Clemson University

TigerPrints

All Dissertations Dissertations

5-2019

Evaporator Modeling and an Optimal Control Strategy

Development of an Organic Rankine Cycle Waste Heat Recovery
System for a Heavy Duty Diesel Engine Application
Dhruvang Rathod
Clemson University, [email protected]

Follow this and additional works at: https://tigerprints.clemson.edu/all_dissertations

Part of the Automotive Engineering Commons

Recommended Citation
Rathod, Dhruvang, "Evaporator Modeling and an Optimal Control Strategy Development of an Organic
Rankine Cycle Waste Heat Recovery System for a Heavy Duty Diesel Engine Application" (2019). All
Dissertations. 2704.
https://tigerprints.clemson.edu/all_dissertations/2704

This Dissertation is brought to you for free and open access by the Dissertations at TigerPrints. It has been
accepted for inclusion in All Dissertations by an authorized administrator of TigerPrints. For more information,
please contact [email protected].
EVAPORATOR MODELING AND AN OPTIMAL CONTROL STRATEGY
DEVELOPMENT OF AN ORGANIC RANKINE CYCLE WASTE
HEAT RECOVERY SYSTEM FOR A HEAVY DUTY
DIESEL ENGINE APPLICATION

A Dissertation
Presented to
the Graduate School of
Clemson University

In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy
Automotive Engineering

by
Dhruvang Rathod
May 2019

Accepted by:
Dr. Zoran Filipi, Committee Chair
Dr. Mark Hoffman
Dr. Ardalan Vahidi
Dr. Robert Prucka

i
 ABSTRACT

The Organic Rankine Cycle (ORC) has proven to be a promising technology for Waste

Heat Recovery (WHR) systems in heavy duty diesel engine applications. However, due to

the highly transient heat source, controlling the working fluid flow through the ORC

system and maximizing the heat recovery is a challenge for real time application. To that

end, this research resulted in the following main developments.

The first new development is in the area of heat exchanger modeling. The heat

exchanger is a key component within the WHR system and it governs the dynamics of the

complete ORC system. The heat exchanger model is extended using a thermal image data

to improve its phase length prediction capability. It’s shown that the new identified

empirical equations help improve the phase length estimation by 43% over a set of transient

experiments. As a result, the model can be used to develop an improved control oriented

moving boundary model and to provide insights into evaporator design.

The second new development is the advancement of the control design of an ORC

system. With advanced knowledge of the heat source dynamics, there is potential to

enhance power optimization from the WHR system through predictive optimal control.

The proposed approach in this this dissertation is a look-ahead control strategy where, the

future vehicle speed is predicted utilizing road topography and V2V connectivity. The

forecasted vehicle speed is utilized to predict the engine speed and torque, which facilitates

estimation of the engine exhaust conditions used in the ORC control model. In the

simulation study, a reference tracking controller is designed based on the Model Predictive

Control (MPC) methodology. Two variants of Non-linear MPC (NMPC) are evaluated: an

ii
NMPC with look-ahead exhaust conditions and a baseline NMPC without the knowledge

of future exhaust conditions. Simulation results show no particular improvement to

working fluid superheat tracking at the evaporator outlet via the look-ahead strategy for a

drive cycle. However, the look-ahead control strategy does provide a substantial reduction

in system control effort via dampening the heavily transient working fluid pump actuation,

enhancing pump longevity, health, and reducing pump power consumption. This reduction

in pump actuation helps the NMPC with preview to maintain the superheat lower than the

NMPC without this feature for certain frequency of the exhaust conditions. Overall, NMPC

with preview feature can help reduce parasitic losses, like pump power and improve power

generation.

