 − γ ) + g g (γ ) − 1⎟⎟ Acs L1 P&e + (ρ f h f − ρ g hg )(1 − γ )Acs L&1
⎝ dPe dPe ⎠
( 2.139 )
⎛ L ⎞
= m& in hin − m& int hint + α i1 Ai ⎜⎜ 1 ⎟⎟(Tw1 − Tr1 )
⎝ LTotal ⎠
⎡⎛ ⎛ ∂ρ ⎞ ⎛ dh ⎞⎛ ⎞⎞ ⎤
⎢⎜ ⎜ 2 ⎟ + ⎛⎜ 1 ⎞⎟⎜ g ⎟⎜ ∂ρ 2 ⎟ ⎟h2 + ⎛⎜ 1 ⎞⎟⎜ g ⎟ ρ 2 − 1⎥ Acs L2 P&e
⎛ dh ⎞
⎢⎜ ⎜ ∂Pe h ⎟ ⎝ 2 ⎠⎜⎝ dPe ⎟⎠⎜ ∂h2 P ⎟ ⎟ ⎝ 2 ⎠⎜⎝ dPe ⎟⎠ ⎥
⎣⎝ ⎝ 2 ⎠ ⎝ e ⎠⎠ ⎦
⎡⎛ ∂ρ ⎞ ⎤ 1
+ ⎢⎜ 2 ⎟h2 + ρ 2 ⎥⎜ ⎟ Acs L2 h&out + (ρ g hg − ρ 2 h2 )Acs L&1
⎛ ⎞
( 2.140 )
⎜
⎢⎝ ∂h2 Pe ⎠ ⎟ ⎥⎝ 2 ⎠
⎣ ⎦
⎛ L ⎞
= m& int hint − m& out hout + α i 2 Ai ⎜⎜ 2 ⎟⎟(Tw 2 − Tr 2 )
⎝ LTotal ⎠
2.6.7.1.3 Conservation of Wall Energy (Two-Phase and Superheat Regions)
(C p ρV )w T&w1 = α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ( 2.141 )
⎡ ⎛ Tw 2 − Tw1 ⎞ & ⎤
(C p ρV )w ⎢T&w 2 − ⎜⎜ ⎟⎟ L1 ⎥ = α i 2 Ai (Tr 2 − Tw 2 ) + α o Ao (Ta − Tw 2 ) ( 2.142 )
⎣ ⎝ L2 ⎠ ⎦
The resulting six differential equations for conservation of refrigerant mass, refrigerant energy, and wall
energy for the two-phase and superheat regions only contain five explicit time derivatives: L&1 , P&e , h&out , T&w1 , and
T&w 2 . The equations can be combined to eliminate the variable m& int . This results in Equation 2.143, which is of the
Z ( x, u ) ⋅ x& = f ( x, u ) form, with states x = [L1 Tw1 Tw 2 ] , and where the elements of the
T
Pe hout
Z ( x, u ) matrix are given in Table 2.11.
40
⎡ ⎛ L1 ⎞ ⎤
0 ⎤ ⎡ L&1 ⎤ ⎢ min (hin − hg ) + α i1 Ai ⎜⎝ L ⎟(Tw1 − Tr1 ) ⎥
&
⎡ z11 z12 0 0
⎢ Total ⎠ ⎥
⎢z z 22 z 23 0 0 ⎥ ⎢ P&e ⎥ ⎢ ⎛ L2 ⎞ ⎥
⎢ 21 ⎥⎢ ⎥
⎢ m& out (hg − hout ) + α i 2 Ai ⎜ ⎟(Tw 2 − Tr 2 )⎥
⎢ z 31 z 32 z 33 0 0 ⎥ ⎢hout ⎥ =
& ⎝ LTotal ⎠ ( 2.143 )
⎢ ⎥⎢ & ⎥ ⎢ m& in − m& out
⎥
⎢0 0 0 z 44 0 ⎥ ⎢ Tw1 ⎥ ⎢ ⎥
⎢⎣ z 51 0 0 0 z 55 ⎥⎦ ⎢⎣Tw 2 ⎥⎦
& ⎢ α A (T − Tw1 ) − α A
i1 i ( Tw1 − T r1 ) ⎥
⎢ ⎥
o o a
⎣ α o Ao (Ta − Tw 2 ) − α i 2 Ai (Tw 2 − Tr 2 ) ⎦
Table 2.11 Matrix Elements of Z ( x, u ) for the Evaporator
z11 [ρ f (h f − h g )](1 − γ ) Acs

⎡⎛ d (ρ f h f ) dρ f ⎞ ⎛ d ( ρ g h g ) dρ g ⎞ ⎤
z12 ⎜
⎢⎜ dP − h ⎟
g ⎟(1 − γ ) + ⎜⎜ − hg ⎟⎟(γ ) − 1⎥ Acs L1
⎣⎝ e dPe ⎠ ⎝ dPe dPe ⎠ ⎦
z 21 ρ 2 (hg − h2 )Acs
⎡⎛ ⎛ ∂ρ 2 ⎞ ⎛ 1 ⎞⎛ ∂ρ 2 ⎞⎛ dh g ⎞ ⎞⎟ ⎛ ρ 2 ⎞⎛ dh g ⎞ ⎤
z 22 ⎢⎜⎜ ⎜⎜ ⎟ + ⎜ ⎟⎜
⎟ ⎝ 2 ⎠⎜ ∂h
⎟⎜⎜
⎟ dP ⎟⎟ ⎟(h2 − h g ) + ⎜ 2 ⎟⎜⎜ dP ⎟⎟ − 1⎥ Acs L2
⎢⎣⎝ ⎝ ∂Pe h2 ⎠ ⎝ 2 Pe ⎠⎝ e ⎠⎠ ⎝ ⎠⎝ e ⎠ ⎥⎦
⎡⎛ 1 ⎞⎛ ∂ρ 2 ⎞ ⎤
z 23 ⎢⎜ ⎟⎜⎜ ⎟(h2 − hg ) + ⎛⎜ ρ 2 ⎞⎟⎥ Acs L2
⎟
⎣⎢⎝ 2 ⎠⎝ ∂h2 Pe ⎠ ⎝ 2 ⎠⎦⎥
z 31 [(ρ g − ρ 2 ) + (ρ f − ρ g )(1 − γ )]Acs
⎡ ⎡⎛ ∂ρ 2 ⎞ 1 ⎛ ∂ρ 2 ⎞⎛ dhg ⎞⎤ ⎡⎛ dρ f ⎞ ⎛ dρ g ⎞ ⎤ ⎤
z 32 ⎢ ⎢⎜⎜ ⎟+ ⎜
⎟ 2 ⎜ ∂h
⎟⎜⎜
⎟ dP ⎟⎟⎥ L2 + ⎢⎜⎜ ⎟⎟(1 − γ ) + ⎜⎜ ⎟⎟(γ )⎥ L1 ⎥ Acs
⎢⎣ ⎢⎣⎝ ∂Pe h2 ⎠ ⎝ 2 Pe ⎠⎝ e ⎠⎥⎦ ⎣⎝ dPe ⎠ ⎝ dPe ⎠ ⎦ ⎥⎦
1 ⎛⎜ ∂ρ 2 ⎞
⎟ Acs L2
z 33
2 ⎜⎝ ∂h2 ⎟
Pe ⎠
z 44 (C p ρV )w
z 51 (C p ρV )w ⎛⎜ Tw1 − Tw2 ⎞⎟
⎝ L2 ⎠
z 55 (C p ρV )w

Recall that the evaporator could be modeled as Z ( x, u )x = f ( x, u ) with states given as Equation 2.146
and where the function f ( x, u ) is defined in Equation 2.144. The model outputs are given as nonlinear functions of
the states and inputs, y = g ( x, u ) . Let the inputs and outputs be defined by Equations 2.147 and 2.148.
41
⎡ ⎛ L1 ⎞ ⎤
⎢ m& in (hin − hg ) + α i1 Ai ⎜⎜ ⎟⎟(Tw1 − Tr1 ) ⎥
⎢ ⎝ LTotal ⎠ ⎥
⎢ ⎛ L2 ⎞ ⎥
⎢ m& out (hg − hout ) + α i 2 Ai ⎜⎜ ⎟⎟(Tw 2 − Tr 2 )⎥
f ( x, u ) = ⎢ ⎝ LTotal ⎠ ⎥ ( 2.144 )
⎢ m& in − m& out ⎥
⎢
⎢ α o Ao (Ta − Tw1 ) − α i1 Ai (Tw1 − Tr1 ) ⎥⎥
⎢⎣ α o Ao (Ta − Tw 2 ) − α i 2 Ai (Tw 2 − Tr 2 ) ⎥⎦
⎡ L1 ⎤ ⎡ L1 ⎤
⎢ P ⎥ ⎢ Pe ⎥
⎢ e ⎥ ⎢ ⎥
⎢ hout ⎥ ⎢ hout ⎥
⎢ ⎥ ⎢ ⎥
⎢ Tw1 ⎥ ⎢ Tw1 ⎥
g ( x, u ) = ⎢ Tw 2 ⎥ = ⎢ Tw 2 ⎥ ( 2.145 )
⎢ ⎥ ⎢ ⎥
⎢Ta ,out ⎥ ⎢
μ
( )
1− μ Ta − 1− μ Ta ,in
1
( )
⎥
⎢T ⎥ ⎢ T (Pe , hout ) ⎥
⎢ r ,out ⎥ ⎢ ⎥
⎢ Tr , sh ⎥ ⎢ T (Pe , hout ) − Tsat (Pe ) ⎥
⎣ e ⎦ ⎣ f [
⎢ m ⎥ ⎢ ρ (1 − γ ) + ρ (γ ) A L + ρ A L ⎥
g cs 1 ]
2 cs 2 ⎦
x = [L1 Tw1 Tw 2 ]
T
Pe hout ( 2.146 )
u = [m& in m& a ]
T
y = [L1 me ]
T
Pe hout Tw1 Tw 2 Ta ,out Tr ,out Tr , sh ( 2.148 )
⎡ n L ⎤
⎣ i =1 LTotal ⎦
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞ ⎡L T LT ⎤
⎜⎜ ⎟⎟⎜⎜ ⎟⎟Ta ,in + ⎢ 1 w,1 + 2 w, 2 ⎥
⎝ 1 − μ ⎠⎝ α o Ao ⎠ ⎣ LTotal LTotal ⎦
Ta = ( 2.150 )
⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
42
α o ⎛ Re l ⎞
m m
⎛ m& ⎞
=⎜ ⎟ = ⎜ air ⎟ ( 2.151 )
α o′ ⎝ Re′l ⎠ ⎝ m& ′air ⎠
∂α o ⎛ m& ⎞ ⎛ α ′ ⎞
m
= m⎜ air ⎟ ⎜ o ⎟ ( 2.152 )
∂m& air ′ ⎠ ⎝ m& air ⎠
⎝ m& air
⎛ Tw,1 − Tw, 2 ⎞
⎜⎜ ⎟
∂Ta ⎝ LTotal ⎟⎠
= ( 2.153 )
∂L1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L1 ⎞
⎜⎜ ⎟
= ( 2.154 )
∂Tw,1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L2 ⎞
⎜⎜ ⎟⎟
∂Ta ⎝ Total ⎠
L
= ( 2.155 )
∂Tw, 2 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟
∂Ta ⎝ 1 − μ ⎠⎝ α o Ao ⎠
= ( 2.156 )
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛1+ μ ⎞ ⎛ 1 ⎞⎛ n Li Tw,i ⎞
⎜⎜ ⎟⎟Ta ,in − ⎜⎜ ⎟⎟⎜⎜ ∑ ⎟⎟
⎟⎟⎜⎜1 −
⎞
⎟⎟
= ⎜α A ( 2.157 )
∂m& air ⎛ ⎛ 1 ⎞⎛ m& air C p ,air ⎞ ⎞
2
⎝ o o ⎠⎝ α o ∂m& air ⎠
⎜⎜ ⎜ ⎟⎟ + 1⎟
⎜ ⎜⎝ 1 − μ ⎟⎟⎠⎜ α A ⎟
⎝ ⎝ o o ⎠ ⎠
∂Tr 2 1 ⎛⎜ dTsat ∂Tr ,out ⎞
⎟,
The following partial derivatives are also used: = +
∂Pe 2 ⎜ dPe ∂Pe ⎟
⎝ hout ⎠
∂Tr 2 1 ⎛⎜ ∂Tr ,out ⎞ ∂ρ

⎟,
⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞ ∂ρ 2 1 ⎛ ∂ρ ⎞
= 2
=⎜ 2 ⎟+ ⎜ 2 ⎟⎜ ⎟⎟ , and = ⎜ 2 ⎟.
∂hout 2 ⎜ ∂hout ⎟ ∂Pe ⎜ ∂Pe ⎟ 2 ⎜ ∂h2 ⎟⎜ dPe ∂hout 2 ⎜ ∂h2 ⎟
⎝ Pe ⎠ ⎝ h2 ⎠ ⎝ Pe ⎠⎝ ⎠ ⎝ Pe ⎠
43
⎡ f x ,11 f x ,12 0 f x ,14 0 ⎤
⎢f f x , 22 f x , 23 0 f x , 25 ⎥⎥
∂f ⎢ x , 21
= Fx = ⎢ 0 0 0 0 0 ⎥ ( 2.158 )
∂x ⎢ ⎥
⎢ f x , 41 f x , 42 0 f x , 44 f x , 45 ⎥
⎢ f x ,51 f x ,52 f x ,53 f x ,54 f x ,55 ⎥⎦
⎣
⎡ f u ,11 0 f u ,13 0 0 ⎤
⎢ 0 f u , 22 0 0 0 ⎥⎥
∂f ⎢
= Fu = ⎢ f u ,31 f u ,32 0 0 0 ⎥ ( 2.159 )
∂u ⎢ ⎥
⎢ 0 0 0 f u , 44 f u , 45 ⎥
⎢ 0 0 0 f u ,54 f u ,55 ⎥⎦
⎣
⎡ 1 0 0 0 0 ⎤
⎢ 0 1 0 0 0 ⎥⎥
⎢
⎢ 0 0 1 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 1 0 ⎥
∂g
= Gx = ⎢ 0 0 0 0 1 ⎥ ( 2.160 )
∂x ⎢ ⎥
⎢ g x ,61 0 0 g x , 64 g x ,65 ⎥
⎢ 0 g x , 72 g x ,73 0 0 ⎥
⎢ ⎥
⎢ 0 g x ,82 g x ,83 0 0 ⎥
⎢g g x ,92 g x ,93 0 0 ⎥⎦
⎣ x ,91
⎡0 0 0 0 0 ⎤
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢ ⎥
∂g ⎢0 0 0 0 0 ⎥
= G u = ⎢0 0 0 0 0 ⎥ ( 2.161 )
∂u ⎢ ⎥
⎢0 0 0 g u , 64 g u , 65 ⎥
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢0 0 0 0 0 ⎥⎦
⎣
α i1 Ai
f x ,11 (Tw1 − Tr1 )
LTotal
dhg L1 ⎛ dTr1 ⎞
f x ,12 − m& in − α i1 Ai ⎜ ⎟
dPe LTotal ⎝ dPe ⎠
44
L1
f x ,14 α i1 Ai
LTotal
α i 2 Ai
f x , 21 − (Tw2 − Tr 2 )
LTotal
dhg L2 ⎛ ∂Tr 2 ⎞
f x , 22 m& out − α i 2 Ai ⎜⎜ ⎟⎟
dPe LTotal ⎝ ∂Pe ⎠
L2 ⎛ ∂Tr 2 ⎞
f x , 23 − m& out − α i 2 Ai ⎜⎜ ⎟⎟
LTotal ⎝ ∂hout ⎠
L2
f x , 25 α i 2 Ai
LTotal
⎛ ∂Ta ⎞
f x , 41 α o Ao ⎜⎜ ⎟⎟
⎝ ∂L1 ⎠
⎛ dTr1 ⎞
f x , 42 α i1 Ai ⎜⎜ ⎟⎟
⎝ dPe ⎠
⎛ ∂T ⎞
f x , 44 − α i1 Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw1 ⎠
⎛ ∂Ta ⎞
f x , 45 α o Ao ⎜⎜ ⎟⎟
⎝ ∂Tw 2 ⎠
⎛ ∂Ta ⎞
f x ,51 α o Ao ⎜⎜ ⎟⎟
⎝ ∂L1 ⎠
⎛ ∂Tr 2 ⎞
f x ,52 α i 2 Ai ⎜⎜ ⎟⎟
⎝ ∂Pe ⎠
⎛ ∂Tr 2 ⎞
f x ,53 α i 2 Ai ⎜⎜ ⎟⎟
⎝ ∂hout ⎠
⎛ ∂Ta ⎞
f x ,54 α o Ao ⎜⎜ ⎟⎟
∂
⎝ w1 ⎠
T
⎛ ∂T ⎞
f x ,55 − α i 2 Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw 2 ⎠
f u ,11 hin − hg
f u ,13 m& in
f u , 22 hg − hout
f u ,31 1
45
f u ,32 −1
⎛ ∂Ta ⎞
f u , 44 α o Ao ⎜⎜ ⎟⎟
⎝ ∂Tai ⎠
⎛ ∂α o ⎞ ⎛ ∂Ta ⎞
f u , 45 ⎜⎜ ⎟⎟ Ao (Ta − Tw1 ) + α o Ao ⎜⎜ ⎟⎟
⎝ ∂m& air ⎠ ⎝ ∂m& air ⎠
⎛ ∂Ta ⎞
f u ,54 α o Ao ⎜⎜ ⎟⎟
⎝ ∂Tai ⎠
⎛ ∂α o ⎞ ⎛ ∂Ta ⎞
f u ,55 ⎜⎜ ⎟⎟ Ao (Ta − Tw 2 ) + α o Ao ⎜⎜ ⎟⎟
⎝ ∂m& air ⎠ ⎝ ∂m& air ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g x , 61 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂L1 ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g x ,64 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tw1 ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g x ,65 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tw 2 ⎠
⎛ ∂Tr ,out ⎞
g x ,72 ⎜ ⎟
⎜ ∂Pe ⎟
⎝ hout ⎠
⎛ ∂Tr ,out ⎞
g x , 73 ⎜ ⎟
⎜ ∂hout ⎟
⎝ Pe ⎠
⎛ ∂Tr ,out ⎞ ⎛ dT ⎞
g x ,82 ⎜ ⎟ − ⎜ sat ⎟⎟
⎜ ∂Pe ⎟ ⎜⎝ dPe ⎠
⎝ hout ⎠
⎛ ∂Tr ,out ⎞
g x ,83 ⎜ ⎟
⎜ ∂hout ⎟
⎝ Pe ⎠
g x ,91 [[ρ f (1 − γ ) + ρ g (γ )] − ρ 2 ]Acs

⎡⎛ dρ f ⎞ ⎛ dρ ⎞ ⎤ ⎛ ∂ρ ⎞
g x 92 ⎢⎜⎜ ⎟⎟(1 − γ ) + ⎜⎜ g ⎟⎟(γ )⎥ Acs L1 + ⎜⎜ 2 ⎟⎟ Acs L2
⎣⎢⎝ dPe ⎠ ⎝ dPe ⎠ ⎦⎥ ⎝ ∂Pe ⎠
1 ⎛⎜ ∂ρ 2 ⎞
⎟A L
g x ,93
2 ⎜ ∂h2 ⎟ cs 2
⎝ Pe ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞ ⎛ μ ⎞
g u , 64 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tai ⎠ ⎝1− μ ⎠
46
⎛ 1 ⎞⎛ ∂Ta ⎞
g u , 65 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂ma ⎠
2.6.8 Condenser
The condenser is modeled with three regions: a superheat region, a two-phase region, and a subcool region.
The moving interface boundaries between these regions are allowed to be dynamic variables. As with the
evaporator, the governing ordinary differential equations (ODEs) are obtained by integrating the governing partial
differential equations (PDEs) (Eqs. 2.114-2.116) along the length of the assumed heat exchanger and assuming
lumped parameters associated with each fluid region.
x=0 x=1
Twall,1(t) Twall,2(t) Twall,3(t)
& in hin
m P(t) & outhout (t)
m
Superheat Two-Phase Subcool
L1(t) L2(t) L3(t)
LTotal
Figure 2.13 Diagram: Condenser with three fluid regions

air temperature across the evaporator is assumed Ta = Ta ,in (μ ) + Ta ,out (1 − μ ) . In the second region, the fluid
properties are determined by assuming a mean void fraction. Thus ρ 2 = ρ f (1 − γ ) + ρ g (γ ) , and so forth. In
hin + hg hout + h f
the first and third regions, average properties are assumed, i.e. h1 = and h3 = , such that
2 2
Tr1 = T (Pc , h1 ) , ρ 3 = ρ (Pc , h3 ) , etc. Again, the time derivative of the mean void fraction is neglected.
2.6.8.1.1 Conservation of Refrigerant Mass (Superheat, Two-phase, and Subcool Regions)
⎡⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞⎤ 1 ⎛ ∂ρ ⎞
(ρ − ρ g )AL&1 + ⎢⎜ 1 ⎟+ ⎜ 1 ⎟⎜
⎟⎜⎝ dPc
⎟⎟⎥ AL1 P&c + ⎜ 1 ⎟ AL h& = m& − m& ( 2.162 )
⎢⎜⎝ ∂Pc ⎟ 2 ⎜ ∂h1 2 ⎜ ∂h1 ⎟ 1 i
1 i int 1
⎣ h1 ⎠ ⎝ Pc ⎠ ⎠⎥⎦ ⎝ Pc ⎠
⎛ dρ dρ ⎞
(ρ g − ρ f )AL&1 + (ρ g − ρ f )γAL&2 + ⎜⎜ f (1 − γ ) + g γ ⎟⎟ AL2 P&c = m& int 1 − m& int 2 ( 2.163 )
⎝ dPc dPc ⎠
47
⎡⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dh f ⎞⎤ 1 ⎛ ∂ρ ⎞
A(ρ f − ρ 3 )(L&1 + L& 2 ) + ⎢⎜ 3 ⎟+ ⎜ 3 ⎟⎜ ⎟⎟⎥ AL3 P&c + ⎜ 3 ⎟ AL h& = m&
int 2 − mo
& ( 2.164 )
⎢⎜⎝ ∂Pc ⎟ 2 ⎜ ∂h3 ⎟⎜⎝ dPc ⎠⎥⎦ 2 ⎜ ∂h3 ⎟ 3 o
⎣ h3 ⎠ ⎝ Pc ⎠ ⎝ Pc ⎠
2.6.8.1.2 Conservation of Refrigerant Energy (Superheat, Two-phase, and Subcool Regions)
⎡⎛ ∂ρ ⎞ ⎛ ⎞⎛ dhg ⎤ ⎡ 1 ⎛ ∂ρ ⎞ ⎤
⎢⎜ 1 ⎟ h + 1 ⎜ ∂ρ1 ⎟⎜ ⎞
⎟⎟ h1 +
1 dhg
ρ1 − 1⎥ AL1 P&c + ⎢ ⎜ 1 ⎟h + 1 ρ ⎥ AL h&
⎢⎜⎝ ∂Pc ⎟ 1 2 ⎜ ∂h1 ⎟⎜⎝ dPc
⎠ 2 dPc ⎥ ⎢ 2 ⎜⎝ ∂h1 ⎟ 1 2 1⎥ 1 i ( 2.165 )
⎣ h1 ⎠ ⎝ Pc ⎠ ⎦ ⎣ Pc ⎠ ⎦
+ (ρ1 h1 − ρ g hg )AL&1 = m& i hi − m& int 1 hg + α i1 Ai 1 (Tw1 − Tr1 )

L
Ltotal
⎡ d (ρ f h f ) d (ρ h ) ⎤
(ρ h
g g − ρ f h f )AL&1 + (ρ g hg − ρ f h f )γ AL&2 + ⎢ (1 − γ ) + g g (γ ) − 1⎥ AL2 P&c
⎣ dPc dPc ⎦ ( 2.166 )
⎛ L ⎞
= m& int 1hg − m& int 2 h f + α i 2 Ai ⎜⎜ 2 ⎟⎟(Tw 2 − Tr 2 )
⎝ Ltotal ⎠
⎡⎛ ∂ρ ⎞ ⎛ ⎞⎛ dh f ⎤ ⎡ 1 ⎛ ∂ρ ⎞ ⎤
⎢⎜ 3 ⎟ h + 1 ⎜ ∂ρ 3 ⎟⎜ ⎞ 1 ⎛ dh ⎞
⎟⎟h3 + ⎜⎜ f ⎟⎟ ρ 3 − 1⎥ AL3 P&c + ⎢ ⎜ 3 ⎟h + 1 ρ ⎥ AL h&
⎢⎜⎝ ∂Pc ⎟ 3
2 ⎜ ∂h3 ⎟⎜⎝ dPc ⎠ 2 ⎝ dPc ⎠ ⎥ ⎢ 2 ⎜⎝ ∂h3 ⎟ 3 2 3⎥ 3 o ( 2.167 )
⎣ h3 ⎠ ⎝ Pc ⎠ ⎦ ⎣ Pc ⎠ ⎦
+ A(ρ f h f − ρ 3 h3 )(L&1 + L& 2 ) = m& int 2 h f − m& o ho + α i 3 Ai 3 (Tw3 − Tr 3 )

L
Ltotal
2.6.8.1.3 Conservation of Wall Energy (Superheat, Two-phase, and Subcool Regions)
⎛ T − Tw 2 ⎞ &
mw C p , wT&w1 + m w C p , w ⎜⎜ w1 ⎟⎟ L1 = α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ( 2.168 )
⎝ L1 ⎠
m C T = α A (T − T ) + α A (T − T )
w p,w
&
w2 i2 i r2 w2 o o a w2
( 2.169 )
⎛ T − Tw 3 ⎞ &
m w C p , wT&w3 + m w C p , w ⎜⎜ w 2 ⎟⎟(L1 + L& 2 ) = α i 2 Ai (Tr 3 − Tw3 ) + α o Ao (Ta − Tw3 ) ( 2.170 )
⎝ L3 ⎠
The resulting nine differential equations for conservation of refrigerant mass, refrigerant energy, and wall
energy for the two-phase and superheat regions only contain seven explicit time derivatives. The equations can be
& int,1 and m& int, 2 . This results in Equation 2.171, which is of the
combined to eliminate the variables m
Z ( x, u ) ⋅ x& = f ( x, u ) form, with states x = [L1 L2 Tw1 Tw 2 Tw3 ] , and where the elements
T
Pc hout
of the Z ( x, u ) matrix are given in Table 2.13.
48
⎡ ⎤
⎢ m& i (hi − hg ) + α i1 Ai L (Tw1 − Tr1 ) ⎥
L1
⎡ z11 0 z13 0 0 0 0 ⎤ ⎡ L&1 ⎤ ⎢ total
⎥
⎢z z 22 z 23 z 24 0 0 0 ⎥⎢ & ⎥ ⎢m& h − m& h + α A L2 (T − T )⎥
⎢ 21 ⎥ ⎢ L2 ⎥ ⎢ i g o f i2 i
Ltotal
w2 r2
⎥
⎢ z 31 z 32 z 33 z 34 0 0 0 ⎥ ⎢ P& ⎥ ⎢ ⎥
⎢
⎢ z 41 z 42 z 43 z 44 0 0
⎥⎢ ⎥
0 ⎥ ⎢ h&o ⎥ = ⎢ m& o (h f − h o ) + α A
i3 i
L3
(T w3 − T r3 ) ⎥ ( 2.171 )
⎢ Ltotal ⎥
⎢ z 51 0 0 0 z 55 0 0 ⎥ ⎢T&w1 ⎥ ⎢ m& i − m& o ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢0 0 0 0 0 z 66 0 ⎥ ⎢T&w 2 ⎥ ⎢ α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ⎥
⎢z
⎣ 71 z 72 0 0 0 0 z 77 ⎥⎦ ⎢⎣T&w3 ⎥⎦ ⎢ α A (T − T ) + α A (T − T ) ⎥
⎢ i2 i r 2 w2 o o a w2
⎥
⎢⎣ α i 3 Ai (Tr 3 − Tw3 ) + α o Ao (Ta − Tw3 ) ⎥⎦
Table 2.13 Matrix Elements of Z ( x, u ) for the Condenser
z11 ρ1 (h1 − hg )Acs

⎡⎛ ⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞ ⎞⎟ ⎤
1 ⎛ dh ⎞
z13 ⎢⎜ ⎜ 1 ⎟+ ⎜ 1 ⎟⎜ ⎟⎟ (h1 − hg ) + ⎜⎜ g ⎟⎟ ρ1 − 1⎥ Acs L1
⎢⎜ ⎜ ∂Pc ⎟ 2 ⎜ ∂h1 ⎟⎜⎝ dPc ⎠⎠
⎟ 2 ⎝ dPc ⎠ ⎥
⎣⎝ ⎝ h1 ⎠ ⎝ Pc ⎠ ⎦
z 21 (ρ h 1 g − ρ 3 h f )Acs
z 22 [(ρ g hg − ρ f h f )γ + (ρ f − ρ 3 )h f Acs ]
⎡⎛ ⎛
⎢⎜ ⎜ ∂ρ1
⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞ ⎞⎟ ⎛ d (ρ f h f ) d (ρ g hg ) ⎞ ⎛ ⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dh f ⎞ ⎞⎟ ⎤
⎟+ ⎜ 1 ⎟⎜ ⎟⎟ hg L1 + ⎜⎜ (1− γ )+ (γ ) − 1⎟⎟ L2 + ⎜ ⎜ 3 ⎟+ ⎜ 3 ⎟⎜ ⎟⎟ h f L3 ⎥ Ac
z 23 ⎢⎜ ⎜ ∂Pc ⎟ 2 ⎜ ∂h1 ⎟⎜⎝ dPc ⎠⎠
⎟
⎝ dP dP ⎠
⎜ ⎜ ∂P ⎟ 2 ⎜ ∂h3 ⎟⎜⎝ dPc ⎠⎠
⎟ ⎥
⎣⎝ ⎝ h1 ⎠ ⎝ Pc ⎠ ⎝⎝ c h3 ⎠ ⎝ Pc ⎠ ⎦
1 ⎛⎜ ∂ρ 3 ⎞
⎟A L h
z 24
2 ⎜ ∂h3 ⎟ cs 3 f
⎝ Pc ⎠
z 31 ρ 3 (h f − h3 )Acs
z 32 ρ 3 (h f − h3 )Acs
⎡⎛ ⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dh f ⎞ ⎞⎟ ⎤
1 ⎛ dh ⎞
z 33 ⎢⎜ ⎜ 3 ⎟+ ⎜ 3 ⎟⎜ ⎟⎟ (h3 − h f ) + ⎜⎜ f ⎟⎟ ρ 3 − 1⎥ Acs L3
⎢⎜ ⎜ ∂Pc ⎟ 2 ⎜ ∂h3 ⎟⎜⎝ dPc ⎠⎠
⎟ 2 ⎝ dPc ⎠ ⎥
⎣⎝ ⎝ h3 ⎠ ⎝ Pc ⎠ ⎦
⎡ 1 ⎛ ∂ρ ⎞ ⎤
z 34 ⎢ ⎜ 3 ⎟(h − h ) + 1 ρ ⎥ A L
⎢ 2 ⎜⎝ ∂h3 ⎟ 3 3 cs 3
2 ⎥
f
⎣ Pc ⎠ ⎦
z 41 (ρ1 − ρ 3 )Acs
z 42 [(ρ g − ρ f )γ + (ρ f − ρ 3 ) Acs ]
⎡⎛ ⎛ ⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞ ⎞⎟ ⎛ dρ dρ ⎞ ⎛ ⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dh f ⎞ ⎞⎟ ⎤
z 43 ⎢⎜ ⎜ ∂ρ1 ⎟+ ⎜ 1 ⎟⎜ ⎟⎟ L1 + ⎜⎜ f (1 − γ ) + g γ ⎟⎟ L2 + ⎜ ⎜ 3 ⎟+ ⎜ 3 ⎟⎜ ⎟⎟ L3 ⎥ Acs
⎢⎜ ⎜ ∂Pc ⎟ 2 ⎜ ∂h1 ⎟⎜ dPc ⎟ ⎝ dP dP ⎠ ⎜ ⎜ ∂P ⎟ 2 ⎜ ∂h3 ⎟⎜ dPc ⎟
⎠ ⎠ ⎥⎦
⎝ Pc ⎠⎝ ⎠⎠ Pc ⎠⎝
⎣⎝ ⎝ h1 ⎠ ⎝⎝ c h3 ⎠ ⎝
49
1 ⎛⎜ ∂ρ 3 ⎞
⎟A L
z 44
2 ⎜ ∂h3 ⎟ cs 3
⎝ Pc ⎠
⎛ T − Tw 2 ⎞
z 51 m w C p , w ⎜⎜ w1 ⎟⎟
⎝ L1 ⎠
z 55 mw C p,w
z 66 mw C p,w
⎛ T − Tw3 ⎞
z 71 m w C p , w ⎜⎜ w 2 ⎟⎟
⎝ L3 ⎠
⎛ T − Tw3 ⎞
z 72 m w C p , w ⎜⎜ w 2 ⎟⎟
⎝ L3 ⎠
z 77 mw C p,w

Recall that the condenser could be modeled as Z ( x, u )x = f ( x, u ) with states given as Equation 2.174
⎡
⎢ m& i (hi − hg ) + α i1 Ai L (Tw1 − Tr1 ) ⎥
L1 ⎤
⎢ total
⎥
⎢m& h − m& h + α A L2 (T − T )⎥
⎢ i g o f i2 i
Ltotal
w2 r2
⎥
⎢ ⎥
f ( x, u ) = ⎢ m& o (h f − h o ) + α A
i3 i
L3
(Tw3 − Tr3 ) ⎥ ( 2.172 )
⎢ Ltotal ⎥
⎢ m& i − m& o ⎥
⎢ ⎥
⎢ α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ⎥
⎢ α i 2 Ai (Tr 2 − Tw 2 ) + α o Ao (Ta − Tw 2 ) ⎥
⎢ ⎥
⎢⎣ α i 3 Ai (Tr 3 − Tw3 ) + α o Ao (Ta − Tw3 ) ⎥⎦
50
⎡ L1 ⎤ ⎡ L1 ⎤
⎢ L ⎥ ⎢ L2 ⎥
⎢ 2 ⎥ ⎢ ⎥
⎢ Pc ⎥ ⎢ Pc ⎥
⎢ ⎥ ⎢ ⎥
⎢ hout ⎥ ⎢ hout ⎥
⎢ Tw1 ⎥ ⎢ Tw1 ⎥
⎢ ⎥ ⎢ ⎥
⎢ Tw 2 ⎥ ⎢ Tw 2 ⎥
g ( x, u ) = ⎢ ⎥ =⎢ ⎥ ( 2.173 )
T Tw3
⎢ w3 ⎥ ⎢ ⎥
⎢Ta ,out ⎥ ⎢
μ
( )
1− μ Ta − 1− μ Ta ,in
1
( ) ⎥
⎢T ⎥ ⎢ T (Pc , hout ) ⎥
⎢ r ,out ⎥ ⎢ ⎥
⎢ Tr , sh ⎥ ⎢ T (Pc , hin ) − Tsat (Pc ) ⎥
⎢ ⎥ ⎢ Tsat (Pc ) − T (Pc , hout )
⎥
⎢ Tr ,sc ⎥ ⎢ ⎥
[ ]
⎢⎣ mc ⎥⎦ ⎢⎣ ρ1 Acs L1 + ρ f (1 − γ ) + ρ g (γ ) Acs L2 + ρ 3 Acs L3 ⎥⎦
x = [L1 L2 Tw1 Tw 2 Tw3 ]

T
Pc hout ( 2.174 )
u = [m& in m& a ]
T
y = [L1 mc ]
T
L2 Pc hout Tw1 Tw 2 Tw3 Ta ,out Tr ,out Tr , sh Tr , sc ( 2.176 )
for Ta and assuming three regions results in Equation 2.178.
⎡n L ⎤
⎣ i =1 LTotal ⎦
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞ ⎡L T LT LT ⎤
⎜⎜ ⎟⎟⎜⎜ ⎟⎟Ta ,in + ⎢ 1 w,1 + 2 w, 2 + 3 w,3 ⎥
⎝ 1 − μ ⎠⎝ α o Ao ⎠ ⎣ LTotal LTotal LTotal ⎦
Ta = ( 2.178 )
⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
α o ⎛ Re l ⎞
m m
⎛ m& ⎞
=⎜ ⎟ = ⎜ air ⎟ ( 2.179 )
α o′ ⎝ Re′l ⎠ ′ ⎠
⎝ m& air
∂α o ⎛ m& ⎞ ⎛ α ′ ⎞
m
= m⎜ air ⎟ ⎜ o ⎟ ( 2.180 )
∂m& air ′ ⎠ ⎝ m& air ⎠
⎝ m& air
51
⎛ Tw1 − Tw3 ⎞
⎜⎜ ⎟
= ( 2.181 )
∂L1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ Tw 2 − Tw3 ⎞
⎜⎜ ⎟
= ( 2.182 )
∂L2 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L1 ⎞
⎜⎜ ⎟⎟
∂Ta ⎝ Total ⎠
L
= ( 2.183 )
∂Tw,1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L2 ⎞
⎜⎜ ⎟⎟
∂Ta ⎝ Total ⎠
L
= ( 2.184 )
∂Tw, 2 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L3 ⎞
⎜⎜ ⎟
= ( 2.185 )
∂Tw,3 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟
∂Ta ⎝ 1 − μ ⎠⎝ α o Ao ⎠
= ( 2.186 )
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α A
o o ⎠
⎛1+ μ ⎞ ⎛ 1 ⎞⎛ n Li Tw,i ⎞
⎜⎜ ⎟⎟Ta ,in − ⎜⎜ ⎟⎟⎜⎜ ∑ ⎟⎟
∂Ta ⎝ 1 − μ ⎠ ⎝ 1 − μ ⎠⎝ i =1 LTotal ⎠ ⎛⎜ C p ,air ⎞⎛ m& air ∂α o ⎞
= ⎜α A ⎟⎟⎜⎜1 − ⎟⎟ ( 2.187 )
∂m& air ⎛ ⎛ 1 ⎞⎛ m& air C p ,air ⎞ ⎞
2
⎝ o o ⎠⎝ α o ∂m& air ⎠
⎜ ⎜⎜ ⎜ ⎟⎟ + 1⎟
⎜ ⎝ 1 − μ ⎟⎟⎠⎜ α A ⎟
⎝ ⎝ o o ⎠ ⎠
52
∂Tr1 1 ⎛⎜ dTsat ∂Tr ,in ⎞ ∂T ⎛ ⎞
The following partial derivatives are also used: = + ⎟ , r1 = 1 ⎜ ∂Tr ,in ⎟,
∂Pc 2 ⎜ dPc ∂Pc ⎟ ∂hin 2 ⎜ ∂hin ⎟
⎝ h1 ⎠ ⎝ Pc ⎠
∂ρ1 ⎛⎜ ∂ρ1 ⎞ 1 ⎛ ∂ρ ⎞⎛ dhg ⎞ ∂ρ1 1 ⎛⎜ ∂ρ1 ⎞ ∂T ⎛ ⎞

= ⎟+ ⎜ 1 ⎟⎜ ⎟⎟ , = ⎟ , r 3 = 1 ⎜ dTsat + ∂Tr ,out ⎟,
∂Pc ⎜ ∂Pc ⎟ 2 ⎜ ∂h1 ⎟⎜⎝ dPc ⎠ ∂hin 2 ⎜⎝ ∂h1 ⎟ ∂Pc 2 ⎜ dPc ∂Pc ⎟
⎝ h1 ⎠ ⎝ Pc ⎠ Pc ⎠ ⎝ h3 ⎠
∂Tr 3 1 ⎛⎜ ∂Tr ,out ⎞ ∂ρ

⎟,
⎛ ∂ρ ⎞ 1 ⎛ ∂ρ ⎞⎛ dh f ⎞ ∂ρ 3 1 ⎛ ∂ρ ⎞
= 3
=⎜ 3 ⎟+ ⎜ 3 ⎟⎜ ⎟⎟ , and = ⎜ 3 ⎟.
∂hout 2 ⎜ ∂hout ⎟ ∂Pc ⎜ ∂Pc ⎟ 2 ⎜ ∂h3 ⎟⎜⎝ dPc ⎠ ∂hout 2 ⎜ ∂h3 ⎟
⎝ Pc ⎠ ⎝ h3 ⎠ ⎝ Pc ⎠ ⎝ Pc ⎠
⎡ f x ,11 0 f x ,13 0 f x ,15 0 0 ⎤