The third development addresses the modeling errors and measurement inaccuracies

for NMPC implementation. NMPC is inherently a state feedback system and for that reason

an Extended Kalman Filter (EKF) is used to estimate unmeasurable states inside the ORC

evaporators based on exhaust gas and working fluid temperatures. Since it is not realistic

to expect that the system model will perfectly describe the behavior of the evaporator

dynamics in all operating conditions, the estimator is therefore augmented with a

disturbance model for offset free MPC tracking. Simulation study shows that the

augmented system is perfectly capable of discarding the model errors and rejecting the

measurement inaccuracies. Moreover, experimental validation confirms that no steady

state error is observed during online implementation of the augmented EKF.

Finally, experimental validation of the designed NMPC control strategy was

conducted. The performance of the NMPC was evaluated on a heavily transient drive cycle,

iii
as well as on a sinusoidal generated heat signals. Both experimental and simulated

sinusoidal exhaust condition shows that evaporator under consideration inherently helps

attenuate the fluctuating exhaust conditions due to its thermal inertia especially for heat

signals of shorter time periods. However for slow changing exhaust conditions, a slower

rate of change of working fluid flow helps in inhibiting temperature overshoot at the

evaporator outlet.

iv
 ACKNOWLEDGEMENT

This dissertation is a result of a team effort. First and for most, I would like to thank

Dr. Mark Hoffman for welcoming me to Clemson University and giving me this

opportunity to pursue my research goal. I am highly grateful towards the invaluable

discussions we had exploring the unknowns and the exceptional mentorship I received

during my time at ICAR. Those were some good times!

I am also highly thankful to my advisor Dr. Zoran Filipi for his willingness and

extraordinary abilities to explain any engineering issues. His unparalleled experience and

expertise were a great help in bringing this thesis to completion. I would also like to thank

Dr. Ardalan Vahidi for explaining all the controls concepts both in class and its application

in my thesis. I was very pleased towards his receptiveness for any questions and issues I

faced during controls implementation.

I am grateful to my colleague Dr. Bin Xu for the many inspiring discussions and

collaboration during last four years. I would also like to appreciate the efforts and technical

support put forward by Jeremy Barnes in helping me and other students in the test lab.

I would like to acknowledge Paul Anschel, Xiaobin (Shawn) Liu from BorgWarner

Inc. for their technical support and financial assistance during this project.

Lastly, I would like to express my immense gratitude towards my family (my father-

Bipinchandra Rathod, my mother- Nirmala Rathod and my brother- Jayendra Rathod) for

all their love and encouragement during these four years. They are the backbone of my

professional and educational growth. Without their support I wouldn’t be standing where I

am today. I cordially thank them for inspiring me every day in all aspects of life!

v
 TABLE OF CONTENTS
TITLE……………………………………………………………………………………...i

ABSTRACT ........................................................................................................................ ii

ACKNOWLEDGEMENT .................................................................................................. v

TABLE OF CONTENTS ................................................................................................... vi

LIST OF TABLES ............................................................................................................. ix

LIST OF FIGURES ............................................................................................................ x

NOMENCLATURE ......................................................................................................... xv

CHAPTER 1. INTRODUCTION ....................................................................................... 1

1.1 Organic Rankine cycle .......................................................................................................... 2
1.2 Open Loop Experimental analysis......................................................................................... 5
1.2.1 Step Change in Uncontrollable Input – Engine Conditions............................................ 6
1.2.2 Step Change in Controllable Input – Working Fluid Pump speed ................................. 8
1.3 Problem statement ............................................................................................................... 10
1.3 Objectives ............................................................................................................................ 12
1.4 Dissertation Outline ............................................................................................................. 12

CHAPTER 2. EVAPORATOR MODELING .................................................................. 14

2.1 Introduction ......................................................................................................................... 14
2.2 Background ......................................................................................................................... 14
2.2.1 Heat Exchanger Modeling ............................................................................................ 14
2.2.2 Importance of Phase length estimation......................................................................... 15
2.2.3 Sensitivity of ORC-WHR Energy Recovery to Working Fluid Phase Length ............. 16
2.3 System Configuration .......................................................................................................... 18
2.4 Thermal Image Processing .................................................................................................. 20