⎢ 0 f x , 22 f x , 23 0 0 f x , 26 0 ⎥⎥
⎢
⎢ f x ,31 f x ,32 f x ,33 f x ,34 0 0 f x ,37 ⎥
∂f ⎢ ⎥
= Fx = ⎢ 0 0 0 0 0 0 0 ⎥ ( 2.188 )
∂x ⎢ f x ,51 f x ,52 f x ,53 0 f x ,55 f x ,56 f x ,57 ⎥
⎢ ⎥
⎢ f x ,61 f x ,62 f x ,63 0 f x ,65 f x , 66 f x ,67 ⎥
⎢f f x ,72 f x ,73 f x ,74 f x ,75 f x , 76 f x ,77 ⎥⎦
⎣ x ,71
⎡ f u ,11 0 f u ,13 0 0 ⎤
⎢f f u , 22 0 0 0 ⎥⎥
⎢ u , 21
⎢ 0 f u ,32 0 0 0 ⎥
∂f ⎢ ⎥
= Fu = ⎢ f u , 41 f u , 42 0 0 0 ⎥ ( 2.189 )
∂u
⎢ 0 0 f u ,53 f u ,54 f u ,55 ⎥
⎢ ⎥
⎢ 0 0 0 f u ,64 f u , 65 ⎥
⎢ 0 0 0 f u ,74 f u , 75 ⎥⎦
⎣
53
⎡ g x ,11 0 0 0 0 0 0 ⎤
⎢ 0 g x , 22 0 0 0 0 0 ⎥⎥
⎢
⎢ 0 0 g x ,33 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 g x , 44 0 0 0 ⎥
⎢ 0 0 0 0 g x ,55 0 0 ⎥
⎢ ⎥
∂g ⎢ 0 0 0 0 0 g x ,66 0 ⎥
= Gx = ⎢ ( 2.190 )
∂x 0 0 0 0 0 0 g x , 77 ⎥
⎢ ⎥
⎢ g x ,81 g x ,82 0 0 g x ,85 g x ,86 g x ,87 ⎥
⎢ 0 0 g x ,93 g x ,94 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 g x ,10,3 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 g x ,11,3 g x ,11, 4 0 0 0 ⎥
⎢⎣ g x ,12,1 g x ,12, 2 g x ,12,3 g x ,12, 4 0 0 0 ⎥⎦
⎡0 0 0 0 0 ⎤
⎢0 0 0 0 0 ⎥⎥
⎢
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢0 0 0 0 0 ⎥
⎢ ⎥
∂g ⎢0 0 0 0 0 ⎥
= Gu = ⎢ ( 2.191 )
∂u 0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 g u ,84 g u ,85 ⎥
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 g u ,10,3 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢⎣0 0 g u ,12,3 0 0 ⎥⎦
⎛ 1 ⎞
f x ,11 α i1 Ai ⎜⎜ ⎟⎟(Tw1 − Tr1 )
⎝ LTotal ⎠
⎛ dhg ⎞ ⎛ L ⎞⎛ ∂Tr1 ⎞
f x ,13 − m& in ⎜⎜ ⎟⎟ + α i1 Ai ⎜⎜ 1 ⎟⎟⎜⎜ − ⎟⎟
⎝ dPc ⎠ ⎝ LTotal ⎠⎝ ∂Pc ⎠
⎛ L1 ⎞
f x ,15 α i1 Ai ⎜⎜ ⎟⎟
⎝ LTotal ⎠
⎛ 1 ⎞
f x , 22 α i 2 Ai ⎜⎜ ⎟⎟(Tw 2 − Tr 2 )
⎝ LTotal ⎠
54
dhg dh f ⎛ L ⎞⎛ dTsat ⎞
f x , 23 m& in − m& out + α i 2 Ai ⎜⎜ 2 ⎟⎟⎜⎜ − ⎟⎟
dPc dPc ⎝ LTotal ⎠⎝ dPc ⎠
⎛ L2 ⎞
f x , 26 α i 2 Ai ⎜⎜ ⎟⎟
⎝ LTotal ⎠
⎛ 1 ⎞
f x ,31 − α i 3 Ai ⎜⎜ ⎟⎟(Tw3 − Tr 3 )
⎝ LTotal ⎠
⎛ 1 ⎞
f x ,32 − α i 3 Ai ⎜⎜ ⎟⎟(Tw3 − Tr 3 )
⎝ LTotal ⎠
dh f ⎛ L ⎞⎛ ∂Tr 3 ⎞
f x ,33 m& out + α i 3 Ai ⎜⎜ 3 ⎟⎟⎜⎜ − ⎟⎟
dPc ⎝ LTotal ⎠⎝ ∂Pc ⎠
⎛ L ⎞⎛ ∂Tr 3 ⎞
f x ,34 − m& out + α i 3 Ai ⎜⎜ 3 ⎟⎟⎜⎜ − ⎟⎟
⎝ LTotal ⎠⎝ ∂ho ⎠
⎛ L3 ⎞
f x ,37 α i 3 Ai ⎜⎜ ⎟⎟
⎝ LTotal ⎠
∂Ta
f x ,51 α o Ao
∂L1
∂Ta
f x ,52 α o Ao
∂L2
⎛ ∂Tr1 ⎞
f x ,53 α i1 Ai ⎜⎜ ⎟⎟
⎝ ∂Pc ⎠
∂Ta
f x ,55 − α i1 Ai − α o Ao + α o Ao
∂Tw1
∂Ta
f x ,56 α o Ao
∂Tw 2
∂Ta
f x ,57 α o Ao
∂Tw3
∂Ta
f x ,61 α o Ao
∂L1
∂Ta
f x ,62 α o Ao
∂L2
⎛ dTsat ⎞
f x ,63 α i 2 Ai ⎜⎜ ⎟⎟
⎝ dPc ⎠
55
∂Ta
f x ,65 α o Ao
∂Tw1
∂Ta
f x ,66 − α i 2 Ai − α o Ao + α o Ao
∂Tw 2
∂Ta
f x ,67 α o Ao
∂Tw3
∂Ta
f x ,71 α o Ao
∂L1
∂Ta
f x ,72 α o Ao
∂L2
⎛ ∂Tr 3 ⎞
f x ,73 α i 3 Ai ⎜⎜ ⎟⎟
⎝ ∂Pc ⎠
⎛ ∂Tr 3 ⎞
f x ,74 α i 3 Ai ⎜⎜ ⎟⎟
⎝ ∂hout ⎠
∂Ta
f x ,75 α o Ao
∂Tw1
∂Ta
f x ,76 α o Ao
∂Tw 2
∂Ta
f x ,77 − α i 2 Ai − α o Ao + α o Ao
∂Tw3
f u ,11 hi − hg
L1 ⎛ ∂Tr1 ⎞
f u ,13 m& i + α i1 Ai ⎜⎜ − ⎟⎟
LTotal ⎝ ∂hin ⎠
f u , 21 hg
f u , 22 − hf
f u ,32 h f − ho
f u , 41 1
f u , 42 −1
⎛ ∂Tr1 ⎞
f u ,53 α i1 Ai ⎜⎜ ⎟⎟
⎝ ∂hin ⎠
∂Ta
f u ,54 Aoα o
∂Tai
56
∂α o ∂T
f u ,55 Ao (Ta − Tw1 ) + Aoα o a
∂ma ∂ma
∂Ta
f u , 64 Aoα o
∂Tai
∂α o ∂T
f u , 65 Ao (Ta − Tw 2 ) + Aoα o a
∂ma ∂ma
∂Ta
f u , 74 Aoα o
∂Tai
∂α o ∂T
f u , 75 Ao (Ta − Tw3 ) + Aoα o a
∂ma ∂ma
g x ,11 1
g x , 22 1
g x ,33 1
g x , 44 1
g x ,55 1
g x ,66 1
g x ,77 1
⎛ 1 ⎞ ∂Ta
g x ,81 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂L1
⎛ 1 ⎞ ∂Ta
g x ,82 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂L2
⎛ 1 ⎞ ∂Ta
g x ,85 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂Tw1
⎛ 1 ⎞ ∂Ta
g x ,86 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂Tw 2
⎛ 1 ⎞ ∂Ta
g x ,87 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂Tw3
⎛ ∂Tr ,out ⎞
g x ,93 ⎜ ⎟
⎜ ∂Pc ⎟
⎝ hout ⎠
⎛ ∂Tr ,out ⎞
g x ,94 ⎜ ⎟
⎜ ∂hout ⎟
⎝ Pc ⎠
57
⎛ ∂Tr ,in ⎞ dT
g x ,10,3 ⎜ ⎟ − sat
⎜ ∂Pc ⎟ dPc
⎝ hin ⎠
dTsat ⎛⎜ ∂Tr ,out ⎞
⎟
g x ,11,3 −
dPc ⎜ ∂Pc ⎟
⎝ hout ⎠
⎛ ∂Tr ,out ⎞
g x ,11, 4 −⎜ ⎟
⎜ ∂hout ⎟
⎝ Pc ⎠
g x ,12,1 (ρ1 − ρ 3 )Acs
g x ,12, 2 (ρ f (1 − γ ) + ρ g γ − ρ 3 )Acs
⎡ ⎛ ∂ρ1 ⎞ ⎡ dρ dρ ⎤ ⎛ ∂ρ ⎞⎤
g x ,12,3 ⎢ L1 ⎜⎜ ⎟⎟ + L2 ⎢ f (1 − γ ) + g (γ )⎥ + L3 ⎜⎜ 3 ⎟⎟⎥ Acs
⎢⎣ ⎝ ∂Pc ⎠ ⎣ dPc dPc ⎦ ⎝ ∂Pc ⎠⎥⎦
⎛ ∂ρ 3 ⎞
g x ,12, 4 ⎜⎜ ⎟⎟ Acs L3
⎝ ∂hout ⎠
⎛ 1 ⎞ ∂Ta ⎛ μ ⎞
g u ,84 ⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂Tai ⎝ 1 − μ ⎠
⎛ 1 ⎞ ∂Ta
g u ,85 ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠ ∂mai
⎛ ∂Tr ,in ⎞
g u ,10,3 ⎜ ⎟
⎜ ∂hin ⎟
⎝ Pc ⎠
⎛ ∂ρ1 ⎞
g u ,12,3 ⎜⎜ ⎟⎟ Acs L1
⎝ ∂hin ⎠
2.6.9 Evaporator with Receiver
The evaporator is modeled with a single two-phase region. The outlet from the evaporator is assumed to be
near the saturated vapor condition. The mean void fraction is used to take into account deviations from the saturated
vapor condition. The outlet fluid from the receiver is saturated vapor. The governing ordinary differential equations
(ODEs) are obtained by integrating the governing partial differential equations (PDEs) (Eqs. 2.144-2.116) along the
length of the assumed heat exchanger and assuming lumped parameters associated with each fluid region.
58
Twall(t)
& inhin
m & inthint (t )
m
P(t)
xin > 0 xout(t) < 1
Two-Phase
LTotal
m& int hint (t ) m& out hg

Pevap
mg hg
mf hf
Figure 2.14 Diagram: Evaporator with one fluid region and receiver

air temperature across the evaporator is assumed Ta = Ta ,in (μ ) + Ta ,out (1 − μ ) . In the two-phase region, the fluid
properties are determined by assuming a mean void fraction. Thus ρ 1 = ρ f (1 − γ ) + ρ g (γ ) , and so forth. For
this derivation, the time derivative of the mean void fraction is not neglected, because this variable captures the
dynamic information regarding the evaporator outlet quality. For this application we will assume a “Slip-Ratio”
x ⎛ ρg ⎞
void fraction correlation, γ = , where μ S = ⎜ ⎟ S . Mean void fraction is calculated using by
x + (1 − x )μ S ⎜ρ
⎝ f
⎟
⎠
1 μS ⎡ μ S + x in (1 − μ S ) ⎤
integrating the expression: γ = + ln ⎢
(1 − μ S ) x Δ (1 − μ S ) ⎣ μ S + x in (1 − μ S ) + x Δ (1 − μ S )⎥⎦
2
, where
x Δ = xint − xin . If the fluid exiting the evaporator is saturated vapor, then xint = 1 , and x Δ = 1 − xin . For small
deviations around this condition, the mean void fraction expression can be approximated using the expression given
in Equation 2.192. Thus, solving for the evaporator outlet quality yields Equation 2.193.
⎡ μ S + xin (1 − μ S ) ⎤
ln ⎢ ⎥ ≈ ln[μ S + xin (1 − μ S )] ( 2.192 )
⎣ μ S + xin (1 − μ S ) + ( xout − xin )(1 − μ S ) ⎦
μ ln[μ S + xin (1 − μ S )]
xout = S +x ( 2.193 )
(1 − μ S )2 γ − (1 − μ S ) in
2.6.9.1.1 Conservation of Refrigerant Mass (Two-phase Region)
⎛ dρ f dρ ⎞
⎜⎜ (1 − γ ) + g (γ )⎟⎟ Acs LTotal P&e + (ρ g − ρ f )Acs LTotal γ& = m& in − m& int ( 2.194 )
⎝ dPe dPe ⎠
59
2.6.9.1.2 Conservation of Refrigerant Energy (Two-phase Region)
⎛ d (ρ f h f ) d (ρ h ) ⎞
⎜⎜ (1 − γ ) + g g (γ ) − 1⎟⎟ Acs LTotal P&e + (ρ g hg − ρ f h f )Acs LTotal γ&
⎝ dPe dPe ⎠ ( 2.195 )
= m& in hin − m& int hint + α i Ai (Tw − Tr )
2.6.9.1.3 Conservation of Wall Energy (Two-phase Region)
(C p ρV )w T&w = α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ( 2.196 )
2.6.9.1.4 Conservation of Mass (Receiver)
m& rec = m& int − m& out ( 2.197 )
2.6.9.1.5 Conservation of Energy (Receiver)
⎡⎛ dρ g dρ f du g du f ⎞ ⎛ ρ gug − ρ f u f ⎞⎛ dρ g dρ f ⎞⎤
⎢⎜⎜ Vg u g + Vf u f + ρ gVg + ρ fVf ⎟⎟ − ⎜ ⎟⎜
⎜ Vg + V f ⎟⎟⎥ P&e
⎜ ⎟ dP
⎣⎢⎝ dPe dPe dPe dPe ⎠ ⎝ ρg − ρ f ⎠⎝ e dPe ⎠⎥⎦ ( 2.198 )
⎡ ρ gug − ρ f u f ⎤
+⎢ ⎥ m& rec = m& int hint − m& out hout − UArec (Trec − Tamb )
⎢⎣ ρ g − ρ f ⎥⎦
The resulting five differential equations for conservation of refrigerant mass, refrigerant energy, and wall
energy for the two-phase and superheat regions only contain four explicit time derivatives: P&e , m
& rec , γ& , and T&w .
& int . This results in Equation 2.199, which is of the
The equations can be combined to eliminate the variable m
Z ( x, u ) ⋅ x& = f ( x, u ) form, with states x = [Pe γ Tw ] , and where the elements of the Z ( x, u )
T
mrec
matrix are given in Table 2.15.
⎡ z11 z12 z13 0 ⎤ ⎡ P&e ⎤ ⎡ m& in − m& out ⎤

⎢ ⎥ ⎢ &
⎢z
⎢ 21 z 22 z 23 0 ⎥ ⎢m& rec ⎥ ⎢ min hin − m& out hg + α i Ai (Tw − Tr ) ⎥⎥
⎥
=
0 ⎥ ⎢ γ& ⎥ ⎢m& out (hg − hout ) − UArec (Trec − Tamb )⎥
( 2.199 )
⎢ z 31 z 32 0
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣0 0 0 z 44 ⎦ ⎢⎣ T&w ⎥⎦ ⎣ α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ⎦
Table 2.15 Matrix Elements of Z ( x, u ) for the Evaporator
⎡⎛ dρ f ⎞ ⎛ dρ ⎞ ⎤
z11 ⎢⎜⎜ ⎟⎟(1 − γ ) + ⎜⎜ g ⎟⎟(γ )⎥ Acs LTotal
⎢⎣⎝ dPe ⎠ ⎝ dPe ⎠ ⎥⎦
z12 1
z13 (ρ g − ρ f )Acs LTotal
60
⎡⎛ d (ρ f h f ) ⎞⎟ ⎛ d (ρ g hg ) ⎞ ⎤
⎟(1 − γ ) + ⎜
z 21 ⎢⎜⎜ ⎜ ⎟(γ ) − 1⎥ Acs LTotal
⎢⎣⎝ dPe ⎠ ⎟ ⎥⎦
⎝ dPe g ⎠
z 22 hint
z 23 (ρ g hg − ρ f h f )Acs LTotal
⎛ dρ g dρ f du g du f ⎞ ⎛ ρ gug − ρ f u f ⎞⎛ dρ g dρ f ⎞
z 31 ⎜⎜ Vg u g + Vf u f + ρ gVg + ρ fVf ⎟⎟ − ⎜ ⎟⎜ V + V ⎟
⎜ ⎟⎜ dP g dP f ⎟
⎝ dPe dPe dPe dPe ⎠ ⎝ ρg − ρ f ⎠⎝ e e ⎠
⎛ ρgug − ρ f u f ⎞
z 32 ⎜ − hint ⎟
⎜ ρ −ρ ⎟
⎝ g f ⎠
z 44 (C p ρV )w

Recall that the evaporator could be modeled as Z ( x, u )x = f ( x, u ) with states given as Equation 2.202
⎡ m& in − m& out ⎤

⎢ m& h − m& h + α A (T − T ) ⎥
f ( x, u ) = ⎢ in in out g i i w r ⎥
⎢m& out (hg − hout ) − UArec (Trec − Tamb )⎥
( 2.200 )
⎢ ⎥
⎣ α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ⎦
⎡ Pe ⎤
⎡ Pe ⎤ ⎢ ⎥
⎢ x ⎥ ⎢ ⎛⎜ μ S ln[μ S + xin (1 − μ S )] ⎞⎟ + x ⎥
⎢ int ⎥ ⎢ ⎜ (1 − μ )2 γ − (1 − μ ) ⎟ in ⎥
⎢ hout ⎥ ⎢ ⎝ S S ⎠
⎥
⎢ ⎥ h
g ( x, u ) = ⎢ Tw ⎥ = ⎢ ⎥
out
( 2.201 )
⎢ Tw ⎥
⎢Ta ,out ⎥ ⎢
⎢ ⎥ ⎢
μ
( )
1− μ Ta − 1− μ Ta ,in
1 ⎥
⎥ ( )
⎢m ⎥ ⎢ f
[
⎢ me ⎥ ⎢ ρ (1 − γ ) + ρ (γ ) A L + ρ A L ⎥
g cs 1 2 cs 2 ]
⎣ rec ⎦ ⎥
⎣ mrec ⎦
x = [Pe γ Tw ]
T
mrec ( 2.202 )
u = [m& in m& a ]
T
y = [Pe mrec ]
T
xint hout Tw Ta ,out me ( 2.204 )
for Ta and assuming a single region results in Equation 2.206.
61
⎡n L ⎤
⎣ i =1 LTotal ⎦
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟Ta ,in + Tw
⎝ 1 − μ ⎠⎝ α o Ao ⎠
Ta = ( 2.206 )
⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
α o ⎛ Re l ⎞
m m
⎛ m& ⎞
=⎜ ⎟ = ⎜ air ⎟ ( 2.207 )
α o′ ⎝ Re′l ⎠ ⎝ m& ′air ⎠
∂α o ⎛ m& ⎞ ⎛ α ′ ⎞
m
= m⎜ air ⎟ ⎜ o ⎟ ( 2.208 )
∂Ta 1
= ( 2.209 )
∂Tw ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α A
o o ⎠
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟
∂Ta ⎝ 1 − μ ⎠⎝ α o Ao ⎠
= ( 2.210 )
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛1+ μ ⎞ ⎛ 1 ⎞
⎜⎜ ⎟⎟Ta ,in − ⎜⎜ ⎟Tw
∂Ta ⎝ 1− μ ⎠ ⎝ 1 − μ ⎟⎠ ⎛ C p ,air ⎞⎛ m& air ∂α o ⎞
= ⎜⎜ ⎟⎟⎜⎜1 − ⎟⎟ ( 2.211 )
∂m& air ⎝ α o Ao α o ∂m& air
2
⎛ ⎛ 1 ⎞⎛ m& air C p ,air ⎞ ⎞ ⎠⎝ ⎠
⎜ ⎜⎜ ⎜ ⎟ ⎟
⎜ ⎝ 1 − μ ⎟⎟⎠⎜ α A ⎟ + 1⎟
⎝ ⎝ o o ⎠ ⎠
⎞ ⎛ dhg dh f ⎞
The following partial derivatives are also used:
∂hint
= (hg − h f )⎛⎜⎜ ∂x⎟⎟ + ⎜⎜
int
−
dh
⎟⎟ xint + f ,
∂Pe ⎝ ∂Pe ⎠ ⎝ dPe dPe ⎠ dPe
∂hint
= (hg − h f )⎛⎜⎜ ∂x
⎞ ∂hint
⎟⎟ ,int
= (hg − h f )⎛⎜⎜ ∂x⎞ ∂xint
⎟⎟ ,
int
= 1−
abc
,
∂γ ⎝ ∂γ ⎠ ∂hin ⎝ ∂hin ⎠ ∂xin (γ − b )2
∂xint ⎛ b 2 c ⎞ ⎛ ac ⎞ ⎛ ab ⎞⎛ 1 − xin ⎞ ∂xint ab ∂xin ⎛⎜ − 1 ⎞⎟⎛ dh f dh fg ⎞
=⎜ ⎟ + ⎜ ⎟γ + ⎜ ⎟⎜ ⎟⎟ , = , = ⎜⎜ + xin ⎟⎟ ,
∂μ S ⎜⎝ d ⎟⎠ ⎝ d 2 ⎠ ⎝ d ⎠⎜⎝ a − xin ⎠ ∂γ (γ − b )(a − x in ) ∂P e
⎜ h ⎟ dP
⎝ fg ⎠⎝ e dP e ⎠
62
∂xin ⎛⎜ 1 ⎞⎟ ∂μ S ⎛ dρ g ⎞⎛⎜ μ S ⎞ ⎛ dρ f
⎟−⎜
⎞⎛ μ S ⎞
⎟ , where a = μ S , b = 1 ,
= , and =⎜ ⎟ ⎟⎟⎜
∂hin ⎜⎝ h fg ⎟⎠ ∂Pe ⎜⎝ dPe ⎟⎠⎜⎝ ρ g ⎟ ⎜ dP
⎠ ⎝ e
⎜
⎠⎝ ρ f
⎟
⎠ 1 − μS 1 − μS
c = ln[μ S + xin (1 − μ S )], and d = (1 − μ S )γ − 1 .
⎡ 0 0 0 0 ⎤
⎢f 0 f x , 23 f x , 24 ⎥⎥
∂f
= Fx = ⎢
x , 21
( 2.212 )
∂x ⎢ f x ,31 0 f x ,33 0 ⎥
⎢ ⎥
⎣ f x , 41 0 0 f x , 44 ⎦
⎡ f u ,11 f u ,12 0 0 0 ⎤
⎢f 0 0 ⎥⎥
∂f f u , 22 f u , 23
= Fu = ⎢
u , 21
( 2.213 )
∂u ⎢ 0 f u ,32 f u ,33 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 f u , 44 f u , 45 ⎦
⎡ g x ,11 0 0 0 ⎤
⎢g 0 g x , 23 0 ⎥⎥
⎢ x , 21
⎢ g x ,31 0 0 0 ⎥
∂g ⎢ ⎥
= Gx = ⎢ 0 0 0 g x , 44 ⎥ ( 2.214 )
∂x
⎢ 0 0 0 g x ,54 ⎥
⎢ ⎥
⎢ g x ,61 0 g x ,63 0 ⎥
⎢ 0 g x , 72 0 0 ⎥⎦
⎣
⎡0 0 0 0 0 ⎤
⎢0 0 g u , 23 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
∂g ⎢ ⎥
= G u = ⎢0 0 0 0 0 ⎥ ( 2.215 )
∂u
⎢0 0 0 g u ,54 g u ,55 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢0 0 0 0 0 ⎥⎦
⎣
Table 2.16 Matrix Elements of Equations 2.212-2.215
∂hint ⎛ dT ⎞
f x , 21 − m& out − α i Ai ⎜⎜ sat ⎟⎟
∂Pe ⎝ dPe ⎠
∂hint
f x , 23 − m& out
∂γ
63
f x , 24 α i Ai
⎛ ∂h dhg ⎞ ⎛ dT ⎞
f x ,31 m& out ⎜⎜ int − ⎟⎟ − UArec ⎜⎜ sat ⎟⎟
⎝ ∂Pe dPe ⎠ ⎝ dPe ⎠
∂h
f x ,33 m& out int
∂γ
⎛ dTsat ⎞
f x , 41 α i Ai ⎜⎜ ⎟⎟
⎝ dPe ⎠
⎛ ∂T ⎞
f x , 44 − α i Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw ⎠
f u ,11 1
f u ,12 −1
f u , 21 hin
f u , 22 − hint
⎛ ∂h ⎞
f u , 23 m& in − m& out ⎜⎜ int ⎟⎟
⎝ ∂hin ⎠
f u ,32 hint − hg
⎛ ∂h ⎞
f u ,33 m& out ⎜⎜ int ⎟⎟
⎝ ∂hin ⎠
⎛ ∂Ta ⎞
f u , 44 α o Ao ⎜⎜ ⎟⎟
⎝ ∂Tai ⎠
⎛ ∂α o ⎞ ⎛ ∂Ta ⎞
f u , 45 ⎜⎜ ⎟⎟ Ao (Ta − Tw ) + α o Ao ⎜⎜ ⎟⎟
⎝ ∂m& air ⎠ ⎝ ∂m& air ⎠
g x ,11 1
⎛ ∂xint ⎞⎛ ∂μ S ⎞ ⎛ ∂xint ⎞⎛ ∂xin ⎞
g x , 21 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ ∂μ S ⎠⎝ ∂Pe ⎠ ⎝ ∂xin ⎠⎝ ∂Pe ⎠
⎛ ∂xint ⎞
g x , 23 ⎜⎜ ⎟⎟
⎝ ∂γ ⎠
⎛ dhg ⎞
g x ,31 ⎜⎜ ⎟⎟
⎝ dPe ⎠
g x , 44 1
64
⎛ 1 ⎞⎛ ∂Ta ⎞
g x ,54 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tw ⎠
⎡⎛ dρ f ⎞ ⎛ dρ ⎞ ⎤
g x , 61 ⎢⎜⎜ ⎟⎟(1 − γ ) + ⎜⎜ g ⎟⎟(γ )⎥ Acs LTotal
⎣⎢⎝ dPe ⎠ ⎝ dPe ⎠ ⎦⎥
g x , 63 (ρ g − ρ f )Acs LTotal
g x ,72 1
⎛ ∂xint ⎞⎛ ∂xin ⎞
g u , 23 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ ∂xin ⎠⎝ ∂hin ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞ ⎛ μ ⎞
g u ,54 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tai ⎠ ⎝1− μ ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g u ,55 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂ma ⎠
2.6.10 Condenser with Receiver

The condenser portion of the condenser with receiver model consists of two fluid regions: superheat and
two-phase. The outlet from the condenser is assumed to be near the saturated liquid condition. The mean void
fraction is used to take into account deviations from the saturated liquid condition. The outlet fluid from the
receiver is saturated liquid. The governing ordinary differential equations (ODEs) are obtained by integrating the
governing partial differential equations (PDEs) (Eqs. 2.114-2.116) along the length of the assumed heat exchanger
and assuming lumped parameters associated with each fluid region.
x=1
& int 2 hint 2 (t )

m
& inhin
m P(t) xint2(t) > 0
Superheat Two-Phase
L1(t) L2(t)
LTotal
m& int 2 hint 2 (t )

Pcond
mg hg
m f hf m& out h f
Figure 2.15 Diagram: Condenser with two fluid regions and receiver
65
The condenser with receiver model uses many of the same assumptions presented in the evaporator model
derivation. For heat transfer a weighted average of the air temperature across the evaporator is assumed
Ta = Ta ,in (μ ) + Ta ,out (1 − μ ) . In the two-phase region, the fluid properties are determined by assuming a mean
void fraction. Thus ρ 1 = ρ f (1 − γ ) + ρ g (γ ) , and so forth. Average properties are assumed in the superheat
region. For this derivation, the time derivative of the mean void fraction is not neglected, because this variable
captures the dynamic information regarding the evaporator outlet quality. For this application we will assume a
x ⎛ ρg ⎞
“Slip-Ratio” void fraction correlation, γ = , where μ S = ⎜ ⎟ S . Mean void fraction is
x + (1 − x )μ S ⎜ρ ⎟
⎝ f ⎠
calculated using by integrating the
1 μS ⎡ μ S + x in (1 − μ S ) ⎤
expression: γ = + ln ⎢ , where x Δ = xint 2 − xint 1 . For
(1 − μ S ) x Δ (1 − μ S ) ⎣ μ S + x in (1 − μ S ) + x Δ (1 − μ S )⎥⎦
2
the condenser, the fluid entering the two-phase region is assumed to be saturated vapor, xint 1 = 1 . If the fluid
exiting the condenser is saturated liquid, then xint 2 = 0 , and x Δ = −1 . For small deviations around this condition,
the mean void fraction expression can be approximated as Equation 2.216. Thus, solving for the evaporator outlet
μS 1
quality yields Equation 2.217, where a = and b = .
1 − μS 1 − μS
1 μS ⎡ 1 ⎤
γ = − ln ⎢
(1 − μ S ) (1 − μ S ) ⎣ μ S + xint 2 (1 − μ S )⎥⎦
2
( 2.216 )
⎛ γ −b ⎞
⎜ ⎟
xint 2 = be ⎝ ab ⎠
−a ( 2.217 )
2.6.10.1.1 Conservation of Refrigerant Mass (Superheat and Two-phase Regions)
⎛ ∂ρ1 1 ∂ρ1 dhg ⎞ ⎛ 1 ∂ρ1 ⎞

⎜⎜ + ⎟⎟ Acs L1 P& + ⎜⎜ ⎟⎟ Acs L1 h&i + (ρ1 − ρ g )Acs L&1 = m& i − m& int 1 ( 2.218 )
∂
⎝ cP 2 ∂h1 dPc ⎠ ⎝ 2 ∂h1 ⎠
⎛ dρ f dρ ⎞
(ρ − ρ f )(1 − γ )Acs L&1 + ⎜⎜
g (1 − γ ) + g (γ )⎟⎟ Acs L2 P&c ( 2.219 )
⎝ dPc dPc ⎠
+ (ρ g − ρ f )Acs L2 γ = m& int 1 − m& int 2
&
66
2.6.10.1.2 Conservation of Refrigerant Energy (Superheat and Two-phase Regions)
⎡⎛ ∂ρ1 1 ∂ρ1 dhg ⎞ ⎛ 1 dhg ⎞ ⎤ ⎡⎛ 1 ∂ρ1 ⎞ ⎛1⎞ ⎤

⎢⎜⎜ + ⎟⎟h1 + ⎜⎜ ⎟⎟ ρ1 − 1⎥ Acs L1 P&c + ⎢⎜⎜ ⎟⎟h1 + ⎜ ⎟ ρ1 ⎥ Acs L1 h&i
⎣⎢⎝ ∂Pc 2 ∂h1 dPc ⎠ ⎝ 2 dPc ⎠ ⎦⎥ ⎣⎝ 2 ∂h1 ⎠ ⎝2⎠ ⎦ ( 2.220 )
⎛ L ⎞
+ (ρ1 h1 − ρ g hg )Acs L&1 = m& i hi − m& int 1 hg + α i1 Ai ⎜⎜ 1 ⎟⎟(Tw1 − Tr1 )
⎝ LTotal ⎠
⎛ d (ρ f h f ) d (ρ h ) ⎞
(ρ g hg − ρ f h f )(1 − γ )Acs L&1 + ⎜⎜ (1 − γ ) + g g (γ ) − 1⎟⎟ Acs L2 P&c
⎝ dP c dPc ⎠ ( 2.221 )
⎛ L ⎞
+ (ρ g hg − ρ f h f )Acs L2 γ& = m& int 1 hg − m& int 2 hint 2 + α i 2 Ai ⎜⎜ 2 ⎟⎟(Tw 2 − Tr 2 )
⎝ LTotal ⎠
2.6.10.1.3 Conservation of Wall Energy (Superheat and Two-phase Regions)
⎛ ⎞
(C P ρV )w ⎜⎜ T&w1 + Tw1 − Tw2 L&1 ⎟⎟ = α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ( 2.222 )
⎝ L1 ⎠
(C P ρV )w T& w2 = α i 2 Ai (Tr 2 − Tw 2 ) + α o Ao (Ta − Tw2 ) ( 2.223 )
2.6.10.1.4 Conservation of Mass (Receiver)
m& rec = m& int 2 − m& o ( 2.224 )
2.6.10.1.5 Conservation of Energy (Receiver)
⎡ dρ g dρ f du g du f ⎛ ρ g u g − ρ f u f ⎞⎛ dρ g dρ f ⎞⎤
⎢ Vg u g + V f u f + mg + mf −⎜ ⎟⎜ V + V ⎟ P&
⎢⎣ dPc dPc ⎜⎝ ρ g − ρ f ⎟⎜ dP g dP f ⎟⎥⎥ c ( 2.225 )
dPc dPc ⎠⎝ c c ⎠⎦
⎛ ρgug − ρ f u f ⎞
+⎜ ⎟m& rec = m& int 2 hint 2 − m& o h f − UArec (Trec − Tamb )
⎜ ρ −ρ ⎟
⎝ g f ⎠
& int 1 and m& int 2 . This

These eight differential equations are combined to eliminate the variables m
combinations results in a model with six states: L1 , Pc , γ , mrec , Tw1 , and Tw 2 . The resulting model is given in
Equation 2.226 and is of the form Z ( x, u ) ⋅ x& = f ( x, u ) , with the elements of the Z matrix given in Table 2.17.
⎡ ⎛ L1 ⎞ ⎤
⎡ z11 z12 0 0 0 0 ⎤ ⎡ L&1 ⎤ ⎢ m& i (h i − h g ) + α i1 A i
⎜
⎜L ⎟
⎟ ( Tw1 − T r 1 ) ⎥
⎢ ⎥ ⎢ ⎝ Total ⎠ ⎥
⎢z 0 ⎥⎥ ⎢ P& ⎥ ⎢
0 ⎛ L2 ⎞ ⎥
( )
⎢ 21 z 22 z 23 z 24
⎢ &
m h − h + α A ⎜⎜ ⎟⎟(Tw 2 − Tr 2 )⎥
⎢ z 31 z 32 z 33 z 34 0 0 ⎥ ⎢ γ& ⎥ ⎢ o g int 2 i 2 i
⎝ LTotal ⎠ ⎥
⎢ ⎥⎢ ⎥=
⎢0 z 42 0 z 44 0 0 ⎥ ⎢m& rec ⎥ ⎢ mi − m& o
& ⎥
0 ⎥ ⎢ T&w1 ⎥ ⎢ m& o (hint 2 − h f ) − UArec (Trec − Tamb ) ⎥
⎢ z 51 ⎢ ⎥
0 0 0 z 55
⎢ ⎥⎢ ⎥
⎢⎣ 0 0 0 0 0 z 66 ⎥⎦ ⎢⎣ T&w 2 ⎥⎦ ⎢ α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ⎥
⎢ ⎥
⎣ α i 2 Ai (Tr 2 − Tw 2 ) + α o Ao (Ta − Tw 2 ) ⎦
( 2.226 )
67
Table 2.17 Matrix Elements of Z ( x, u ) for the Condenser with Receiver
z11 (ρ h − ρ h )A
1 1 1 g cs
⎡⎛ ∂ρ1 1 ∂ρ1 dhg ⎞ ⎛ 1 dhg ⎞ ⎤

z12 ⎢⎜⎜ + ⎟⎟(h1 − hg ) + ⎜⎜ ⎟⎟ ρ1 − 1⎥ Acs L1
⎣⎝ ∂P 2 ∂h1 dP ⎠ ⎝ 2 dP ⎠ ⎦
z 21 (ρ h − ρ h )(1 − γ )A
f g f f cs
⎛ d (ρ f h f ) (1 − γ ) + d (ρ h ) γ − dρ h (1 − γ ) − dρ ⎞
⎜⎜ hg γ − 1⎟⎟ Acs L2
g g f g
z 22 g
⎝ dP dP dP dP ⎠
z 23 (ρ f hg − ρ f h f )Acs L2
z 24 (h int 2 − hg )
z 31 [ρ 1 − ρ f (1 − γ ) − ρ g γ Acs ]
⎡⎛ dρ f dρ ⎞ ⎛ dh ⎞ ⎤
z 32 ⎢⎜⎜ (1 − γ ) + g γ ⎟⎟ L2 + ⎜⎜ ∂ρ1 + 1 ∂ρ1 g ⎟⎟ L1 ⎥ Acs
⎣⎝ dP dP ⎠ ⎝ ∂P 2 ∂h1 dP ⎠ ⎦
z 33 (ρ g − ρ f )Acs L2
z 34 1
⎡ dρ g dρ du du f ⎛ ρ g u g − ρ f u f ⎞⎛ dρ g dρ f ⎞⎤
⎢ Vg u g + f V f u f + m g g + m f −⎜ ⎟⎜ + ⎟
⎟⎜⎝ dP g dP f ⎟⎠⎥⎥
z 42 V V
⎢⎣ dP dP dP dP ⎜⎝ ρ g − ρ f ⎠ ⎦
⎛ ρ gug − ρ f u f ⎞
z 44 ⎜ − hint 2 ⎟
⎜ ρ −ρ ⎟
⎝ g f ⎠
⎛ Tw1 − Tw 2 ⎞
z 51 (c p ρV )w ⎜⎜ ⎟⎟
⎝ L1 ⎠
z 55 (c ρV )
p w
z 66 (c ρV )
p w

Recall that the condenser could be modeled as Z ( x, u )x = f ( x, u ) with states given as Equation 2.229
68
⎡ ⎛ L1 ⎞ ⎤
⎢ m& i (hi − hg ) + α i1 Ai ⎜⎜ ⎟⎟(Tw1 − Tr1 ) ⎥
⎢ ⎝ LTotal ⎠ ⎥
⎢ ⎛ L2 ⎞ ⎥
⎢m& o (hg − hint 2 ) + α i 2 Ai ⎜⎜ ⎟⎟(Tw 2 − Tr 2 )⎥
f ( x, u ) = ⎢ ⎝ LTotal ⎠ ⎥ ( 2.227 )
⎢ m& i − m& o ⎥
⎢ m& o (hint 2 − h f ) − UArec (Trec − Tamb ) ⎥
⎢ ⎥
⎢ α i1 Ai (Tr1 − Tw1 ) + α o Ao (Ta − Tw1 ) ⎥
⎢ ⎥
⎣ α i 2 Ai (Tr 2 − Tw 2 ) + α o Ao (Ta − Tw 2 ) ⎦
⎡ L1 ⎤
⎡ L1 ⎤ ⎢ ⎥
⎢ P ⎥ ⎢ Pc ⎥
⎛ γ −b ⎞
⎢ c ⎥ ⎢ ⎜ ⎟ ⎥
⎢ xint 2 ⎥ ⎢ be ⎝ ab ⎠ − a ⎥
⎢ ⎥ ⎢ h ⎥
⎢ hout ⎥ ⎢ f
⎥
⎢ Tw1 ⎥ ⎢ Tw1 ⎥
g ( x, u ) = ⎢ ⎥=⎢ ⎥ ( 2.228 )
⎢ Tw 2 ⎥ ⎢ Tw 2
⎥
⎢ T ⎥ ⎢⎛ 1 ⎞ ⎛ μ ⎞ ⎥
⎢ a ,o ⎥ ⎢⎜⎜ 1 − μ ⎟⎟Ta − ⎜⎜ 1 − μ ⎟⎟Ta ,in ⎥
⎢Tr , sh ⎥ ⎢⎝ ⎠ ⎝ ⎠ ⎥
⎢m ⎥ ⎢ T r ,in − Tsat ⎥
⎢ c ⎥ ⎢ (ρ L + ρ L )A ⎥
⎢⎣ mrec ⎥⎦ ⎢ 1 1 2 2 cs
⎥
⎣⎢ mrec ⎦⎥
x = [L1 P γ Tw1 Tw 2 ]
T
mrec ( 2.229 )
u = [m& i m& a ]
T
m& o hi Tai ( 2.230 )
y = [L1 mrec ]
T
Pc xint 2 hout Tw1 Tw 2 Ta ,o Tr , sh mc ( 2.231 )
⎡n L ⎤
⎣ i =1 LTotal ⎦
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞ ⎡L T LT ⎤
⎜⎜ ⎟⎟⎜⎜ ⎟⎟Ta ,in + ⎢ 1 w,1 + 2 w, 2 ⎥
⎝ 1 − μ ⎠⎝ α o Ao ⎠ ⎣ LTotal LTotal ⎦
Ta = ( 2.233 )
⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
69
α o ⎛ Re l ⎞
m m
⎛ m& ⎞
=⎜ ⎟ = ⎜ air ⎟ ( 2.234 )
α o′ ⎝ Re′l ⎠ ⎝ m& ′air ⎠
∂α o ⎛ m& ⎞ ⎛ α ′ ⎞
m
= m⎜ air ⎟ ⎜ o ⎟ ( 2.235 )
⎛ Tw,1 − Tw, 2 ⎞
⎜⎜ ⎟
= ( 2.236 )
∂L1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L1 ⎞
⎜⎜ ⎟
= ( 2.237 )
∂Tw,1 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ L2 ⎞
⎜⎜ ⎟⎟
∂Ta ⎝ Total ⎠
L
= ( 2.238 )
∂Tw, 2 ⎛ 1 ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛ 1 + μ ⎞⎛ m& air C p ,air ⎞
⎜⎜ ⎟⎟⎜⎜ ⎟⎟
∂Ta ⎝ 1 − μ ⎠⎝ α o Ao ⎠
= ( 2.239 )
⎜⎜ ⎟⎟⎜⎜ ⎟⎟ + 1
⎝ 1 − μ ⎠⎝ α o Ao ⎠
⎛1+ μ ⎞ ⎛ 1 ⎞⎛ n Li Tw,i ⎞
⎜⎜ ⎟⎟Ta ,in − ⎜⎜ ⎟⎟⎜⎜ ∑ ⎟⎟
⎟⎟⎜⎜1 −
⎞
⎟⎟
= ⎜α A ( 2.240 )
∂m& air ⎛ ⎛ 1 ⎞⎛ m& air C p ,air ⎞ ⎞
2
⎝ o o ⎠⎝ α o ∂m& air ⎠
⎜ ⎜⎜ ⎜ ⎟⎟ + 1⎟
⎜ ⎝ 1 − μ ⎟⎟⎠⎜ α A ⎟
⎝ ⎝ o o ⎠ ⎠
∂Tr1 1 ⎛⎜ dTsat ∂Tr ,in ⎞ ∂T ⎛ ⎞
The following partial derivatives are also used: = + ⎟ , r1 = 1 ⎜ ∂Tr ,in ⎟,
∂Pc 2 ⎜ dPc ∂Pc ⎟ ∂hin 2 ⎜ ∂hin ⎟
⎝ h1 ⎠ ⎝ Pc ⎠
∂ρ1 ⎛⎜ ∂ρ1 ⎞ 1 ⎛ ∂ρ
⎟+ ⎜ 1
⎞⎛ dhg
⎟⎜ ⎞ ∂ρ1 1 ⎛⎜ ∂ρ1 ⎞
⎟,
= ⎟⎟ , =
∂Pc ⎜ ∂Pc ⎟ 2 ⎜ ∂h1 ⎟⎜⎝ dPc ⎠ ∂hin 2 ⎜⎝ ∂h1 ⎟
⎝ h1 ⎠ ⎝ Pc ⎠ Pc ⎠
⎞ ⎛ dhg dh f ⎞
∂hint 2
= (hg − h f )⎛⎜⎜ ∂x⎟⎟ + ⎜⎜
int 2
−
dh ∂h
⎟⎟ xint 2 + f , int 2 = (hg − h f )⎛⎜⎜ ∂x
⎞
⎟⎟ , int 2
∂Pc ⎝ ∂Pc ⎠ ⎝ dPc dPc ⎠ dPc ∂γ ⎝ ∂γ ⎠
70
∂xint 2
⎛ γ −b ⎞
⎜ ⎟ ⎡ ⎛1⎞ ⎛a+b⎞ ⎛ −⎛⎜ γ −b ⎞⎟ ⎞⎤⎛ ∂μ ⎞ ∂xint 2 xint 2 1
= b 2e⎝ ab ⎠
⎢1 − ⎜ ⎟ + ⎜ 2 ⎟(γ − b ) − ⎜ e ⎝ ab ⎠ ⎟⎥⎜⎜ S ⎟⎟ , = + , and
∂Pc ⎢⎣ ⎝ a ⎠ ⎝ a ⎠ ⎜ ⎟⎥⎝ ∂Pc ⎠ ∂γ ab b
⎝ ⎠⎦
∂μ S ⎛ dρ g ⎞⎛⎜ μ S ⎞ ⎛ dρ f
⎟−⎜
⎞⎛ μ S ⎞
⎟ , where a = μ S and b = 1 .
=⎜ ⎟ ⎟⎟⎜
∂Pc ⎜⎝ dPc ⎟⎠⎜⎝ ρ g ⎟ ⎜ dP
⎠ ⎝ c
⎜
⎠⎝ ρ f
⎟
⎠ 1 − μS 1 − μS
⎡ f x ,11 f x ,12 0 0 f x ,15 0 ⎤

⎢f f x , 22 f x , 23 0 0 f x , 26 ⎥⎥
⎢ x , 21
∂f ⎢ 0 0 0 0 0 0 ⎥
= Fx = ⎢ ⎥ ( 2.241 )
∂x ⎢ 0 f x , 42 f x , 43 0 0 0 ⎥
⎢ f x ,51 f x ,52 0 0 f x ,55 f x ,56 ⎥
⎢ ⎥
⎣⎢ f x ,61 f x ,62 0 0 f x ,65 f x ,66 ⎦⎥
⎡ f u ,11 0 f u ,13 0 0 ⎤
⎢ 0 f u , 22 0 0 0 ⎥⎥
⎢
∂f ⎢ f u ,31 f u ,32 0 0 0 ⎥
= Fu = ⎢ ⎥ ( 2.242 )
∂u ⎢ 0 f u , 42 0 0 0 ⎥
⎢ 0 0 f u ,53 f u ,54 f u ,55 ⎥
⎢ ⎥
⎣⎢ 0 0 0 f u , 64 f u , 65 ⎦⎥
⎡ g x ,11 0 0 0 0 0 ⎤
⎢ 0 g x , 22 0 0 0 0 ⎥⎥
⎢
⎢ 0 g x ,32 g x ,33 0 0 0 ⎥
⎢ ⎥
⎢ 0 g x , 42 0 0 0 0 ⎥
∂g ⎢ 0 0 0 0 g x ,55 0 ⎥
= Gx = ⎢ ⎥ ( 2.243 )
∂x ⎢ 0 0 0 0 0 g x ,66 ⎥
⎢g 0 0 0 g x ,75 g x ,76 ⎥
⎢ x ,71 ⎥
⎢ 0 g x ,82 0 0 0 0 ⎥
⎢g g x ,92 g x ,93 0 0 0 ⎥
⎢ x ,91 ⎥
⎢⎣ 0 0 0 g x ,10, 4 0 0 ⎥⎦
71
⎡0 0 0 0 0 ⎤
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 0 0 0 ⎥
∂g ⎢0 0 0 0 0 ⎥
= Gu = ⎢ ⎥ ( 2.244 )
∂u ⎢0 0 0 0 0 ⎥
⎢0 0 0 g u , 74 g u , 75 ⎥
⎢ ⎥
⎢0 0 g u ,83 0 0 ⎥
⎢0 0 g u ,93 0 0 ⎥
⎢ ⎥
⎢⎣0 0 0 0 0 ⎥⎦
Table 2.18 The elements of Equations 2.241 through 2.244.
α i1 Ai
f x ,11 (Tw1 − Tr1 )
LTotal
dhg L1 ⎛ ∂Tr1 ⎞
f x ,12 − m& in − α i1 Ai ⎜⎜ ⎟⎟
dPc LTotal ⎝ ∂Pc ⎠
L1
f x ,15 α i1 Ai
LTotal
α i 2 Ai
f x , 21 − (Tw2 − Tr 2 )
LTotal
⎛ dhg ∂hint 2 ⎞ L ⎛ dTsat ⎞
f x , 22 m& out ⎜⎜ − ⎟⎟ − α i 2 Ai 2 ⎜⎜ ⎟⎟
⎝ dPc ∂Pc ⎠ LTotal ⎝ dPc ⎠
⎛ ∂h ⎞
f x , 23 − m& out ⎜⎜ int 2 ⎟⎟
⎝ ∂γ ⎠
L
f x , 26 α i 2 Ai 2
LTotal
⎛ ∂h dh f ⎞ ∂T
f x , 42 m& out ⎜⎜ int 2 − ⎟⎟ − UArec rec
⎝ ∂Pc dPc ⎠ ∂Pc
⎛ ∂h ⎞
f x , 43 m& out ⎜⎜ int 2 ⎟⎟
⎝ ∂γ ⎠
⎛ ∂Ta ⎞
f x ,51 α o Ao ⎜⎜ ⎟⎟
⎝ ∂L1 ⎠
⎛ ∂Tr1 ⎞
f x ,52 α i1 Ai ⎜⎜ ⎟⎟
⎝ ∂Pc ⎠
72
⎛ ∂T ⎞
f x ,55 − α i1 Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw1 ⎠
⎛ ∂Ta ⎞
f x ,56 α o Ao ⎜⎜ ⎟⎟
∂
⎝ w2 ⎠
T
⎛ ∂Ta ⎞
f x ,61 α o Ao ⎜⎜ ⎟⎟
⎝ ∂L1 ⎠
⎛ ∂Tr 2 ⎞
f x ,62 α i 2 Ai ⎜⎜ ⎟⎟
⎝ ∂Pc ⎠
⎛ ∂Ta ⎞
f x ,65 α o Ao ⎜⎜ ⎟⎟
⎝ ∂Tw1 ⎠
⎛ ∂T ⎞
f x ,66 − α i 2 Ai − α o Ao + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂Tw 2 ⎠
f u ,11 hin − hg
⎛ L ⎞⎛ ∂Tr1 ⎞
f u ,13 m& in − α i1 Ai ⎜⎜ 1 ⎟⎟⎜⎜ ⎟⎟
⎝ LTotal ∂
⎠⎝ in ⎠
h
f u , 22 (h g − hint 2 )
f u ,31 1
f u ,32 −1
f u , 42 hint 2 − h f
⎛ ∂Tr1 ⎞
f u ,53 α i1 Ai ⎜⎜ ⎟⎟
⎝ ∂hin ⎠
⎛ ∂Ta ⎞
f u ,54 α o Ao ⎜⎜ ⎟
⎟
⎝ ∂Ta ,i ⎠
⎛ ∂Ta ⎞ ⎛ ∂α o ⎞
f u ,55 α o Ao ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟ Ao (Ta − Tw1 )
⎝ ∂m& a ⎠ ⎝ ∂m& a ⎠
⎛ ∂Ta ⎞
f u , 64 α o Ao ⎜⎜ ⎟
⎟
⎝ ∂Ta ,i ⎠
⎛ ∂α o ⎞ ⎛ ∂T ⎞
f u , 65 ⎜⎜ ⎟⎟ Ao (Ta − Tw 2 ) + α o Ao ⎜⎜ a ⎟⎟
⎝ ∂m& a ⎠ ⎝ ∂m& a ⎠
g x ,11 1
g x , 22 1
73
⎛ ∂xint 2 ⎞
g x ,32 ⎜⎜ ⎟⎟
⎝ ∂Pc ⎠
⎛ ∂xint 2 ⎞
g x ,33 ⎜⎜ ⎟⎟
⎝ ∂γ ⎠
dh f
g x , 42
dPc
g x ,55 1
g x , 66 1
⎛ 1 ⎞⎛ ∂Ta ⎞
g x ,71 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂L1 ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g x ,75 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ∂
⎠⎝ w1 ⎠
T
⎛ 1 ⎞⎛ ∂Ta ⎞
g x , 76 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tw 2 ⎠
⎛ ∂Tr ,in ⎞ ⎛ dT ⎞
g x ,82 ⎜ ⎟ − ⎜ sat ⎟⎟
⎜ ∂Pc ⎟ ⎜⎝ dPc ⎠
⎝ hin ⎠
g x ,91 (ρ1 − ρ 2 )Acs
⎛ ∂ρ1 ⎞ ⎡⎛ dρ ⎞ ⎛ dρ ⎞ ⎤
g x ,92 ⎜⎜ ⎟⎟ Acs L1 + ⎢⎜⎜ f ⎟⎟(1 − γ ) + ⎜⎜ g ⎟⎟(γ )⎥ Acs L2
⎝ ∂Pc ⎠ ⎢⎣⎝ dPc ⎠ ⎝ dPc ⎠ ⎥⎦
g x ,93 (ρ g − ρ f )Acs L2
g x ,10, 4 1
⎛ 1 ⎞⎛ ∂Ta ⎞ ⎛ μ ⎞
g u , 74 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂Tai ⎠ ⎝1− μ ⎠
⎛ 1 ⎞⎛ ∂Ta ⎞
g u , 75 ⎜⎜ ⎟⎟⎜⎜ ⎟⎟
⎝ 1 − μ ⎠⎝ ∂ma ⎠
⎛ ∂Tr ,in ⎞
g u ,83 ⎜ ⎟
⎜ ∂hin ⎟
⎝ Pc ⎠
⎛ ∂ρ1 ⎞
g u ,93 ⎜⎜ ⎟⎟ Acs L1
⎝ ∂hin ⎠
74
2.6.11 Internal Heat Exchanger
The internal heat exchanger is assumed to be a single-phase, counterflow heat exchanger. This component
is also modeled using a lumped parameter approach. However, because the mode of heat transfer is significantly
simpler, this component uses lumped capacitance assumptions to simplify the derivation. This results in a 3rd order
model.
Hot Fluid
m& h hh ,in P high Th,ave (t) m& h hh ,out (t )
Twall(t)
m& c hc ,out (t ) P low Tc,ave (t) m& c hc ,in
Cold Fluid
Figure 2.16 Diagram: Internal Counterflow Heat Exchanger

The three differential equations are formed by using the unsteady state form of the conservation of energy
equation for the hot fluid (Equation 2.245), the cold fluid (Equation 2.246), and the heat exchanger wall (Equation
Th,in + Th,out Tc ,in + Tc ,out
2.247). Average fluid temperatures are assumed and defined as Th , ave = and Tc ,ave = .
2 2
Uniform refrigerant pressure is assumed for each side of the heat exchanger, and the inlet and outlet temperatures
and enthalpies are then related in terms of the thermodynamic equation of state: Th ,in = T (Ph , hh ,in ) ,
Tc ,in = T (Pc , hc ,in ) , hh ,out = h(Ph , Th,out ) , and hc ,out = h(Pc , Tc ,out ) .
m& h (hh,in − hh,out ) − α h Ah (Th, ave − Twall ) = (ρVC p )h T&h, ave ( 2.245 )
m& c (hc ,in − hc ,out ) − α c Ac (Tc , ave − Twall ) = (ρVC p )c T&c , ave ( 2.246 )
α c Ac (Tc,ave − Twall ) + α h Ah (Th,ave − Twall ) = (ρVC p )wall T&wall ( 2.247 )

Recall that the internal heat exchanger was modeled with three differential equations, where
Th ,in + Th ,out Tc ,in + Tc ,out
Th , ave = and Tc ,ave = . Thus the outlet temperatures are calculated as
2 2
Th ,out = 2Th , ave − Th ,in and Tc ,out = 2Tc ,ave − Tc ,in . The states are assumed to be x = [Th ,ave Twall ] .
T
Tc ,ave
To simplify the implicit nature of these equations for linearization, the assumptions are made that
hh ,in − hh ,out ≈ C p h (Th ,in − Th ,out ) and hc ,in − hc ,out ≈ C p c (Tc ,in − Tc ,out ) . This assumption of average specific
heats will admittedly fail near the critical point. This problem will be addressed in future models of the heat
exchanger. After substitution, the differential equations are given in Equations 2.248 – 2.250.
2m& h C p h (Th ,in − Th , ave ) − α h Ah (Th ,ave − Twall ) = (ρVC p )h T&h ,ave ( 2.248 )
75
2m& c C p c (Tc ,in − Tc ,ave ) − α c Ac (Tc ,ave − Twall ) = (ρVC p )c T&c , ave ( 2.249 )
α c Ac (Tc,ave − Twall ) + α h Ah (Th,ave − Twall ) = (ρVC p )wall T&wall ( 2.250 )

The inputs and outputs are given in Equation 2.252 and 2.253. The partial derivatives of the differential
equations f ( x, u ) and output equations g ( x, u ) with respect to the states and inputs are defined in Equations
2.254 – 2.257, with the matrix elements listed in Table 2.19.
x = [Th ,ave Twall ]

T
Tc ,ave ( 2.251 )
u = [m& h hc ,in ]
T
m& c Ph Pc hh ,in ( 2.252 )
y = [hh ,out Tc ,out ]

T
hc ,out Th ,out ( 2.253 )
⎡ f x ,11 0 f x ,13 ⎤
∂f ⎢ ⎥
= Fx = ⎢ 0 f x , 22 f x , 23 ⎥ ( 2.254 )
∂x
⎢⎣ f x ,31 f x ,32 f x ,33 ⎥⎦
⎡ f u ,11 0 f u ,13 0 f u ,15 0 ⎤
∂f
= Fu = ⎢⎢ 0 f u , 22 0 f u , 24 0 f u , 26 ⎥⎥ ( 2.255 )
∂u
⎢⎣ 0 0 0 0 0 0 ⎥⎦
⎡ g x ,11 0 0⎤
⎢ 0 0⎥⎥
∂g g x , 22
= Gx = ⎢ ( 2.256 )
∂x ⎢ g x ,31 0 0⎥
⎢ ⎥
⎣ 0 g x , 42 0⎦
⎡0 0 g u ,13 0 g u ,15 0 ⎤
⎢0 0 0 0 g u , 26 ⎥⎥
∂g g u , 24
= Gu = ⎢ ( 2.257 )
∂u ⎢0 0 g u ,33 0 g u ,35 0 ⎥
⎢ ⎥
⎣0 0 0 g u , 44 0 g u , 46 ⎦
Table 2.19 Matrix Elements for Equations 2.254 – 2.257
⎡ α h Ah (Th , ave − Twall ) ⎛⎜ ∂C p h ⎞⎟ ⎤

⎢− 2m& hC p h − α h Ah + ⎥
1 ⎢ (C p h ) ⎜ ∂Th , ave ⎟
⎝ Ph ⎠ ⎥
f x ,11 ⎢ ⎥
(ρVC p )h ⎢ 2m& hC p h (Th , ave − Th ,in ) + α h Ah (Th , ave − Twall ) ⎛⎜ ∂ρ h ⎞⎥
⎢+ ⎟⎥
⎢⎣ ( ρh ) ⎜ ∂Th , ave ⎟
⎝ Ph ⎠ ⎥
⎦
α h Ah
f x ,13
(ρVC p )h
76
⎡ α c Ac (Tc, ave − Twall ) ⎛⎜ ∂C p c ⎞⎟ ⎤
⎢− 2m& cC p c − α c Ac + ⎥
1 ⎢ (C p c ) ⎜ ∂Tc , ave ⎟
⎝ Pc ⎠ ⎥
f x , 22 ⎢ ⎥
(ρVC p )c ⎢ 2m& cC p c (Tc , ave − Tc ,in ) + α c Ac (Tc , ave − Twall ) ⎜ ∂ρc
⎛ ⎞⎥
⎢+ ⎟⎥
⎢⎣ ( ρc ) ⎜ ∂Tc , ave ⎟
⎝ Pc ⎠ ⎥
⎦
α c Ac
f x , 23
(ρVC p )c
α h Ah
f x ,31
(ρVC ) p wall
α c Ac
f x ,32
(ρVC ) p wall
− (α h Ah + α c Ac )
f x ,33
(ρVC p )wall
2C p h (Th,in − Th ,ave )
f u ,11
(ρVC ) p h
⎡α h Ah (Th, ave − Twall ) ⎛ ∂C p ⎞ ⎤

⎢ ⎜ h ⎟ ⎥
1 ⎢ (C p h ) ⎜ ∂Ph
⎝ T
⎟
h , ave ⎠
⎥
f u ,13 ⎢ ⎥
(ρVC p )h ⎢ 2m& hC p h (Th, ave − Th,in ) + α h Ah (Th, ave − Twall ) ⎛ ∂ρh ⎞ ⎛ ∂Th ,in ⎞⎥
⎢+ ⎜ ⎟ + 2m& C ⎜ ⎟⎥
( ρh ) ⎜ ∂Ph ⎟ h ph
⎜ ∂Ph ⎟⎥
⎣⎢ ⎝ Th , ave ⎠ ⎝ hh , in ⎠ ⎦
2m& hC p h ⎛⎜ ∂Th ,in ⎞⎟

(ρVC p )h ⎜⎝ ∂hh,in Ph ⎟⎠
f u ,15
2C p c (Tc ,in − Tc ,ave )

f u , 22
(ρVC ) p c
⎡α c Ac (Tc , ave − Twall ) ⎛ ∂C p ⎞ ⎤

⎢ ⎜ c ⎟ ⎥
1 ⎢ (C p c ) ⎜
⎝ ∂P c T
⎟
c , ave ⎠
⎥
f u , 24 ⎢ ⎥
(ρVC p )c ⎢ 2m& cC p c (Tc , ave − Tc ,in ) + α c Ac (Tc , ave − Twall ) ⎛⎜ ∂ρc ⎞ ∂T ⎥
⎢+ ⎟ + 2m& C c , in
⎥
( ρc ) ⎜ ∂Pc ⎟ c pc
∂
⎣⎢ ⎝ Tc , ave ⎠
P c ⎦ ⎥
2m& cC p c ⎛⎜ ∂Tc ,in ⎞

⎟
(ρVC p )c ⎜⎝ ∂hc,in
f u , 26
⎟
Pc ⎠
⎛ ∂h ⎞
2⎜⎜ ⎟
h , out
g x ,11
⎜ ∂Th, out ⎟⎟
⎝ Ph ⎠
77
⎛ ∂h ⎞
2⎜ ⎟
c ,out
g x , 22
⎜ ∂T ⎟
⎝ c ,out Pc ⎠
g x ,31 2
g x , 42 2
⎛ ∂h ⎞⎛ ∂T ⎞ ⎛ ∂h ⎞
−⎜ ⎟⎜ h ,in ⎟ + ⎜ h ,out ⎟
h ,out
g u ,13
⎜ ∂T ⎟⎜ ∂P ⎟ ⎜ ∂Ph ⎟
⎝ h ,out Ph ⎠⎝ h hh , in ⎠ ⎝ Th , out ⎠
⎛ ∂h ⎞⎛ ∂T ⎞
g u ,15 − ⎜ h, out ⎟⎜ h,in ⎟
⎜ ∂Th , out ⎟⎜ ∂hh ,in ⎟
⎝ Ph ⎠⎝ Ph ⎠
⎛ ∂hc ,out ⎞⎛ ∂Tc ,in ⎞ ⎛ ∂hc ,out ⎞

g u , 24 −⎜ ⎟⎜ ⎟+⎜ ⎟
⎜ ∂Tc ,out ⎟⎜ ∂Pc ⎟ ⎜ ∂Pc ⎟
⎝ Pc ⎠⎝ hc , in ⎠ ⎝ Tc , out ⎠
⎛ ∂h ⎞⎛ ∂T ⎞
g u , 26 − ⎜ c , out ⎟⎜ c ,in ⎟
⎜ ∂Tc , out ⎟⎜ ∂hc ,in ⎟
⎝ Pc ⎠⎝ Pc ⎠
⎛ ∂T ⎞
g u ,33 − ⎜ h ,in ⎟
⎜ ∂Ph ⎟
⎝ hh ,in ⎠
⎛ ∂T ⎞
g u ,35 − ⎜ h ,in ⎟
⎜ ∂hh ,in ⎟
⎝ Ph ⎠
⎛ ∂T ⎞
g u , 44 − ⎜ c ,in ⎟
⎜ ∂Pc ⎟
⎝ hc ,in ⎠
⎛ ∂T ⎞
g u , 46 − ⎜ c ,in ⎟
⎜ ∂hc ,in ⎟
⎝ Pc ⎠
78
Chapter 3. System Simulation
This chapter describes a set of simulation tools developed for the purposes of model validation, analysis,
and controller design. This library was developed for use with MATLAB and its associated simulation program
Simulink®. At the time of publication of this dissertation, this software was entitled the Thermosys™ Toolbox for
MATLAB and available to the companies sponsoring this research effort (member companies of the Air Conditioning
and Refrigeration Center at the University of Illinois at Urbana-Champaign).
3.1 Software Introduction

The primary purpose for the development of the Thermosys™ Toolbox was to facilitate this research by
providing a means for model validation, dynamic analysis, and controller design. Much of the information included
in this chapter can be found in greater detail in the Thermosys™ User’s Manual [80].
The Thermosys™ Toolbox was developed for use in MATLAB. This software platform offers several
distinct advantages. First, the widespread use of this program among industrial and academic institutions as a
simulation and control design tool increases the applicability of the Thermosys™ Toolbox. Second, the open
architecture style of programming allows users to customize, extend, and otherwise modify the tools provided.
Third, the availability of additional toolboxes such as system identification, GUI creation, dynamic analysis, control
design, etc. that are available for use with MATLAB can also be used in conjunction with the Thermosys™ Toolbox.
3.2 Library Structure

Thermosys™ is a library of models and tools for simulating vapor compression systems. These models are
created using the visual programming package Simulink®, while making extensive use of the commands and
capabilities of MATLAB. These simulation tools can be easily accessed from the Simulink® Library Browser. The
tools are organized into three directories: auxiliary tools, components, and fluid properties.
79
Figure 3.1 Overview of Thermosys™ Library from the Simulink® Browser
3.2.1 Auxiliary Tools

These blocks are miscellaneous subroutines used repeatedly by the component models. Blocks are
included for calculating fluid quality, heat transfer coefficients using the Dittus Boelter Correlation, scaling of heat
transfer coefficients with Reynold’s number, and calculating mean void fraction based on a slip ratio correlation. A
modified memory block (see Simulink® manual) that permits the initial condition to be set externally is also
included.
80
Figure 3.2 Overview of the Auxiliary Tools in the Thermosys™ Toolbox
3.2.2 Components
These blocks include the component models developed in this dissertation. This includes the evaporator,
condenser, gas cooler, evaporator with receiver, condenser with receiver, internal heat exchanger, compressor, and
several types of expansion valves. Both nonlinear and linearized versions of these components are available. The
nonlinear models are based on the nonlinear governing equations (Chapter 2), and generally require more
computational time to simulate. The linear models use the equations resulting from a local linearization of the
governing equations (Chapter 2). These models require less computational time while approximating closely the
nonlinear models. Reduced order models are linear models that require fewer dynamic states and can be obtained
using advanced features of Thermosys™. These models require the least computational time and yield results
indistinguishable from the full order linear models.
81
Figure 3.3 Overview of the Components in the Thermosys™ Toolbox
3.2.3 Fluid Properties

Fluid properties are calculated using tables, based on predefined interpolation tables. Subroutines for
iteratively solving empirical equations of state were included in previous releases of Thermosys™, but have been
eliminated due to their excessively slow computation time. These are discussed in detail in Section 3.19.
82
Figure 3.4 Overview of Fluid Properties in the Thermosys™ Toolbox
3.3 Component Models

Each component model consists of three essential parts: the graphical user interface, the Simulink® block
diagram, and a component S-function.
3.3.1 Component GUIs

By double clicking on any component, an interactive graphical user interface (GUI) is activated which
allows the user to specify the component’s physical parameters, operating condition, and other necessary
information. The author developed the initial versions of these GUIs; Joel Jeddolah and Brian Eldredge contributed
to the development of the most recent versions. As an example, Figure 3.5 shows the graphical user interface for the
evaporator model. The different parts of these heat exchanger GUIs consist of the component name, physical
parameters, operating condition, heat transfer coefficients, recorded outputs, solver convergence error, and save/load
profile. Additionally, a description is included for the pipe loss blocks, and an expansion device.
83
Figure 3.5 Evaporator GUI
3.3.1.1 Component Name

This allows the user to specify the “name” of the component by assigning it a number. This becomes
critical for simulation of multi-component systems when several identical component blocks are used in the same
system model.
Figure 3.6 GUI: Component Name Section
3.3.1.2 Physical Parameters

The user is required to enter the effective parameters required by the component model. These are
specified in SI units.
84
Figure 3.7 GUI: Physical Parameters Section
3.3.1.3 Operating Condition

The component models require appropriate initial conditions to begin the simulation. Unfortunately, many
of the dynamic variables are not measurable, such as the effective length of two-phase flow. Therefore the user is
asked for commonly measured variables that define the operating condition, such as pressures and temperatures.
The GUI then uses this information to calculate the necessary initial conditions of the dynamic variables to match
the user-specified operating condition. Also calculated and displayed for the heat exchanger models are the
refrigerant mass inventory and the suggested upper and lower bounds for the slip ratio used in the mean void
fraction correlation.
Figure 3.8 GUI: Operating Condition Section
3.3.1.4 Heat Transfer Coefficients

The heat exchanger components use correlations to calculate the necessary heat transfer coefficients.
However, because of the unlimited number and variety of such correlations, it is impossible to include even a
fraction of those desired. Therefore options are included to specify a constant value, a user-defined correlation, or to
calculate the value from entered data. Note that these values are only used as initial values. During simulation all
heat transfer coefficients are assumed to scale with Reynold’s number.
An addition made by Brian Eldredge permits the user to select the external heat transfer coefficient using
industry standard J-factor correlations [50]. These relationships give a relationship between heat transfer coefficients
and Reynold’s number (Fig. 3.10).
85
Figure 3.9 GUI: Heat Transfer Coefficients Section
Figure 3.10 J-factor Correlation
3.3.1.5 Recorded Outputs

This section provides a simple interface for specifying which outputs are to be recorded during the
simulation as well as the option to decimate this data. This provides a simple means for the user to select only the
desired variables to be stored.
Figure 3.11 GUI: Recorded Outputs Section
86
3.3.1.6 Steady Solver Convergence Error
After specifying the physical parameters, operating condition, and heat transfer coefficients, the GUI
attempts to determine the initial conditions for system simulation using steady state energy balances. This requires
searching for a solution to a set of nonlinear equations. A constrained minimization algorithm is used to search for
an appropriate solution. The convergence error indicates to the user if the GUI was successful in finding a solution.
The “OK” button applies the changes made and closes the GUI without displaying the new calculated values. The
“Cancel” button closes the GUI without applying the changes, and the “Apply” button applies the changes and
displays the new calculated values without closing the GUI.
Figure 3.12 Steady State Solver Convergence Error
3.3.1.7 Save/Load Profile

Perhaps the most important feature of the GUIs is the ability to save the current profile to a file or load a
previously stored profile. This enables the user to quickly switch between commonly used component types or
operating conditions.
Figure 3.13 GUI: Save/Load Profile Section
3.3.1.8 Pipe Losses

The user specifies the boundary conditions for the fluid entering and exiting each component. Generally
the operating conditions would be taken from an experimental system with energy losses between components. To
resolve the discrepancy between the user-specified boundary conditions, the GUI calculates the necessary static
pressure drop and temperature (enthalpy) drop between the current and next component to ensure that the boundary
conditions are met. Alternatively the user can disable this property by specifying the losses directly. Pipe loss
models based on turbulent flow friction factor correlations can also be selected to determine the pressure drop
between components. Additionally, the model can determine the transport delay due to long pipe lengths and
parasitic heat gain.
87
Figure 3.14 Pressure and Temperature Drop Block
3.3.1.9 Expansion Valve and Compressor GUIs

The expansion valve and compressor models also require the user to enter information regarding operating
condition, component name, and desired recorded outputs. Additionally, the user specifies an empirical map (e.g.
discharge coefficient, volumetric efficiency, etc.) and any rate limit or delay information.
Figure 3.15 Expansion Valve GUI
3.3.2 Simulink® Block Diagram

Each of the component models in organized as a masked subsystem. The component “mask” simply
displays an image of the component type. Underneath the mask, a Simulink® block diagram arranges and passes the
component inputs, physical parameters, and calculated thermodynamic properties to the component S-function
88
(Figure 3.16). The outputs of the S-function are then converted into the desired component outputs. Generally the
outputs of the component S-function are state derivatives, which are integrated at each time step (Figure 3.17).
3.3.3 S-function
The component S-function is sequential code that performs operations more easily understood as text code
rather than visual-based programming like block diagrams (Figure 3.18). This generally consists of matrix
calculations and inversions. This code is compiled for faster execution.
Figure 3.16 Sample Simulink® Block Diagram
89
Figure 3.17 Integration of S-function Outputs
Figure 3.18 Sample S-function Code
3.4 System Models

To build a system simulation, the inputs and outputs of each component can be connected in a logical
manner with a basic understanding of vapor compression systems. Color-coded “From” and “Go To” tags can be
used to avoid confusing wire diagrams. The following figure demonstrates how the components of a subcritical
vapor compression system can be connected together (Figure 3.19). Pressure, enthalpy, and mass flow rate are used
to define the inputs and outputs of each component. Proper connection between all the components is required for
correct simulation outcome. Sample models for subcritical and transcritical vapor compression systems with R134a
90
and CO2 as refrigerants, respectively, are included in the library. Various external inputs can be applied to the
system. These include:
• Compressor speed
• Expansion valve opening
• Mass flow rate of inlet air to the evaporator
• Temperature of inlet air to the evaporator
• Mass flow rate of inlet air to the condenser/gas cooler
• Temperature of inlet air to the condenser/gas cooler
Figure 3.19 Sample System Simulation Model
3.5 Fluid Properties

The Thermosys™ Toolbox currently contains thermo-physical properties and data for five different
refrigerants:
• R134a
• R744 (CO2)
• R22
• R404a
• R410a
91
3.5.1 Calculation
1-D and 2-D look-up tables were generated in EES (Engineering Equation Solver) and saved as comma-
delimited files (csv). A MATLAB m-file loads the comma-delimited files and stores the data in matrix form. Partial
derivatives of thermodynamic functions are also evaluated and stored as matrices. The tables can be used to
calculate thermodynamic properties by using the interpolation routines included with MATLAB/Simulink®. The
Thermosys™ Toolbox contains pre-programmed Simulink® 1-D and 2-D interpolation blocks for fluid property
calculation (Figure 3.20). These are independent of the type of fluid. Whichever fluid properties are currently
loaded in the MATLAB workspace are used.
Figure 3.20 Sample Simulink® 2-D Interpolation Table
3.6 Library Limitations

The toolbox has a few notable limitations. The most important of these is that the simulation results are
only valid when the assumptions applied in the modeling approach are valid. An obvious example: If the system
being simulated has a 2-region evaporator model (a two-phase region and a superheat region) and the length of the
superheat region goes to zero, the simulation will fail. For this purpose the component models have programming
which halts the simulation if the model violates any of these constraints. Also, if the steady state conditions supplied
to the system result in a solution that is physically impossible (e.g. negative lengths or pressures), the simulation will
likewise stop.
Figure 3.21 Sample Interpolation Table: Temperature as a Function of Pressure and Enthalpy for Carbon
Dioxide
92
Figure 3.22 Sample Interpolation Table: Specific Heat as a Function of Pressure and Temperature for Carbon
Dioxide
The primary method for fluid property calculation also has limitations. First, the calculated result is only as
accurate as the mesh used to create the interpolation table (Figure 3.21). Second, when evaluating properties near the
critical point, incorrect results are more likely. A good example of this can be seen when plotting specific heat near
the critical point (Figure 3.22). Because specific heat approaches infinity at the critical point by definition, using
interpolation tables near this point can be problematic. Because this is generally only a common problem for
transcritical cycles, iterative subroutines for calculating the exact fluid properties based on the equation of state for
carbon dioxide (CO2) are included in the toolbox. These routines are extremely slow and only are recommended
when reliable results are not possible using the look-up tables.
3.7 Sample Systems

The Thermosys™ Toolbox includes several sample system simulations. These include:
• R134a critically charged system with EEV
• R134a critically charged system with orifice tube
• R134a critically charged system with TEV
• R134a system with EEV and low-side accumulator
• R134a system with EEV and high-side receiver
• R134a system with EEV and dual evaporators
• R744 (CO2) system with EEV and internal heat exchanger
This set of sample simulations is meant to be representative of the versatility of the toolbox, and provide a template
for creating more complex and alternative system simulations.
93
Chapter 4. Experimental System
This chapter describes an experimental air-conditioning and refrigeration system used for dynamic model
validation and control research. Much of the model validation research by the author that preceded this dissertation
was conducted on experimental facilities available as part of the Air-Conditioning and Refrigeration Center (ACRC)
at the University of Illinois at Urbana-Champaign. However, these systems lacked some of the sensing and
actuating abilities necessary for the design and implementation of advanced control strategies. To this end an
experimental system was designed and constructed to meet the needs of this and future research efforts.
4.1 General System Description

The experimental system is a dual-evaporator “trainer” system, with tube-and-fin heat exchangers, semi-
hermetic compressor, liquid line receiver, suction line accumulator, internal heat exchanger, an assortment of
expansion devices, and full suite of sensors. Numerous manual valves allow different components to be
included/excluded as desired, resulting in many different system configurations. A notable deficiency of the system
is the inability to control ambient temperatures. However, the system is deemed sufficient for its primary objectives
of transient testing and control without this ability.
A photo of the original system is shown in Figure 4.1. This photo predates many modifications, including
the interfacing of the system to the data acquisition and control system, the addition of Electronic Expansion Valves
(EEV), the addition of immersion thermocouples, the addition of variable fan speed control, the relocation of the
compressor and variable frequency drive to minimize electrical noise, and the addition of insulation to minimize
parasitic heat gain/losses. A recent photo of the modified system is shown in Figure 4.2.
Figure 4.1 Photo of original system
94
Figure 4.2 Photo of modified system
95
Figure 4.3 Schematic of Experimental System
A schematic of the system is shown in Figure 4.3. Major components are identified by name, while
explanation of the valve acronyms are given in Table 4.1. The component models and manufacturers is listed in
Table 4.2.
Table 4.1 Valve Designations

Valve Designation Description
CTV Capillary Tube Valve
TEV Thermostatic Expansion Valve
AEV Automatic Expansion Valve
96
EEV Electronic Expansion Valve
SV Solenoid Valve
SVB Solenoid Valve Bypass
EPR Evaporator Pressure Regulating Valve
EPRB Electronic Pressure Regulating Valve Bypass
ES2 Evaporator Side #2 Valve
HXV Internal Heat Exchanger Valve
HXB Internal Heat Exchanger Bypass
LRI Liquid Line Receiver Inlet
LRO Liquid Line Receiver Outlet
LRB Liquid Line Receiver Bypass
SAI Suction Line Accumulator Inlet
SAO Suction Line Accumulator Outlet
SAB Suction Line Accumulator Bypass
MV Manual Valve
Table 4.2 Component Name, Manufacturer, Model, and URL
Component Manufacturer Model URL

Evaporator Fan and
Larkin (HeatCraft) VAK-17A www.heatcraftrpd.com
Casing
Evaporator #1 HeatCraft 52601301 (VAK-17A) www.heatcraftrpd.com
Evaporator #2 Blissfield BH517 www.blissfield.com
Condenser and Fan Tecumseh (Blissfield) 50803-1 (66001-3) www.blissfield.com
Fan Control Boards Control Resources Inc. Nimbus 240BJW00 www.controlres.com
Internal Heat Exchanger Superior (Sherwood) HXSV-1/2 www.sherwoodvalve.com
AC&R Components, Inc. (Henry
Liquid Line Receiver S-8064 www.henrytech.com
Tech.)
Suction Line Accumulator S-7043 www.henrytech.com
Tech.)
Oil Separator S-5581 www.henrytech.com
Tech.)
Compressor Copeland KANA-006E-TAC-800 www.copeland-corp.com
Variable Frequency Drive Baldor ID15J101-ER www.baldor.com
Capillary Tubing Sealed Unit Parts Co., Inc. BC-4 www.supco.com
TEV Sporlan Valve Co. FJ ¼ C www.sporlan.com
AEV Parker-Hannefin A2 www.parker.com
EEV Sporlan Valve Co. SEI-0.5 www.sporlan.com
EEV Control Board Sporlan Valve Co. IB1, TCB www.sporlan.com
EPR Sporlan Valve Co. ORIT 6-0/50-1/2” www.sporlan.com
Manual Valves Mueller Brass Co. 14838, 14841 www.muellerindustries.com
Filter-Dryer Sporlan Valve Co. C-052 www.sporlan.com
Sight Glasses Sporlan Valve Co. SA-14S, SA-12FM www.sporlan.com
Pressure Transducers Cole-Palmer 07356-53, 07356-54 www.coleparmer.com
Pressure Gauges Ritchie Engineering Co., Inc. 49051, 49052 www.yellowjacket.com
Mass Flow Transducers McMillan Company 102-6P www.mcmflow.com
Mass Flow Gauges Brooks Instrument (Emerson) 1350 EPIPMEAIA www.emersonprocess.com/brooks
Immersion Thermocouple Omega GTMQSS-062U-6 www.omega.com
Welded Thermocouple Omega FF-T-20-100 www.omega.com
Watt Meter Ohio Semitronics, Inc. GW5-019D www.ohiosemitronics.com
Pressure Switches Ranco 010-1402, 011-1711 www.ranco.invensys.com
Line Reactors MTE (Galco Industrial Electronics) MTE RL-00402 www.galco.com
Analog Input Board Measurement Computing, Inc. PCI-DAS1200/JR www.measurementcomputing.com
97
Analog Output Board Measurement Computing, Inc. PCI-DDA-08/12 www.measurementcomputing.com
Thermocouple Board Measurement Computing, Inc. PCI-DAS-TC www.measurementcomputing.com
Terminal Boards Measurement Computing, Inc. CIO-MINI50 www.measurementcomputing.com
Signal Conditioners Omega OM5 Series www.omega.com
4.2 Sensors
The experimental system is equipped with a wide range of sensing capabilities. Temperature measurements
are obtained using type T thermocouples, and include air, surface, and refrigerant measurements. The presence of
thermowells also provides the ability to take manual measurements using a thermometer. Refrigerant pressure is
measured using strain-gauge based pressure transducers, while needle-based pressure gauges allow for visual
corroboration. Turbine-based mass flow sensors are used to measure liquid mass flow prior to the expansion device.
Visual mass flow gauges are also present. Finally, electric power consumed by the compressor and/or heat
exchanger fans is measured using a watt-meter. Table 4.3 details the location of the surface thermocouples,
immersion thermocouples, thermowells, pressure transducers, pressure gauges, and flow transducers.
Table 4.3 Temperature, Pressure, and Flow Measurement Designations
Surface Immersion
Thermowell Location
Thermocouple Thermocouple
T-1A TW-1A Evaporator #1 Inlet
T-1B T-1B-IM TW-1A (removed) Evaporator #1 Outlet
T-2A TW-2A Evaporator #2 Inlet
T-2B TW-2B Evaporator #2 Outlet
T-3A T-3A-IM TW-3A (removed) Condenser Inlet
T-3B T-3B-IM TW-3B (removed) Condenser Outlet
T-4A Evaporator #1 Air Inlet
T-4B Evaporator #1 Air Outlet
T-5A Evaporator #2 Air Inlet
T-5B Evaporator #2 Air Outlet
T-6A Condenser Air Inlet
T-6B Condenser Air Outlet
T-7A T-7A-IM TW-7A (removed) Compressor Inlet
T-7B T-7B-IM TW-7B (removed) Compressor Outlet
T-8A TW-8A Internal Heat Exchanger – Liquid Inlet
T-8B TW-8B Internal Heat Exchanger – Liquid Outlet
T-9A TW-9A Internal Heat Exchanger – Vapor Inlet
T-9B TW-9B Internal Heat Exchanger – Vapor Outlet
T-10A Ambient
Pressure
Pressure Gage Flow Transducer Location
Transducer
PT-1 PG-1 Evaporator #1 Inlet
PT-2 PG-2 Evaporator #1 Outlet
PT-3 PG-3 Compressor Inlet
PT-4 PG-4 Compressor Outlet
PT-5 PG-5 Evaporator #2 Inlet
PT-6 PG-6 Evaporator #2 Outlet
FT-1 Expansion Device #1 Inlet
FT-2 Expansion Device #2 Inlet
98
4.2.1 Temperature Measurement
Type T thermocouples are used throughout the system for temperature measurement. Welded tip
thermocouples are used for air and surface measurements, while ungrounded thermocouples with stainless steel
sheaths are used for immersed refrigerant measurement (Omega GTMQSS-062U-6, see www.omega.com). Figures
4.4-4.6 depict air, surface, and immersion thermocouples.
Figure 4.4 Welded Tip Thermocouple for Air Temperature Measurement
Figure 4.5 Welded Tip Thermocouple for Surface Temperature Measurement
Figure 4.6 Immersion Thermocouple for Refrigerant Temperature Measurement
99
4.2.2 Pressure Measurement
Pressure measurements are made using a strain-gage based pressure sensor (Cole-Parmer 07356-53 and
07356-54, see www.coleparmer.com). Of the six pressure transducers, five are 07356-53 sensors that have a range
of 0 to 100 psig, while the sixth sensor is 07356-54 and has a range of 0 to 300 psig and is located on the high
pressure side of the compressor. Both sensor models output a 4-20 mA signal, have a listed accuracy of ±0.4%, a
response time of less than 5 ms, and are rated for -20 to 160ºF (-29 to 71ºC) . Using a 497Ω resistor, the 4-20 mA
signal is converted to the 2-5V range accepted by the data acquisition system. The transducers are located at the inlet
and outlet of each evaporator and the compressor. There are also six pressure gauges in corresponding locations for
visual corroboration.
Figure 4.7 Pressure Transducer
Figure 4.8 Pressure Gauge
4.2.3 Mass Flow Measurement

Liquid mass flow before each of the valve arrays is measured using turbine-based mass flow meter
(McMillan 102-6P, see www.mcmflow.com). These sensors have a range of 100-1000 mL/minute with ±1%
accuracy, and are rated to 500 psig and from 41 to 122ºF (5 to 50ºC). The sensors provide a 0-5V output compatible
with the data acquisition system. Two visual mass flow meters manufactured by Brooks (1350 Sho-Rate flow meter)
were placed in corresponding locations, but ruptured and were removed from the system.
100
Figure 4.9 Mass Flow Transducer
Figure 4.10 Mass Flow Gauge
4.2.4 Power Measurement

Power consumed by the compressor and/or heat exchanger fans is measured using an AC watt-transducer
(Ohio Semitronics GW5-019D, see www.ohiosemitronics.com). The transducer outputs a 0 to 10 V signal with
±0.2% accuracy. The power transducer is shown in Figure 4.11.
101
Figure 4.11 Power Transducer
4.3 Actuators
The experimental system is designed for complete computer control of compressor speed, heat exchanger
fan speed, and electronic valve position. Other types of expansion devices such as the TEVs, AEV, etc. operate as
mechanical feedback devices and are not computer controlled.
4.3.1 Compressor
The compressor is a Copeland KANA-006E-TAC-800 semi-hermetic reciprocating compressor (see
www.copeland-corp.com). The design of a reciprocating compressor uses a piston driven from a crankshaft to pull
refrigerant vapor in from the low pressure side, and discharge the compressed vapor to the high pressure side. In the
case of a semi-hermetic compressor, an electric motor is mounted directly on the compressor crankshaft, and the two
are enclosed in a shell. The semi-hermetic compressor has the advantages of design simplicity, durability, high
efficiencies at high compression ratios, and relatively low cost. This compressor model has two cylinders with a
bore diameter of 1.375 in and a stroke of 0.625 in. Thus the total compressor volume is 1.856 in3 (30.42 cm3). The
compressor is shown in Figure 4.12.
Figure 4.12 Compressor
102
Figure 4.13 shows the Baldor ID15J101-ER variable frequency drive (VFD) that is used to vary the
compressor speed and to convert single-phase 120VAC to the three-phase 240VAC required by the compressor (see
www.baldor.com). Two MTE RL-00402 line reactors (Figure 4.14) are used to filter high frequency electric
fluctuations before and after the VFD (see www.galco.com). Ranco pressure switches (models 010-1402 and 011-
1711, see www.ranco.invensys.com) are used to cut off power to the compressor if the pressures are either too high
or too low (Figure 4.15).
Figure 4.13 Variable Frequency Drive
Figure 4.14 Line Reactors
103
Figure 4.15 Pressure Switches
The empirical compressor models depend on maps for volumetric efficiency and adiabatic efficiency.
These maps are shown in Figures 4.18 and 4.20, and were generated from extensive data taken on the experimental
system.
3000
High
Com
pres
sor P
2500 ower
ssure
2000
r Pre
orato
RPM
1500
Evap
at
erhe
Sup
Low
rator
1000 a po
w Ev
Lo
500
0
6 8 10 12 14 16 18
Valve Opening [%]
Figure 4.16 Operating Envelope
104
3000 700
650
2500
600
2000 550
ΔP [kPa]
500
RPM
1500
450
1000 400
350
500
300
0 250
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6 8 10 12 14 16 18
(P /P ) [kPa] Valve Opening [%]
ko ki
1000
950
900
Pc [kPa]
850
800
750
700
650
100 150 200 250 300 350 400 450
P [kPa]
e
Figure 4.17 Range of Operating Conditions
ηvol = (0.65127) + (0.00027681)*RPM + (−0.031338)*Pr