vi
 2.4.1 Still Image Processing .................................................................................................. 20
2.4.2 Video Processing .......................................................................................................... 21
2.4.3 Phase Change Detection ............................................................................................... 25
2.5 Evaporator Modeling ........................................................................................................... 26
2.5.1 Heat Transfer Coefficients ........................................................................................... 29
2.6 Sensitivity Analysis: FVM Discretization ........................................................................... 31
2.7 Enhancing FVM Phase Prediction Accuracy ...................................................................... 34
2.7.1 Manual Parameter Tuning ............................................................................................ 35
2.7.2 Particle Swarm Optimization Parameter Identification ................................................ 41
2.8 Results ................................................................................................................................. 46
2.8.1 Transient Test Sequence #1 .......................................................................................... 46
2.8.2 Transient Test Sequence #2 .......................................................................................... 52
2.9 Conclusions ......................................................................................................................... 54

CHAPTER 3. DYNAMIC CHARACTERIZATION OF EVAPORATOR ..................... 56

3.1 Introduction ......................................................................................................................... 56
3.2 Background ......................................................................................................................... 56
3.3 Evaporator Thermal Inertia ................................................................................................. 57
3.4 Working fluid properties ..................................................................................................... 61
3.5 ORC System dynamics evaluation ...................................................................................... 62
3.5.1 Uncontrollable Inputs: Exhaust Conditions .................................................................. 64
3.5.2 Controllable Inputs: Working fluid mass flow ............................................................. 66
3.6 Conclusion ........................................................................................................................... 68

CHAPTER 4. CONTROL DESIGN ................................................................................. 69

4.1 Introduction ......................................................................................................................... 69
4.2 Background ......................................................................................................................... 69
4.2.1 Look-ahead ORC control ............................................................................................. 70
4.2.2 Opportunity and Uniqueness of the current work......................................................... 71
4.3 Moving Boundary Control Model ....................................................................................... 72
4.4 NMPC Problem Formulation .............................................................................................. 76
4.5.1 ACADO Implementation.............................................................................................. 77
4.6 Disturbance Model and Observer Design............................................................................ 78

vii
 4.7 Simulation Results ............................................................................................................... 79
4.7.1 NMPC Cost function formulation ................................................................................ 80
4.7.2 Disturbance Rejection .................................................................................................. 84
4.7.3 Look-ahead NMPC ....................................................................................................... 88
4.8 Conclusions ....................................................................................................................... 100

CHAPTER 5. NMPC EXPERIMENTAL VALIDATION ............................................ 102

5.1 Introduction ....................................................................................................................... 102
5.2 Background ....................................................................................................................... 102
5.3 Experimental Set-up .......................................................................................................... 104
5.4 Online Model Calibration .................................................................................................. 105
5.4.1 EKF state estimation error .......................................................................................... 107
5.5 Ramp-up and Ramp-down Events ..................................................................................... 110
5.6 Drive cycle ........................................................................................................................ 112
5.7 Sinusoidal Inputs ............................................................................................................... 114
5.7.1 Time period of 20 s..................................................................................................... 115
5.7.2 Time period of 60 s..................................................................................................... 116
5.7.3 Time period of 120 s................................................................................................... 118
5.8 Conclusions ....................................................................................................................... 120

Chapter 6. Conclusion, Contributions and Future Work ................................................ 122

6.1 Conclusions and Contributions.......................................................................................... 122
6.2 Future Work ...................................................................................................................... 125

APPENDIX ..................................................................................................................... 127

REFERENCES ............................................................................................................... 128

PUBLICATIONS ............................................................................................................ 136

viii
 LIST OF TABLES

Table 2. 1. Steady state experimental conditions for evaluating FVM discretization. ..... 31

Table 2. 2. Error summary for different levels of discretization of the FVM model........ 33

Table 2. 3. Analysis the effects of additional multipliers on phase lengths for test conditions
in Table 1. [Negative error means under-estimation and positive error is over-estimation]
........................................................................................................................................... 36