2 2
+ (3.0221e−005)*RPM*Pr + (−1.1905e−007)*RPM + (−0.0081256)*Pr
0.8
vol
0.6
η
0.4
0.2
3000
6
2000 5
4
1000 3
2
0 1
RPM (P /P ) [kPa]
ko ki
Figure 4.18 Compressor Volumetric Efficiency Map
105
errorRMS = 0.11556 g/s, errormax = 0.49796 g/s
Mass Flow Rate [g/s] 15
10
0
3000
6
2000 5
4
1000 3
2
0 1
RPM (P /P ) [kPa]
ko ki
Figure 4.19 Compressor Mass Flow Prediction
η = (0.5119) + (−0.0030112)*RPM + (1.5437)*P

adiabatic r
+ (−0.00089251)*RPM*P + (1.5795e−006)*RPM2 + (0.066539)*P2
r r
20
15
adiabatic
10
η
0
3000
6
2000 5
4
1000 3
2
0 1
RPM (P /P ) [kPa]
ko ki
Figure 4.20 Compressor Adiabatic Efficiency Map
4.3.2 Heat Exchanger Fans

The fans associated with each of the heat exchangers are controlled using a Nimbus 240BJW00 fan control
board from Control Resources, Inc. (see www.controlres.com). These boards take a 0-10V signal, and alter the AC
signal powering the fans to alter the fan speed. An approximate air flow rate map was generated for each of the fans
by measuring the average air flow rate using a hot ball anemometer at locations represented regions of equal area.
106
The evaporator fan was characterized using 20 measurement locations (shown in Fig. 4.24 with the center
measurement location repeated four times), and 15 locations were used for the condenser fan. This was repeated at
different fan speeds to generate the empirical flow rate maps shown in Figure 4.26. The evaporator fans are rated
with a maximum flow rate of 245 cfm (0.116 m3/s), while the maximum rate flow rate of the condenser fan is not
available.
Figure 4.21 Evaporator Fan Figure 4.22 Condenser Fan
Figure 4.23 Fan Control Boards
107
17
16
15
5.5
14 5
4.5
Air Velocity [m/s]

4
3.5
1 2 3 4 5 6 7 8 9
3
2.5
1.5
13 4
Vertical Position
2 4
12 0 2
0
11 −2
−2
10 Vertical Position [in] −4 −4 Horizontal Position [in]
Horizontal Position
Figure 4.24 Evaporator Velocity Measurement Figure 4.25 Velocity Profile for Evaporator #1 at
Locations 100% Fan Speed
0.25
Evaporator #1
Evaporator #2
Condenser
0.2
Volumetric Flow Rate [m /s]
3
0.15
0.1
0.05
0
40 50 60 70 80 90 100
Fan Speed [%]
Figure 4.26 Evaporator and Condenser Fan Volumetric Flow Maps
4.3.3 Electronic Expansion Valve

The electronic expansion valve (EEV) is manufactured by Sporlan (SEI-0.5, see www.sporlan.com), and
uses a stepper motor to open/close the valve. The stepper motor has a range of 1596 steps, and a slew rate of 200
steps/sec. The EEV can be controlled using either the Sporlan Series IB control board, or its predecessor, the TCB
board, which has more features.
The stepper motor has both hysteresis and quantization effects. The hysteresis of the first valve (EEV #1)
is experimentally determined to be approximately 1.1% of the total travel length (Figure 4.29), and 0.1% for EEV #2
(Figure 4.30). Since for this experimental system the valve normally operates between 6-20% open, this hysteresis
is significant and requires compensation in the software (e.g. Figure 4.31). The EEVs also exhibits quantization
effects of slightly less than 0.1% of the travel length. This corresponds roughly to a single step of the stepper motor,
108
and is expected (Figure 4.32). The mass flow rate maps for each of the EEVs is determined using extensive
empirical data, and are shown in Figures 4.33-4.36.
Figure 4.27 Photo of EEV Figure 4.28 Photo of IB Control Board
−3 −3
x 10 EEV Hysteresis − Uncompensated x 10 EEV #2 Hysteresis − Uncompensated
7 7
6.5
6.5
6
5.5
6
Mass Flow Rate [kg/s]
4.5 5.5
5
3.5
3
4.5
2.5
2 4
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 9 9.5 10 10.5 11 11.5 12 12.5 13
Valve Opening [%] Valve Opening [%]
Figure 4.29 EEV #1 Hysteresis Figure 4.30 EEV #2 Hysteresis
109
−3 −3
x 10 EEV Hysteresis − Compensated (1.1%) x 10 EEV Quantization
7

6.5 4
6 3.5
5.5
3
5
2.5
5600 5800 6000 6200 6400 6600 6800
4.5
9
4
8.5
Valve Opening [%]

3.5
8
3 7.5
7
2.5
6.5
2 6
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 5600 5800 6000 6200 6400 6600 6800
Valve Opening [%] Time [s]
Figure 4.31 EEV #1 with Hysteresis Compensation Figure 4.32 EEV Quantization
C (10−6) = (−9.5984) + (2.0481)u + (0.0054106)ΔP + (−0.00074909)u ΔP + (−0.037775)u

d v v
Effective Discharge Coef. [10−6]
15
10
0
20
700
15 600
500
10 400
300
5 200
Valve Opening [%] (P −P ) [kPa]
vi vo
Figure 4.33 EEV #1 Effective Discharge Coefficient Map
110
0
20
700
15 600
500
10 400
300
5 200
vi vo
Figure 4.34 EEV #1 Mass Flow Rate Prediction
C (10−6) = (−9.7959) + (1.8027)u + (0.0074994)ΔP + (−0.00093102)u ΔP + (−0.018605)u

d v v
Effective Discharge Coef. [10−6]
15
10
0
20
800
15 700
600
10 500
400
5 300
vi vo
Figure 4.35 EEV #2 Effective Discharge Coefficient Map
111
0
20
800
15 700
600
10 500
400
5 300
vi vo
Figure 4.36 EEV #2 Mass Flow Rate Prediction
4.3.4 Thermostatic Expansion Valve

The Thermostatic Expansion Valve (TEV) is also manufactured by Sporlan (www.sporlan.com). This
system uses model FJ-¼C, and is shown in Figure 4.37. The valve opening is determined by a force balance of the
evaporator inlet pressure, the saturation pressure of the evaporator outlet temperature, and an adjustable spring force.
A sensing bulb is placed at the outlet of the evaporator as shown in Figure 4.38. The pressure in this bulb is the
saturation pressure corresponding to the evaporator outlet temperature. This mechanism results in a mechanical
feedback loop that controls the mass flow rate of refrigerant in an effort to maintain a constant pressure difference,
or superheat. The model for this component requires geometric information about the valve mechanism, and an
empirical map for the discharge coefficient. Although sufficient empirical data has been recorded, at the time of
writing the necessary information about valve geometry was not available. Therefore, an empirical map for
discharge coefficient will be constructed and reported in later publications.
112
Figure 4.37 Photo of TEV
Figure 4.38 Diagram of TEV Operation (from [1])
113
4.3.5 Capillary Tube Expansion
The capillary tube is the simplest expansion device. A fixed length of tubing with a small diameter is used
as a fixed orifice expansion device. Although there are no moving parts, the fixed nature of the device results in
relatively inflexible rate of refrigerant flow. Liquid line receivers are not used with capillary tube systems, and
proper system operation requires the correct amount of refrigerant charge. However, even with proper charge, the
capillary tube may at times starve or flood the evaporator.
The capillary tubing in use in the system is manufactured by Sealed Unit Parts Co., Inc. (www.supco.com).
The capillary tube is 8 ft of BC-4, with outer diameter of 0.125 in and inner diameter 0.064 in. The empirical mass
flow rate map for this component is given in Figures 4.40-4.41.
Figure 4.39 Photo of Capillary Tube
Cd = (0.018742) + (−7.7665e−005)ΔP + (1.4969e−007)ΔP2 + (−8.9518e−011)ΔP3
7.4
7.2
7
Discharge Coef. [−]
6.8
6.6
6.4
6.2
5.8
5.6
250 300 350 400 450 500 550 600 650 700
(Pvi−Pvo) [kPa]
Figure 4.40 Capillary Tube Discharge Coefficient Map
114
error = 0.142 g/s, error = 0.93253 g/s
RMS max
10.5
10
9.5
Mass Flow Rate [g/s]
8.5
7.5
6.5
250 300 350 400 450 500 550 600 650 700
(Pvi−Pvo) [kPa]
Figure 4.41 Capillary Tube Mass Flow Rate Prediction
4.3.6 Automatic Expansion Valve

The automatic expansion valve (AEV) is better described as a constant evaporator pressure expansion
valve. The valve consists of a diaphragm that opens/closes a ball valve based on a force balance of the evaporator
pressure, atmospheric pressure, and spring force that is adjustable using a screw. The valve prevents liquid
refrigerant from entering the evaporator, but rather sprays a mist of refrigerant (leading to the description of a “dry”
system). However, the AEV has a tendency to starve the evaporator during high load conditions, and flood the
evaporator during low load conditions. The AEV is model A2 manufactured by Parker (www.parker.com). The
empirical mass flow rate map for this component is given in Figure 4.44.
115
Figure 4.42 Photo of AEV
Figure 4.43 Diagram of AEV Construction (from Parker Bulletin 21-02D)
116
C = (−1.5796e−005) + (2.2945e−008)ΔP + (3.2426e−008)P
d e
Effective Discharge Coef. [−]

0.015
0.01
0.005
0
400
700
300
600
200 500
400
100 300
P [kPa] (P −P ) [kPa]
ei vi vo
Figure 4.44 AEV Effective Discharge Coefficient Map
error = 0.2903 g/s, error = 1.8447 g/s

RMS max
15
Mass Flow Rate [g/s]
10
0
400
700
300
600
200 500
400
100 300
Pei [kPa] (P −P ) [kPa]
vi vo
Figure 4.45 AEV Mass Flow Rate Map
4.3.7 Evaporator Pressure Regulating Valve

In a multiple evaporator system, an evaporator pressure regulator (EPR) is used to maintain different
evaporator pressures (and hence temperatures) between the different evaporators. Its operation is similar to the AEV
with a diaphragm that adjusts a valve plunger based on evaporation pressures. The EPR is model ORIT 6-0/50-1/2”
manufactured by Sporlan (www.sporlan.com).
117
Figure 4.46 Photo of EPR
Figure 4.47 Diagram of EPR Construction (from [2])
4.4 Data Acquisition

The experimental system is interfaced with a data acquisition and control system. This system consists of a
Dell Optiplex GX240 computer with a Pentium 4 1.5 GHz processor and 256Kb of memory (www.Dell.com),
various signal conditioning equipment, and a 16 channel thermocouple board (PCI-DAS-TC), an 8 channel 12 bit
analog output board (PCI-DDA08/12), a 16 channel 12 bit analog input board (PCI-DAS1200/JR), and terminal
boards (CIO-MINI50) from Measurement Computing (www.measurementcomputing.com). The signal conditioning
equipment can be purchased from a number of sources, however, the Omega series is listed here for convenience.
Isolation modules with the associated back plane and screw terminal boards (OM5-IV series, www.omega.com) are
used to isolate the analog inputs and outputs from the computer. This is particularly important for electrically
isolating the VFD from the main system to prevent electro-magnetic noise from contaminating the thermocouple
signals.
118
Figure 4.48 Photo of Computer and Data Acquisition
Figure 4.49 Data Acquisition Boards
Figure 4.50 Terminal Boards
119
Figure 4.51 Signal Conditioning Equipment
There are 19 surface thermocouples, and 5 immersion thermocouples. Since not all of these sensors are
essential for any given test, only the 16 critical thermocouples are connected to the thermocouple board. Likewise,
the analog input board accepts 16 single-ended analog inputs, or 8 double-ended inputs. With 6 pressure
transducers, 2 mass flow rate transducers, and 1 power meter, one of these sensors has to be omitted during each
test. Fortunately, the 8 channel analog output board is more than sufficient for controlling the two EEVs, the three
heat exchanger fans, and the compressor.
4.5 Heat Exchangers and Auxiliary Components

The experimental system has four different heat exchangers: two evaporators, a condenser and an internal
heat exchanger. Various auxiliary components are also described in this section.
4.5.1 Evaporator
The evaporator fan and casing are Larkin model VAK-17A Reach-in Cooler units manufactured by
HeatCraft (www.heatcraftrpd.com). Both evaporators are tube-and-fin with copper tubes and aluminum fins.
Evaporator #1 is the original HeatCraft evaporator that is the standard with Larkin VAK-17A units. Evaporator #2
was replaced with a similar evaporator with an alternative tube configuration. While evaporator #1 is typical of
industrial evaporators in that there are effectively two slabs, meaning that the fluid enters at the top of the heat
exchanger and flows back and forth down the front side, and then back up the second slab, to finally exit at the top.
The tube configuration of evaporator #2 is more typical of condensing units in that the fluid enters at the top and
then flows back and forth from top to bottom, exiting at the bottom. While evaporator #1 is more typical of
industrial configurations, the configuration of evaporator #2 is simpler and requires fewer assumptions regarding
lumped air temperatures for dynamic modeling. Thus both types are available for model validation and control
design.
A diagram and photo of the cooling unit are shown in Figures 4.52 and 4.53. Photos of the evaporators are
shown in Figures 4.54 and 4.55, the schematic of evaporator #2 is given in Figure 4.56, and Table 4.4 summarizes
the important dimensions of each heat exchanger.
120
Figure 4.52 Diagram of the Evaporator Unit (from Larkin Bulletin RI-04)
Figure 4.53 The Evaporator Unit
Figure 4.54 Photo of the Evaporator #1
121
Figure 4.55 Photo of the Evaporator #2
Figure 4.56 Schematic of the Evaporator #2
4.5.2 Condenser
The condenser is a Tecumseh model #50803-1, manufactured by Blissfield under the model #66001-3
(www.blissfield.com). The condenser is an all-steel tube-and-fin design. A diagram and photo of the unit are shown
in Figures 4.57 and 4.58, while Table 4.4 summarizes the important dimensions of the specific unit.
122
Figure 4.57 The Condenser Unit
Figure 4.58 Schematic of the Condenser (from www.blissfield.com)
123
Table 4.4 Principal Heat Exchanger Parameters
Physical Parameter Evaporator #1 Evaporator #2 Condenser Units

Heat Exchanger Mass 1.4999 2.7438 4.6564 kg
Heat Exchanger Specific Heat 0.6252 0.4877 0.4670 kJ/kg/K
Heat Exchanger Frontal Area 653.2245 583.8440 842.7080 cm2
Hydraulic Diameter of Tubes 7.3279 8.0126 8.1026 mm
Fluid Flow Length 11.8734 11.4579 10.6895 m
Cross Sectional Area for Flow 0.4217 0.5156 0.5156 cm2
Internal Volume 500.7527 590.8063 557.1600 cm3
Internal Surface Area 0.2733 0.2917 0.2750 m2
External Surface Area 2.9090 3.0680 2.7927 m2
4.5.3 Internal Heat Exchanger

The internal heat exchanger is a Superior model HXSV-1/2 (see www.sherwoodvalve.com). The cold
vapor exiting the evaporator passes through the center of the internal heat exchanger, while the hot liquid exiting
from the condenser/receiver flows through a concentric tube surrounding the vapor line. In this way, the cold vapor
is preheated before entering the compressor, while the hot liquid is subcooled before entering the expansion device.
This prevents possible compressor failure, and valve choking, while increasing the overall capacity of the system.
Figure 4.59 Diagram of the Internal Heat Exchanger (from Sherwood Product Catalog)
Figure 4.60 Internal Heat Exchanger
4.5.4 Auxiliary Components

The experimental system has both a liquid line receiver and a suction line accumulator. The presence of
manual bypass valves allows each of these components to be included/excluded as desired. Both components are
manufactured by AC&R Components, Inc. (www.henrytech.com). The receiver is model S-8064 with a calculated
internal volume of 2866.51 cm3. The accumulator is model S-7043 with a calculated internal volume of 773.23 cm3.
124
Figure 4.61 Liquid Line Receiver
Figure 4.62 Suction Line Accumulator
125
An oil separator is placed immediately after the compressor discharge line in an attempt to separate the oil
from the refrigerant and return it to the compressor oil sump. The oil separator is model S-5581 and is also
manufactured by AC&R Components, Inc. (www.henrytech.com).
Figure 4.63 Oil Separator
A Sporlan Catch-All filter/drier (model C-052) is used to remove moisture, acids, and other particulates
that may be in the system (www.sporlan.com). These contaminants can result from oil breakdown, refrigerant
impurities, etc. and may cause problems if not removed. The system also contains several Sporlan See-All moisture
indicators (models SA-12FM, and SA-14S), to verify that the filter is functioning.
Figure 4.64 Diagram of Filter/Drier (from Sporlan Bulletin 40-10)
126
Figure 4.65 Filter /Drier
Figure 4.66 Sight Glass
The system uses manual valves (Mueller Brass Co. models 14838 and 14841, see
www.muellerindustries.com) to isolate various parts of the system. This allows the system to be configured in a
various ways, to emulate many different types of AC&R systems.
Figure 4.67 Manual Valve
127
Finally, the pipe lengths between the principal components can contribute significantly to the system
dynamics. Table 4.5 summarizes the lengths and diameters of the principal pipe sections in the system.
Table 4.5 Principal Pipe Sections
Location Length [m] Diameter [mm]

Compressor to Condenser 4.04 9.5
Condenser to Valve 4.79 9.5
Valve to Evaporator 1.63 9.5
Evaporator to Compressor 6.53 9.5
128
Chapter 5. System Identification and Model Validation
In this chapter, the perennial problem of model validation is addressed. Although this term is essentially
misleading in that it is impossible to truly validate any model, confidence in a model’s fidelity can be gained
through a variety of methods. These methods include visual comparison between model and data, statistical
measures of model residuals, and dynamic uncertainty based metrics.
These methods apply not only to first principles based models, but data-driven models as well. This chapter
attempts first to report the work of data based system identification and validation. This includes time based
identification of the experimental system described in the preceding chapter. Particular attention is given to
evaluating how dynamics change as a function of operating condition. First, some experimental data is presented to
illustrate the general dynamic response of the system and how the dynamics change as a function of operating
condition. Then following the development of empirical models, the first principles models derived in Chapter 2 are
validated. Finally, estimates of model uncertainty are determined as a precursor to the control design process that is
implemented in Chapter 6.
5.1 System Dynamics

This section presents experimental data illustrating the major dynamic responses. The data is taken over
several operating conditions, and clearly shows the changes in dynamic behavior as a function of operating
condition. The two principle actuators of the system are the electronic expansion valve (EEV) and the compressor.
These devices modulate the mass flow rate on one side of the system. The temporarily imbalance is mass flow
entering and exiting the heat exchanger results in changes in operating pressures, and relative lengths of two-phase
and superheat refrigerant regions in the heat exchangers. Changes in evaporator and condenser fan speed also result
in changes in the operating condition, but with significantly slower, and smaller dynamic responses. Therefore, this
section shows the dynamic response for step changes in compressor speed and expansion valve.
The experimental system presented in Chapter 3 is configured to operate with a high side receiver, as
shown in Figures 5.1-5.2. The system operating with both types of evaporator are identified over a range of
operating conditions (test matrices shown in Tables 5.1-5.2). The reader should note that when operating with only
evaporator #1, evaporator #2 is completely isolated from the system. However, when operating with only evaporator
#2, some back flow of refrigerant vapor into evaporator #1 is possible, thus affecting the pressure response slightly.
129
Evaporator Fan
4 1
Evaporator
Expansion Valve Compressor
Pressure
3 2
Filter/Dryer Oil Seperator
4 1
Condenser
3 2
Receiver
Enthalpy
Condenser Fan
Figure 5.1 System Diagram - Subcritical Vapor Figure 5.2 P-h Diagram - Subcritical Vapor Compression
Compression Cycle with High Side Receiver Cycle with High Side Receiver
5.1.1 Evaporator #1
When evaluating evaporator #1, a sequence of step changes in expansion valve opening and compressor
(Figure 5.3) were applied to the system over a wide range of operating conditions (Table 5.1). All relevant system
pressures, temperatures, mass flow rates, etc. were measured. As an example of characteristic system dynamics, the
responses of evaporator superheat, evaporator pressure, and condenser pressure are shown.
Table 5.1
Evaporator Compressor EEV Opening Evaporator Condenser

Fan Speed [%]
Superheat [C] [rpm] [%] Pressure [kPa] Pressure [kPa]
900 12.9 381 885
1200 15.0 350 940
8
1500 18.0 329 1005
Condenser Fan 100% 1800 22.5 323 1069
Evaporator Fan 100% 900 10.5 307 835
1200 12.2 303 911
16
1500 14.2 285 957
1800 14.5 284 1009
900 10.4 315 850
Condenser Fan 75% 1200 10.5 294 902
16
Evaporator Fan 100% 1500 13.9 292 990
1800 14.0 281 1011
900 9.6 273 785
Condenser Fan 100% 1200 9.5 232 777
16
Evaporator Fan 75% 1500 11.5 254 897
1800 12.5 250 934
130
900 9.6 266 801
Condenser Fan 75% 1200 10.2 229 809
16
Evaporator Fan 75% 1500 12.0 239 877
1800 12.2 243 932
Valve Opening Signal Compressor Speed Signal

1.5 200
150
1
100
Compressor Speed [rpm]

0.5
EEV Opening [%]
50
0 0
−50
−0.5
−100
−1
−150
−1.5 −200
0 500 1000 1500 0 500 1000 1500
Time [s] Time [s]
Figure 5.3 System Dynamics: Actuator Step Changes
Examination of the responses to step changes in valve opening and compressor speed reveals some
interesting trends. First, when operating with a nominal superheat of 16 degrees, changes in the resulting gain of the
step response increase as the nominal compressor speed decreases (Figure 5.4). Because the nominal mass flow rate
is lower at lower compressor speed, an equivalent change in valve opening or compressor speed results in a larger
relative change in mass flow rate, and thus a greater change in superheat. This effect becomes less noticeable
around lower superheat values.
A similar trend is still visible in the response of evaporator pressure (Figure 5.5), albeit to a lesser degree
for lower nominal superheat values. The speed of response of evaporator pressure also increases as the nominal
superheat increases, due to the relatively less amount of refrigerant charge stored in the evaporator. The response of
condenser pressure is relatively unaffected (Figure 5.6), although the reader will note that large pressure oscillations
are observed at selected compressor speeds (e.g. 1000 rpm). These are a result of compressor induced pressure
oscillations whose frequency is below the bandwidth of the low pass filters applied to the data before downsampling
the signal to 1 Hz.
131
Evaporator Superheat [C]
Valve Step Changes Compressor Step Changes
8 degrees nominal superheat

8 8
@ 900 RPM
@ 1200 RPM
6 6 @ 1500 RPM
@ 1800 RPM
4 4
2 2
0 0
−2 −2
0 500 1000 1500 0 500 1000 1500
8 8
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 0 500 1000 1500
Time [s] Time [s]
Speed and Evaporator Superheat
Evaporator Pressure [kPa]

20 20
0 0
−20 −20
−40 −40 @ 900 RPM

@ 1200 RPM
@ 1500 RPM
−60 −60 @ 1800 RPM
0 500 1000 1500 0 500 1000 1500
20 20
0 0
−20 −20
−40 −40
−60 −60
0 500 1000 1500 0 500 1000 1500

Time [s] Time [s]
Figure 5.5 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Compressor Speed
and Evaporator Superheat
132
Condenser Pressure [kPa]

20 20
0 0
−20 −20
−40 −40 @ 900 RPM

@ 1200 RPM
@ 1500 RPM
−60 −60 @ 1800 RPM
0 500 1000 1500 0 500 1000 1500

20 20
0 0
−20 −20
−40 −40
−60 −60
0 500 1000 1500 0 500 1000 1500

Time [s] Time [s]
Figure 5.6 System Dynamics: Condenser Pressure Response for Different Nominal Values of Compressor Speed
and Evaporator Superheat
Comparisons of the dynamics can also be made for changes in nominal air flow rates. The responses shown
are for a compressor speed of 1200 rpm and a nominal superheat of 16º C. Decreasing the evaporator air flow rate
accentuates the superheat response (Figure 5.7) and evaporator pressure response (Figure 5.8) for valve changes.
However, decreasing the condenser air flow rate results in negligible differences in the dynamic responses. In either
case, the differences in responses for compressor changes is negligible, as is the condenser pressure response for
either input (Figure 5.9).
8 8
Nominal Air Flow
75% Evaporator Air
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 0 500 1000 1500
8 8
Nominal Air Flow
75% Condenser Air
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 0 500 1000 1500
Time [s] Time [s]
Figure 5.7 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Air Flow Rate
133

20 20 Nominal Air Flow
75% Evaporator Air
0 0
−20 −20
−40 −40
−60 −60
0 500 1000 1500 0 500 1000 1500

20 20 Nominal Air Flow

75% Condenser Air
0 0
−20 −20
−40 −40
−60 −60
0 500 1000 1500 0 500 1000 1500

Time [s] Time [s]
Figure 5.8 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Air Flow Rates

20 20
0 0
−20 −20
−40 −40
Nominal Air Flow
−60 −60 75% Evaporator Air
0 500 1000 1500 0 500 1000 1500

20 20
0 0
−20 −20
−40 −40
Nominal Air Flow
−60 −60 75% Condenser Air
0 500 1000 1500 0 500 1000 1500

Time [s] Time [s]
Figure 5.9 System Dynamics: Condenser Pressure Response for Different Nominal Values of Air Flow Rates
5.1.2 Evaporator #2
When evaluating evaporator #2, a similar sequence of step changes in expansion valve opening and
compressor (Figure 5.10) were applied to the system over a range of operating conditions (Table 5.2) similar to
those for evaporator #1. Again the responses of evaporator superheat, evaporator pressure, and condenser pressure
are shown, in addition to the response of evaporator exit air temperature.
Recall from Chapter 4 that evaporator #2 is similar to evaporator #1 in every respect except the tube
configuration. Evaporator #1 is configured so that the fluid flows back and forth down the front of the evaporator,
134
and then back and forth up the back. In this way the evaporator acts virtually as two heat exchangers in series, where
the air is cooled passing through the first, and then cooled further as it flows across the second. Evaporator #2 has a
simpler tube configuration where the fluid passes back and forth as it flows from top to bottom. This single-slab
configuration is more representative of the modeling assumptions.
Table 5.2
Evaporator Compressor EEV Opening Evaporator Condenser

Fan Speed [%]
Superheat [C] [rpm] [%] Pressure [kPa] Pressure [kPa]
900 11.4 362 876
1200 13.1 334 931
10
1500 14.7 307 970
Condenser Fan 100% 1800 16.0 288 978
Evaporator Fan 100% 900 10.4 321 831
1200 12.3 311 897
16
1500 13.4 279 920
1800 14.8 272 966
Condenser Fan 75% 10 14.3 308 1004
1500
Evaporator Fan 100% 16 13.1 292 976
Condenser Fan 100% 10 12.2 302 875
1200
Evaporator Fan 75% 16 11.3 280 855
Condenser Fan 75% 10 12.5 336 976
1200
Evaporator Fan 75% 16 11.0 283 885
Valve Opening Signal Compressor Speed Signal

1.5 200
150
1
100
0.5
EEV Opening [%]
50
0 0
−50
−0.5
−100
−1
−150
−1.5 −200
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.10 System Dynamics: Actuator Step Changes
Examination of the responses of evaporator #2 to step changes in valve opening and compressor speed
reveals slight differences to those of evaporator #1. First, the response of evaporator superheat is more consistent
135
across a range of compressor speeds, and from low to high nominal superheat values, although the same trends are
observed (Figure 5.11). Second, the response of evaporator outlet air temperature also is somewhat consistent
across operating conditions (Figure 5.12). The evaporator pressure response again demonstrates a clear trend of
decreased gain with increasing compressor speed, and increased speed of response with higher values of nominal
superheat (Figure 5.13), while the condenser pressure remains relatively unaffected (Figure 5.14).
6 6
@ 900 RPM
@ 1200 RPM
4 4 @ 1500 RPM
@ 1800 RPM
2 2
0 0
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Evaporator Air Outlet Temperature [C]

2 2
@ 900 RPM
1.5 1.5 @ 1200 RPM
@ 1500 RPM
1 1 @ 1800 RPM
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
2 2
1.5 1.5
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Compressor Speed and Evaporator Superheat
136

30 30
@ 900 RPM
20 20 @ 1200 RPM
10 10 @ 1500 RPM
@ 1800 RPM
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]

30 30
20 20
10 10
0 0
−10 −10
−20 −20
@ 900 RPM
−30 −30 @ 1200 RPM
−40 −40 @ 1500 RPM
@ 1800 RPM
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Comparisons of the dynamics can also be made for changes in nominal air flow rates. The responses shown
are for a compressor speed of 1200 rpm and nominal superheat values of 10º C and 16º C. Decreasing the
evaporator air flow rate accentuates the superheat (Figure 5.15), evaporator exit air temperature (Figure 5.16),
evaporator pressure responses (Figure 5.17), and condenser pressure (Figure 5.18) slightly, with greater differences
observed at higher nominal values of superheat.
137

6 6
4 4
2 2
0 0
Nominal Air Flow
80% Evaporator Air
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.15 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Evaporator Air
Flow Rate

2 2
Nominal Air Flow
1.5 1.5 80% Evaporator Air
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
2 2
1.5 1.5
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Evaporator Air Flow Rate
138

30 30
Nominal Air Flow
20 20 80% Evaporator Air
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.17 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Evaporator Air
Flow Rate

30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40 Nominal Air Flow
80% Evaporator Air
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.18 System Dynamics: Condenser Pressure Response for Different Nominal Values of Evaporator Air
Flow Rate
Responses for reduced condenser air flow rate are shown for a nominal compressor speed of 1500 rpm and
nominal superheat values of 10º C and 16º C. Although there appears to be slight differences in dynamic response
(Figures 5.19-5.22), these are believed to be partly due to frosting effects that were observed for high compressor
speeds and high superheat.
139
10 degrees nominal superheat 6 6
Nominal Air Flow
75% Condenser Air
4 4
2 2
0 0
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
6 6
4 4
2 2
0 0
−2 −2
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.19 System Dynamics: Evaporator Superheat Response for Different Nominal Values of Condenser Air
Flow Rate

2 2
Nominal Air Flow
1.5 1.5 75% Condenser Air
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
2 2
1.5 1.5
1 1
0.5 0.5
0 0
−0.5 −0.5
−1 −1
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Condenser Air Flow Rate
140

30 30
Nominal Air Flow
20 20 75% Condenser Air
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.21 System Dynamics: Evaporator Pressure Response for Different Nominal Values of Condenser Air
Flow Rate

30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40 Nominal Air Flow
75% Condenser Air
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
30 30
20 20
10 10
0 0
−10 −10
−20 −20
−30 −30
−40 −40
−50 −50
0 500 1000 1500 2000 0 500 1000 1500 2000
Time [s] Time [s]
Figure 5.22 System Dynamics: Condenser Pressure Response for Different Nominal Values of Condenser Air
Flow Rate
In summary, the most significant changes is dynamic response was observed for different compressor speed
and superheat settings rather than partial reduction in air flow rates. The dynamic differences appeared to be largely
a change in the steady state gain and speed of the response, while the qualitative behavior remained largely the
141
same. The dynamic differences are significant, and the eventual control strategy will need to account for these
changes in order to achieve high performance over a wide range of operating conditions.
5.2 System Identification

In this section, linear time-invariant (LTI) data-driven models are constructed to predict the dynamic
response of the vapor compression system at different operating conditions. The resulting models confirm that the
differences between local linear dynamic models of the system at different conditions are significant.
5.2.1 Background and Identification Approach

The field of data-driven model development is extensive. A standard textbook regarding the process of
constructing linear time invariant (LTI) dynamic models is given in [60]. The process consists of applying a
dynamic signal to the system inputs, measuring the dynamic response at the system outputs, and then determining a
set of model parameters that result in minimal prediction error. Inherent in this procedure are the assumptions that
1) the input signal “excites” the system dynamics of interest (persistence of excitation), and 2) the structure of the
empirical model correctly represents the underlying physical dynamics.
Frequency based system identification consists of applying a frequency dense excitation signal to the
experimental system, and then attempting to identify the frequency content of the resulting system outputs. An
intuitive choice of input signal is a chirp signal or a sine-sweep. These signals apply a sinusoid with a specific
frequency and amplitude, and then measure the amplitude and phase lag of the resulting sinusoid measured at the
system output.
Time based system identification can be approached using a variety of methods. In simplest form, the
identification procedure seeks to find a set of model parameters that minimize the difference between model and
measured data (residuals) in a least squares sense. More recent methods employ subspace methods for identifying
empirical models, and offer some advantages when identifying multi-input multi-output (MIMO) systems.
For identification purposes, it is necessary to excite the system by varying each of the inputs. In this case,
the data was generated by varying each of the system inputs separately, and simultaneously using a Pseudo-Random
Binary Sequence (PRBS). Although not ideal for nonlinear identification, this signal is persistently exciting of
adequate order, and is sufficient for identifying approximate linear models. The choice of input amplitudes was
approximately 10% of the actuator range, generally resulting in a 5-10% change in the outputs, which was sufficient
to discern the dynamic behavior of the system.
Individual input-output models could be constructed offline assuming an ARMAX model structure,
A(q ) y (t ) = B(q )u (t ) + C (q )e(t ) . The coefficients of the A(q ) , B(q ) , and C (q ) polynomials are found
using a standard nonlinear least squares iterative search method (Gauss-Newton) [25], that minimizes a quadratic
prediction error criterion, V (θ , Z ) as defined in Equation 5.1, while imposing constraints to ensure that only
models with stable predictors are used [60]. This is a common approach for modeling air conditioning systems (e.g.
[17] and [73]).
1 N 1 2
V (θ , Z ) = ∑ ε (t,θ ) ( 5.1 )
N t =1 2
142
Applying this approach to vapor compression systems reveals that the salient dynamics of each input-
output pair can be captured by low order SISO models [81]. In [81] the minimum model order necessary to
adequately model the dynamics, while ensuring whiteness and independence of the model residuals, was 3rd order or
lower. The data sets were divided into estimation and validation sets and the models were cross-validated to ensure
that the models were not over-fitted to a specific data set, and repeated for several operating conditions. For a few of
the input/output combinations the resulting models were 2nd order, while others were observed to be unaffected by
changes in the inputs.
Given the results of the SISO identification, it would be tempting for someone unfamiliar with system
identification to assume that only a 3rd order physical model is needed. However, the states of the various SISO
models need not be the same, and therefore a MIMO model would be a more appropriate description of the system.
Indeed, from our first principles modeling approach, a 6th order MIMO model is expected to be necessary for
capturing the essential dynamic behavior. Although to achieve such a model the individual SISO models could be
combined, scaled, and then reduced using a balanced realization/truncation approach, a more direct approach for
constructing a MIMO model (subspace algorithms) is preferred.
A state-space model structure is selected of the form of Equation 5.2, where Ts is the sampling time,
{A, B, C , D, K } are constant matrices, and x(t ) , u(t ) , y(t ) , and e(t ) are the time sequences of states, inputs,
outputs, and model residuals respectively. This model structure differs notably from those in previous studies that
identified air conditioning systems using multivariable ARX models [73]. The particular advantage of the model
structure in Equation 5.2 is its close relationship to physically-based continuous time state space models. Other
multivariable methods (i.e multivariable ARX, vector difference equation, etc.) are less easily related to their first
principles counterparts [64].
x(t + Ts ) = Ax(t ) + Bu(t ) + Ke(t )

( 5.2 )
y (t ) = Cx(t ) + Du(t ) + e(t )
The algorithm used to identify the {A, B, C , D, K } matrices was a combined prediction error method and
subspace algorithm. The initial guess values of these matrices are determined by an N4SID algorithm [106].
Contrary to the classical identification methods which determined the system matrices first and then the system
states, the N4SID subspace method identifies the state vector first, and then determines the system matrices using a
linear least squares approach. As discussed in [64],[107] the general steps of subspace algorithms are: 1) construct
the extended observability matrix from input-output data, 2) select appropriate weighting matrices (possibilities are
listed in [107]) and perform a singular value decomposition (SVD), 3) determine the state vector from the SVD, 4)
determine the {A, B, C, D} matrices using a linear least squares approach, and 5) determine the {K } matrix from
{A, B, C, D} and the covariance of the residuals. For this application, an initial guess for the system matrices is
obtained using an N4SID algorithm, and the model is then adjusted by improving the prediction error fit using an
approach similar to the SISO algorithm. As discussed in [64], the initial input/output values are assumed to be close
143
to the equilibrium values and are unnecessary for identification of a linear model. Therefore, the sample means are
removed prior to the offline identification and the results are plotted with a zero mean value.
Empirical models are constructed for the test conditions given in Tables 5.1-5.2. Prior to identification, the
inputs and outputs are scaled such that the magnitude of input and output variations are approximately equal. Thus
the resulting units are: expansion valve opening in [%], compressor speed in [100 rpm], temperatures [ºC], and
pressures in [10 kPa]. As a demonstration of the model fit three basic statistics about the model residuals are
presented. The model residuals, ε , are defined as ε (t ) = yˆ (t ) − y (t ) with ŷ as the predicted output and y as the
measured output. The maximum residual and the average residual are calculated as given in Equation 5.3 and
Equation 5.4 with N being the number of measurements. Additionally, a relative error measure is calculated as
shown in Equation 5.5, where y is the mean value of y . The percentage of model fit can then be calculated as:
% Fit = 100(1 − S r ) .
S1 = max ε (t ) ( 5.3 )
1⎛ N 2 ⎞
S2 = ⎜ ∑ ε (t )⎟ ( 5.4 )
N ⎝ t =1 ⎠
⎛ N ⎞ ⎛ N ⎞ yˆ − y
Sr = ⎜ ∑ ε 2 (t ) ⎟ ⎜ ∑ ( y(t ) − y ) ⎟=
2 2
( 5.5 )
⎜ ⎟ ⎜ ⎟ y−y
⎝ t =1 ⎠ ⎝ t =1 ⎠ 2
As an example of the identification process detailed results are given for one of the operating conditions
listed (Evaporator #2, nominal air flow, 10º C superheat, 1200 rpm). Models of order 1-10 were estimated and cross-
validated. Models of order 1-3 yielded poor prediction, while models of order 8-10 were overfitted to the particular
data set used for estimation. Models of order 4-7 had roughly similar predictive capabilities, but analysis of the
auto- and cross-correlation of the errors revealed that models of order 4 and 7 were unacceptable. Finally, models of
order 5 and 6 yielded similar predictive capabilities and correlation errors for the temperature outputs, while the 6th
order model resulted in lower correlation errors for the pressure outputs. Based on these results and the expectation
from the first principles modeling results a 6th order model is selected.
The excitation inputs include a section of PRBS and a section of standard step responses (Figure 5.23). As
mentioned, the input amplitudes were approximately 10% of the actuator range, generally resulting in a 5-10%
change in the output. The model’s predictive capabilities are shown in Figure 5.24, and the statistical measures of
model fit are tabulated in Table 5.3. To evaluate the uncertainty of the estimated model, step responses are
presented in Figure 5.25 with a confidence interval for 3 standard deviations shown. Finally, a Bode plot of the
model is given in Figures 5.26-5.27, while the discrete time version of the model is given in Equations 5.6-5.9
assuming a sampling frequency of 1 Hz.
144
Table 5.3 Statistical Model Fits of the 6th Order Model
Max Error RMS Error % Model Fit

Evaporator Superheat [C] 1.41 0.62 60.3
Evaporator Exit Air Temp. [C] 0.78 0.23 25.0
Evaporator Pressure [10 kPa] 0.99 0.30 71.9
Condenser Pressure [10 kPa] 3.21 0.63 31.4
1
Valve Opening
Expansion
0.5
[%]
−0.5
−1
0 1000 2000 3000 4000 5000 6000 7000
2
Compressor
[100 rpm]
1
Speed
−1
−2
0 1000 2000 3000 4000 5000 6000 7000
Figure 5.23 Identification Inputs
5
Measured Output
Evaporator
Superheat
6th Order Model

[C]
−5
0 1000 2000 3000 4000 5000 6000 7000
2
Exit Air Temp.
Evaporator
[C]
−2
0 1000 2000 3000 4000 5000 6000 7000
5
Evaporator
Pressure
[10 kPa]
−5
0 1000 2000 3000 4000 5000 6000 7000
5
Condenser
Pressure
[10 kPa]
−5
0 1000 2000 3000 4000 5000 6000 7000
Figure 5.24 Identification Results: Model Prediction
145
Step from Valve Opening [1%] Step from Compressor Speed [100 rpm]
Evaporator
0 4
Superheat
[C]
2
−5
0
−10
Outlet Air Temp. −400 −200 0 200 400 600 800 1000 1200 −400 −200 0 200 400 600 800 1000 1200
1 0.5
Evaporator
0
[C]
0
−1
−2 −0.5
−400 −200 0 200 400 600 800 1000 1200 −400 −200 0 200 400 600 800 1000 1200
Evaporator
4 0
Pressure
[10 kPa]
2 −2
0 −4
−400 −200 0 200 400 600 800 1000 1200 −400 −200 0 200 400 600 800 1000 1200
2
Condenser
4
Pressure
[10 kPa]
1
2
0
0
−1
−400 −200 0 200 400 600 800 1000 1200 −400 −200 0 200 400 600 800 1000 1200
Figure 5.25 Identification Results: Step Responses within 3 Standard Deviations
Bode Diagram
From: Valve Opening From: Compressor Speed
50
To: Evaporator Superheat
−50
−100
180
Magnitude (dB) ; Phase (deg)
−180
To: Evaporator Outlet Air Temp.
40
20
−20
−40
−60
360
180
0
−4 −3 −2 −1 0 −4
1 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 5.26 Identification Results: Bode Plot - Temperatures
146
Bode Diagram
20
To: Evaporator Pressure

0
−20
−40
−60
180
0
−180
−360
20
To: Condenser Pressure
−20
−40
−60
360
180
−180
−4 −3 −2 −1 0 −4
1 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 5.27 Identification Results: Bode Plot - Pressures
⎡- 0.3042 0.2171 0 0 ⎤0 0
⎢- 0.2171 - 0.3042 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0.9949 0.0020 0 0 ⎥
Aid = ⎢ ⎥ ( 5.6 )
⎢ 0 0 - 0.0020 0.9949 0 0 ⎥
⎢ 0 0 0 0 0.9722 0 ⎥
⎢ ⎥
⎣⎢ 0 0 0 0 0 0.9578⎦⎥
⎡ 7.0115 8.9233⎤
⎢ 0.6525 6.7280⎥
⎢ ⎥
⎢ 0.8732 - 0.2192⎥ −3
Bid = ⎢ ⎥ ⋅ 10 ( 5.7 )
⎢ - 1.1584 0.5562 ⎥
⎢ 1.2409 - 1.6651⎥
⎢ ⎥
⎢⎣- 0.2009 - 1.0323⎥⎦
⎡ 1.8560 0.1673 8.1508 29.8921 - 5.5381 - 2.2777 ⎤
⎢- 0.0362 0.0695 1.8964 3.7616 - 0.8217 8.5067 ⎥⎥
C id = ⎢ ( 5.8 )
⎢- 2.9092 0.0378 5.2696 - 7.7861 15.8280 7.0390 ⎥
⎢ ⎥
⎣- 0.3648 12.7381 20.3440 - 11.4359 - 25.3879 - 8.1621 ⎦
Did = [0 4 x 2 ] ( 5.9 )
147
5.2.2 System Identification Results: Evaporator #1
The system identification approach presented in the previous section was applied to construct empirical
dynamic models of the system across a wide range of operating conditions. This section presents the results of the
identification for the system operating with evaporator #1. The statistical measures of model fit are presented in
Tables 5.4-5.6. The results confirm many of the conclusions drawn from visual evaluation of the step responses.
First, the models confirm that the dynamics change significantly as compressor speed increases (Figures 5.28-5.29),
with the differences being accentuated at high nominal levels of superheat (Figures 5.30-5.31). Finally, that reducing
the air flow rates does affect the dynamics at low compressor speeds (Figure 5.32), but has relatively little effect at
high compressor speeds (Figure 5.33).
Table 5.4 Model Fit of the Identified Models
Model Fit
Evaporator
Compressor
Fan Speed [%] Superheat Evaporator Evaporator Condenser Comments
[rpm]
[C] superheat Pressure Pressure
[C] [10 kPa] [10 kPa]
900 74.0 85.2 46.8
1200 61.3 75.7 56.2
8 1500 44.4 60.8 41.9
1800 33.6 32.1 2.6 Poor initial conditions; adequate
Condenser Fan 100%
prediction otherwise
Evaporator Fan 100%
900 79.2 81.9 56.1
1200 75.0 78.2 63.6
16
1500 72.7 75.0 58.3
1800 69.0 70.9 64.1
900 83.5 83.5 65.7
Condenser Fan 75% 1200 80.5 80.8 74.5
16
Evaporator Fan 100% 1500 73.5 77.4 66.7
1800 64.2 72.4 67.5
900 78.8 85.0 61.6
Condenser Fan 100% 1200 66.8 72.5 32.5
16
Evaporator Fan 75% 1500 81.7 82.0 70.7
1800 75.2 76.1 65.2
900 76.1 79.9 66.2
Condenser Fan 75% 1200 60.8 67.9 29.1
16
Evaporator Fan 75% 1500 59.6 84.5 77.1
1800 73.2 77.1 57.2
148
Table 5.5 Max Error of the Identified Models
Max Error
Evaporator
Compressor
[rpm]
[C] [10 kPa] [10 kPa]
900 1.40 0.99 4.57
1200 1.59 1.13 2.03
8 1500 5.18 5.10 7.67
Condenser Fan 100% prediction otherwise
Evaporator Fan 100%
900 1.26 1.24 7.63
1200 1.86 1.94 1.25
16
1500 3.13 3.19 1.55
1800 3.22 3.79 4.80
900 0.93 1.10 3.42
Condenser Fan 75% 1200 0.96 0.87 1.23
16
Evaporator Fan 100% 1500 3.08 3.44 3.55
1800 3.60 4.09 3.66
900 1.83 1.31 3.15
Condenser Fan 100% 1200 1.32 1.02 2.83
16
Evaporator Fan 75% 1500 2.50 2.47 1.43
1800 3.28 3.24 2.30
900 2.87 2.29 2.97
Condenser Fan 75% 1200 2.40 1.11 3.40
16
Evaporator Fan 75% 1500 1.20 0.45 1.60
1800 2.48 2.92 4.36
Table 5.6 RMS Error of the Identified Models
RMS Error
Evaporator
Compressor
[rpm]
[C] [10 kPa] [10 kPa]
900 0.36 0.20 0.56
1200 0.49 0.23 0.49
8 1500 0.87 0.24 0.40
Condenser Fan 100% prediction otherwise
Evaporator Fan 100%
900 0.41 0.33 0.55
1200 0.27 0.26 0.34
16
1500 0.22 0.22 0.42
1800 0.22 0.22 0.38
900 0.32 0.30 0.43
Condenser Fan 75% 1200 0.25 0.24 0.27
16
Evaporator Fan 100% 1500 0.21 0.21 0.37
1800 0.28 0.23 0.39
900 0.60 0.33 0.47
Condenser Fan 100% 1200 0.58 0.35 1.14
16
Evaporator Fan 75% 1500 0.16 0.16 0.32
1800 0.20 0.21 0.36
900 0.81 0.49 0.52
Condenser Fan 75% 1200 0.81 0.48 1.45
16
Evaporator Fan 75% 1500 0.42 0.15 0.27
1800 0.21 0.18 0.40
149
30
20

10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
@ 900 RPM
20
@ 1200 RPM
10
@ 1500 RPM
0 @ 1800 RPM
−10
−20
−30
−40
−50
−6 −5 −4 −3 −2 −1 0 1 −6 −5 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
Superheat)

30
20
10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
@ 900 RPM
20
@ 1200 RPM
10
@ 1500 RPM
0 @ 1800 RPM
−10
−20
−30
−40
−50
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Superheat)
150
30
20

10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
@ 8 degrees Superheat
20
10
0
−10
−20
−30
−40
−50
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Compressor Speed)

30
20
10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Compressor Speed)
151
30
20

10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
Nominal Air Flow
20
75% Condenser Air Flow

10
75% Evaporator Air Flow
0 75% Condenser/Evaporator Air Flow
−10
−20
−30
−40
−50
−6 −4 −2 0 −6 −4 −2 0
10 10 10 10 10 10 10 10
Superheat)

30
20
10
0
−10
−20
−30
−40
−50
30
20
10
0
−10
−20
−30
−40
−50
30
Nominal Air Flow
20
75% Condenser Air Flow

10
0 75% Condenser/Evaporator Air Flow
−10
−20
−30
−40
−50
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Superheat)
These differences in system dynamics are significant. Assuming a nominal model (low superheat,
moderate compressor speed, nominal air flow rates), and depending on the input-output considered, the difference in
152
dynamics identified for other operating conditions is equivalent to a multiplicative uncertainty of 100-500% at low
frequencies and 1000% or greater at high frequencies. Attempting to accommodate such drastic differences by
synthesizing robust controllers would lead to extremely conservative control strategies. However, the change in
system dynamics due to operating condition can be accommodated using the gain-scheduled control techniques
presented in the following chapter.
Essential to this approach is identifying the parameters that most effectively capture the system
nonlinearities. From the step responses and the frequency responses of the identified models, it is observed that the
compressor speed has the most significant effect on changing the system dynamics. However, this may be an
indirect cause since the nominal value of superheat and reducing the air flow rates at low compressor speeds also
affect the system dynamics. Current research regarding the parametric sensitivity of the first principles models
indicate that the operating pressures significantly affect the system dynamics [51], and these would be the most
direct way to capture the system nonlinearities. However, for most applications with constant air flow rates, and
reasonable regulation of evaporator superheat, a reasonable approximation would be to use compressor speed to
schedule the control strategies.
5.2.3 System Identification Results: Evaporator #2

The system identification results for the system operating with evaporator #2 are similar to those discussed
previously. The only exception is that frosting was observed for high compressor speeds, and high nominal
superheat values or reduced air flow rates, resulting in less repeatable results and making the system identification
process more difficult and thus less reliable. The statistical measures of model fit are presented in Tables 5.7-5.9.
The frequency responses are presented here for completeness without further discussion.
Table 5.7 Model Fit of the Identified Models
Model Fit
Evaporator Compressor
Fan Speed [%] Evaporator Condenser
Superheat [C] [rpm] Evaporator Evaporator Exit
Pressure Pressure
Superheat [C] Air Temp. [C]
[10 kPa] [10 kPa]
900 68.2 40.9 76.7 31.0
1200 65.3 46.8 78.5 58.4
10
1500 56.2 26.9 74.3 65.7
Condenser Fan 100% 1800 51.5 42.5 60.9 33.6
Evaporator Fan 100% 900 62.2 35.0 64.9 30.9
1200 59.9 22.0 80.1 31.4
16
1500 52.9 7.8 70.8 54.9
1800 54.0 27.8 64.7 61.6
Condenser Fan 75% 10 49.0 39.1 58.2 47.8
1500
Evaporator Fan 100% 16 73.8 45.8 76.4 65.4
Condenser Fan 100% 10 61.3 17.4 79.0 49.3
1200
Evaporator Fan 75% 16 75.1 48.2 78.8 53.4
Condenser Fan 75% 10 53.2 45.5 74.8 43.6
1200
Evaporator Fan 75% 16 65.5 38.0 74.9 48.6
153
Table 5.8 Max Error of the Identified Models
Max Error
Pressure Pressure
[10 kPa] [10 kPa]
900 1.87 0.56 1.16 3.83
1200 2.17 0.49 0.74 2.48
10
1500 2.45 0.61 0.77 1.32
Condenser Fan 100% 1800 3.21 0.54 0.72 1.96
Evaporator Fan 100% 900 1.63 0.56 1.32 3.41
1200 1.73 0.81 0.80 4.19
16
1500 1.88 0.71 0.87 2.33
1800 1.83 0.61 0.74 1.08
Condenser Fan 75% 10 2.62 0.57 1.11 2.52
1500
Evaporator Fan 100% 16 1.07 0.41 0.78 1.46
Condenser Fan 100% 10 2.75 0.83 0.77 1.68
1200
Evaporator Fan 75% 16 1.28 0.63 0.75 1.85
Condenser Fan 75% 10 2.99 0.85 1.10 3.84
1200
Evaporator Fan 75% 16 1.35 0.95 0.86 2.35
Table 5.9 RMS Error of the Identified Models
RMS Error
Pressure Pressure
[10 kPa] [10 kPa]
900 0.57 0.17 0.31 0.76
1200 0.63 0.14 0.23 0.39
10
1500 0.57 0.19 0.17 0.27
Condenser Fan 100% 1800 0.68 0.12 0.18 0.42
Evaporator Fan 100% 900 0.52 0.20 0.46 0.80
1200 0.59 0.33 0.20 0.52
16
1500 0.50 0.22 0.19 0.42
1800 0.49 0.20 0.21 0.28
Condenser Fan 75% 10 0.79 0.17 0.30 0.66
1500
Evaporator Fan 100% 16 0.29 0.12 0.17 0.30
Condenser Fan 100% 10 0.75 0.30 0.20 0.68
1200
Evaporator Fan 75% 16 0.38 0.16 0.22 0.53
Condenser Fan 75% 10 1.03 0.16 0.28 0.57
1200
Evaporator Fan 75% 16 0.48 0.39 0.29 0.84
154
Bode Diagram
20
To: Evap Superheat

0
−20
−40
To: Evap Exit Air Temp

20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
@ 900 RPM
20
@ 1200 RPM
To: Cond Pressure
@ 1500 RPM
0 @ 1800 RPM
−20
−40
−6 −5 −4 −3 −2 −1 0 −6
1 −5 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Superheat)
Bode Diagram
20
To: Evap Superheat
−20
−40
20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
20 @ 900 RPM
@ 1200 RPM
To: Cond Pressure
@ 1500 RPM
0 @ 1800 RPM
−20
−40
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Superheat)
155
Bode Diagram
20
To: Evap Superheat

0
−20
−40

20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
20 @ 8 degrees Superheat
To: Cond Pressure
−20
−40
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Compressor Speed)
Bode Diagram
20
To: Evap Superheat
−20
−40
20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
20 @ 8 degrees Superheat
To: Cond Pressure
−20
−40
−4 −3 −2 −1 0 1 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Compressor Speed)
156
Bode Diagram
20
To: Evap Superheat

0
−20
−40

20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
Nominal Air Flow

20
To: Cond Pressure
75% Condenser/Evaporator Air Flow

0
−20
−40
−6 −5 −4 −3 −2 −1 0 −6
1 −5 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Superheat)
Bode Diagram
20
To: Evap Superheat
−20
−40
20
−20
Magnitude (dB)
−40
20
To: Evap Pressure
−20
−40
20 Nominal Air Flow

To: Cond Pressure
75% Condenser/Evaporator Air Flow

0
−20
−40
−6 −5 −4 −3 −2 −1 0 −6
1 −5 −4 −3 −2 −1 0 1
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec)
Superheat)
5.3 First Principles Model Validation

This section presents comparison of the first principles models with experimental data. As possible, the
component models are discussed and validated individually, and then the validation of the overall system model is
157
presented. Particular attentions is given to discussing the methods to for specifying the operating condition, physical
parameters, and lumped parameters required by the individual models. Comparison of the nonlinear and linearized
models is given.
Valve Opening Compressor Speed

16 2000
15.5
1900
15
14.5 1800
[rpm]
[%]
14
13.5 1700
13
1600
12.5
12 1500
1000 1500 2000 2500 1000 1500 2000 2500
Time [s] Time [s]
Condenser Fan Evaporator Fan

110 110
105 105
100 100
[%]
[%]
95 95
90 90
85 85
80 80
1000 1500 2000 2500 1000 1500 2000 2500
Time [s] Time [s]
Figure 5.40 System Model Validation: Actuator Step Changes
The validation data set shown in this section is obtained by applying step changes in valve opening,
compressor speed, and air flow rates, and recording the component and system responses. The actuator changes are
shown in Figure 5.40. The nominal operating condition is selected at operating pressures of 270 and 970 kPa, with a
compressor speed of 1500 rpm and 100% air flow rates across the evaporator and condenser. For individual
component validation, the measured values of pressure, temperature, and mass flow rate are input to the individual
component models, and the component outputs are then compared to the experimental data. For the system
validation, the individual component models are linked to form the overall system, with pressures, temperatures, and
mass flow rates all being predicted by the component models; only the step changes in compressor speed, valve
opening, and air flow rates are applied as inputs to the overall system.
5.3.1 Expansion Valve

The expansion valve model requires a static functional relationship between the inlet pressure, outlet
pressure, and inlet temperature, and the resulting mass flow rate through the valve. In this case, this relationship is
given in terms of an empirical map (see Chapter 4). Because the heat exchanger models are extremely sensitive to
the entering and exiting mass flow rates, a highly accurate empirical valve map is essential.
158
−3
x 10 Mass Flow Rate Prediction
Data (Valve)
Nonlinear (Valve)
8.5
8
[kg/s]
7.5
6.5
6
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.41 Component Model Validation: Expansion Valve Mass Flow Rate
Because the fluid exiting the valve is two-phase, it is impractical to validate the isenthalpic assumption.
However, Figure 5.41 gives the mass flow rate prediction of the empirical model compared to recorded data. There
is excellent agreement between model data for the changes in compressor speed and air flow rates. This agreement
reflects the accuracy of valve map in terms of the dependence on the pressure differential. There is a slight mismatch
in prediction for the step change in valve opening, indicating an opportunity for refinement of the valve map in
terms of its dependence on valve position.
5.3.2 Compressor
Similar to the expansion valve model, the compressor requires a static functional relationship between the
inlet pressure, outlet pressure, and inlet temperature, and the resulting mass flow rate. Again, this relationship is
given in terms of an empirical map (see Chapter 4), that needs to be highly accurate. Additionally, the prediction of
the outlet enthalpy is given in terms of isentropic efficiency. If necessary, a first order time constant can be
implemented to simulate the thermal capacity of the compressor itself.
159
−3
Data (Valve)
Nonlinear (Compressor)
8.5
8
[kg/s]
7.5
6.5
6
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.42 Component Model Validation: Compressor Mass Flow Rate
Figure 5.42 gives the mass flow rate prediction of the empirical model compared to data measured at the
expansion valve. The lack of a mass flow sensor at the compressor is a notable gap in the available measurements.
The challenge of measuring vapor flow with rapid and significant changes in pressures and temperatures is
recognized by industry and academia alike. Ongoing efforts include a search for an appropriate solution to this
difficulty. However, the model and data do match after the transient responses have mostly concluded, suggesting
that the map is accurate. However, there is a significant spike in mass flow rate associated with sudden changes in
compressor speed. The size and shape of this transient response depends on the rate limit of the compressor speed
change. However, because the heat exchanger models are sensitive to differences in mass flow rate this unknown
transient could result in significant changes the pressure and temperature responses.
5.3.3 Evaporator
Unlike the expansion valve and compressor models, which are relatively simple, and largely semi-
empirical, the heat exchanger models are first principles based, and require knowledge of the physical parameters
and operating condition. Because of the simplifying model assumptions, the model relies extensively on lumped
parameters. Several of these are known with certainty (e.g. heat exchanger mass), whereas others are only known
within an order of magnitude (e.g. lumped two-phase heat transfer coefficient). To obtain a satisfactory model fit,
several of the parameters are adjusted within acceptable ranges. A discussion of each of the model parameters and
the tuning procedure follows.
160
5.3.3.1 Physical Parameters
Hydraulic Diameter – This value is used only for calculating heat transfer coefficients. As such, its value
has minimal impact on the dynamic model.
Fluid Flow Length – This value is defined as the length that the fluid travels from the entrance to the exit of
the heat exchanger. All possible fluid flow paths are assumed to have the same length. Many heat exchangers use a
series of tubes or plates arranged in a serpentine manner for fluid flow. Often these tubes or plates will join at a
“header” and the fluid is redistributed before entering the next “pass” or series of or tubes or plates. The user is
required to make appropriate assumptions or simplifications to calculate an effective fluid flow length if necessary.
Cross-sectional Area – This value along with the effective fluid flow length determines the effective
internal volume of the heat exchanger. Clearly, this could be calculated using the hydraulic diameter. However, for
most heat exchangers the cross-sectional area is not constant. The number of tubes or plates per pass generally
increases as the fluid evaporates or decreases as the fluid condensates, thus changing the cross-sectional area. For
the purposes of modeling a constant cross-sectional area is assumed. Furthermore, this parameter most directly
affects the dynamic response of evaporator pressure. This is intuitive, since a larger internal volume will result in a
smaller and slower buildup of pressure. On an actual system, the buildup of pressure occurs not only in heat
exchanger itself, but also in the headers, entering and exiting pipe lengths, etc. Thus to match the dynamic response
of evaporator pressure the effective cross sectional area is increased to include the internal volume of these
additional components (Note: this may result in a value of cross-sectional area 2 or 3 times larger than the cross-
sectional calculated from the hydraulic diameter).
Internal or External Surface Area –This parameter is either calculated from the known geometry, or
obtained from the manufacturer. Because this parameter almost always appears as a product of area and heat transfer
coefficient, tuning both this parameter and the heat transfer coefficients would be redundant.
Mass – This parameter is easily obtained from the manufacturer or measured. Because the header pipes do
not play a critical role in heat transfer, the mass of these may be included or neglected. In general, large changes in
this parameter are required before noticeably affecting the system response.
Specific Heat – The value of this parameter is easily obtained with knowledge of the heat exchanger
material. For heat exchangers constructed of multiple materials, the use of a weighted average of specific heats or
thermal capacities would be appropriate.
5.3.3.2 Empirical Parameters

Mean Void Fraction – Many correlations are available as mentioned previously. For these simulations, a
general slip ratio correlation is assumed. Slip ratio is defined as the ratio of the velocities of the vapor and liquid
phases in a two-phase flow. The lower bound on this parameter is given by the homogeneous relationship, with
S = 1 . A suggested upper bound is given by the Zivi correlation, where the slip ratio is given as
S = (ρ f ρ g ) . This upper bound is not a hard bound, but a guideline to be considered when tuning this
1/ 3
parameter. In general, the slip ratio has the greatest effect on the response of evaporator superheat.
Heat Transfer Coefficients – This value can be estimated using an empirical correlation or simply a
nominal value chosen by the user. For the single phase regions, an appropriate estimate can be calculated using the
161
Dittus-Boelter correlation [44]. For the two-phase region, the heat transfer coefficient will vary significantly
depending on pressure, fluid quality, and mass flow rate. The lumped two-phase heat transfer coefficient again
could be selected by correlation or as a nominal value significantly higher than the single phase region or the air side
heat transfer coefficient. The air side heat transfer coefficient can be determined using the measured air data or using
manufacturer data (i.e. J-factor correlation data [50]). The absolute and relative value of the heat transfer
coefficients will affect the response of each of the model outputs, most notably that of outlet enthalpy or outlet
temperature, which becomes critical to correctly predicted evaporator superheat.
5.3.3.3 Operating Condition

Evaporation Pressure – Although the evaporator model assumes isobaric evaporation, an actual evaporator
will have some measure of pressure drop from inlet to outlet. Because a principal goal of the model is to predict
evaporator superheat at the evaporator outlet, the measured outlet pressure is assumed to be the evaporation
pressure. The pressure drop from inlet to outlet is then lumped with the pipe losses between components.
Inlet Enthalpy – Although the valve is assumed to be isenthalpic, there exist losses in the pipe between the
valve and the evaporator, including pressure drop, parasitic heat gain, etc. Because the fluid quality entering the
evaporator cannot be measured, the inlet enthalpy is assumed to be approximately equal to the enthalpy entering the
valve. This value can be adjusted slightly without violating the limits of feasibility, but in general, small changes do
not significantly affect the dynamic responses.
Fluid Temperatures, Mass Flow Rates – The values of the outlet temperature of the refrigerant, the inlet and
outlet temperature of the external fluid (air), and the mass flow rates of the refrigerant and air are measured from
data.
5.3.3.4 Pipe Losses

Between components there are both momentum losses associated with friction, as well as thermal
losses/gains due to heat transfer to the environment. These losses are accounted for, as well as the possible transport
delay due to longer pipe lengths. Although the pressure and temperature drops are necessary, the transport delay is
insignificant for the particular experimental system considered.
5.3.3.5 Component Validation

In other sections, the comparison of experimental data and predicted outputs for individual component
models is presented. Measured values of inlet and outlet mass flow rate, inlet enthalpy, etc. are applied as inputs to
these component models. Due to the lack of a measured value for mass flow rate at the compressor, the compressor
model is necessarily a part of the individual validation of the heat exchangers. This isolated component validation is
not possible, however, for the evaporator model. This model is extremely sensitive to mass flow variations, and the
noisy mass flow signal from data, coupled with the small biases in the mass flow predicted by the compressor result
in undesirable fluctuations that obscure the salient dynamics. Therefore, component validation of this model is
deferred to a later publication.
5.3.4 Condenser with Receiver

The condenser model requires the same parameters as the evaporator model, with a few additions. The
internal volume and lumped heat transfer coefficient of the receiver are required, as is the initial mass of the
162
refrigerant in the receiver. This is chosen to match so that the total refrigerant charge in the simulation model
matches the actual refrigerant charge.
Condenser Pressure Prediction

1060
Data (Inlet)
1040 Nonlinear
Linear
1020
[kPa]
1000
980
960
940
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Condenser Air Temperature Prediction

34
Data (Outlet)
Data (Inlet)
32 Nonlinear
Linear
30
[C]
28
26
24
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.43 Component Model Validation: Condenser Pressure and Air Outlet Temperature
Unlike the evaporator, the presence of the receiver makes the model more robust to mass flow rate
variations, and an individual model validation can be conducted. The pressure response matches extremely well,
with a slight mismatch in steady state prediction. The response of the linear model differs only slightly from the
nonlinear model. Similarly, the shape of the outlet air temperature response matches well, again with a slight
mismatch in steady state.
5.3.5 System Validation

Having demonstrated the efficacy of each of the individual component models, we proceed to validate the
combined system model. The final choice of most parameters is straightforward; the values used in the simulation
are shown in Table 5.10. The effective cross-sectional area was chosen to reflect the internal volume of the heat
exchanger, inlet and outlet pipe lengths, and the vapor filled accumulator present in the system, thus accounting for
the all the low-side volume that participates in the pressure response. This is consistent with experimental studies
and results in the appropriate transient behavior. The value for slip ratio was chosen near the rough upper bound
given by the Zivi correlation. The empirical relationships for mass flow through the compressor and expansion valve
were developed using steady state data. Because of the sensitivity of the models to small changes in mass flow, the
accuracy of these maps is essential to good prediction.
163
Table 5.10 Parameter Values for Model Validation
Component Parameter Units Value Comment
Heat Exchanger Mass kg 1.548 Measured
Heat Exchanger Specific Heat kJ/kg-K 0.567 Aluminum Fin with Copper Pipe
Heat Exchanger Frontal Area cm^2 583.8 From Manufacturer
Hydraulic Diameter mm 8.915 Measured
Fluid Flow Length m 11.46 Calculated from Internal Volume
Cross Sectional Area for Flow cm^2 2.062 Calculated from Hydraulic Diameter
Evaporator
Internal Volume cm^3 500.8 From Manufacturer

Internal Surface Area m^2 0.321 Measured
External Surface Area m^2 3.068 Measured
Two-phase Flow Heat Transfer Coefficient kW/m^2/K 2.000 Tuned Parameter
Superheat Flow Heat Transfer Coefficient kW/m^2/K 0.040 Tuned Parameter
Exterior Fluid Heat Transfer Coefficient kW/m^2/K 0.058 Calculated from J-Factor Correlations
Exterior Fluid Specific Heat kJ/kg/K 1.05 Air
Slip Ratio [-] 4.6 Zivi Correlation
Heat Exchanger Mass kg 4.656 Measured
Heat Exchanger Specific Heat kJ/kg-K 0.467 Steel
Heat Exchanger Frontal Area cm^2 842.7 From Manufacturer
Hydraulic Diameter of Tubes mm 8.103 Measured
Fluid Flow Length m 10.69 Calculated from Internal Volume
Cross Sectional Area for Flow cm^2 0.516 Calculated from Hydraulic Diameter
Condenser
Internal Volume cm^3 557.2 From Manufacturer

Internal Surface Area m^2 0.275 Measured
External Surface Area m^2 2.793 Measured
Superheat Flow Heat Transfer Coefficient kW/m^2/K 0.388 Calculated using Dittus-Boelter Correlation
Two-phase Flow Heat Transfer Coefficient kW/m^2/K 1.000 Tuned Parameter
Exterior Fluid Heat Transfer Coefficient kW/m^2/K 0.126 Calculated from J-Factor Correlations
Exterior Fluid Specific Heat kJ/kg/K 0.960 Humid Air with Effective Specific Heat
Slip Ratio [-] 2.900 Zivi Correlation
Receiver Volume m^3 0.003 From Manufacturer
Compressor Displacement m^3 0.000 From Manufacturer

Compressor
Mass Flow Rate Empirical Map [-] - See Chapter 4
Rate Limit rpm/s 1000 Estimated

Expansion
Mass Flow Rate Empirical Map [-] - See Chapter 4

Valve
Rate Limit %/s 12.5 From Manufacturer
Compressor Condenser m 4.04 Measured

Lengths
Condenser to Valve m 4.79 Measured

Pipe
Valve to Evaporator m 1.63 Measured

Evaporator to Compressor m 6.53 Measured
The figures show generally good agreement between model and data. Although the mass flow rate is only
measured at the expansion valve, the difference in mass flow rates between the inlet and outlet of the heat
exchangers is a driving mechanism of the system dynamics. The peak in compressor mass flow rate with a step
change in compressor speed is expected even though it is not measured, and the remaining predictions coincide with
the system response (Figure 5.44).
The prediction of condenser pressure matches extremely well. Evaporator pressure response matches
excepting the speed of the response (Figure 5.45). With further tuning of the evaporator parameters, the speed of the
164
response could be improved. As mentioned previously, the effective volume of the evaporator may need to be
increased to reflect the entire low side volume.
The response of the air temperatures match well, although the changes are small (Figure 5.46). The
prediction of evaporator superheat is a key variable for control. Because evaporator superheat is the difference
between the saturation temperature and the outlet refrigeration temperature, correct prediction of both outlet
temperature and evaporator pressure is essential. Similar to the evaporator pressure, the response of refrigerant outlet
temperature, and thus superheat, matches generally well, excepting that the predicted response is faster than
measured (Figure 5.47). There also is an anomalous spike associated with the step change in compressor speed.
This problem may be correctable by adjusting the compressor mass flow rate prediction. This is part of current and
future efforts to include a sensor to measure transient mass flow rate at the compressor. With this added ability the
predicted response of mass flow rate under compressor speed changes could be verified, and the model validation
improved.
−3
Data (Valve)
Nonlinear (Compressor)
Linear (Compressor)
Nonlinear (Valve)
8.5 Linear (Valve)
8
[kg/s]
7.5
6.5
6
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.44 System Model Validation: Mass Flow Rates
165
Evaporator Pressure Prediction
340
Data (Outlet)
Data (Inlet)
320 Nonlinear
Linear
300
[kPa]
280
260
240
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Condenser Pressure Prediction

1060
Data (Inlet)
1040 Nonlinear
Linear
1020
[kPa]
1000
980
960
940
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.45 System Model Validation: System Pressures
Evaporator Air Temperature Prediction

28
Data (Outlet)
26 Data (Inlet)
Nonlinear
24
22
[C]
20
18
16
14
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Condenser Air Temperature Prediction

34
Data (Outlet)
Data (Inlet)
32 Nonlinear
Linear
30
[C]
28
26
24
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.46 System Model Validation: Air Temperatures
166
Evaporator Refrigerant Outlet Temperature Prediction
19
Data
18.5 Nonlinear
18
17.5
[C]
17
16.5
16
15.5
15
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Evaporator Superheat Temperature Prediction

23
Data
22 Nonlinear
21
20
[C]
19
18
17
16
15
800 1000 1200 1400 1600 1800 2000 2200 2400
Time [s]
Figure 5.47 System Model Validation: Evaporator Temperatures
5.4 Summary
Overall the initial model validation demonstrated agreement between model and data. The minor
discrepancies can be explained, and some improvements to the model are suggested. Most of the parameters
required by the model are known values, and experience has shown that the transient response of the system was
relatively insensitive to changes in most parameters. Moderate changes in other parameters, however, did result in
different transient responses. Specifically, the choice of the two-phase flow heat transfer coefficient changed the
shape and magnitude of the output responses. This effect is due, in part, to how the choice of this parameter affects
the calculated initial conditions of the system. The choice of slip ratio also affects the transient response notably.
The value for slip ratio determines the value of the void fraction, and thus the amount of liquid and vapor
refrigerant. A larger slip ratio results in a smaller void fraction and more liquid mass in the evaporator. The amount
of refrigerant mass inventory in the evaporator appears to affect the transient response much more than the values of
the physical geometry of the heat exchanger.
The model is also extremely sensitive to the algebraic relationships for mass flow as given for the
compressor and expansion valve. The principal dynamics of the system appear to be caused by the redistribution of
mass inventory, and the transient differences between inlet and outlet mass flow rate into the heat exchangers. The
simplified dynamic model does not include some of the small, fast transient behavior that would dampen and
stabilize the system. Like many nonlinear dynamic systems, the dynamic state variables of a vapor compression
cycle seem to exist on slow and fast dynamic manifolds. The slow dynamic manifolds are determined by the
dominant system dynamics. The fast dynamic manifolds that force the dynamic system to remain on the slow
dynamic manifold in the physical system are neglected in the model for simplicity. This simplicity comes at the
price of being sensitive to small changes in mass flow.
167
Chapter 6. Gain-Scheduled Control
As illustrated by the experimental results of Chapter 5, the dynamics of air-conditioning and refrigeration
systems are observed and analyzed to be highly nonlinear. To effectively control these systems, the control strategy
must account for these nonlinearities, and preferably provide guarantees of acceptable performance across the entire
range of possible operating conditions. This chapter presents gain-scheduled control as a practical and effective
solution to the nonlinear control problem. In the past, guaranteeing the stability of gain-scheduled controlled
systems has posed a challenging research problem. For this reason, special attention is given to the tools and
methods available for evaluating the stability of the gain-scheduled closed loop.
6.1 Introduction
Many physical systems exhibit dynamics that vary appreciably over the typical operating regime; a single
linear controller may fail to achieve acceptable performance throughout the envelope of conditions. One of the most
popular solutions to this nonlinear control design problem is the use of gain-scheduling, in which a nonlinear
controller is constructed by interpolating a family of local linear controllers. Thus the nonlinear control design
problem can be divided into several control problems where linear design tools are generally employed. This
“divide-and-conquer” [58],[15] methodology has found success in a variety of applications. The success of the gain-
scheduled approach seems directly attributable to its simplicity in design and implementation. In the literature, a
wide variety of approaches are termed “gain-scheduling.” For the purpose of this paper, we consider a gain-
scheduled controller as one that interpolates between local linear control laws as a function of scheduling variables,
which capture the system nonlinearities.
Despite the overwhelming success in industrial practice, the principal difficulty of using gain-scheduled
control is guaranteeing stability. By blending several linear controllers, the resulting global controller is nonlinear,
and results in the addition of “hidden coupling terms” [85] or additional dynamics due to the interpolation functions.
Thus, although the gain-scheduled controller may be stable at every fixed value of the scheduling variable, the true
global closed loop may not be stable.
Rugh and Shamma discuss gain-scheduling in terms of a four-step process [85]. First, determine a linear
parameter-varying model of the plant either from first principles, or by interpolating identified models. Second, use
linear design methods to design controllers. Third, determine the method of interpolating the linear controllers as a
function of the scheduling variable(s). Two common rules of thumb for selecting the scheduling variables is that
they should 1) capture the plant nonlinearities, and 2) they should vary slowly [92]. Fourth, assess the stability and
performance of the overall system, often by extensive simulations or experiments.
Stability of the closed loop system is generally addressed by one of the following three methods. First,
proving that the gain-scheduled controller is stable for all fixed values of the scheduling variable, and then inferring
overall stability by assuming slowly varying scheduling parameters. Although this is appropriate for exogenously
scheduled systems, it may not be possible for endogenously scheduled systems where the scheduling parameters
may vary arbitrarily fast. Second, stability for time-varying, but often rate-limited, scheduling parameters is
guaranteed by construction of the gain-scheduled controller using LPV or LFT methods. This generally assumes an
168
explicit model of the nonlinear system, which may not be available. Third, stability is verified through extensive
simulation and experiments.
In 2000, two separate survey articles appeared detailing the history and scope of gain-scheduling [85],[58].
The acknowledged motivation of gain-scheduled controllers is the ability to solve a nonlinear control problem by
using available linear control tools. Application areas have included flight control, engine control, power plants,
vehicle control, and many others.
The remaining sections of this chapter will first address the topics of problem formulation and stability
analysis in sections 6.2 and 6.3 respectively. Then an alternative framework for implementing gain-scheduled
controllers based on the Youla parameterization will presented in section 6.4. The final sections will address design
issues and present illustrative examples of the overall approach.
6.2 Problem Formulation

In this chapter we consider the nonlinear control problem shown in Figure 6.1. The plant and controller are
assumed to be nonlinear. Common notation is used whenever possible with d representing system disturbances, u
the control outputs, y the control inputs, and z the performance outputs.
Figure 6.1 General Nonlinear Closed Loop
6.2.1 Plant Representation

6.2.1.1 Nonlinear Plant Representation
A general representation of a nonlinear dynamic system is given as:
x& = f ( x, u , d )
y = g ( x, u , d ) ( 6.1 )
z = h( x, u , d )
where x ∈ R , u ∈ R , d ∈ R , y ∈ R , z ∈ R . The functions f, g, and h are assumed to be sufficiently
n m l p q
smooth and continuously differentiable:
f : Rn × Rm × Rl → Rn
g : Rn × Rm × Rl → R p .
h : Rn × Rm × Rl → Rq
Let (x0, u0, d0) denote an equilibrium point of Equation 6.1 such that
0 = f (x0 , u 0 , d 0 )
y 0 = g (x0 , u 0 , d 0 ) ,
z 0 = h( x 0 , u 0 , d 0 )
and define deviation variables δx = x − x 0 , δu = u − u 0 , δd = d − d 0 , δy = y − y 0 , δz = z − z 0 .
169
6.2.1.2 Jacobian Linearization
The nonlinear plant model can be linearized about (x0, u0, d0) by performing a first-order Taylor expansion:
⎛ ∂f ⎞ ⎛ ⎞ ⎛ ⎞
x& = ⎜ ⎟δx + ⎜ ∂f ⎟δu + ⎜ ∂f ⎟δd
⎜ ∂x x ,u ,d ⎟ ⎜ ∂u x ,u ,d ⎟ ⎜ ∂d x ,u ,d ⎟
⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠
⎛ ∂g ⎞ ⎛ ⎞ ⎛ ⎞
δy = ⎜ ⎟δx + ⎜ ∂g ⎟δu + ⎜ ∂g ⎟δd ( 6.2 )
⎜ ∂x x ,u ,d ⎟ ⎜ ∂u x ,u ,d ⎟ ⎜ ∂d x ,u ,d ⎟
⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠
⎛ ∂h ⎞ ⎛ ⎞ ⎛ ⎞
δz = ⎜ ⎟δx + ⎜ ∂h ⎟δu + ⎜ ∂h ⎟δd
⎜ ∂x x ,u ,d ⎟ ⎜ ∂u x ,u ,d ⎟ ⎜ ∂d x ,u ,d ⎟
⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠ ⎝ 0 0 0 ⎠
The Jacobian matrices can be a function of all or some of the variables (x, u, d). The subset of the variables that
parameterize the Jacobian matrices are denoted as θ ⊂ {x, u , d } . Thus an approximation of the nonlinear model
around the equilibrium point is given by the linear time invariant model:
⎡ x& P ⎤ ⎡ AP (θ 0 ) BP1 (θ 0 ) BP 2 (θ 0 ) ⎤ ⎡δx P ⎤

⎢ δy ⎥ = ⎢ C (θ ) D (θ ) D (θ )⎥ ⎢ δu ⎥ ( 6.3 )
⎢ ⎥ ⎢ P1 0 P11 0 P12 0 ⎥⎢ ⎥
⎢⎣ δz ⎥⎦ ⎢⎣C P 2 (θ 0 ) DP 21 (θ 0 ) DP 22 (θ 0 )⎥⎦ ⎢⎣ δd ⎥⎦
⎛ ∂f ⎞ ⎛ ⎞ ⎛ ⎞
where AP (θ 0 ) = ⎜ ⎟ , B P1 (θ 0 ) = ⎜ ∂f ⎟ , BP 2 (θ 0 ) = ⎜ ∂f ⎟ , etc.
⎜ ∂x ⎟ ⎜ ∂u ⎟ ⎜ ∂d ⎟
⎝ x0 , u 0 , d 0 ⎠ ⎝ x0 , u 0 , d 0 ⎠ ⎝ x0 , u 0 , d 0 ⎠
Using this linearization approach, the nonlinear model can be decomposed into several linear
representations around specified operating points suitable for use with linear control design techniques. Perhaps
more common in practice, the linear models used for control design will be identified models constructed from
experimental data. In either case, the methods for analysis to be presented later in this chapter will require a
representation of the nonlinear plant in terms of a family of linear models parameterized by a set of variables termed
scheduling signals.
However, note that Equation 6.3 is a Linear Time Invariant (LTI) local approximation of the nonlinear
dynamics about a given equilibrium point, not a Linear Parameter Varying (LPV) representation of the global
nonlinear dynamics. There would be no physical significance if one substitutes the constant θ0 with a time-varying
θ (t ) .
A method of constructing a valid nonlinear representation from a set of local linear models will be
presented later. However, since many of the control design techniques assume a LPV representation of the plant
dynamics, this topic will be addressed first.
6.2.1.3 Linear Parameter Varying Representation

A Linear Parameter Varying or LPV system is a special class of nonlinear systems defined as “a linear
system whose dynamics depend on exogenous parameters with values that are unknown a priori but can be
measured on-line” [92]. A true Linear Parameter Varying (LPV) representation of a system is given in Equation 6.4.
170
⎡ x& P ⎤ ⎡ AP (θ (t )) BP1 (θ (t )) BP 2 (θ (t )) ⎤ ⎡ x P ⎤
⎢ y ⎥ = ⎢ C (θ (t )) D (θ (t )) D (θ (t ))⎥ ⎢ u ⎥ ( 6.4 )
⎢ ⎥ ⎢ P1 P11 P12 ⎥⎢ ⎥
⎢⎣ z ⎥⎦ ⎢⎣C P 2 (θ (t )) DP 21 (θ (t )) DP 22 (θ (t ))⎥⎦ ⎢⎣ d ⎥⎦
If a nonlinear system does not naturally have an LPV representation, an alternative method for representing
the system as a parameterized family of linear models may exist, and has been termed the quasi-LPV representation.
In this method, nonlinear terms are “hidden with newly defined, time-varying parameters that are then included in
the scheduling variable.” The reader should note that quasi-LPV representations are not unique, and there may be
difficulty in finding a suitable representation for controller design. In particular, rapid variations in the augmented
set of scheduling variables may compromise stability. The interested reader is referred to [85] for detailed
information regarding quasi-LPV representations.
6.2.2 Controller Design

Gain-scheduled control design methods can be roughly classified into two groups: 1) local linear control
design and 2) LPV control design methods. The former approach is extensively used in practice and has tremendous
freedom in the design process. However it suffers from a general lack of suitable tools for stability analysis. The
latter has the advantage that some level of stability is guaranteed by construction of the controller, with the
disadvantage that the controllers are designed en masse with common size and structure, and no guarantee of
existence. Thus some freedom in the design process is lost, while at the same time being computationally
expensive.
6.2.2.4 Local Linear Control Design

With linearized models of the dynamics available, the nonlinear control design problem can be divided into
several control problems where any of the many linear design tools can be employed. This is the fundamental
“divide-and-conquer” approach of gain-scheduling. The local linear controllers are then denoted as:
⎡ x& K ⎤ ⎡ AK (θ 0 ) BK (θ 0 )⎤ ⎡δx K ⎤
⎢ δu ⎥ = ⎢C (θ ) D (θ )⎥ ⎢ δy ⎥ ( 6.5 )
⎣ ⎦ ⎣ K 0 K 0 ⎦⎣ ⎦
Reported methods of designing local controllers range from PID to H∞ techniques [42]. Since virtually no
restrictions are imposed during the design phase, research has focused on determining an appropriate means of
interpolating between these local linear controllers as a function of the scheduling variables. This topic will be
addressed in the following subsections. But first the second method of designing gain-scheduled controllers will be
presented.
6.2.2.5 LPV Control Design