Table 2. 4. Summary of Sensitivity analysis of different PSO targets. Negative error in the
above Table indicates under-estimation whereas positive error corresponds to over-
estimation. ......................................................................................................................... 43

Table 2. 5. PSO identified variables for Case #5 in Table 2.4. ......................................... 44

Table 2. 6. Comparison of the phase length estimation between baseline FVM model and
calibrated FVM model. ..................................................................................................... 49

Table 3. 1. Summary of the control objective of the ORC-WHR system. ....................... 64

Table 3. 2. Exhaust conditions for sensitivity analysis on the uncontrollable inputs ....... 64

Table 4. 1. Quantified NMPC temperature error comparison with and without preview
capability for the drive cycle in Figure 4.9. ...................................................................... 92

Table 4. 2. NMPC temperature error for different preview times of Figure 4.10. ........... 94

Table 4. 3. NMPC temperature error comparison for different time periods of the sinusoidal
input of Figure 4.11........................................................................................................... 97

Table 4. 4. Percentage reduction in control input amplitude for the look-ahead NMPC
(‘with preview’) for different time periods of the sinusoidal input in Figure 4.11. .......... 98

Table 5. 1. NMPC superheat tracking error for different time period sinusoidal heat signal.
......................................................................................................................................... 120

ix
 LIST OF FIGURES

Figure 1. 1. Performance indicators of each waste heat recovery technology for an

automotive application (5 being the best) [8] ..................................................................... 2

Figure 1. 2. Schematic of ORC-WHR system coupled with Engine. (x – Vapor fraction) 3

Figure 1. 3. T-S diagram for Rankine cycle........................................................................ 4

Figure 1. 4. (a) Engine condition for a step change in torque and (b) Exhaust condition and
working fluid temperature response at the evaporator outlet for a step change in engine
torque .................................................................................................................................. 7

Figure 1. 5. Working fluid evaporator outlet temperature response to the step-change in

pump-actuator position. ...................................................................................................... 9

Figure 2. 1. Vapor phase length control diagram.............................................................. 16

Figure 2. 2. Organic Rankine cycle waste heat recovery system power comparison between
two working fluid boundary layer estimation accuracies: (a) vapor phase length over the
total boiler length as a percentage, and (b) normalized turbine power and normalized
cumulative energy. In the legends, 7% model error means the results from a simulation
subject to a vapor phase length estimation error of 7% prior to the feedback control. In
contrast, 2% model error means the results from a simulation with a 2% vapor phase length
estimation error prior to the feedback control................................................................... 18

Figure 2. 3. Evaporator design schematic. ........................................................................ 19

Figure 2. 4. An example of the Thermal Image from FLIR A300 camera utilizing an 18mm
lens displaying the line chosen for detailed temperature analysis from the thermal images.
In consideration of the sponsor’s intellectual property rights, some features of the
evaporator have been intentionally blurred....................................................................... 19

Figure 2. 5. Comparison of RGB intensities on each point on the temperature detection

line..................................................................................................................................... 22

Figure 2. 6. Temperature comparison data between FLIR tools and Matlab code operating
on the same still thermal image. ....................................................................................... 24

x
Figure 2. 7. Extracted temperatures along the top surface of the evaporator (along the
defined line of interest shown in Figure 2.4) with respect to normalized working fluid tube
length in the evaporator. Also shown are the three regions of interest (A-C), corresponding
to the different working fluid phases. ............................................................................... 24

Figure 2. 8. Schematic representation of Finite Volume Method used for evaporator

modeling. .......................................................................................................................... 28

Figure 2. 9. Working Fluid temperature across the length of evaporator tube for different
FVM discretization. .......................................................................................................... 32

Figure 2. 10. Comparison of simulated working fluid temperature from FVM model 500
cells discretization and corresponding thermal image. ..................................................... 33

Figure 2. 11. Effects of varying Uvap values on phase lengths and outlet temperature. .... 37

Figure 2. 12. Effects of varying Uliq values on phase lengths and outlet temperature...... 38