An alternative method for designing gain-scheduled controllers has been termed “Gain Scheduling the LPV
Way” [76]. In this paradigm an LPV representation of the nonlinear plant is assumed to be available, and controllers
are designed en masse, with common size, structure, and state variables, such that stability of the nonlinear closed
loop is guaranteed if some basic assumptions are met. In particular, the assumptions that the scheduling variables be
measurable and bounded, and in most cases, have time derivatives that are measurable and bounded. These
assumptions can be restrictive, and current research is attempting to alleviate some of the resultant difficulties.
171
Although this approach does not have the tremendous design freedom, it remains attractive due to the
guarantees of stability. In essence, stability can be guaranteed in either of two ways. First, the problem can be
formulated with the parameter variations cast as uncertainty, where a single LTI controller is sought such that a
small-gain condition is met. If found, stability can be guaranteed for arbitrarily fast variations in the scheduling
parameter, but generally resulting in a poorly performing controller due to the extremely conservative nature of the
analysis. Second, an LPV controller can be sought such that stability of the closed loop system is guaranteed using a
constant quadratic Lyapunov functions (e.g. [76],[9]). Often the synthesis of such controllers can be framed as
solving a set of Linear Matrix Inequalities (LMIs) which can be solve efficiently. However, [8] indicates that
approaches that rely on a fixed Lyapunov function are “potentially very conservative because they allow for
arbitrary rates of variation in the scheduled variables.” Hence, recent efforts have employed parameter dependant
Lyapunov functions that can reduce some of this conservatism as long as the time derivative of the scheduling
variable can bounded [8],[33],[113],[29].
Overall, gain scheduling the LPV way has previously been shown to be successful for several examples,
with some guarantees of stability. However, the potential drawbacks of design inflexibility, computational intensity,
and restrictive assumptions have apparently limited its industrial application.
6.2.3 Scheduling Variables and Interpolation Methods

6.2.3.1 Scheduling Variables
Although the linearization of the plant and the associated controller is parameterized by θ ⊂ {x, u , d } ,
only a subset of these parameters may be necessary or desirable for scheduling controllers. Weak dependence,
inability to measure, or fast variations might prompt the controller engineer to neglect a parameter when scheduling
controllers. Thus we denote the scheduling variables as ρ ⊂ {y, u, d } , and the vector space defined by the
scheduling variables as the scheduling space. Note that an equilibrium point in the scheduling space need not be an
equilibrium point of the nonlinear plant.
6.2.3.2 Interpolation Methods

In the literature a variety of methods for interpolating local controllers are reported, including the
interpolation of poles, zeros, and gains, interpolation of H∞ controllers by interpolating Riccati equations,
interpolation of balanced state space matrix coefficients, interpolation of state and observer gains, and the
interpolation of eigenvalue placement state feedback gains (for a list of references, see [85]). Alternatively, several
authors have reported the use of controller blending, where the system signals of each of the linear controllers is
blended as a function of operating condition to form a global nonlinear controller [83],[19],[70].
The latter type of gain-scheduling, termed “output-blending,” is similar to the Tagaki-Sugeno
models/controllers in the field of fuzzy logic [103]. The principal benefit of this approach is the ability to simply
gain-schedule between controllers of different sizes and structures; no restrictions are placed on the design of the
individual controllers.
This output blending approach generally assumes the configuration shown in Figure 6.2. This diagram
illustrates the closed loop dynamics in the standard framework. The input disturbance w1, and the control output
signals u, are inputs to the nonlinear plant model, whose output y and the reference disturbance w2, are inputs to the
172
each of the linear controllers. The nonlinear controller is formed by weighting the individual outputs of several
linear controllers. These weighting or blending functions are a function of a scheduling variable ρ , as
α (t ) = f ( ρ (t )) . Common assumptions include α i (t ) ∈ [0,1] and ∑α (t ) = 1 , with the magnitude based on
i
the relative distance to the respective design point in the scheduling space.
This approach is sometimes termed a Local Controller Network (LCN) and is attractive because of the
simplicity of the controller implementation. A controller is constructed using standard linear techniques and a linear
approximation of the nonlinear model, either using a linearized first principles model or a data-driven identified
model. This process is repeated for several key operating conditions, and the resulting controllers are computed in
parallel, while a weighted sum of their outputs is applied to the nonlinear plant (Figure 6.2). For analysis purposes,
the nonlinear plant is often assumed to be adequately represented by a Local Model Network (LMN), where the
local linear representations of the plant are the result of the linearization of a nonlinear model, or simply data-driven
empirical models (Figure 6.3). The interconnection of these two is depicted in Figure 6.4. We will denote the LMN
as Pα (s ) = ∑ α i Pi (s ) , and similarly the LCN as Kα (s ) = ∑ α i K i (s ) (depicted in Figure 6.5).
P(s) LMN
e1 y
P1(s)
w1 w2
u e2
Pn(s) n
K1(s)
e1 y
w1 w2
n Kn(s)
u e2
LCN K(s)
Figure 6.2 Controller Blending using a Local Figure 6.3 Plant Blending using a Local Model
Controller Network Network
173
Figure 6.5 Simplified Diagram of the LCN/LMN
Figure 6.4 Diagram of the Combined LCN/LMN
Interconnection
6.2.4 Closed Loop Representation

Given a set of local linear models of the plant Pi (s ) ∈ RH 2
⎡ x& pi ⎤ ⎡ Api B pi ⎤ ⎡ x pi ⎤
⎢ y ⎥ = ⎢C 0 ⎥⎦ ⎢⎣ e1 ⎥⎦
( 6.6 )
⎣ pi ⎦ ⎣ pi
and given a priori a set of local linear controllers K i (s ) ∈ RH 2
⎡ x& ki ⎤ ⎡ Aki Bki ⎤ ⎡ x ki ⎤

⎢u ⎥ = ⎢C 0 ⎥⎦ ⎢⎣ e2 ⎥⎦
( 6.7 )
⎣ ki ⎦ ⎣ ki
the closed loop system (Figure 6.3) from w1 [ w 2 ] to [u
T
y ] is given in state-space form in Equation 6.10,
T
and denoted Gα (s ) . Note that this system can be represented as a system affinely parameterized in α with the
constraint ∑α (t ) = 1 . Alternatively, the system can be recognized as a polytopic model formed from a convex
i
set of individual models
⎡ x& p1 ⎤ ⎡ Ap1 0 ⎤ ⎡ x p1 ⎤ ⎡ B p1 ⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ M ⎥=⎢ O ⎥ ⎢ M ⎥ + ⎢ M ⎥ e1
⎢ x& pn ⎥ ⎢ 0 ⎥ ⎢ ⎥ ⎢ B pn ⎥
⎣ ⎦ 1 ⎣ 4424A pn ⎣ x pn ⎦
43⎦ 1 23 1 ⎣ 23⎦
AP xP BP ( 6.8 )
⎡ x p1 ⎤
[ ⎢ ⎥
y = α1C p1 L α nC pn ⎢ M ⎥
14442444 3
]
⎢ x pn ⎥
CP
⎣ ⎦
174
⎡ x& k1 ⎤ ⎡ Ak1 0 ⎤ ⎡ x k1 ⎤ ⎡ Bk 1 ⎤
⎢ M ⎥=⎢ O ⎥ ⎢ M ⎥ + ⎢ M ⎥e
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ 2
⎢⎣ x& kn ⎥⎦ ⎢⎣ 0 A ⎥ ⎢⎣ x kn ⎥⎦ ⎢⎣ Bkn ⎥⎦
144244kn3⎦ 1 23 123
AK xK BK ( 6.9 )
⎡ x k1 ⎤
u = [α 1C k 1 L α n C kn ]⎢⎢ M ⎥⎥
144424443
CK ⎢⎣ x kn ⎥⎦
⎡ x& P ⎤ ⎡ AP B P C K ⎤ ⎡ x P ⎤ ⎡ B P 0 ⎤ ⎡ w1 ⎤
⎢ x& ⎥ = ⎢ B C +
AK ⎥⎦ ⎢⎣ x K ⎥⎦ ⎢⎣ 0 B K ⎥⎦ ⎢⎣ w2 ⎥⎦
⎣ K ⎦ ⎣14 4~2443
K P
142 43
~
A (α ) B
( 6.10 )
⎡u ⎤ ⎡ 0 C P ⎤ ⎡ x P ⎤
⎢ y ⎥ = ⎢C 0 ⎥⎦ ⎢⎣ x K ⎥⎦
⎣ ⎦ ⎣14 K
2 43
~
C (α )
6.3 Stability Analysis

The stability analysis of a gain-scheduled controller in closed loop with a nonlinear plant can pose a
challenging problem. Indeed, Rugh and Shamma note that “there have been relatively few research papers on the
stability and performance of gain scheduled systems” [85] and Apkarian and Adams state that the classical
approaches connecting locally designed controllers “lack supporting theories that guarantee the behavior of the
scheduled controller” [8]. Furthermore, the nature of endogenous scheduling results in additional challenges when
evaluating stability. For this purpose, stability of the gain-scheduled closed loop system receives specific attention in
this dissertation.
6.3.1 Stability Characterizations

Stability of the gain-scheduled closed loop can be examined at several levels. As depicted in Figure 6.6,
the simplest form of stability analysis is termed frozen parameter stability, where the stability of the closed loop of
the linearized model and linearized controller are evaluated using the standard criteria for linear systems that all
poles lie in the closed left half plane, Re (λ (A~ α )) < 0 . An improvement upon this is to consider the scheduling
⎛⎜
⎝
⎞⎟
0⎠
dynamics by evaluating instead, the linearization of the nonlinear closed loop. Although this guarantees stability for
some ball, possibly small, around the equilibrium point, it is not a true guarantee of the stability of the nonlinear
system. To achieve this level of stability, common or parameter dependent Lyapunov functions can be searched for
that will guarantee asymptotic stability of the equilibrium point. This guarantees that for finite disturbances all
internal signals, and hence all outputs, will remain bounded. By extending Lyapunov methods, input-output
performance bounds can be determined resulting in guarantee of practical stability for these nonlinear systems. The
following sections will address each of these stability characterizations in more detail.
175
Figure 6.6 Successive Levels of Gain-Scheduled Stability
6.3.1.1 Frozen Parameter Stability

Most analytical methods for guaranteeing stability focus on achieving frozen-parameter stability, or closed
loop stability for any fixed value of the scheduling parameter (i.e. no scheduling dynamics are considered).
Although individual controllers may be stabilizing at the design points, the interpolated controller may not be
stabilizing at intermediate points. A simple example of this phenomenon can be given as follows. Let a plant and
two stabilizing controllers be defined as in Equation 6.11. An interpolated controller could be defined as in
Equation 6.12 where α ∈ [0,1] . Although both K1 and K 2 stabilize the plant, the blended controller K b
destabilizes the plant for α ∈ [0.25, 1] .
1 −1 −1
P (s ) = , K 1 (s ) = , K 2 (s ) = ( 6.11 )
s +1 s + 0.5 s − 0 .5
K b (s ) = αK 1 + (1 − α )K 2 ( 6.12 )
Interpolation methods that guarantee frozen parameter stability have been termed “stability-preserving”
interpolation methods [98]. Recent research has focused on guaranteeing this level of stability by design [96],[69].
Then by assuming slowly varying scheduling variables, the stability of the true system is inferred.
Shamma reported research that derived bounds on the allowable rate of change for the specific case of
scheduling on a reference trajectory or scheduling on a state variable [91], but in general, efforts to relax the “slowly
varying” assumption are ongoing. Thus, while these types of approaches may be valid for a class of systems where
the controller is scheduled using a purely exogenous signal, with known limits on the rate of change, the nonlinear
behavior of many systems is more appropriately captured by scheduling parameters that are functions of plant
outputs or controlled inputs. Guaranteeing the stability of these endogenously scheduled systems is a more
challenging problem, because a bound on the rate of change may not be known a priori. Furthermore, practical
stability necessitates that, despite reference changes or disturbances, the scheduling variables and system outputs
remain within prescribed bounds.
176
6.3.1.2 Stability of the Linearization of the Closed Loop
Note that the previous analysis also neglects the “hidden coupling terms” [92] due to the scheduling
dynamics. This is due to the fact that the closed loop of the linearization and the linearization of the closed loop are
different (see [117] or [85]). Lawrence and Rugh presented an interpolation scheme such that about any operating
point, the gain-scheduled controller linearizes to the linear controller designed for that point [55], thus eliminating
the hidden coupling terms. Although, conditions for such an interpolation to exist are given in [85], Rugh states that
in general these conditions are quite restrictive. Thus the next level of stability considers not simply the closed loop
of the plant/controller linearizations, but the linearization of the closed loop.
6.3.1.3 Nonlinear Stability

Any stability analysis of a linearization of the closed loop merely evaluates whether the system is
asymptotically stable for infinitesimal deviations about the equilibrium point [52]. To guarantee stability for a larger
region, perhaps globally, the stability of the nonlinear closed loop system must be evaluated. The most common
technique for guaranteeing the global stability of a nonlinear system is using Lyapunov functions.
Theorem 6.1
Let x& = f ( x ) be a nonlinear autonomous system with an equilibrium point x = 0 . Further more, let V
be a continuously differentiable function such that
V (0 ) = 0 and V ( x ) > 0 ∀x ≠ 0 ( 6.13 )

V& ( x ) < 0 ∀x ( 6.14 )
Then x = 0 is a globally asymptotically stable equilibrium point of the nonlinear system.
Proof
See [52].
A function V ( x ) that satisfies these conditions is called a Lyapunov function. Perhaps the most common
choice of a Lyapunov function is the quadratic function V ( x ) = x Px, P > 0 . For linear autonomous systems,
T
( )
x& = Ax , this choice of function leads to the condition V& ( x ) = x T AT P + PA x < 0 , such that the condition for
stability is equivalent to finding a solution to the well-known Lyapunov Equation A P + PA = −Q .
T
6.3.1.4 Nonlinear Stability with Performance Bounds

In terms of gain scheduled control, simply a guarantee of bounded signals is not sufficient for assuring
practical stability. Consider that the LMN or LPV representation of the nonlinear plant is likely to only be valid
within a specified operating envelope. Thus, any stability guarantees will only be valid if there is an accompanying
guarantee that the system does not leave the operating envelope during the transient being considered. Furthermore,
guarantees of worst case performance are often essential when considering the efficacy of a given control strategy or
practical limitations of a physical system. Thus, the next logical step in analysis is to extend the Lyapunov
techniques to generate worst case bounds on system outputs, controller outputs, or scheduling variables for a
177
specific class of disturbances. To address these issues, the following section will introduce methods for efficiently
determining the nonlinear stability of the gain scheduled closed loop using Linear Matrix Inequalities (LMIs).
6.3.2 Stability Analysis via Linear Matrix Inequalities

In this section we introduce Linear Matrix Inequalities (LMIs). Portions of this section are drawn from
[27],[16],[28]. LMIs have important applications in control theory because many of the control synthesis or analysis
problems can be framed as LMIs, and because these LMIs can be solved efficiently (i.e. in polynomial time) as
convex optimization problems.
An LMI is of the form: F ( x ) = F0 + ∑x F i i > 0 with symmetric known matrices Fi . This vector form
is not as restrictive as it may appear, and indeed, many problems can be transformed into LMIs. A particularly
useful result is known as the Schur complement formula, where for symmetric matrices Q and R , the following
two statements are equivalent:
Q > 0 , and Q − SR −1 S T > 0 ( 6.15 )
⎡Q S⎤
⎢S T > 0. ( 6.16 )
⎣ R ⎥⎦
At this point it becomes necessary to formally define a polytopic system in terms of the convex hull of
several linear systems.
Definition 6.2
Given a set of points x1 , K , x n the convex hull of these points is defined as:
⎧ n n
⎫
Co({x1 ,K , x n }) = ⎨ x ∈ R x = ∑ α i xi , α i ∈ [0,1], ∑α i = 1⎬ ( 6.17 )
⎩ i =1 i =1 ⎭
A standard linear dynamic system in RH2 is given as:
⎡ x& ⎤ ⎡ A B ⎤ ⎡ x ⎤
⎢ z ⎥ = ⎢C 0 ⎥ ⎢ w⎥ ( 6.18 )
⎣ ⎦ ⎣ ⎦⎣ ⎦
A linear polytopic system as:
⎡ x& ⎤ ⎡ A(t ) B(t )⎤ ⎡ x ⎤

⎢ z ⎥ = ⎢C (t ) 0 ⎥ ⎢ w⎥ ( 6.19 )
⎣ ⎦ ⎣ ⎦⎣ ⎦
where A(t ) = Co({A1 , K , An }) , B (t ) = Co({B1 , K , Bn }) , etc. Under these conditions, the closed loop
dynamics of the LMN/LCN are clearly a polytopic system, and can be analyzed as such. The following subsections
present LMI based analysis methods for polytopic systems.
6.3.2.1 Quadratic Stability

Assuming a quadratic Lyapunov function V (t ) = x (t ) Px(t ) where P > 0 , a necessary and sufficient
T
condition for asymptotic stability of the linear system (Eq. 6.18) is determined by the existence of a solution to the
178
LMI A P + PA < 0 . For polytopic systems, asymptotic stability is guaranteed by solving the set of LMIs
T
Ai P + PAi < 0, i = 1, K , m for the common Lyapunov function P > 0 .

T
6.3.2.2 H∞ Performance
The first of the performance metrics we consider in this section is referred to the H∞ performance. Among
[
several interpretations of this metric is that the H∞ norm of a dynamic system is equal to the induced L2 0, ∞ ) to
L2 [0, ∞ ) gain
z
G ∞
= sup 2
( 6.20 )
w∈L2 w 2
w≠ 0
With this definition, we proceed to give LMI conditions for an upper bound on the H∞ gain.
Theorem 6.3
The LTI system (Eq. 6.18) is asymptotically stable and has H∞ norm less than γ if and only if there exists
P > 0 such that
1
AT P + PA + PBB T P + C T C < 0 . ( 6.21 )
γ2
Proof
This is the standard Ricatti inequality and is presented without proof. However, using Schur complement
this can be transformed into an LMI as
⎛ AT P + PA + C T C PB ⎞
⎜ ⎟ < 0. ( 6.22 )
⎜ BT P − γ 2 I ⎟⎠
⎝
Thus determining the H∞ gain of the system becomes a semidefinite optimization problem, searching for the
minimum γ such that a solution to the LMI (Equation 6.22) exists.
This LMI formulation can be easily extended to polytopic systems as the set of LMIs
⎛ Ai T P + PAi + C i T C i PBi ⎞
⎜ ⎟ < 0, i = 1, K, m . ( 6.23 )
⎜ T
− γ 2 I ⎟⎠
⎝ Bi P
The reader should also note that for both LTI and polytopic systems, the H∞ gain for systems in RH∞ can be upper
bounded by the solving LMI
⎛0 ⎞⎛ I 0 ⎞
T
⎛I 0⎞ P 0 0
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜A B⎟ ⎜P 0 0 0 ⎟⎜ A B ⎟
⎜0 <0 ( 6.24 )
⎜0 I ⎟ 0 − γI 0 ⎟⎜ 0 I ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜C D⎟ ⎜0 0 0 1 ⎟⎜
I ⎟
⎝ ⎠ ⎝ γ ⎠⎝ C D ⎠
179
6.3.2.3 Generalized H2 Performance
The second performance metric we consider in this section is known as the generalized H2 performance.
[ [
This metric has the interpretation of the induced L2 0, ∞ ) to L∞ 0, ∞ ) gain a dynamic system is equal to
z
G 2→∞
= sup ∞
( 6.25 )
w∈L2 w 2
w≠ 0
Again we proceed to give LMI conditions for an upper bound on the H2 gain.
Theorem 6.4
The LTI system (Eq. 6.18) is asymptotically stable and has H2 norm less than γ if and only if there exists
P > 0 such that
⎛ AT P + PA PB ⎞ ⎛ P CT ⎞
⎜ ⎟ < ⎜ ⎟
⎜ BT P − γ ⎟ 0 and
⎜ C γI ⎟ > 0 ( 6.26 )
⎝ I ⎠ ⎝ ⎠
or equivalently
⎛ AT P + PA PC T ⎞ ⎛ P B⎞
⎜ ⎟ < 0 and ⎜⎜ T ⎟>0 ( 6.27 )
⎜ CP
⎝ − γI ⎟⎠ ⎝B γI ⎟⎠
Proof
First note that:
d
dt
[ T T
]
x(t ) Px(t ) = x& (t ) Px(t ) + x(t ) Px& (t )
T
= ( Ax(t ) + Bw(t )) Px(t ) + x(t ) P( Ax(t ) + Bw(t ))

T T
( 6.28 )
T
( )
= x(t ) AT P + PA x(t ) + w(t ) B T Px(t ) + x(t ) PBw(t )
T T
Assuming the existence of P > 0 that satisfies Equation 6.26, and then left and right multiplying Equation 6.26 by
[x(t ) w(t )] leads to:

T
T
( )
x(t ) AT P + PA x(t ) + x(t ) PBw(t ) + w(t ) B T Px(t ) − γw(t ) w(t ) < 0
T T T
d
dt
x(t ) Px(t ) − γw(t ) w(t ) < 0
T T
[ ] ( 6.29 )
Setting w(t ) = 0 would guarantee asymptotic stability, while integrating from 0 to T leads to:
x(T ) Px(T ) < γ ∫ w(t ) w(t ) = γ w

T 2
∀w ∈ L2 , T ≥ 0
T T
2
( 6.30 )
0
which guarantees that at every time instant, the states of the system remain within the ellipsoid:
{
Ω = x ∈ R n x T Px < γ w
2
2
} ( 6.31 )
The second part of Equation 6.26 leads to:
1 1
P− C T C > 0 or x(t ) Px(t ) > x(t ) C T Cx(t )
T T
( 6.32 )
γ γ
180
Since
z (t ) z (t ) = x(t ) C T Cx(t ) ,
T T
( 6.33 )
then
z (T ) z (T ) = x(T ) C T Cx(T )
T T
< γx(T ) Px(T )

T
( 6.34 )
2
<γ2 w 2
= sup z (T ) z (T ) yields
2 T
and taking the supremum z ∞
T ≥0
2 2
z ∞
<γ2 w 2. ( 6.35 )
Again determining the H2 gain of the system is a semidefinite optimization problem, searching for the
minimum γ such that a solution to the LMI (Equation 6.26) exists. This LMI formulation can be easily extended to
polytopic systems as the set of LMIs
⎛ Ai T P + PAi PBi ⎞ ⎛P T
Ci ⎞
⎜ ⎟ < 0 and ⎜ ⎟ > 0 for i = 1,K, m ( 6.36 )
⎜ BTP ⎟
− γI ⎠ ⎜C γI ⎟
⎝ i ⎝ i ⎠
or equivalently
⎛ Ai T P + PAi PC i
T
⎞ ⎛ P Bi ⎞
⎜ ⎟ < 0 and ⎜ T ⎟ > 0 for i = 1,K, m ( 6.37 )
⎜ − γI ⎟ ⎜B γI ⎟⎠
⎝ Ci P ⎠ ⎝ i
6.3.2.4 Peak-to-Peak Performance
The final performance metric we consider in this section is known as the peak-to-peak gain of the system,
or in other words, the maximum amplification from a finite magnitude disturbance to a finite magnitude output. The
solution to bounding the peak-to-peak norm is similar to the previous case, and is based on seeking for an invariant
ellipsoid that contains the set of all reachable states with peak input, and then the peak output given the set of
possible states. Similarly, for a polytopic system, these conditions are easily formed as LMIs. The formal theorem
follows.
Theorem 6.5
The LTI system (Eq. 6.18) is asymptotically stable and has peak-to-peak norm less than γ (i.e.
G α (s ) L∞ →L∞
< γ ) if and only if there exists P > 0 such that
⎡I 0⎤
T
⎡ηP P 0 ⎤⎡ I 0⎤ ⎡ηP 0 Ci ⎤
T
⎢A ⎢ T ⎥
⎢ i Bi ⎥⎥ ⎢P
⎢ 0 0 ⎥⎥ ⎢⎢ Ai Bi ⎥⎥ < 0 and ⎢ 0 (γ − μ )I Di ⎥ > 0 ( 6.38 )
⎢⎣ 0 I ⎥⎦ ⎢⎣ 0 0 − μI ⎥⎦ ⎢⎣ 0 I ⎥⎦ ⎢C Di − γI ⎥⎦
⎣ i
181
Proof
The proof follows a similar procedure as the generalized H2 performance problem and is not presented.
However, the reader should note that the conditions in Equation 6.38 imply that for every time instant the states of
the system remain within the ellipsoid
⎧ μ 2⎫
Ω = ⎨ x ∈ R n x T Px < w ∞⎬, ( 6.39 )
⎩ η ⎭
and that the peak output is bounded by the peak input
2 2
z ∞
<γ 2 w ∞. ( 6.40 )
Although the reader will note that the conditions given in Equation 6.38 are not a strict set of LMIs, this
challenge may be overcome by solving the LMIs for fixed η , and doing a line search over
[
η ∈ 0,−2 ⋅ max Re λ j Ai( ( ( ~ )))]. Again we refer the reader to [16] and [28] for more information.
With computationally efficient means of determining upper bounds on these norms for polytopic systems,
the designer can determine the class of disturbances or reference changes that will result in acceptable values of u
and y, and ensure that ρ = f (u, y ) stays within an acceptable region in the scheduling space.
6.3.2.5 Decay Rate
Beyond characterizing the input-output properties of a polytopic system, the ability to bound the decay rate
of the system states or outputs is of particular interest to the problem of gain scheduling. In particular, the ability to
determine an allowable rate of change on reference signals is extremely useful in practice. By combining results
from the previous sections, this problem can be formulated as an LMI.
Definition 6.6
The decay rate of the system (Eq. 6.18) is defined to be the largest α such that for all trajectories x
lim e −αt x(t ) = 0 ( 6.41 )

t →∞
Theorem 6.7
]
For a disturbance w(t ) such that w(t ) ∈ L∞ (− ∞,0 applied to the system (Eq. 6.18), the outputs z (t )
can be bounded as
z (t ) ≤ γe −ηt w ∞
( 6.42 )
where γ and η are given by the equations defined by the LMIs in Equation 6.38.
Proof
Recall the LMI conditions for determining an upper bound on the peak-to-peak gain (Eq. 6.45-6.46).
Furthermore, recall that the inequality in Equation 6.45 infers that for every time instant the system states lie within
the ellipsoid
⎧ μ 2⎫
Ω = ⎨ x ∈ R n x T Px < w ∞⎬ ( 6.43 )
⎩ η ⎭
182
Thus for a disturbance that is only nonzero for t ≤ 0 , the set of possible state values at t = 0 can be bounded as
μ 2
x0 Px0 <
T
w ∞.
η
Secondly, note that the LMI AT P + PA + 2ηP < 0 infers that V ( x(t )) ≤ V ( x(0 ))e −2ηt , thus
x(t ) Px(t ) ≤ x0 Px0 e −2ηt . Finally, we relate the bound on the outputs to the states by writing
T T
z (t ) z (t ) = x(t ) C T Cx(t ) . Transforming the LMI given in Eq. 6.38 using Schur’s complement formula,
T T
1
C T C < ηP , the outputs can be bounded by z (t ) z (t ) < ηγx(t ) Px(t ) . Combining these results in
T T
z (t ) z (t ) < ηγx(t ) Px(t )

T T
< ηγx0 Px0 e − 2ηt .

T
( 6.44 )
2
< γμe − 2ηt w ∞
Since γ > μ , then z (t ) ≤ γe − λt w ∞ , which completes the proof. A similar proof holds for polytopic systems.
6.3.3 Stability Analysis via LMIs

With the closed loop system being represented as polytopic system, stability can be determined using
Lyapunov based approaches. Specifically, the internal stability of the interconnected LMN/LCN system can be
guaranteed for arbitrarily fast variations in the scheduling parameter if a common Lyapunov function can be found
~
for each of the A (α i ) of the closed loop system. For exogenously gain-scheduled systems, this condition would be
sufficient for guaranteeing global stability throughout the operating envelope, although no indications are given
regarding performance. However, for endogenously scheduled systems, the above conditions are still insufficient.
For practical stability, it is necessary to ensure the scheduling parameter remains within acceptable bounds for an
assumed class of disturbances.
For endogenously scheduled systems, α = h(ρ ) and ρ = f (u, y ) . Although the existence of a
common Lyapunov function guarantees the boundedness of u and y, large values of u and y could drive the system
outside of the region for which the LMN, and hence the analysis, is valid. To quantify the worst case amplification
from disturbances to u and y requires the Lyapunov extensions presented, namely the finite energy to peak deviation
G α (s ) L2 → L∞
, and peak-to-peak deviation Gα (s ) L∞ → L∞
. Finally, a bound on the decay rate of the states and
outputs can also be determined, allowing the user to determine an approximate rate of change on reference signals
by combining the bound on peak deviation with a bound on the rate of convergence of the output. This approach will
be illustrated later in this chapter. All of these problems can be formulated as LMIs for polytopic systems.
183
6.4 Youla-based Gain-Scheduling
In this section we introduce in detail the Youla parameterization and an approach to gain scheduling on the
Youla parameters. A detailed discussion of its properties is necessary prior to discussing its applications to gain-
scheduled design.
6.4.1 Youla Parameterization

The Youla parameterization is named for Dante Youla who introduced the idea in 1976 along with J.
Bongiorno and H. Jabr [114],[115]. Later in 1979 Kucera presented the discrete time counterpart [54]. Thus, it is
sometimes termed the Youla-Bongiorno-Jabr-Kucera (YBJK) parameterization. The crux of the Youla
parameterization is the ability to describe all stabilizing controllers in terms of a single parameter (often termed the
Youla parameter), Q. Moreover, the closed loop relationship, assuming a controller parameterized in this way, is
affine in the parameter Q, permitting the problem of searching for an optimal stabilizing controller to be posed as a
convex optimization problem.
At this point we introduce coprime factors. A dynamic system can be decomposed into right and left
~ ~
coprime factors G (s ) = XY
−1
= Y −1 X (the reader is referred to [27] or [119] for a formal definition of coprime
~ −1 ~
factors). We denote the decomposition of a controller into left and right coprime factors as K (s ) = UV = V U
−1
~ ~ ~ −1 ~
where U , U , V , V ∈ RH ∞ . Similarly we may decompose a plant model as P(s ) = NM = M N . Assuming
−1
that K 0 stabilizes P0 and that the coprime factors satisfy the double Bezout identity (Equation 6.45), all stabilizing
controllers may be parameterized as in Equation 6.46 where Q ∈ RH ∞ and is termed the Youla parameter.
Similarly, all plants stabilizable by the controller K 0 can be parameterized in terms as in Equation 6.47 where
S ∈ RH ∞ and is termed the dual Youla parameter. For a formal proof of these results, see [119] or [104].
~ ~
⎡M U ⎤⎡ V − U ⎤ ⎡ I 0⎤
⎢N ⎢ ~ ~ ⎥=⎢ ( 6.45 )
⎣ V ⎥⎦ ⎣− N M ⎦ ⎣0 I ⎦
⎥
K (Q ) = (U 0 + M 0 Q )(V0 + N 0 Q )
−1
( 6.46 )
(
~ ~ −1 ~
)(
= V0 + QN 0 U 0 + QM 0
~
)
P(S ) = ( N 0 + V0 S )(M 0 + U 0 S )
−1
( 6.47 )
(~ ~ −1 ~
)(
= M 0 + SU 0 N 0 + SV0
~
)
This parameterization of all stabilizing controllers or all stabilizable plants can be conveniently represented
as a linear fractional transformation (LFT). For example, for a standard feedback control loop with stabilizing
controller K 0 (Figure 6.7), the class of all stabilizing controllers can be represented as a lower LFT of J K and Q ,
denoted Fl ( J K , Q ) , as shown in Figure 6.8. Similarly, the class of all stabilizable plants can be represented as an
upper LFT of J P and S , denoted Fu (J P , S ) (Figure 6.9). Together the dual Youla parameterization of controller
and plant lead to the diagram shown in Figure 6.10.
184
⎡U V −1 ~ −1 ⎤
J K (s ) = ⎢ 0 −01
V0
−1 ⎥ ( 6.48 )
⎣ V0 − V0 N 0 ⎦
⎡ N 0 M 0 −1 ~ −1 ⎤
J P (s ) = ⎢
M0
−1 −1 ⎥ ( 6.49 )
⎣ M0 − M0 U0 ⎦
Figure 6.8 Feedback Loop Depicting the Class

Figure 6.7 Feedback Control Loop
of All Stabilizing Controllers
S(s)
r1 s1
e1 y
JP(s)
w1 w2
u e2
JK(s)
r2 s2
Q(s)
Figure 6.9 Feedback Loop Depicting the Class Figure 6.10 Feedback Loop Depicting the Dual
of All Stabilizable Plants Youla Parameterization
The performance of a feedback control loop is generally of significant interest. The closed loop
relationship of the standard feedback loop given in Figure 6.7 can be written as:
−1
⎡ e1 ⎤ ⎡ I − K ⎤ ⎡ w1 ⎤
⎢e ⎥ = ⎢ − P I ⎥⎦ ⎢⎣ w2 ⎥⎦
. ( 6.50 )
⎣ 2⎦ ⎣
These disturbance-to-output relationships can be written in the more familiar form:
⎡ (I − KP )−1 K (I − PK ) ⎤
−1 −1
⎡ I − K⎤
⎢− P =⎢ ⎥, ( 6.51 )
⎣ I ⎥⎦ ⎣ P(I − KP )
−1
(I − PK )−1 ⎦
or in terms of the coprime factors:
−1 −1
⎡ I − K⎤ ⎡M 0 ⎤ ⎡M U⎤
⎢− P =⎢ . ( 6.52 )
⎣ I ⎥⎦ ⎣0 V ⎥⎦ ⎢⎣ N V ⎥⎦
The stability of the closed loop is guaranteed if:
185
−1
⎡ I − K⎤
⎢− P ∈ RH ∞ , ( 6.53 )
⎣ I ⎥⎦
or equivalently:
−1
⎡M U⎤
⎢N ∈ RH ∞ . ( 6.54 )
⎣ V ⎥⎦
The performance of the closed loop system parameterized by a nominally stabilizing controller K 0 and the
Youla parameter Q (Figure 6.8)can be given as:
− K (Q )⎤
−1 −1
⎡ I
⎢− P I ⎥⎦
⎡ I
=⎢
− K0 ⎤
⎡M ⎤ ~
+ ⎢ 0 ⎥Q N 0 M 0 .
I ⎥⎦
~
[ ] ( 6.55 )
⎣ 0 ⎣− P0 ⎣ N0 ⎦
Thus the closed loop performance is affine in the Youla parameter Q , allowing the search for an optimal controller
to be framed as a convex search for the optimal Q . This also reaffirms the fact that the closed loop will remain
stable if the plant is nominally stabilized by K 0 , and Q ∈ RH ∞ .

The performance of the closed loop system parameterized by the dual Youla parameters (Figure 6.10) can
be similarly written as:
⎡ I − K (Q )⎤
−1
⎡ I − K0 ⎤
−1
⎡ M U 0 ⎤ ⎪⎧⎡ I − Q⎤
−1
⎫⎪⎡ V~ ~
U0 ⎤ .
=⎢ +⎢ 0 ⎥ ⎨⎢ − I ⎬⎢ ~0 ~ ⎥
⎢ − P (S )
⎣ I ⎥⎦ ⎣− P0 I ⎥⎦ ⎣ N 0 V0 ⎦ ⎪⎩⎣− S I ⎥⎦ ⎪⎭⎣ N 0 M0⎦
( 6.56 )
Again, this reveals the fact that the closed loop is stable if:
−1
⎡ I − K0 ⎤ ⎡ I − Q⎤
⎢− P ⎥ ∈ RH ∞ and ⎢ ∈ RH ∞ . ( 6.57 )
⎣ 0 I ⎦ ⎣− S I ⎥⎦
This is confirmed by the examination of the interconnection of matrices J K and J P (given in Equations 6.48 and
6.49), which can be verified as given in Equation 6.58, using the double Bezout identity. Thus the system of Figure
6.11 is internally stable if and only if the system in Figure 6.12 is stable.
⎡0 I ⎤
Fl (J P , J K ) = ⎢ ⎥ ( 6.58 )
⎣ I 0⎦
186
S(s)
r s
JP(s)
u y
JK(s)
r s
Q(s)
Figure 6.11 Feedback Loop Depicting the Dual Figure 6.12 Feedback Loop Depicting the
Youla Parameterization Equivalent Stability Condition
A more complete discussion of the foregoing facts is presented in [104]. In this text the Youla
parameterization is discussed in detail, and several interesting applications are suggested. Besides a presentation of
the classical design methods that rely on the Youla parameterization (e.g. LQG, H∞, etc.), this text suggest several
online control design methods including iterative nested Q-designs and adaptive-Q control strategies. Several other
authors have proposed the additional uses of the Youla parameterization. For example, Niemann and Stoustrup have
used the Youla parameterization for smooth implementation of unstable controllers [102], switching between
multiple controllers [69],[14], and robust or fault tolerant controllers [68],[71],[101].
The use of Youla parameter based gain-scheduling has been suggested by a few authors [97],[70],[100].
The stated objective to achieve an interpolated controller such that the closed loop is stable at every intermediate
point, i.e. frozen parameter stable. However, the dual Youla parameterization offers significant advantages to gain-
scheduled control approach beyond frozen parameter stability.
6.4.2 Q-Blending
While the Local Controller Network is a straightforward and intuitive framework for creating a gain-
scheduled controller, the blended controller may not be stabilizing for fixed intermediate values of the scheduling
variable (see the example given in Equations 6.11 and 6.12). Furthermore, the framework requires that the each of
the local plant and controller models be open-loop stable. However, an alternative framework for blending the
controller efforts results in the recovery of the original local controllers at the design points, but increases the
stability of the gain-scheduled system. This framework has been termed J-Q interpolation [98], or blending of the
Youla parameters [69]. This framework was proposed as a means of guaranteeing frozen parameter stability at
intermediate design points, while recovering the original local controllers at the design points.
This method is proposed simply as a different method of controller interpolation. Given a set of local linear
controllers K i , i = 1K n and a set of local linear models Pi , i = 1K n , a Local Control Network (LCN) and
Local Model Network (LMN) could be constructed as Kα (s ) = ∑ α i K i (s ) and Pα (s ) = ∑ α i Pi (s ) , and
187
depicted in Figure 6.13. Even with the assumptions that 1) each K i and Pi are open loop stable, and 2) the
feedback interconnection of each K i and Pi is stable, the closed loop at intermediate points need not be stable.
However, assuming there exists a controller K 0 that stabilizes each Pi , then an alternative network of controllers
can be formed that which guarantees frozen point stability at all intermediate points. Using the coprime factors of
the nominal and local controllers, the Youla parameter required to recover the local controllers can be found as:
~ ~
Q i = U i V 0 − Vi U 0 . ( 6.59 )
The blended controller is then constructed as:
[ ][
K (Qα ) = U 0 + M 0 (∑ α i Qi ) V0 + N 0 (∑ α i Qi ) , ]
−1
( 6.60 )
which results in a Local Q-Network (LQN) as depicted in Figure 6.14. This requires the choice of a nominal plant
−1
P0 = N 0 M 0 , which is stabilizable by each of the local controllers K i , and the coprime factors are chosen to
satisfy the double Bezout identity. Note that at α i = 1 the original local controller K i is recovered. Moreover,
because the K (Q ) is stabilizing for any Q ∈ RH ∞ , then K (Qα ) is stabilizing for every frozen value of α , since
Qi ∈ RH ∞ and thus Qα = ∑ α i Qi ∈ RH ∞ .
Figure 6.13 Feedback Loop Depicting a Local Figure 6.14 Feedback Loop Depicting a Local
Controller Network Q-Network
The logical extension of these ideas leads to a network of dual Youla parameters (LSN) that is used to
capture the nonlinear plant behavior. These are similarly defined as:
~ ~
Si = Ni M 0 − M i N 0 ( 6.61 )
[ ][
P(Sα ) = N 0 + V0 (∑ α i S i ) M 0 + U 0 (∑ α i S i ) ]−1
( 6.62 )
and depicted in Figures 6.15-6.16.
188
S (s)
r1 s1
JP(s)
e1 y
w1 w2
u e2
JK(s)
r2 s2
Q (s)
Figure 6.15 Feedback Loop Depicting the Figure 6.16 Feedback Loop Depicting the
Local S-Network LQN/LSN Interconnection
To demonstrate that the local controllers are recovered under the LQN/LSN framework, first note that:
~ ~
Q i = U i V 0 − Vi U 0
~ ~ −1 ~ ~ −1
Qi = ViVi U iV0 − ViU 0V0 V0 ( 6.63 )
~
Qi = Vi (K i − K 0 )V0
Using this fact and employing the results of the Bezout identity, recovery of the local controllers is shown.
189
K i = K (Qi )
K i = [U 0 + M 0 Qi ][V0 + N 0 Qi ]
−1
K i [V0 + N 0 Qi ] = [U 0 + M 0 Qi ]
−1
K iV0 + K i N 0 Qi = U 0V0 V0 + M 0 Qi
K iV0 + K i N 0 Qi = K 0V0 + M 0 Qi
(K i − K 0 )V0 = M 0 Qi − K i N 0 Qi
~ ~ ~
Vi (K i − K 0 )V0 = Vi M 0 Qi − Vi K i N 0 Qi ( 6.64 )
~ ~
Q i = Vi M 0 Q i − Vi K i N 0 Q i
~ ~
I = Vi M 0 − V i K i N 0
~ ~
0 = U i N 0 − Vi K i N 0
~ ~
0 = U i − Vi K i
~ −1 ~
K i = Vi U i
Ki = Ki
The stated assumptions requiring a nominal stabilizing controller, and nominal stabilizable plant may
appear restrictive, but in reality are less restrictive than the requirements of the original LCN/LMN formulation, in
that each of the local models need not be open loop stable, but merely stabilizable by a common controller (similarly
with each of the local controllers). Moreover, the traditional LCN/LMN framework can be shown to be merely a
special case of the more general LQN/LSN. By choosing K 0 = P0 = 0 , the choice of coprime factors can be
selected as U 0 = 0 , V0 = I , N 0 = 0 , and M 0 = I . Thus our choice of Youla parameter reduces to individual

local controllers:
K i = [U 0 + M 0 Qi ][V0 + N 0 Qi ]
−1
= [0 + (I )Qi ][I + (0 )Qi ]

−1
, ( 6.65 )
= Qi
and the choice of dual Youla parameters reduces to the local models:
Pi = [N 0 + V0 S i ][M 0 + U 0 S i ]
−1
= [0 + (I )S i ][I + (0 )S i ]
−1
. ( 6.66 )
= Si
Thus by choosing the nominal controller and nominal model as K 0 = P0 = 0 , the LQN becomes the
LCN, and the LSN becomes the LMN. In place of the assumption that each of the local models is open loop stable,
the framework merely requires that each of the plant models be stabilizable by a single controller. Interestingly, an
industrial survey reported that 95% of control applications surveyed were open loop stable {Takatsu & Itoh 1999
#1120}, and for such plants this criteria is trivially satisfied. The choice of the nominal controller and nominal plant
also present an element of design freedom. Future efforts will focus on analytical methods of exploiting this
freedom. However, initial intuition would indicate that the nominal controller should be chosen to incorporate the
190
common elements of the local linear controllers, such as integrators. In this way, the Local Q-Network comprises
only the dynamic elements that differ among the local controllers. The nominal plant model could be chosen with a
similar objective in view.
Before continuing to present the representation of the closed loop system under the LQN/LSN
configuration, it is important to emphasize that this framework does not alter the nature of the blended controller at
the design points. The framework merely alters how the nonlinear controller transition between the design points.
As per the discussion earlier, the system of Figure 6.17 is internally stable if and only if the system in Figure 6.18 is
stable. Furthermore, the interconnection in Figure 6.18 is given in state-space form in Equation 6.74, and is denoted
H α (s ) .
[ w 2 ] to [e1 e 2 ] or [u y ] for the modified framework are