Figure 2. 13. Effects of varying Umix,1 and Umix,2 values on phase lengths and outlet
temperature. ...................................................................................................................... 39

Figure 2. 14. Effects of varying Umix,1 and Umix,2 values on heat transfer coefficient. ..... 40

Figure 2. 15. Comparison of working fluid heat transfer coefficient from the FVM baseline
model and the FVM with PSO identified values. ............................................................. 44

Figure 2. 16. Comparison of working fluid temperature along the length of the evaporator
from the FVM baseline model and the FVM with PSO identified values. ....................... 46

Figure 2. 17. Exhaust and Working Fluid conditions for Test 1....................................... 47

Figure 2. 18. Baseline FVM simulation results for the evaporator outlet temperatures for
Test 1................................................................................................................................. 48

Figure 2. 19. Phase length comparison between the baseline FVM model and experimental
thermal video for Test sequence #1. ................................................................................. 48

Figure 2. 20. Phase length comparison between the enhanced FVM and the experimental
thermal video for Test #1. ................................................................................................. 50

Figure 2. 21. Phase length error comparison for the baseline and enhanced FVMs relative
to the thermal imaging results during transient Test 1. ..................................................... 51

xi
Figure 2. 22. Simulated working fluid and exhaust gas outlet temperature results from the
enhanced FVM plotted relative to the experimental values for transient Test #1. ........... 52

Figure 2. 23. Exhaust and Working Fluid conditions for Test 2....................................... 53

Figure 2. 24. Phase lengths comparison between the enhanced FVM model and the thermal
videos for Transient Test #2. ............................................................................................ 53

Figure 2. 25. Simulated working fluid and exhaust outlet temperatures using the enhanced
FVM plotted relative to the experimental values during transient Test #2....................... 54

Figure 3. 1. Typical TP and EGR exhaust conditions....................................................... 57

Figure 3. 2. Evaporator size comparison .......................................................................... 58

Figure 3. 3. Thermal response of TP and EGR evaporators to sinusoidal exhaust conditions

with varying time period, T .............................................................................................. 59

Figure 3. 4. Summary of thermal response of TP and EGR evaporator. .......................... 60

Figure 3. 5. (a) ORC system pressure as a linear function of Working fluid mass flow (b)
Saturation temperature vs system pressure correlation for ethanol as working fluid ....... 61

Figure 3. 6. Open loop plant model with inputs, outputs and control objective. .............. 63

Figure 3. 7. Open loop simulation to characterize ORC system time constant for step
changes in exhaust mass flow and temperature. ............................................................... 65

Figure 3. 8. Open loop simulation to characterize ORC system time constant for step
changes in working fluid flow. ......................................................................................... 67

Figure 4. 1. Schematic representation of Moving Boundary Model used for evaporator

control-oriented modeling (counter flow). ........................................................................ 73

Figure 4. 2. Generated sinusoidal exhaust mass flow with varying time periods to test the
NMPC controller............................................................................................................... 80

Figure 4. 3. Power maximization MPC performance for the exhaust condition shown in
Figure 4.2. Plot (a) shows the response of the working fluid temperature at the evaporator
outlet to different sinusoidal inputs with the superheat plot shown in the bottom graph. Plot

xii
(b) shows the controller generated control input and corresponding response of the
saturation pressure. ........................................................................................................... 82

Figure 4. 4. Plot (a) Superheat tracking MPC performance for the exhaust condition shown
in Figure 4.2. Plot (b) shows the controller generated control input and corresponding
response of the saturation pressure. .................................................................................. 84

Figure 4. 5. Step sequence of the exhaust test conditions for plant and control model
mismatch comparison ....................................................................................................... 85

Figure 4. 6. NMPC performance comparison with and without the augmented disturbance
model during a period of mismatch between the plant and control models. .................... 86

Figure 4. 7. Disturbance augmented NMPC performance comparison for various constant

temperature offsets between plant and control model. ..................................................... 88

Figure 4. 8. Proposed control law for the ORC-WHR system.......................................... 89

Figure 4. 9. HDDE Drive Cycle exhaust test conditions for NMPC analysis .................. 90

Figure 4. 10. Look-ahead NMPC performance (‘with preview) compared to the baseline
NMPC (‘without preview’) during a drive cycle. ............................................................. 91

Figure 4. 11. Look-ahead NMPC performance comparison with different preview window
sizes (total preview time). ................................................................................................. 93

Figure 4. 12. Sinusoidal exhaust test conditions for comparison of the look ahead and
baseline NMPC. ................................................................................................................ 95

Figure 4. 13. Performance comparison for the look-ahead NMPC (‘with preveiw’) and the
baseline NMPC (‘without preview’) while subjected to the sinusoidal heat source. H p=60
s. SH Reference = 20˚C .................................................................................................... 96

Figure 4. 14. Performance comparison for the look-ahead NMPC (‘with preveiw’) and the
baseline NMPC (‘without preview’) while subjected to the sinusoidal heat source. H p=60
s. SH Reference = 20˚C .................................................................................................... 99

Figure 5. 1. Experimental set-up showing the Engine, after-treatment system and the
ORC rig in a transient capable heavy duty engine dynamometer at Clemson University’s
Automotive research facility. .......................................................................................... 104

Figure 5. 2. (a) AVL’s PUMA and INCA data collection interface (b) DSPACE
MicroAutoBox for implementing the NMPC algorithm .................................................. 105

xiii
Figure 5. 3. System aging characterization for the same engine conditions with data set 1
being the green ORC system and data set 2 representing the aged ORC system. .......... 106

Figure 5. 4. Online tuning of heat transfer coefficients for MBM of TP evaporator

resulted in stable state estimation by the EKF ................................................................ 108

Figure 5. 5. Proposed control law (Figure 4.8) incorporated with system aging adaptation
variable, 𝜀(𝑡). .................................................................................................................. 109

Figure 5. 6. Tuned NMPC performance for a 30s ramp in engine conditions. ............... 110

Figure 5. 7. Tuned NMPC performance for a ramp input of rise time 5 sec .................. 111

Figure 5. 8. (a) Engine conditions for the constant speed drive cycle (b) Tuned NMPC
superheat tracking performance and generated working fluid flow rate ........................ 113

Figure 5. 9. NMPC Computation time for the constant speed drive cycle ..................... 114

Figure 5. 10. (a) Engine conditions for the sinusoidal cycle with time period of 20s (b)
Tuned NMPC superheat tracking performance and generated working fluid flow rate,
𝑊𝑑𝑢=10 .......................................................................................................................... 116

Figure 5. 11. (a) Engine conditions for the sinusoidal cycle with time period of 60s (b)
Tuned NMPC superheat tracking performance and generated working fluid flow rate,
𝑊𝑑𝑢=10 .......................................................................................................................... 117

Figure 5. 12. (a) Engine conditions for the sinusoidal cycle with time period of 120s (b)
Tuned NMPC superheat tracking performance comparison with weights, 𝑊𝑑𝑢=10 and
𝑊𝑑𝑢=60 with generated working fluid flow rate ........................................................... 119

xiv
 NOMENCLATURE

𝑎 transverse pitch ratio

A area [m2]
𝑏 longitudinal pitch ratio
B Blue
𝑐 heat capacity [J/kg∙K]
𝐶 constant of two-phase multiplier correlation
𝑑 diameter [m]
𝑓 friction factor
F force [N]
𝐺 Green
h enthalpy [J/kg]
𝐻 height [m]
𝐻̇ enthalpy flowrate [J/s]
k Thermal conductivity
𝑙, 𝐿 length [m]
m mass [kg]
𝑚̇ mass flow rate [kg/s]
p pressure [Pa]
t time [s]
T temperature [K]
u velocity [m/s], internal energy [J/kg]
U heat transfer coefficient [J/kg∙K]
𝜈 flow velocity [m/s]
𝜐 kinematic viscosity [m2/s]
𝑣 volume [m3]
𝑥 vapor quality
z space coordinate [m]
𝜓 void fraction
𝑅 Red
𝑅𝑒 Reynolds number
𝑃𝑟 Prandtle number
𝑁𝑢 Nusselt number
𝛾 specific heat ratio
𝜌 density [kg/m3]
𝜕 partial derivative operator
𝜉 friction factor