T T T
Note that the relationship from w1
given by Equations 6.68-6.69, where T1 , T2 , T3 , and T4 depend only on the choice of K 0 and P0 .
S (s)
r1 s1
JP(s)
e1 y
w1 w2
u e2
JK(s)
r2 s2
Q (s)
Figure 6.17 Feedback Loop Depicting the Figure 6.18 Simplified LQN/LSN
LQN/LSN Interconnection Interconnection
⎡ x& S ⎤ ⎡ AP BP C K ⎤ ⎡ x S ⎤ ⎡ BP 0 ⎤ ⎡ d1 ⎤
⎢ x& ⎥ = ⎢ ⎢ ⎥+
⎣ Q ⎦ ⎣1 BK C P AK ⎥⎦ ⎣ xQ ⎦ ⎢⎣ 0 B K ⎥⎦ ⎢⎣d 2 ⎥⎦
44~2443 142 ~
43
AQS (α ) BQS
( 6.67 )
⎡r ⎤ ⎡ 0 C P ⎤ ⎡ x S ⎤
⎢ s ⎥ = ⎢C ⎢ ⎥
0 ⎥⎦ ⎣ xQ ⎦
⎣ ⎦ 1 ⎣ 4K
24 3
~
CQS (α )
⎡ e1 ⎤ ⎡ w1 ⎤
⎢e ⎥ = (T1 + T2 H α T3 )⎢ w ⎥ ( 6.68 )
⎣ 2⎦ ⎣ 2⎦
⎡u ⎤ ⎡ w1 ⎤
⎢ y⎥ = (T4 + T2 H α T3 )⎢w ⎥ ( 6.69 )
⎣ ⎦ ⎣ 2⎦
~
⎡M U 0 ⎤ ⎡ 0 −U0 ⎤
T1 = ⎢ 0 ⎥⎢ ~ ⎥, ( 6.70 )
⎣ N 0 V0 ⎦ ⎣ − N 0 0 ⎦
191
⎡M U 0 ⎤
T2 = ⎢ 0 ⎥, ( 6.71 )
⎣ N 0 V0 ⎦
~ ~
⎡ V0 U 0 ⎤
T3 = ⎢ ~ ~ ⎥, ( 6.72 )
⎣N 0 M 0 ⎦
~
⎡M U 0 ⎤ ⎡ 0 −U0 ⎤
T4 = ⎢ 0 ⎥⎢ ~ ⎥−I ( 6.73 )
⎣ N 0 V0 ⎦ ⎣ − N 0 0 ⎦
6.5 Design Aspects

The alternative LQN/LSN network offers several elements of design freedom that deserve reiteration.
First, the framework and analysis tools have been presented assuming that the linear controllers designed at each
operating point do not have any common elements or similar structure. This allows for tremendous freedom for
iterative controller design at particularly important operating points. Should the nonlinear blended controller fail to
achieve the desired performance, some or all of the local controllers could be redesigned.
Second, the analysis of the nonlinear closed loop is independent on the shape of the blending functions.
Intuition dictates that the interpolation of the controllers should mimic the behavior of the nonlinear plant. While a
smoothly varying plant might be best controlled by a smoothly blended controller, abrupt transitions in plant
dynamics may be more appropriately handled by a similar change in controllers. Possible choices of these weighting
functions for multiple scheduling dimensions has been addressed in [117] and are shown in Figures 6.19 and 6.20.
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
1 1
0.8 1 0.8 1
0.6 0.8 0.6 0.8
0.4 0.6 0.4 0.6
0.4 0.4
0.2 0.2
0.2 0.2
0 0 0 0
Figure 6.19 Exponential Weighting Function Figure 6.20 Quadratic Weighting Function
The research in [117] also discusses the use of a scheduling filter. In practice, fast, but short lived
deviations from a design condition may result in poor performance due to fast scheduling dynamics. The use of a
low pass filter on the scheduling variable was found to increase the robust stability of the closed loop. While it is
impossible to restrict how fast the true plant transitions between operating conditions, the scheduling signal
governing the transition between the local linear controllers can be filtered. In this case, the LCN may lag behind
the LMN in the scheduling space. If this lag is significant, performance, and even stability, may be lost. However,
this practice has been reportedly successful in practice. Note that the analysis techniques presented in this
192
dissertation may be used, but with the undesirable necessity of doubling the size of the parameter space to represent
independent scheduling of the LMN and the LCN, thus leading to more conservative results.
Finally, for the case of gain scheduling on the Youla parameters, the choice of the nominal controller and
nominal plant offer significant design freedom. Although the optimization problem in this case is not convex,
engineering intuition may be applied and the common elements among the controllers or plants may be isolated,
leaving the scheduling of only the necessary dynamic differences.
6.6 Illustrative Example: Nonlinear Mass-Spring-Damper System

The analysis techniques presented in this chapter are illustrated using a simple example with 1-dimensional
gain scheduling. The application of these techniques to vapor compression cycles will be presented in the following
chapter.
6.6.1 Nonlinear Plant Model

To demonstrate the proposed method for stability analysis, a nonlinear mass-spring-damper system is used
[ ] [
(Figure 6.21). The governing equations are given in Equation 6.74, where c ∈ 0.3, 0.7 , k ∈ 0.15, 0.05 , ]
and m = 1 . The damping coefficient and the spring constant are assumed to vary linearly as a function of the
displacement. This results in a system whose dynamics vary significantly as a function of displacement. In this case
scheduling variable is selected as the output ρ = y , and design points are chosen at five points along the one-
dimensional scheduling space. The root locus for the system is shown in Figure 6.22 for varying y, and the Bode
plots for each of the five design points are given in Figure 6.23.
k(y)
w
m u
c(y)
y
Figure 6.21 Nonlinear Mass-Spring-Damper System
⎡ 0 1 ⎤ ⎡0 ⎤
x& = ⎢ ⎥ x + ⎢ ⎥ (u + w)
⎣− k ( y ) − c( y )⎦ ⎣1 ⎦ ( 6.74 )
y = [1 0]x
193
Pole-Zero Map
0.4
0.3
0.2
0.1
Imaginary Axis
-0.1
-0.2
-0.3
-0.4
-0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0
Real Axis
Figure 6.22 Root Locus for Values of y with Five Design Points
Step Response
20
18
16
14
12
Amplitude
10
0
0 10 20 30 40 50 60 70
Time (sec)
Figure 6.23 Step Response Plots for Five Local Linear Models
194
Bode Diagram
40
20
Magnitude (dB)
0
-20
-40
-60
-80
0
-45
Phase (deg)
-90
-135
-180
-3 -2 -1 0 1 2
10 10 10 10 10 10
Frequency (rad/sec)
Figure 6.24 Bode Plots for Five Local Linear Models
A LMN representation of the plant is formed using weighting functions that vary linearly with position
(Figure 6.25) and blending the five local plant models. The predictive capabilities of the LMN are compared to that
of the actual nonlinear plant in simulation (Figure 6.27). Alternatively, a LSN can be constructed by assuming a
nominal plant. For simplicity, the nominal plant is chosen to be the local plant model at the center design point, P3.
Bode plots of the resulting five S models is given in Figure 6.26, while the predictive capabilities of the LSN is
found to be similar to that of the LMN (Figure 6.27).
Figure 6.25 Weighting Functions
195
Bode Diagram
0
Magnitude (dB)
-50
-100
-150
360
180
Phase (deg)
-180
-360
-3 -2 -1 0 1 2
10 10 10 10 10 10
Frequency (rad/sec)
Figure 6.26 Bode Plots for Five S Models (P0 = P3)
Model Prediction Nonlinear Plant

0.6 LMN
LSN
0.4
P1
0.2 P5
0
y
-0.2
-0.4
-0.6
0 10 20 30 40 50 60 70 80 90 100
Tim e [s]
Figure 6.27 Simulated Comparison of Nonlinear Plant, LMN, and LSN
6.6.2 Nonlinear Control Design

H∞ controllers are designed for each of the five design points using frequency dependent performance
weightings on the tracking error and controller effort. The resulting controllers are shown in Figure 6.28. The LCN
is formed using the identical weighting functions given in Fig. 6.25.
196
Bode Diagram
60
40
Magnitude (dB)
20
-20
270
225
Phase (deg)
180
135
90
-4 -2 0 2
10 10 10 10
Frequency (rad/sec)
Figure 6.28 Bode Plots for Five Local Linear Controllers
Step Response
1
0.5
Error
0
Amplitude
-0.5
2
0
Control Input
-2
-4
-6
0 2 4 6 8 10 12 14 16 18
Time (sec)
Figure 6.29 Closed Loop Step Response for Five Local Linear Controllers
The increase in tracking performance using the LCN is demonstrated through simulation. Figure 6.30
compares the tracking performance of several of the local controllers operating over the entire ranges versus the
LCN. While the local controllers perform acceptably near their respective design points, the LCN clearly performs
better over the entire range.
197
Referenc e Tracking
0.6 Ref
0.4 LCN
K1
0.2
K3
0 K5
y
-0.2
-0.4
-0.6
0 10 20 30 40 50 60 70 80 90 100
Time [s]
Figure 6.30 Simulated Comparison of Local Controllers and LCN
Referenc e Trac king Referenc e Trac king
0
Ref Ref
0.6 -0.1
LCN LCN
0.5 K1 -0.2 K5
0.4 -0.3
y
y
0.3 -0.4
0.2 -0.5
0.1 -0.6
0
0 5 10 15 20 25 30 40 45 50 55 60 65 70
Tim e [s ] Tim e [s ]
Figure 6.31 Simulated Comparison of Local Controllers and LCN (Detailed View)
W eighting Func tion

1
0.5
α1
0
0 10 20 30 40 50 60 70 80 90 100
1
0.5
α2
0
0 10 20 30 40 50 60 70 80 90 100
1
LCN
0.5
α3
LQN
0
0 10 20 30 40 50 60 70 80 90 100
1
0.5
α4
0
0 10 20 30 40 50 60 70 80 90 100
1
0.5
α5
0
0 10 20 30 40 50 60 70 80 90 100
Time [s]
Figure 6.32 Controller Weighting Functions
198
Alternatively, the nonlinear controller can be constructed using a Local Q-Network. This requires a choice
for a nominal controller, which for simplicity is chosen as the local controller at the center design point, K3. A Bode
plot of the resulting five Q controllers is shown in Figure 6.33. The tracking performance using the LQN is
comparable to that of the LCN (Figure 6.34). However, the principal benefit of this alternative framework is found
in the guarantees of stability and performance.
Bode Diagram
0
-20
Magnitude (dB)
-40
-60
-80
540
360
Phase (deg)
180
-180
-4 -2 0 2
10 10 10 10
Frequency (rad/sec)
Figure 6.33 Bode Plots for Five Q Controllers (K0 = K3)
Reference Tracking
0.6 Ref
0.4 LCN
LQN
0.2
0
y
-0.2
-0.4
-0.6
0 10 20 30 40 50 60 70 80 90 100
Tim e [s]
Figure 6.34 Simulated Comparison of LCN and LQN
6.6.3 Stability Analysis

Using the LMI machinery presented previously, performance bounds for the LCN/LMN system or the
LQN/LSN framework. Note that this inherently guarantees internal stability, while also guaranteeing that the peak
deviation in the scheduling space can be bounded for disturbances with finite energy or finite peak magnitude. One
of the drawbacks of the Lyapunov based LMI machinery is the added computational cost of additional design points.
This also can result in more conservative performance bounds. Figure 6.35 illustrates the growth in performance
bounds as the number of design points is increased for the LCN/LMN. Results are shown for two design points
199
{ρ1 , ρ 5 }, three design points {ρ1 , ρ 3 , ρ 5 }, and five design points {ρ1 , ρ 2 , ρ 3 , ρ 4 , ρ 5 }. However, this increase
in performance bound is significantly less for the LQN/LSN. Realizing that these results are dependent on the
choice of nominal plant and controller (P1 and K1 in this case), the corresponding increase in bounds with additional
design points is markedly less.
The freedom in choosing the nominal plant and controller within the LQN/LSN framework can be
exploited to achieve lower performance bounds. Although any stabilizing controller or stabilizable plant could be
selected, a logical choice would be one of the local controllers or plants, i.e. K 0 = K 1 or P0 = P4 . Figure 6.36
demonstrates the change in performance bounds for K 0 = K i and P0 = Pi . Note that there are infinitely many
choices possible, and the performance bounds may vary significantly depending on this choice revealing the
flexibility in the design process. Furthermore, a comparison of the results of the LCN/LMN vs. the LQN/LSN
reveals that, in general, the LQN/LSN framework results in much lower performance bounds (Figure 6.37). A
complete listing of results is given in Tables 6.1-6.2.
Bounds on the Peak−to−Peak Gain
50
LCN/LMN ||r||→||e||
LQN/LSN ||r||→||e||
45
40
35
30
25
20
15
10
0
2 3 5
Number of Design Points
Figure 6.35 Comparison of Performance Bounds for Additional Design Points
Bounds on the H∞ Gain for 5 Design Points

Bounds on the Peak−to−Peak Gain for 5 Design Points
9
LQN/LSN ||w||→||y|| LQN/LSN ||w||→||y||
LQN/LSN ||r||→||y|| LQN/LSN ||r||→||y||
1.65 LQN/LSN ||r||→||e|| LQN/LSN ||r||→||e||
8
1.6
1.55
7
1.5
1.45 6
1.4
5
1.35
1.3
4
1.25
1.2 3
K_0 = K_1 K_0 = K_2 K_0 = K_3 K_0 = K_4 K_0 = K_5 K_0 = K_1 K_0 = K_2 K_0 = K_3 K_0 = K_4 K_0 = K_5
Choice of Nominal Controller/Plant Choice of Nominal Controller/Plant
Figure 6.36 Comparison of Performance Bounds for Different Choices of Nominal Plant/Controller in the
LQN/LSN Framework
200
Bounds on the H∞ Gain
2
LCN/LMN ||w||→||y||
LCN/LMN ||r||→||y||
1.9 LCN/LMN ||r||→||e||
LQN/LSN ||w||→||y||
LQN/LSN ||r||→||y||
1.8 LQN/LSN ||r||→||e||
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1
2 3 5
Figure 6.37 Comparison of H∞ Performance Bounds for LCN and LQN
Bounds on the H2 Gain

2
1.8 LQN/LSN ||w||→||y||
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0
2 3 5
Figure 6.38 Comparison of H2 Performance Bounds for LCN and LQN
201
Bounds on the Peak−to−Peak Gain
50
45 LCN/LMN ||r||→||e||
LQN/LSN ||w||→||y||
40 LQN/LSN ||r||→||e||
35
30
25
20
15
10
0
2 3 5
Figure 6.39 Comparison of Peak-to-Peak Performance Bounds for LCN and LQN
Table 6.1 Performance Bounds for LCN/LMN Framework
2 Design 3 Design
5 Design Points
Points Points
Disturbance-to-Output ⎛ y ⎞
sup ⎜ 2 ⎟ 1.474 1.526 1.510
H∞ Gain 0 < w 2 < ∞⎜ w ⎟
⎝ 2⎠
Reference-to-Output ⎛ y ⎞
sup ⎜ 2 ⎟ 1.552 1.668 1.676
H∞ Gain ⎜ r ⎟
⎝ 2⎠
0< w 2 <∞
Reference-to-Error ⎛ e ⎞
sup ⎜ 2 ⎟ 1.696 1.827 1.814
H∞ Gain 0 < w 2 < ∞⎜ r ⎟
⎝ 2⎠
sup ⎜ ∞ ⎟ 0.778 0.801 0.803
H2 Gain 0 < w 2 < ∞⎜ w ⎟
⎝ 2⎠
Reference-to-Output H2 ⎛ y ⎞
sup ⎜ ∞ ⎟ 1.326 1.417 1.668
Gain ⎜ r ⎟
0< r
2
< ∞
⎝ 2⎠
sup ⎜ ∞ ⎟ NA NA NA
H2 Gain 0 < r 2 < ∞⎜ r ⎟
⎝ 2⎠
sup ⎜ ∞ ⎟ 23.950 24.261 35.943
Peak-to-Peak Gain 0 < w ∞ < ∞⎜ w ⎟
⎝ ∞⎠
Reference-to-Output ⎛ y ⎞
sup ⎜ ∞ ⎟ 32.647 33.445 46.403
Peak-to-Peak Gain 0 < r ∞ < ∞⎜ r ⎟
⎝ ∞⎠
202
sup ⎜ ∞ ⎟ 33.682 34.454 46.386
Peak-to-Peak Gain 0 < r ∞ < ∞⎜ r ⎟
⎝ ∞⎠
Decay Rate η 0.00190036 0.00189328 0.00188636
Table 6.2 Performance Bounds for LQN/LSN Framework
3 Design
2 Design
Points 5 Design Points
Points
(K0 = K1, K3, (K0 = K1,…,K5)
(K0 = K1 or K5)
K5)
Disturbance-to-Output ⎛ y ⎞ {1.228, 1.242, {1.228, 1.242, 1.278,
sup ⎜ 2 ⎟ {1.230, 1.618}
H∞ Gain 0 < w 2 < ∞⎜ w ⎟ 1.277} 1.391, 1.605}
⎝ 2⎠
Reference-to-Output ⎛ y ⎞ {1.388, 1.357, {1.396, 1.365, 1.365,
sup ⎜ 2 ⎟ {1.377, 1.568}
H∞ Gain ⎜ r ⎟ 1.359} 1.429, 1.571}
⎝ 2⎠
0< w 2 <∞
Reference-to-Error ⎛ e ⎞ {1.516, 1.489, {1.526, 1.517, 1.467,

sup ⎜ 2 ⎟ {1.508, 1.458}
H∞ Gain 0 < w 2 < ∞⎜ r ⎟ 1.456} 1.468, 1.486}
⎝ 2⎠
sup ⎜ ∞ ⎟ {0.699, 0.800}
H2 Gain 0 < w 2 < ∞⎜ w ⎟ 0.758} 0.780, 0.832}
⎝ 2⎠
sup ⎜ ∞ ⎟ {1.217, 1.204}
H2 Gain 0 < r 2 < ∞⎜ r ⎟ 1.212} 1.205, 1.250}
⎝ 2⎠
sup ⎜ ∞ ⎟ NA NA NA
H2 Gain ⎜ r ⎟
0< r2
< ∞
⎝ 2⎠
sup ⎜ ∞ ⎟ {4.225, 3.507}
Peak-to-Peak Gain 0 < w ∞ < ∞⎜ w ⎟ 3.859} 3.618, 3.179}
⎝ ∞⎠
sup ⎜ ∞ ⎟ {6.886, 4.051}
Peak-to-Peak Gain 0 < r ∞ < ∞⎜ r ⎟ 5.644} 4.977, 4.225}
⎝ ∞⎠
Reference-to-Error ⎛ e ⎞ {8.006, 7.204, {8.030, 7.240, 6.632,
sup ⎜ ∞ ⎟ {7.923, 5.157}
Peak-to-Peak Gain ⎜ r ⎟ 6.643} 5.975, 5.221}
0< r∞
< ∞
⎝ ∞⎠
Decay Rate η 0.0988107 0.0644418 0.0558495
6.6.4 Rate Limit Bound on Reference Changes

Finally, for this special case of gain scheduling on the output, a rate limit bound on the reference signal can
be derived. Consider the case of one- dimensional scheduling, shown in Figure 6.40 where the distance between the
scheduling points is denoted d (the two-dimensional case is depicted in Figure 6.41). Assume that the system is
operating around design point #2, and the operator desires to stably transition to design point #4 while guaranteeing
that the system never leaves the region in scheduling space between design points #1 and #5. In reality this is
another requirement for system stability, since the analysis of our LMN/LCN is only valid in this region. Should the
system leave this region in the scheduling space, the LMN may not be valid and the analysis results guaranteeing
stability need not hold. Hence the necessity of not only guaranteeing the internal stability of the system within this
region, but also determining the class of acceptable disturbances and reference changes the system can handle with
the guarantee that the trajectories never leave the acceptable region in the scheduling space.
203
Figure 6.40 Depiction of the One-Dimensional Scheduling Space with Prescribed Region of Stability
Figure 6.41 Depiction of the Two-Dimensional Scheduling Space with Prescribed Region of Stability
Using the performance bound:
⎛ e ⎞
sup ⎜ ∞ ⎟<γ ( 6.75 )
0< r ∞ < ∞⎜ r ⎟
⎝ ∞ ⎠
the largest step change in reference the system can endure while being guaranteed not to leave the envelope shown
in Figure 6.40 is given as r ∞

< (d γ ) . Thus a series of reference step changes could be applied where each step
change is separated by a sufficient time period for the system to stabilize at its new equilibrium. The necessary time
period can be determined using the LMI methods outlined earlier which calculates a bound on the decay rate of the
system outputs. In particular, the tracking error of the system is guaranteed to decay as e(t ) ≤ γe
−ηt
r ∞
for
single step change in reference. Assuming that a series of step reference changes of magnitude R are applied to the
system, separated by interval of time T, the total response is guaranteed to be bounded by:
( )
e(t ) ≤ e −ηt u (t ) + e −η (t −T )u (t − T ) + ... + e −η (t − nT )u (t − nT ) γR . ( 6.76 )
The maximum will occur at t = nT , and for an infinite summation, the magnitude is given by
⎛ ∞ ⎞ ⎛ 1 ⎞
e(t ) ≤ ⎜ ∑ e −iηT ⎟γR = ⎜ −ηT ⎟
γR . ( 6.77 )
⎝ i =0 ⎠ ⎝1− e ⎠
204
Thus, to guarantee the tracking error never exceeds the minimum allowable distance d then
⎛ 1 ⎞
⎜ −ηT ⎟γR < d . ( 6.78 )
⎝1− e ⎠
As the magnitude of the step changes decreases and the interval time decreases, this results in an effective rate limit.
Defining the rate limit as M = (R T ) , the stability condition becomes
M <
(
d 1 − e −ηT
.
) ( 6.79 )
Tγ
Taking the limit as T → 0 yields
M < lim ⎢
(
⎡ d 1 − e −ηT ⎤ dη )
⎥= . ( 6.80 )
T →0
⎣ Tγ ⎦ γ
Thus the allowable rate limit increases as the acceptable magnitude of deviation d increases, as the decay rate of the
system η increases, and as the peak-to-peak gain from reference change to tracking error decreases.
Applying these results to simple case of two design points with d = 0.5 , the LCN/LMN framework yields
values of γ = 33.682 , and η = 1.90 ⋅ 10 -3 . Thus the necessary rate limit on the reference signal is
M < 28.2 ⋅ 10 -6 . For the LQN/LSN the performance bounds are much more reasonable, yielding γ = 5.157 ,
and η = 9.88 ⋅ 10 -2 . Thus the alternative framework allows a much higher rate limit of M < 9.58 ⋅ 10 -3 .
Simulation confirms that the system can stably and smoothly transition between operating points using this rate limit
(Figure 6.42).
Rate-Limited Reference Tracking
0.5
0
y
Ref
-0.5 LQN
50 100 150 200 250

Time [s]
0.02
0.01
0
e
-0.01
-0.02
50 100 150 200 250
Time [s]
Figure 6.42 Simulated Transition in Operating Condition using Rate Limited Reference Signal
205
6.7 Summary
This chapter explores guaranteeing global stability of a gain-scheduled system for exogenous and
endogenous scheduling with arbitrarily fast changes in the scheduling parameter. Using Local Model Networks and
Local Controller Networks, bounds on the peak deviation in the scheduling space for disturbances with finite energy
or finite peak magnitude can be efficiently determined using LMIs. A modified formulation of the gain-scheduled
system using the Youla parameterization is shown to result in significantly lower performance bounds (e.g. bound
on reference-to-error peak-to-peak value), and an element of design freedom that can be exploited when designing
the gain-scheduled controller.
206
Chapter 7. Control Design and Implementation
As mentioned in the introduction, advanced control has the potential to significantly improve the
operational efficiency and performance of vapor compression systems. This chapter reports the experimental
evaluation of several types of control strategies as applied to the subcritical air-conditioning system described in
Chapter 4.
7.1 Introduction and Background

Control of HVAC systems covers a broad range of applications, objectives, and methods, and of necessity,
must be subdivided before an appropriate overview can be given. First, a distinction needs to be made between
controlling the environmental dynamics vs. the refrigerant circuit dynamics. The former area of study focuses on
controlling the ambient conditions within a confined space, such as a room, car cabin, or refrigerator chambers.
Often the cooling capacity of the air conditioner or refrigerator is assumed to be a controllable input, and is treated
as an actuator. This can be an appropriate assumption based on the principle of time-scale separation discussed in
Chapter 2, since the dynamics of the environmental chamber are generally much slower than the refrigerant circuit
dynamics. The latter area of research focuses on issues such as expansion valve control, capacity control using
variable speed compressors, head pressure control using fan speed control, etc. In this case, the cooling capacity of
the system is an output of concern.
Second, distinction should be made between application areas, such as automotive air conditioning or
residential refrigeration. This is of secondary importance particularly when the focus is on the refrigerant circuit
dynamics, but still necessary since the application will dictate the control objectives, typical transient scenarios, and
the constraints on controllable actuators. With particular attention to these distinctions, we proceed to give an
overview regarding past research efforts regarding control of HVAC systems.
7.1.1 Environmental Control

Because this area is not the focus of this research, only a basic overview is given. The fundamental
objective of heating or cooling systems is to control the ambient conditions within a confined space, often for
occupant comfort, but also for other reasons (e.g. food refrigeration). Possible mechanisms for altering the ambient
conditions include: 1) changing the system cooling/heating capacity directly or 2) modulating the flow a secondary
fluid (e.g. valve control of chilled water supply as in [49], control of air dampers for automotive systems [32] or in
refrigerators [36],[37].
The bulk of this research focuses on building heating/cooling systems [23]. Although there is continuing
research on classical SISO control methods, such as PID [49],[72],[72],[47], the many possibilities for system
configurations and of the cooling “zones” have led to research on adaptive control strategies
[63],[21],[34],[75],[94],[89], or fuzzy logic or neural network approaches. If a model of the system is used, a first-
order plus time-delay model is almost universally assumed (Equation 7.1).
Ke − sτ
P (s ) = ( 7.1 )
s+a
207
7.1.2 Vapor Compression Cycle Control
In contrast with the development of environmental control strategies, research efforts regarding controlling
the refrigerant circuit dynamics have focused almost exclusively on developing models for analysis of the feedback
control strategies, which have remained generally simple. As discussed in Chapter 1, there are several possible
actuators on a typical vapor compression cycle. However, the application being considered may dictate whether this
device is a controllable input, disturbance, or neither.
First, consider the evaporator and condenser fans. In an automotive setting, the air flow across the
condenser is largely dependant on the vehicle speed, while the evaporator fan speed is generally selected by the
passenger. In a household refrigerator, the condenser rejects heat without fans, while the interior fans operate in an
on/off fashion with the designer dictating the switching logic. For older large industrial air conditioners, the fans
generally again in a binary fashion, while newer units equipped with variable speed controls the fan speeds may be
altered to control operating pressures or humidity.
Second, consider the compressor. In a typical automotive setting, the compressor speed is tied to the
engine speed and thus uncontrollable excepting by using a clutch to disengage the compressor completely. In a
household refrigerator, the compressor operates at a fixed speed, and cycles on and off as needed to provide the
necessary capacity. Older large industrial air conditioners will generally use multiple fixed speed compressors
allowing the system to operate at full or partial capacity, whereas newer units with variable speed or variable
displacement compressors provide a continuum of possible speed settings.
Despite the many and varied possibilities for advanced compressor or fan speed control, the bulk of the
literature on vapor compression cycle control focuses on the expansion device. The typical refrigerant metering
devices have been discussed in previous chapters. A fixed orifice valve or capillary tube meters refrigerant flow
simply as a function of operating pressures. A thermostatic expansion valve (TEV) is a mechanical feedback device
that senses the superheat at the exit of the evaporator, and then varies the area of the valve opening in an attempt to
regulate the system to a constant superheat. Unfortunately, the dynamics of the system using this device often result
in hunting, prompting considerable research. The automatic expansion valve (AEV) is another mechanical feedback
device that meters refrigerant to maintain a constant evaporator pressure, which can result in starving or flooding of
the evaporator in off-design conditions. Finally, the electronic expansion valve (EEV) allows much greater freedom
in valve control, generally using an electronically controlled stepper motor to adjust the valve opening.
As mentioned, much of the original research regarding control of vapor compression cycles focused on the
performance of the thermostatic expansion valve (TEV). The marginally stable behavior of this type of control is
well documented. The observed sinusoidal fluctuations in superheat temperature are generally termed “valve
hunting”. In 1966 W. Stoecker presented a simplified dynamical analysis of valve hunting [99],[110], and in 1973
Najork attempted to give optimal parameter settings to minimize hunting [67]. In 1980 Broersen and van der Jagt
used a mean void fraction model of the evaporator to show that valve hunting is caused by the interaction between
the TEV and evaporator dynamics [18], and two years later Broersen and Napel presented an identified model of the
TEV’s sensing bulb dynamics [17]. Gruhle and Isermann used a discretized model of the evaporator to show similar
results and contrasted the TEV with the proportional-integral (PI) controlled valve in 1985 [39]. Other authors also
208
develop TEV and evaporator models for dynamic analysis, but provide no validation [45],[43]. Recent research
continues to analyze and explain the instabilities of a TEV controlled evaporator [118].
A few publications report results with SISO control of an EEV. For example, in [74] the authors use a
simplified identified model for the evaporator, and then compare PID and optimal control algorithms. Similarly in
[31] the authors design a PID controller based on a simplified identified model for the evaporator, and compare PID
control to the TEV. Interestingly, both report that it is necessary to schedule the controller gains as a function of
operating condition.
The first research to attempt model-based MIMO control was the work of Xiang-Dong He. This research
employed the moving boundary, lumped parameter approach to model a subcritical vapor compression cycle, and
then design MIMO control strategies for a residential air conditioning system [40],[41]. Other researchers have used
similar model based approaches to conduct simulation studies [90] or rudimentary control experiments [57]. The
objective of this chapter is to provide a clear comparison of several different control strategies, both SISO and
MIMO, for industrial relevant scenarios. In doing so, the potential of advanced model-based control strategies will
be demonstrated.
7.2 Expansion Valve Control

Before addressing the more challenging case of coordinated expansion valve-compressor control, the
classical subject of SISO expansion valve control is evaluated. The automotive drive cycle test known as the FTP
test is selected as an industrially relevant transient testing scenario. This drive cycle is intended to simulate a 10
minute commute in Los Angeles, where the changing vehicle speed alters the compressor speed and condenser fan
speed. The SISO control strategy is to modulate the valve position so as to regulate evaporator superheat. This
drive cycle represents the harshest of the transient conditions that an air conditioning or refrigeration system will
experience. As a precursor to this industrial test, the ability to regulate superheat is evaluated under simple step
response tests.
7.2.1 Superheat Regulation Step Response Tests

Before evaluating the control strategies using the drive cycle, the control strategies are evaluated using
simple regulation tests. First the TEV is tuned for three a moderate capacity conditions. Second, using the EEV a
classic PID controller is tuned. The following plots illustrate superheat regulation for a step disturbance in
compressor speed for the TEV (Figure 7.1) and EEV (Figure 7.2). The TEV can effectively prevent liquid flow to
the compressor by maintaining a minimum level of superheat. The dynamics of the TEV are essentially a
proportional-type controller with sensor dynamics. As expected, this type of controller is not able to regulate the
superheat to a particular level, and the dynamic response is somewhat oscillatory. The EEV with PID control can
regulate superheat to the desired level, and provides a more damped response.
209
TEV Regulation of Superheat

12
Setpoint
TEV
10
8
2000 2500 3000 3500 4000 4500
Valve Opening [%]
16
14 NA
12
2000 2500 3000 3500 4000 4500

Time [s]
340
320
300
280
2000 2500 3000 3500 4000 4500
Time [s]
2000
1500
1000
2000 2500 3000 3500 4000 4500
Time [s]
Figure 7.1 TEV Regulation Evaporator Superheat
EEV Regulation of Superheat (PID kp=2, ki=0.01, kd=2)

12
Setpoint
EEV
10
8
2000 2500 3000 3500 4000 4500
Valve Opening [%]
16
14
12
2000 2500 3000 3500 4000 4500

Time [s]
340
320
300
280
2000 2500 3000 3500 4000 4500
Time [s]
2000
1500
1000
2000 2500 3000 3500 4000 4500
Time [s]
Figure 7.2 EEV Regulation Evaporator Superheat
210
7.2.2 FTP Automotive Drive Cycle
To further evaluate SISO control strategies for superheat regulation, a transient disturbance loading cycle
known as the SFTP-SCO3 is selected. This 10-minute test is used by the EPA when testing vehicle emissions, and
is meant to resemble a standard commute in the Los Angeles area. The vehicle speed profile is the same as the FTP-
75 emissions test, but with the specific requirement that the a/c system be running. The vehicle speed profile is
shown in Figure 7.3.
The implementation of this loading test requires the conversion of the vehicle speed profile to profiles of
compressor speed and condenser air flow rate. The conversion of vehicle speed to engine speed depends on the
transmission and the shifting schedule. Using a well-known vehicle software program, ADVISOR [5], a reasonable
engine speed profile can be created given the SCO3 vehicle speed profile. For this exercise the transmission of a
typical small car was assumed. The conversion from engine speed to compressor speed is simply a gain of 1 to 1.5
depending on the clutch and pulley system of the vehicle considered. For this exercise, a factor 1 was chosen for
simplicity. The resulting compressor speed profile is shown in Figure 7.4. This profile is scaled to reflect the range
of compressor speeds permitted on the experimental facility. To determine the condenser air flow rate profile, a
linear dependence of air flow rate to vehicle speed is assumed.
Vehicle Speed [mph]
60
50
40
30
20
10
0
0 100 200 300 400 500 600
Time [sec]
Figure 7.3 Vehicle Speed for the SFTP-SCO3 Test
3500
Compressor [rpm]
3000
2500
2000
1500
1000
500
0 100 200 300 400 500 600
Time [sec]
Figure 7.4 Compressor Speed for the SFTP-SCO3 Test (Assuming the Transmission of a Small-Sized Vehicle)
Under this loading cycle the compressor speed and condenser air flow rate are considered to be
disturbances, while the expansion valve is selected as a controllable input. During the test, the evaporator fan speed
is set to 100%. The control objective is to maximize efficiency while preventing liquid flow into the compressor,
which are roughly translated into the regulation of evaporator superheat. Figure 7.5 shows the regulation for the
211
TEV controls (tuned for high, moderate, and low capacities) and the PID controlled EEV. The COP shown includes
only compressor power; fan power is not included.

Regulation of Superheat Setpoint
EEV
15
TEV − Low
TEV − Med
10
TEV − High
5
0 500 1000 1500 2000 2500 3000
Valve Opening [%]
20
10
0
0 500 1000 1500 2000 2500 3000

Time [s]
400
350
300
250
0 500 1000 1500 2000 2500 3000
Time [s]
2000
1500
1000
0 500 1000 1500 2000 2500 3000
Time [s]
5
COP [−]
2
0 500 1000 1500 2000 2500 3000
Time [s]
Figure 7.5 Evaporator Superheat Regulation for SCO3 Drive Cycle
As mentioned, the FTP drive cycle subjects the system to large transient disturbances. The response of the
TEV shows that if properly tuned for the testing conditions, the TEV can effectively regulate evaporator superheat.
However, the response of the TEV is dependant on the changing operating conditions. Should the conditions
change, or if the TEV is poorly tuned, the result will be a periodic loss of superheat and hunting behavior, or
regulation to a high value of superheat. This is another reflection of the nonlinear dynamic behavior of the system,
and the need to account for these changes in the control strategy. In both of these scenarios, the poorly tuned TEV
results in approximately a 2% decrease in average cooling capacity over the cycle. Because the compressor speed
profile in this case is repeated, there is virtually no difference in the power consumed. The PID control provides
superior superheat regulation. There are still several instances with significant superheat deviations. These reflect
the most challenging portions of the transient tests, and the bandwidth and control authority of the EEV is
insufficient to overcome the faster effects of the compressor speed changes. Although the tighter control of
evaporator superheat allows for marginally lower evaporator pressures, and approximately 2% increase in cooling
capacity, significant increases in transient efficiency are possibly only with the incorporation of additional actuators,
such as the compressor.
7.3 Coordinated Compressor-Expansion Valve Control

The results of the previous section illustrate two overriding points. First, SISO control of the expansion
valve can provide tighter regulation of evaporator superheat, but only small increases in efficiency when compared
212
to the more traditional TEV control strategy. Second, there is a clear need for a gain-scheduled approach to feedback
control.
While advanced control strategies for the expansion valve alone would not likely yield large increases in
efficiency as compared to the TEV, the coordinated control of compressor and expansion valve has the potential for
dramatic increases in efficiency as well as performance. This is largely due to the inherently better efficiencies of a
continuously controlled compressor vs. alternative capacity control strategies. In a survey paper by Qureshi and
Tassou [79], a review of the application of variable speed capacity control to vapor compression systems yielded the
following key conclusions. First, although the advantages of variable speed capacity control have been known for
several decades, its implementation has been limited due to several technological issues (e.g. “insufficient
information from manufacturers” or “insufficient development and integration of compressors and variable-speed
drives”), and a single methodological issue, namely, “poor reliability… caused by unsophisticated and inadequately
developed control systems” [79].
The gains in efficiency result at partial load conditions. Because systems are often oversized to provide
fast pull-down (initial cooling of the desired space), these systems rarely operate at full-load during off-peak
conditions. Under partial load conditions, a fixed speed compressor must cycle on/off or rely on other strategies
(e.g. discharge gas bypass) that result in significantly less efficient operation. In contrast, a variable speed or
variable displacement compressor can use feedback control strategies to meet changing demands for cooling
capacity while providing partial load efficiency close to the theoretical ideal.
These gains can be clearly demonstrated for the experimental system described in Chapter 4. Figure 7.6
shows cooling capacity versus compressor power for variable speed operation and on/off control. The on/off control
results were obtained by using a fixed cycle length (5, 10, 15, or 30 minutes) and varying the duty cycle (20, 40, 60,
or 80%). From the plot, there is a dramatic increase of cooling capacity using the variable speed control strategies.
Although the efficiency of the on/off strategies increase with longer cycle times (longer pseudo-steady-state
operation, less start-up inefficiencies), these gains are negligible when compared to the variable speed strategies.
In particular, it is interesting to note that the governmentally-mandated SEER Test D, evaluates the system
efficiency for a 20% duty cycle, 30 minute cycle length, where this inefficiency is extremely pronounced. This test
(Test D) is known as the Cyclic Dry Coil Performance (Test D), which
“…shall be conducted by cycling the unit ‘on’ and ‘off’ by manual or automatic operation of the
normal control circuit of the unit. The unit shall cycle with the compressor ‘on’ for 6 minutes and
‘off’ for 24 minutes. The indoor fan shall also cycle ‘on’ and ‘off’, the duration of the indoor fan
‘on’ and ‘off’ periods being governed by the automatic controls which the manufacturer normally
supplies with the unit.” [3]
Under these conditions, SEER Test D effectively exposes a key inefficiency of the standard on/off control
strategies. However, this also means that variable speed control strategies are offered no advantage in SEER rating,
although they offer significant energy savings.
213
2
Variable Speed
5 min Cycle
1.8 10 min Cycle
15 min Cycle
SEER Test D
1.6
1.4
Cooling Capacity [kW]
1.2
80% Duty Cycle
0.8
60% Duty Cycle
0.6
40% Duty Cycle

0.4
0.2 20% Duty Cycle
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Compressor Power [kW]
Figure 7.6 Partial Load Efficiency
7.3.1 MIMO Control Design

This section details the design of coordinated control strategies for an air-conditioning system. For the
purposes of this chapter, this coordinated strategy modulates valve position and compressor speed to control
evaporator superheat and evaporator pressure or temperature. Although fan speed control could possibly be
exploited for controlling other system outputs, such as humidity, this is beyond the scope of the current work. The
choice of actuators may initially appear counterintuitive since both the expansion valve and compressor alter mass
flow rate of the refrigerant. However, recall that one of the driving mechanisms of the heat exchanger dynamics is
the transient difference in inlet and outlet mass flow rate. By independently manipulating the mass flow rate at the
entrance or exit of the heat exchanger, the actuators are both redistributing the refrigerant charge and the
temperatures at which heat is rejected and absorbed. However, in the simplest sense, the valve opening is used to
alter evaporator superheat (an indirect measure of efficiency) and the compressor speed is used to alter evaporator
pressure (an indirect measure of capacity), with significant coupling between the two inputs and two outputs.
A logical first attempt for attempting capacity and superheat control would be using multiple SISO control
loops. In addition to the EEV-superheat PID controller previously designed, a compressor speed-evaporator
pressure PID controller is independently tuned. Figure 7.7 shows evaporator pressure regulation for a step
disturbance in valve position, while Figure 7.8 shows evaporator pressure tracking for a step change in reference.
However, the dynamic coupling of the system is significant. An evaluation of the Relative Gain Array (RGA) [95] at
median capacity reveals large off diagonal terms, and thus substantial dynamic coupling. When implemented, the
interaction of these controllers leads to large oscillations, as seen in Figure 7.9.
214
Compressor Regulation of Evaporator Pressure (PID)

15
10
5
400 600 800 1000 1200 1400 1600 1800 2000
Valve Opening [%]
17
16
15
14
400 600 800 1000 1200 1400 1600 1800 2000
Time [s]
350
300
250
400 600 800 1000 1200 1400 1600 1800 2000
Time [s]
2500
2000
1500
1000
400 600 800 1000 1200 1400 1600 1800 2000
Time [s]
Figure 7.7 Evaporator Pressure Regulation
Compressor Tracking of Evaporator Pressure (PID)

15
10
5
2000 2500 3000 3500 4000 4500
17
Valve Opening [%]
16
15
14
2000 2500 3000 3500 4000 4500
Time [s]
350
300
250
2000 2500 3000 3500 4000 4500
Time [s]
2500
2000
1500
1000
2000 2500 3000 3500 4000 4500
Time [s]
Figure 7.8 Evaporator Pressure Tracking
215
Dual PID Control
15
10
5
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Valve Opening [%]
20
10
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time [s]
400
300
200
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time [s]
2000
1000
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time [s]
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time [s]
Figure 7.9 Dual SISO Control Interaction
Thus a true MIMO design approach is necessary. Of the multitude of linear control synthesis approaches,
an H∞ design approach is utilized here. This approach minimizes the worst case performance, with the designer
selecting the relative penalties on control action and performance. The general control problem as shown in Figure
7.10, with controllable inputs (e.g. valve opening, compressor speed), disturbances (e.g. air flow rate), outputs (e.g.
evaporator superheat, evaporator pressure), and measurement noise. Essential to successful control design is the
proper scaling of inputs/outputs. Based on the discussion in [95], prior to the control design process, the model
inputs/outputs were scaled according to their maximum expected deviation, such that u scaled ∈ [− 1,1] .
The standard format for H∞ control design procedure is shown in Figure 7.11, where z is the weighted
collection of signals we wish to minimize, v is the set of signals used by the controller, w is the collection of all
exogenous signals (disturbances, references, and noise), and u is the controller outputs. By including frequency
weightings on w and z, the control problem is rearranged as shown in Figure 7.12.
216
d
Gd
e u
r K Gu y
ym
n
Figure 7.10 The General Control Problem
w z
P
u v
K
Figure 7.11 Standard H∞ Control Formulation
P Wu z2
d Wd Gd
Gu
y
n Wn
_ ym
r Wr
+
We z1
u v
K
Figure 7.12 Standard Formulation with Frequency Weighting
217
From Figure 7.12, note that v = r − y m = Wr r − Gu u − Gd Wd d − Wn n and the collection of signals to
be minimized as a weighted vector of errors and control inputs: z = We v Wu u [ ]T . We define the exogenous
⎡z⎤ ⎡w⎤
w = [r d n] . Thus the system matrix ⎢ ⎥ = [P]⎢ ⎥ can be defined as in Equation 4.
T
signals as:
⎣v ⎦ ⎣u ⎦
⎡WeWr − We Gd Wd − WeWn − We Gu ⎤
⎡z⎤ ⎢ ⎡ w⎤
⎢v ⎥ = ⎢ 0 0 0 Wu ⎥⎥ ⎢ ⎥ ( 7.2 )
⎣ ⎦ ⎢ W − Gd Wd − Wn − Gu ⎥⎦ ⎣ ⎦
u
⎣ r
The choice of the weighting matrices is an important design step, and merits explanation. To emphasize
⎛ s 2+ω ⎞
low frequencies, we use the general form: Wi = diag ⎜⎜ Bi
⎟⎟ where Ai << 1 . This is used for We to
⎝ s + ω Bi Ai ⎠
penalize principally steady state errors. To emphasize higher frequencies, we simply use the inverse, of the above.
This is used for Wu to avoid high frequency controller actions. The other weighting functions were chosen simply
as static weights. The bandwidth for We and Wu are chosen iteratively, after evaluating controller performance
experimentally. Scaled versions of these weighting matrices are shown in Figures 7.13-7.14.
Bode Diagram
From: In(1) From: In(2)
0
To: Out(1)
Low Capacity
−20
High Capacity
−40
0
To: Out(1)
−45
−90
50
To: Out(2)
−50
0
To: Out(2)
−45
−90
−6 −4 −2 0 −6 −4 −2 0
10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.13 Performance Frequency Weighting
218
Bode Diagram
100
50
To: Out(1)
Low Capacity
0 High Capacity
−50
−100
90
To: Out(1)
45
0
0
To: Out(2)
−50
−100
90
To: Out(2)
45
0
0 0
10 10
Frequency (rad/sec)
Figure 7.14 Control Effort Frequency Weighting
Using the system in Equation 7.2, an H∞ controller is synthesized using the appropriate tools available in
MATLAB. The resulting dynamic controller is high order, and for implementation on an experimental system a low
order controller is preferred. The Hankel singular values of the resulting system are evaluated, and a balanced Schur
model reduction approach [87] is employed to remove less significant dynamic modes.
The synthesis procedure is repeated at several operating conditions. The magnitude response of two of the
plant models (high/low capacity) are shown in Figure 7.15 and the associated controllers are shown in Figure 7.16,
while the singular values of the controllers are given in Figure 7.17. In all these plots the inputs are valve opening
and compressor speed, and the outputs are evaporator superheat and pressure. The performance of the controllers is
shown in Figures 7.18-7.19. At low capacity the system responds slower and is more sensitive to actuator changes,
necessitating a less aggressive controller. As a demonstration of the controller performance both controllers are
implemented at each operating condition. The failure of each controller to adequately regulate the system at the off-
design points is another illustration of the necessity of a gain-scheduled approach.
219
Bode Diagram
20
0 Low Capacity
High Capacity
−20
To: Out(1)
−40
−60
Magnitude (dB)
−80
−100
20
−20
To: Out(2)
−40
−60
−80
−100
−120
−4 −2 0 −4 −2 0
10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.15 Magnitude Response of the Plant Models
Bode Diagram
40
20 Low Capacity
High Capacity
0
To: Out(1)
−20
−40
Magnitude (dB)
−60
−80
100
50
To: Out(2)
−50
−100
−4 −2 0 2 −4 −2 0 2
10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.16 Magnitude Response of the H∞ Controllers
220
Singular Values
80
Low Capacity
High Capacity
60
40
20
Singular Values (dB)
−20
−40
−60
−80
−100
−4 −3 −2 −1 0 1 2
10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.17 Singular Values of the H∞ Controllers

Low Capacity Operating Condition

14
12
10 Setpoint
H∞ Design Low Capacity
8
0 500 1000 1500 2000 2500 3000 3500
H∞ Design High Capacity
Valve Opening [%]
20
10
0
0 500 1000 1500 2000 2500 3000 3500

Time [s]
360
350
340
330
0 500 1000 1500 2000 2500 3000 3500
Time [s]
2000
1000
0
0 500 1000 1500 2000 2500 3000 3500
Time [s]
1.5
0.5
0 500 1000 1500 2000 2500 3000 3500
Time [s]
Figure 7.18 MIMO Control at Low Capacity
221
High Capacity Operating Condition
14
12
10 Setpoint
H Design Low Capacity
8 ∞
0 500 1000 1500 2000 2500 3000
H Design High Capacity
Valve Opening [%] ∞
20
10
0
0 500 1000 1500 2000 2500 3000

Time [s]
310
300
290
280
0 500 1000 1500 2000 2500 3000
Time [s]
2000
1000
0
0 500 1000 1500 2000 2500 3000
Time [s]
1.5
0.5
0 500 1000 1500 2000 2500 3000
Time [s]
Figure 7.19 MIMO Control at High Capacity
7.3.2 Gain-Scheduling
Although individual H∞ controllers perform well at their respective design points, none of the controllers
provides adequate regulation across all design points. To achieve the desired performance levels across the entire
operating regime a gain scheduled controller is constructed using the local linear controllers, as per the LMN/LCN
framework and LQN/LSN framework outlined in Chapter 6. The identified models indicate that the dynamics of the
system change with increasing evaporator superheat and increasing compressor speed. Assuming that superheat is
regulated within a sufficiently tight band, the dynamics could be scheduled using only compressor speed. However,
the difference in dynamic response is more likely due to the changes in evaporator pressure that are closely tied to
compressor speed. For this reason, the controllers are scheduled on evaporator pressure, which is one of the system
outputs. Similar to the example problem of Chapter 6, the interpolation function are chosen to vary linearly with the
scheduling variable.
To demonstrate the efficacy of a gain scheduled approach, two controllers (representing high and low
capacity conditions) are blended and analyzed using the LMIs given in Chapter 6. Unfortunately, in the LCN/LMN
case the computational algorithms fail to converge to a solution guaranteeing quadratic stability of the gain-
scheduled closed loop for arbitrarily fast scheduling. When analyzing the LQN/LSN framework for the given choice
of nominal plant/controller, the algorithms again fail to find. However, the existence of a common quadratic
Lyapunov function is merely a sufficient condition for stability. Thus while the stability of the gain-scheduled
closed loop system cannot be guaranteed, the system may indeed be stable in practice, as shown in Figure 7.20. The
222
experimental results demonstrate that with LCN/LMN framework the gain-scheduled controller can safely transition
between design points, as well as off-design points.
Although the superheat regulation is more oscillatory at off-design points, the performance is still
acceptable. At the design point, regulation of both superheat and pressure is excellent. The reader should note that
the experimental results shown are for rate-limited reference changes.
The alternative LSN/LQN framework can also be explored. The necessary decomposition of the controllers
can be performed, selecting K 0 = K 2 and P0 = P2 . This leads to Q2 = 0 , while Q1 and J K are shown in
Figures 7.21-7.22. The analysis tools of Chapter 6 are again employed to determine the associated peak-to-peak
gain and decay rate. While this approach is inherently more stable, the analysis is again inconclusive and fails to find
a common quadratic Lyapunov function. However, in this framework there is more design freedom for altering the
gain-scheduled controller to achieve a guaranteed stable closed loop. This would include exploring alternative
choices of the nominal plant/controller. However, because of the infinite possibilities, such a search is beyond the
scope of this dissertation. Future work will explore the possibility of searching for the optimal choice of nominal
plant/controller for guaranteed stability. Again experimental evaluation demonstrates the stability that is not yet
guaranteed by the analysis. The control performance is similar, if not slightly better than the traditional LCN/LMN
framework.
Transitioning Between Operating Conditions

15
Setpoint
14 LCN
LQN
13
12
11
10
8
0 1000 2000 3000 4000 5000 6000
360
350
Off−Design Point
340
330 Off−Design Point

Design Point
320
Design Point
310
Off−Design Point
300
290
280
0 1000 2000 3000 4000 5000 6000
Time [s]
Figure 7.20 Gain Scheduled Controller Transitioning Between Design Points and Off-Design Points
223
Bode Diagram
From: In(1) From: In(2) From: In(3) From: In(4)
200
To: Out(1)
0
−200
−400
500
To: Out(2)
0
Magnitude (dB)
−500
200
To: Out(3)
−200
200
To: Out(4)
−200
0 5 0 5 0 5 0 5
10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.21 Decomposition of the Controllers, J K
Bode Diagram
20
−20
To: Out(1)
−40
−60
Magnitude (dB)
−80
−100
60
40
To: Out(2)
20
−20
−40
−4 −2 0 2 −4 −2 0 2
10 10 10 10 10 10 10 10
Frequency (rad/sec)
Figure 7.22 Decomposition of the Controllers, Q1
7.4 Summary
This chapter reports the experimental evaluation of several types of control strategies as applied to the
subcritical air-conditioning system described in Chapter 4. While the regulation of evaporator superheat can be
improved by feedback control, the resulting energy savings vs. the standard TEV are not significant. However, the
coordinated control of compressor speed/displacement and valve opening has the potential to yield tremendous
energy savings relative to the current cycling strategies used. The dynamic coupling of the system requires a MIMO
control design approach, and the changing dynamics necessitate a gain-scheduled approach. The gain scheduling
224
framework presented in Chapter 6 is experimentally demonstrated to yield a nonlinear control strategy that can
extend the performance benefits across the entire operating regime by allowing safe and smooth transitions between
operating points. While the analysis developed in Chapter 6 is yet inconclusive for this particular example, future
improvements in LMI solvers, alternative Lyapunov formulations (e.g. parameter-dependent Lyapunov functions),
and techniques for the optimal selection of the Youla-based decomposition of controllers/plants will all lead to
improved results.
225
Chapter 8. Conclusions and Future Work
8.1 Summary of Research Contributions
This dissertation makes contributions in two distinct areas. First, with respect to vapor compression cycle
control, this dissertation presents a validated dynamic modeling framework and the experimental evaluation of
advanced model based control algorithms. Second, with respect to nonlinear control, this dissertation provides tools
for stability analysis of the nonlinear gain-scheduled system, and an alternative framework for implementing gain-
scheduled controllers with improved stability and a degree of design freedom.
8.1.1 Dynamic Modeling and Control of Vapor Compression Cycles

The first contribution of this dissertation is the presentation of a dynamic modeling framework for
subcritical and transcritical vapor compression cycles. The models are derived using a lumped parameter, moving
boundary model that captures the salient dynamic behavior, while remaining sufficiently tractable to be useful for
model based control synthesis. The dominant dynamic modes are identified and then a physics based model
reduction approach is applied to achieve a minimal representation of the dynamics. The models are validated with
experimental data, and are then used for designing and evaluating model based control strategies. These control
strategies are then implemented on an experimental system for several industrial relevant transient scenarios,
demonstrating improved performance and efficiency.
8.1.2 Gain Scheduling

Gain scheduling is perhaps the most the widely used nonlinear control technique used in industry.
However, there has been a lack of supporting theory to guarantee stability or performance of the nonlinear gain
scheduled closed loop. This dissertation presents a method of interpolating between local linear controllers termed
Local Controller Network, and then methods for determining the practical stability of the gain scheduled system in
terms of Linear Matrix Inequalities. Furthermore, an alternative method of interpolation is presented based on the
dual Youla parameterization, which results in improved stability and a degree of design freedom. The approach is
demonstrated in simulation and in experiment.
8.2 Future Work

This research has many aspects that have yet to be explored. A few of these are mentioned here, including
model validation and extensions, control research and fault detection, and further gain-scheduling research.
8.2.1 Model Validation and Extensions

Additionally, the modeling approach presented has been validated on a automotive subcritical cycle, an
experimental subcritical cycle, and a prototype transcritical system. Future validation efforts on additional systems
will demonstrate the efficacy of the modeling approach. Furthermore, modeling extensions such as humidity
modeling, condensation, and start-up and shut-down transients will make the model more useful.
8.2.2 Control Research and Fault Detection

This dissertation compares only a few control strategies, under selected scenarios. With the models
available, all the many different control technologies can be brought to bear on the problem. The nonlinear
dynamics of these systems, and the multitude of possible system configurations will provide almost endless
opportunities for development and implementation of classical and novel control strategies. In particular, multi-
226
evaporator systems afford a network-like platform for control development with tremendous potential for efficiency
and performance improvements. Perhaps of even greater interest is the potential for dynamic model based fault
detection strategies, that could detect and categorize system problems such as refrigerant leaks, heat exchanger
fouling, valve, fan or compressor failure, etc.
227
List of References
[1] "Sporlan Bulletin 10-9," Sporlan Valve Company, Washington, MO, Jun 2001.
[2] "Sporlan Bulletin 90-20," Sporlan Valve Company, Washington, MO, Nov 2001.
[3] "2005 Standard for Performance Rating of Unitary Air-Conditioning and Air-Source Heat Pump Equipment,"
Air-Conditioning & Refrigeration Institute, Arlington, VA, ARI 210/240-2005, 2005.
[4] "EASY5 Multiphase Fluid Library," http://www.mscsoftware.com/products/products_detail.cfm?PI=492,
2005.
[5] "Vehicle Systems Analysis," http://www.ctts.nrel.gov/analysis/, 2005.
[6] Anand, G., Mahajan, M., Jain, N., Maniam, B., and Tumas, T.M., "e-Thermal: Automobile Air Conditioning
Module," in Society of Automotive Engineers 2004 World Congress, Detroit, Michigan, 2004.
[7] Anderson, B.D.O., "Controller Design: Moving From Theory to Practice," IEEE Control Systems Magazine,
vol. 13, no. 4, pp. 16-25, Aug, 1993.
[8] Apkarian, P. and Adams, R.J., "Advanced gain-scheduling techniques for uncertain systems," IEEE
Transactions on Control Systems Technology, vol. 6, no. 1, pp. 21-32, 1998.
[9] Apkarian, P. and Gahinet, P., "Convex characterization of gain-scheduled H∞ controllers," IEEE Transactions
on Automatic Control, vol. 40, no. 5, pp. 853-864, 1995.
[10] Astrom, K.J. and Bell, R.D., "Drum-Boiler Dynamics," Automatica, vol. 36, no. 3, pp. 363-78, Mar, 2000.
[11] Beck, B.T. and Wedekind, G.L., "A Generalization of the System Mean Void Fraction Model for Transient
Two-Phase Evaporating Flows," ASME Journal of Heat Transfer, vol. 103, no. 1, pp. 81-85, Feb, 1981.
[12] Bendapudi, S. and Braun, J.E., "Development and Validation of a Mechanistic, Dynamic Model for a Vapor
Compression Centrifugal Chiller," ASHRAE Report #4036-4, May 2002.
[13] Bendapudi, S. and Braun, J.E., "A Review of Literature on Dynamic Models of Vapor Compression
Equipment," ASHRAE Report #4036-5, May 2002.
[14] Bendtsen, J.D., Stoustrup, J., and Trangbaek, K., "Bumpless Transfer Between Advanced Controllers with
Applications to Power Plant Control," in 42nd IEEE Conference on Decision and Control, Maui, HI, vol. 3,
2003, pp. 2059-2064.
[15] Bontempi, G. and Birattari, M., "From Linearization to Lazy Learning: A survey of Divide-and-Conquer
Techniques for Nonlinear Control," International Journal of Computational Cognition, vol. 3, no. 1, pp. 56-
73, Mar, 2005.
[16] Boyd, S., El Ghaoui, L., Feron, E., and Balakrishnan, V. Linear Matrix Inequalities in System and Control
Theory, Philadelphia, Pa: SIAM, 1994.
[17] Broersen, P.M.T. and ten Napel, J., "Identification of a Thermostatic Expansion Valve," in Identification and
System Parameter Estimation, Proceedings of the Sixth IFAC Symposium, vol. 1, 1983, pp. 415-20.
[18] Broersen, P.M.T. and van der Jagt, M.F.G., "Hunting of Evaporators Controlled by a Thermostatic Expansion
Valve," ASME Journal of Dynamic Systems Measurement & Control, vol. 102, no. 2, pp. 130-135, Jun, 1980.
[19] Buschek, H. , "Full Envelope Missile Autopilot Design Using Gain Scheduled Robust Control," Journal of
Guidance, Control, and Dynamics, vol. 22, no. 1, pp. 115-22, Jan, 1999.
[20] Cengel, Y. A. and Boles, M. A. Thermodynamics: An Engineering Approach, Boston, MA: WCB McGraw-
Hill, 1998.
[21] Chen, Y.H., Lee, K.M., and Wepfer, W.J., "Adaptive robust control scheme applied to a single-zone HVAC
system," in Proceedings of the 1990 American Control Conference, 1990, pp. 1076-1081.
[22] Chi, J. and Didion, D., "A Simulation Model of the Transient Performance of a Heat Pump," International
Journal of Refrigeration, vol. 5, no. 3, pp. 176-184, May, 1982.
228
[23] Coffin, M. Direct Digital Control for Building HVAC Systems, Boston, MA: Kluwer Academic Publishers,
1998.
[24] Cullimore, B.A. and Hendricks, T.J., "Design and Transient Simulation of Vehicle Air Conditioning
Systems," in Society of Automotive Engineers 2004 World Congress, Detroit, Michigan.
[25] Dennis, J. E. and Schnabel, R. B. Numerical Methods for Unconstrained Optimization and Nonlinear
Equations, Englewood Cliffs, New Jersey: Prentice Hall, 1983.
[26] Dhar, M. and Soedel, W., "Transient Analysis of a Vapor Compression Refrigeration System," in
Proceedings of the XVth International Congress of Refrigeration, Venice, Italy, vol. 2, 1979 , pp. 1031-1067.
[27] Dullerud, G. E. and Paganini, F. A Course in Robust Control Theory, A Convex Approach, New York, NY:
Springer-Verlag, 2000.
[28] El Ghaoui, L. and Niculescu, S. Advances in Linear Matrix Inequality Methods in Control, Philadelphia, Pa:
SIAM, 2000.
[29] Fan Wang and Balakrishnan, V., "Improved Stability Analysis and Gain-Scheduled Controller Synthesis for
Parameter-Dependent Systems," IEEE Transactions on Automatic Control, vol. 47, no. 5, pp. 720-34, May,
2002.
[30] Fernando, K.V. and Nicholson, H., "Singular Perturbational Model Reduction of Balanced Systems," IEEE
Transactions on Automatic Control, vol. 27, no. 2, pp. 466-468, Apr, 1982.
[31] Finn, D.P. and Doyle, C.J., "Control and Optimization Issues Associated with Algorithm-Controlled
Refrigerant Throttling Devices," in 2000 ASHRAE Winter Meeting, vol. 106 (PA, 2000, pp. 524-533.
[32] Forrest, W.O. and Bhatti, M.S., "Energy Efficient Automotive Air Conditioning System," in Society of
Automotive Engineers 2002 World Congress, Detroit, MI.
[33] Gahinet, P. , Apkarian, P., and Chilali, M., "Affine Parameter-Dependent Lyapunov Functions and Real
Parametric Uncertainty," IEEE Transactions on Automatic Control, vol. 41, no. 3, pp. 436-442, 1996.
[34] Georgescu, C., Afshari, A., and Bornard, G., "Optimal adaptive predictive control and fault detection of
residential building heating systems," in Proceedings of the 1994 IEEE Conference on Control Applications,
vol. 3, 1994, pp. 1601-1606.
[35] Grald, E.W. and MacArthur, J.W., "A Moving-Boundary Formulation for Modeling Time-Dependent Two-
Phase Flows," International Journal of Heat & Fluid Flow, vol. 13, no. 3, pp. 266-272, Sep, 1992.
[36] Graviss, K.J. and Collins, R.L., "Control of Household Refrigerators. Part I: Modeling Temperature Control
Performance," HVAC&R Research, vol. 4, no. 4, pp. 427-443, 1998.
[37] Graviss, K.J. and Collins, R.L., "Control of Household Refrigerators. Part II: Alternate Control Approaches
for Improving Temperature Performance and Reducing Energy Use," HVAC&R Research, vol. 4, no. 4, pp.
445-464, 1998.
[38] Greitzer, E.M. and Moore, F.K., "Theory of Post-Stall Transients in Axial Compression Systems: Part II -
Application," ASME Journal of Engineering for Gas Turbines & Power, vol. 108, no. 2, pp. 231-239, Apr,
1986.
[39] Gruhle, W.D. and Isermann, R., "Modeling and Control of a Refrigerant Evaporator," ASME Journal of
Dynamic Systems Measurement & Control, vol. 107, no. 4, pp. 235-240, Dec, 1985.
[40] He, X.D., Asada, H., Liu, S., and Itoh, H., "Multivariable Control of Vapor Compression Systems," HVAC&R
Research, vol. 4, no. 3, pp. 205-230, Jul, 1998.
[41] He, X.D., Liu, S., and Asada, H., "Modeling of Vapor Compression Cycles for Multivariable Feedback
Control of HVAC Systems," ASME Journal of Dynamic Systems Measurement & Control, vol. 119, no. 2, pp.
183-191, Jun, 1997.
[42] Hyde, R.A. and Glover, K., "Application of Scheduled H∞ Controllers to a VSTOL Aircraft," IEEE
Transactions on Automatic Control, vol. 38, no. 7, pp. 1021-1039, 1993.
229
[43] Ibrahim, G.A., "Effect of Sudden Changes in Evaporator External Parameters on a Refrigeration System with
an Evaporator Controlled by a Thermostatic Expansion Valve," International Journal of Refrigeration, vol.
24, no. 6, pp. 566-576, 2001.
[44] Incropera, F. P. and DeWitt, D. P. Fundamentals of Heat and Mass Transfer, New York: John Wiley & Sons,
1996.
[45] James, K.A. and James, R.W., "Transient Analysis of Thermostatic Expansion Valves for Refrigeration
System Evaporators Using Mathematical Models," Transactions of the Institute of Measurement & Control,
vol. 9, no. 4, pp. 198-205, 1987.
[46] Jensen, J.M. and Tummescheit, H., "Moving Boundary Models for Dynamic Simulations of Two-Phase
Flows," in Proceedings of the 2nd International Modelica Conference, 2002, pp. 235-244.
[47] Kamimura, K., Hashimoto, Y., Yamazaki, T., Noda, Y., and Kurosu, S., "A Comparison of Controller
Tuning Methods from a Design Viewpoint of the Potential for Energy Savings," in ASHRAE Transactions
2002, Jun 22-26 2002, vol. 108 Part 2, 2002, pp. 155-164.
[48] Kapadia, M. and Wolgemuth, C.H., "A Dynamic Model of a Condenser in a Closed Rankine Cycle Power
Plant," in Proceedings of the 1984 American Control Conference, New York, NY, vol. 1, 1984, pp. 79-84.
[49] Kasahara, M., Matsuba, T., Kuzuu, Y., Yamazaki, T., Hashimoto, Y., Kamimura, K., and Kurosu, S., "Design
and Tuning of Robust PID Controller for HVAC Systems," ASHRAE Transactions, vol. 105, pp. 154-166,
Jun, 1999.
[50] Kays, W. M. and London, A. L. Compact Heat Exchangers, New York: McGraw-Hill Book Company, 1984.
[51] Keir, M., Rasmussen, B., and Alleyne, A., "Parameter Sensitivity Analysis and Model Tuning Applied to
Vapor Compression Systems," in 2005 ASME International Mechanical Engineering Congress, Orlando, FL.
[52] Khalil, H. K. Nonlinear Systems, Upper Saddle River, New Jersey: Prentice-Hall Inc., 1996.
[53] Kokotovic, P. V. Singular Perturbation Methods in Control: Analysis and Design, Orlando, Florida:
Academic Press Inc., 1986.
[54] Kucera, V. Discrete Linear Control: The Polynomial Equation Approach, New York: John Wiley and Sons,
1979.
[55] Lawrence, D.A. and Rugh, W.J., "Gain Scheduling Dynamic Linear Controllers for a Nonlinear Plant,"
Automatica, vol. 31, no. 3, pp. 381-390, 1995.
[56] Lebrun, J. and Bourdouxhe, J.-P., "Reference Guide for Dynamic Models of HVAC Equipment," ASHRAE,
Atlanta, GA, Project 738-TRP, 1998.
[57] Leducq, D., Guilpart, J., and Trystram, G., "Low Order Dynamic Model of a Vapor Compression Cycle for
Process Control Design," Journal of Food Process Engineering, vol. 26, no. 1, pp. 67-91, 2003.
[58] Leith, D.J. and Leithead, W.E., "Survey of Gain-Scheduling Analysis and Design," International Journal of
Control, vol. 73, no. 11, pp. 1001-25, Jul 20, 2000.
[59] Lenger, M.J., Jacobi, A.M., and Hrnjak, P.S., "Superheat Stability of an Evaporator and Thermostatic
Expansion Valve, " Air Conditioning and Refrigeration Center, University of Illinois at Urbana Champaign,
ACRC TR-138, Jul 1998.
[60] Ljung, L. System Identification: Theory for the User, Upper Saddle River, New Jersey: Prentice Hall PTR,
1999.
[61] MacArthur, J.W., "Transient Heat Pump Behaviour: a Theoretical Investigation," International Journal of
Refrigeration, vol. 7, no. 2, pp. 123-132, Mar, 1984.
[62] MacArthur, J.W. and Grald, E.W., "Unsteady Compressible Two-Phase Flow Model for Predicting Cyclic
Heat Pump Performance and a Comparison With Experimental Data," International Journal of Refrigeration,
vol. 12, no. 1, pp. 29-41, Jan, 1989.
230
[63] Magnuski, M.J., "Review of ASHRAE Symposium Session on the Application of Adaptive Control to Real
HVAC Applications," International Journal of Refrigeration, vol. 11, no. 1, pp. 11-15, 1988.
[64] Mithraratne, P. and Wijeysundera, N.E., "An Experimental and Numerical Study of the Dynamic Behaviour
of a Counter-Flow Evaporator," International Journal of Refrigeration, vol. 24, no. 6, pp. 554-565, 2001.
[65] Mithraratne, P., Wijeysundera, N.E., and Bong, T.Y., "Dynamic Simulation of a Thermostatically Controlled
Counter-Flow Evaporator," International Journal of Refrigeration, vol. 23, no. 3, pp. 174-189, 2000.
[66] Moore, F.K. and Greitzer, E.M., "Theory of Post-Stall Transients in Axial Compression Systems: Part I -
Development of Equations," ASME Journal of Engineering for Gas Turbines & Power, vol. 108, no. 1, pp.
68-76, Jan, 1986.
[67] Najork, H., "Investigations on the Dynamical Behavior of Evaporators with Thermostatic Expansion Valve,"
in Proceedings of the 13th International Congress of Refrigeration, Washington, 1973, pp. 759-769.
[68] Niemann, H. , "Dual Youla Parameterisation," IEE Proceedings: Control Theory and Applications, vol. 150,
no. 5, pp. 493-497, 2003.
[69] Niemann, H. and Stoustrup, J., "An Architecture for Implementation of Multivariable Controllers," in
Proceedings of the 1999 American Control Conference, vol. 6, 1999, pp. 4029-33.
[70] Niemann, H. and Stoustrup, J., "Gain Scheduling Using the Youla Parameterization," in The 38th IEEE
Conference on Decision and Control, vol. 3, 1999, pp. 2306-2311.
[71] Niemann, H. and Stoustrup, J., "Reliable Control Using the Primary and Dual Youla Parameterizations," in
41st IEEE Conference on Decision and Control, vol. 4, 2002, pp. 4353-4358.
[72] Noda, Y., Yamazaki, T., Matsuba, T., Kamimura, K., and Kurosu, S., "Comparison in Control Performance
Between PID and H∞ Controllers for HVAC Control," in Technical and Symposium Papers Presented at the
2003 Winter Meeting of The ASHRAE, vol. 109 Part 1, 2003, pp. 3-11.
[73] Nohara, T., Ohyagi, S., Mizuno, N., and Fujii, S., "Identification of Air Conditioner as a Multi-Dimensional
Auto-Regressive Model," in IFAC Proceedings Series, vol. 2, no. 7, 1985, pp. 1853-1857.
[74] Outtagarts, A., Haberschill, P., and Lallemand, M., "Transient Response of an Evaporator Fed Through an
Electronic Expansion Valve," International Journal of Energy Research, vol. 21, no. 9, pp. 793-807, 1997.
[75] Ozsoy, C., "Self-tuning Control of a Heating, Ventilating and Air-Conditioning System," Journal of Systems
& Control Engineering, vol. 207, no. 4, pp. 243-251, 1993.
[76] Packard, A. and Kantner, M., "Gain Scheduling the LPV Way," in Proceedings of the 35th IEEE Conference
on Decision and Control, vol. 4, 1996, pp. 3938-3941.
[77] Pettit, N.B.O.L., Willatzen, M., and Ploug-Sorensen, L., "General Dynamic Simulation Model for Evaporators
and Condensers in Refrigeration. Part II: Simulation and Control of an Evaporator," International Journal of
Refrigeration, vol. 21, no. 5, pp. 404-414, Aug, 1998.
[78] Pfafferott, T. and Schmitz, G., "Numeric Simulation of an Integrated CO2 Cooling System," in Proceedings
of the Modelica 2000 Workshop, Lund, Sweden, 2000, pp. 89-92.
[79] Qureshi, T.Q. and Tassou, S.A., "Variable-Speed Capacity Control in Refrigeration Systems.," Applied
Thermal Engineering, vol. 16, no. 2, pp. 103-13, Feb, 1996.
[80] Rasmussen, B.P. "Thermosys Toolbox User's Manual," http://mr-roboto.me.uiuc.edu/thermosys, 2005.
[81] Rasmussen, B.P., "Control-Oriented Modeling of Transcritical Vapor Compression Systems," Dept. of
Mechanical and Industiral Engineering, University of Illinois, Urbana, IL, Oct. 2002.
[82] Rasmussen, B.P. and Alleyne, A., "Control-Oriented Modeling of Transcritical Vapor Compression Systems,"
ASME Journal of Dynamic Systems, Measurement, and Control, vol. 126, no. 1, pp. 54-64, Mar, 2004.
[83] Rhong Zhang , Alleyne, A.G., and Carter, D.E., "Robust Gain Scheduling Control of an Earthmoving Vehicle
Powertrain," in Proceedings of 2003 American Control Conference, vol. 6, 2003, pp. 4969-74 .
231
[84] Rice, C.K., "Effect of Void Fraction Correlation and Heat Flux Assumption on Refrigerant Charge Inventory
Predictions," ASHRAE Transactions, vol. 93, no. 1, pp. 341-367, 1987.
[85] Rugh, W.J. and Shamma, J.S., "Research on Gain Scheduling," Automatica, vol. 36, no. 10, pp. 1401-25, Oct,
2000.
[86] Rugh, W.J., "Analytical Framework for Gain Scheduling," IEEE Control Systems Magazine, vol. 11, no. 1,
pp. 79-84, 1991.
[87] Safonov, M.G. and Chiang, R.Y., "Model Reduction for Robust Control: a Schur Relative Error Method,"
International Journal of Adaptive Control & Signal Processing, vol. 2, no. 4, pp. 259-272, Dec, 1988.
[88] Sami, S.M. and Comeau, M.A., "Development of a Simulation Model for Predicting Dynamic Behaviour of
Heat Pump With Nonazeotropic Refrigerant Mixtures," International Journal of Energy Research, vol. 16, no.
5, pp. 431-444, Jul, 1992.
[89] Seem, J.E., "New pattern recognition adaptive controller with application to HVAC systems," Automatica,
vol. 34, no. 8, pp. 969-982, 1998.
[90] Shah, R., Rasmussen, B.P., and Alleyne, A.G., "Application of a Multivariable Adaptive Control Strategy to
Automotive Air Conditioning Systems," International Journal of Adaptive Control and Signal Processing,
vol. 18, no. 2, pp. 199-221, 2004.
[91] Shamma, J.S. and Athans, M., "Analysis of Gain Scheduled Control for Nonlinear Plants," IEEE
Transactions on Automatic Control, vol. 35, no. 8, pp. 898-907, 1990.
[92] Shamma, J.S. and Athans, M., "Gain scheduling: Potential Hazards and Possible Remedies," in Proceedings
of the 1991 American Control Conference, vol. 1, 1991, pp. 516-521.
[93] Shoureshi, R. and McLaughlin, K., "Modeling and Dynamics of Two-Phase Flow Heat Exchangers Using
Temperature-Entropy Bond Graphs," in Proceedings of the 1984 American Control Conference, 1984, pp. 93-
8.
[94] Singh, G., Zaheer-uddin, M., and Patel, R.V., "Adaptive Control of Multivariable Thermal Processes in
HVAC Systems," Energy Conversion and Management, vol. 41, no. 15, pp. 1671-1685, 2000.
[95] Skogestad, S. and Postlethwaite, I. Multivariable Feedback Control: Analysis and Design, New York: John
Wiley & Sons, 1996.
[96] Stilwell, D.J., "J-Q interpolation for Gain Scheduled Controllers," in Proceedings of 1999 Conference on
Decision and Control, vol. 1, 1999, pp. 749-54.
[97] Stilwell, D.J., "State-Space Interpolation for a Gain-Scheduled Autopilot," Journal of Guidance, Control, and
Dynamics, vol. 24, no. 3, pp. 460-465, 2001.
[98] Stilwell, D.J. and Rugh, W.J., "Stability Preserving Interpolation Methods for the Synthesis of Gain
Scheduled Controllers," Automatica, vol. 36, no. 5, pp. 665-671, 2000.
[99] Stoecker, W.F., "Stability of an Evaporator-Expansion Valve Control Loop," ASHRAE Transactions, vol. 72,
no. 2007, 1966.
[100] Stoustrup, J., Andersen, P., Pedersen, T.S., and Hangstrup, M.E., "Gain-Scheduling Procedures for Unstable
Controllers: a Behavioral Approach," in Proceedings of the 1999 American Control Conference, vol. 6, 1999,
pp. 4004-8.
[101] Stoustrup, J. and Blondel, V.D., "A Simultaneous Stabilization Approach to (Passive) Fault Tolerant
Control," in Proceedings of the 2004 American Control Conference, vol. 2, 2004, pp. 1817-22.
[102] Stoustrup, J. and Niemann, H., "Starting Up Unstable Multivariable Controllers Safely," in Proceedings of
the 36th IEEE Conference on Decision and Control, vol. 2, 1997, pp. 1437-8.
[103] Takagi, T. and Sugeno, M., "Fuzzy Identification of Systems and its Applications to Modelling and Control,"
IEEE Transactions on Systems, Man and Cybernetics, vol. SMC-15, no. 1, pp. 116-32, Jan, 1985.
[104] Tay, T.-T., Mareels, I., and Moore, J. B. High Performance Control, Ann Arbor, MI: Birkhauser, 1998.
232
[105] Tummescheit, H. and Eborn, J., "Design of a Thermo-Hydraulic Model Library in Modelica™," in
Simulation: Past, Present and Future. 12th European Simulation Multiconference, San Diego, CA, pp. 132-6.
[106] Van Overschee, P. and De Moor, B., "N4SID: Subspace Algorithms for the Identification of Combined
Deterministic-Stochastic Systems," Automatica, vol. 30, no. 1, pp. 75-93, January 1994.
[107] Viberg, M. , Wahlberg, B., and Ottersten, B., "Analysis of State Space System Identification Methods Based
on Instrumental Variables and Subspace Fitting," Automatica, vol. 33, no. 9, pp. 1603-1616, Sep, 1997.
[108] Wedekind, G.L., "An Experimental Investigation Into the Oscillatory Motion of the Mixture-Vapor Transition
Point in Horizontal Evaporating Flow," ASME Journal of Heat Transfer, vol. 93, no. 1, pp. 47-54, Feb, 1971.
[109] Wedekind, G.L., Bhatt, B.L., and Beck, B.T., "A System Mean Void Fraction Model for Predicting Various
Transient Phenomena Associated With Two-Phase Evaporating and Condensing Flows," International
Journal of Multiphase Flow, vol. 4, no. 1, pp. 97-114, Mar, 1978.
[110] Wedekind, G.L. and Stoecker, W.F., "Transient Response of the Mixture-Vapor Transition Point in
Horizontal Evaporating Flow," ASHRAE Transactions, vol. 72, no. 1, 1966.
[111] Willatzen, M., Pettit, N.B.O.L., and Ploug-Sorensen, L., "General Dynamic Simulation Model for Evaporators
and Condensers in Refrigeration. Part I: Moving-Boundary Formulation of Two-Phase Flows with Heat
Exchange," International Journal of Refrigeration, vol. 21, no. 5, pp. 398-403, Aug, 1998.
[112] Wilson, M.J., Newell, T.A., and Chato, J.C., "Experimental Investigation of Void Fraction during Horizontal
Flow in Larger Diameter Refrigeration Applications," Air Conditioning and Refrigeration Center, University
of Illinois at Urbana Champaign, ACRC TR-140, Jul 1998.
[113] Wu, F., "A Generalized LPV System Analysis and Control Synthesis Framework," International Journal of
Control, vol. 74, no. 7, pp. 745-759, 2001.
[114] Youla, D.C., Bongiorno, J.J.Jr., and Jabr, H.A., "Modern Wiener-Hopf Design of Optimal Controllers Part I:
The Single-Input-Output Case," IEEE Transactions on Automatic Control, vol. AC-21, no. 1, pp. 13-3, 1976.
[115] Youla, D.C., Jabr, H.A., and Bongiorno, J.J.Jr., "Modern Wiener-Hopf Design of Optimal Controllers Part II:
The Multivariable Case," IEEE Transactions on Automatic Control, vol. AC-21, no. 3, pp. 338-319, 1976.
[116] Zahn, W.R., A Visual Study of Two-Phase Flow While Evaporating in Horizontal Tubes ASME Paper No.
63-WA-166, 1963.
[117] Zhang, R., Alleyne, A., and Carter, D.E., "Generalized Multivariable Gain Scheduling with Robust Stability
Analysis," ASME Journal of Dynamic Systems, Measurement, and Control, vol. 27, Dec, 2005.
[118] Zhijiu, C. , Ruiqi, Z., Yezheng, W., and Chen, W., "Experimental Investigation of a Minimum Stable
Superheat Control System of an Evaporator," International Journal of Refrigeration, vol. 25, no. 8, pp. 1137-
1142, 2002.
[119] Zhou, K., Doyle, J. C., and Glover, K. Robust and Optimal Control, New Jersey: Prentice Hall, 1996.
233

Dynamic Modelling and Advanced Control of Air Conditioning and Refrigeration Systems

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dynamic Modelling and Advanced Control of Air Conditioning and Refrigeration Systems

Uploaded by

Copyright:

Available Formats

University of Illinois at Urbana-Champaign

Dynamic Modeling and Advanced Control

ACRC TR-244 June 2006

For additional information:

Air Conditioning and Refrigeration Center

For additional information:

Air Conditioning & Refrigeration Center

217 333 3115

1.1 Vapor Compression Systems

Liquid Two-Phase Vapor

Liquid Two-Phase Vapor

Gas Cooler Enthalpy

1.2 Control-Oriented Modeling of Vapor Compression Systems

1.3 Control of Vapor Compression Cycles

1.4 Gain Scheduling

1.5 Organization of Dissertation

2.2 Modeling Paradigms

2.2.1 Lumped Parameter Models

2.2.2 Discretized Models

2.2.3 Moving Boundary Models

2.3 Vapor Compression Systems

Liquid Two-Phase Vapor

Liquid Two-Phase Vapor

Figure 2.3 P-h Diagram - Actual Vapor Compression Cycle

Filter/Dryer Oil Seperator

Figure 2.6 System Diagram - Subcritical Vapor

2.5 Heat Exchangers

2.5.1 Moving Boundary Approach

2.5.1.1 Modeling Assumptions

2.5.1.2 Governing Partial Differential Equations

(C p ρA)w ∂(Tw ) = piα i (Tr − Tw ) + poα o (Ta − Tw ) ( 2.5 )

pi inner perimeter (interior surface area per unit length)

po outer perimeter (exterior surface area per unit length)

Tw tube wall temperature

m& mass flow rate of refrigerant flowing along the tubes

(C p ρA)w thermal capacitance of tube wall per unit length

2.5.1.3 Dynamic Conservation Equations

volume by a change in the moving boundary is negative, − PV& .

E& w = Q& w − Q& a − E& int ( 2.10 )

m& = m& in − m& out ( 2.11 )

Equation 2.14, and with the substitution δx = x − x o , becomes Equation 2.15.

⎡ x& ⎤ ⎡ A11 A12 ⎤ ⎡ x ⎤ ⎡ B1 ⎤

2.5.4 Example Derivation: Single Phase Heat Exchanger

2.5.4.1 Conservation of Refrigerant Mass

dz = Acs ⎡⎢ ( ρ c Ltotal )⎤⎥

∫ dz = Acs Ltotal [ ρ& c ] ( 2.34 )

∫ dz = Acs Ltotal ⎢⎜ ⎟h&c ⎥ ( 2.35 )

2.5.4.2 Conservation of Refrigerant Energy

∂( ρAcs h − Acs P ) ∂(m& h )

dz = Acs ⎡⎢ ( ρ c hc Ltotal )⎤⎥

∫ dz = Acs Ltotal [ρ& c hc + ρ c h&c ] ( 2.42 )

∫ dz = Acs Ltotal ⎢ ⎢⎜ ⎟h&c ⎥ hc + ρ c h&c ⎥ ( 2.43 )

∫ dz = ⎢⎜ ⎟hc + ρ c ⎥ Acs Ltotal h&c ( 2.44 )

+ m& out hout − m& in hin = α i Ai (Tw − Tr )

(C p ρA)w ∂(Tw ) = piα i (Tr − Tw ) + poα o (Ta − Tw ) ( 2.52 )

(C p ρA)w Ltotal ∂(Tw ) = pi Ltotalα i (Tr − Tw ) + po Ltotalα o (Ta − Tw ) ( 2.53 )

(C p ρV )w ∂(Tw ) = α i Ai (Tr − Tw ) + α o Ao (Ta − Tw ) ( 2.54 )

are given in Table 2.2.

⎡ z11 z12 0 ⎤ ⎡ P&c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤

Table 2.2 Matrix Elements of Z ( x, u ) for the Gas Cooler

U& = H& in − H& out − Q& w − W& ( 2.56 )

E& w = Q& w − Q& a − E& int ( 2.57 )

m& = m& in − m& out ( 2.58 )

⎡U& c ⎤ ⎡m& in hin − m& out hout − α i Ai (Tr − Tw )⎤