Abbreviations
ORC organic Rankine cycle
DOE Department of Energy (USA)

xv
WHR waste heat recovery
HDD heavy duty diesel
MBM moving boundary method
FVM finite volume method
TP tail pipe
EGR exhaust gas recirculation
HTC heat transfer coefficient
FLIR Forward looking Infrared

Subscripts and superscripts

f working fluid
w wall
e exhaust gas
v vapor
l liquid
i ith discretized cell
in inlet/ upstream
out outlet
η efficiency
M,m mix point
V vapor point
𝑝 pressure
𝑓𝑟 friction
𝑔 gravitation
𝑣𝑎𝑝 saturated vapor
𝑠𝑎𝑡 saturated liquid
𝑡𝑝 two phase
𝑠 single phase
𝑈 heat transfer coefficient
𝑠𝑖𝑚 simulation
𝑒𝑥𝑝 experimental
𝑇𝑃 tail pipe
𝐸𝐺𝑅 exhaust gas recirculation
𝑐𝑟𝑜𝑠𝑠 cross or sectional surface

xvi
 CHAPTER 1. INTRODUCTION

The trucking industry is the lifeblood of the US (United States) economy. In a survey

conducted by University of Michigan and in cooperation by ATRI (American Transportation

Research Institute), with a total of 31 million commercial registered trucks, the total fuel consumed

in 2015 was 54 billion gallons [1]. Heavy duty trucks which are powered by diesel fuel, accounted

for 75% of this fuel consumption and represents the 2nd highest expense of total operating costs.

The other side-effect of fuel consumption is the diesel exhaust emissions which are detrimental to

both environment and human health. With the growing economy and consumer demand these

numbers are projected to increase every year. Therefore, in response to a Presidential

Memorandum, EPA in coordination with NHTSA issued greenhouse gas emissions and fuel

economy standards for heavy duty trucks [2]. The ultimate goal of this proposed policy was to

demonstrate 55% or greater brake thermal efficiency engine in the SuperTruck II program funded

by Department of Energy (DOE).

Since the United States DOE super truck program commenced in 2010, companies and

research institutions have continually explored a wide array of cutting-edge technologies to

improve HDD (Heavy Duty Diesel) engine efficiency. Waste heat recovery (WHR) has proven to

be a crucial technique in improving the fuel economy [3-4]. Since exhaust heat energy comprises

nearly 45% of the total fuel energy [7], WHR techniques have become top contenders for reducing

fuel consumption and CO2 emissions. Some research studies produced fuel economy

improvements of 4-8% when utilizing exhaust energy recovery [5-6]. One proven WHR technique

is the utilization of an Organic Rankine Cycle (ORC).

1
 (a) (b)
Figure 1. 1. Performance indicators of each waste heat recovery technology for an automotive
application (5 being the best) [8]

Figure 1.1 shows that turbo-compounding and Rankine cycle are most profitable WHR systems

to be most likely feasible in an automotive application. Although the weight to power ratio of

turbo-compounding system is favorable, the higher back pressure reduces the overall engine +

WHR system efficiency. Utilizing one or more heat sources for Rankine cycle [5, 9-10] will help

improve weight to power ratio. Other research works also demonstrates high thermal efficiency of

Rankine cycle in comparison with other waste heat recovery techniques [11-12]. These results are

encouraging and provide impetus towards increasing research efforts in investigating feasible

control strategies towards ORC implementation.

1.1 Organic Rankine cycle

ORC operation is similar to the stationary Rankine cycles used in power plants, except that the

working fluid in ORC is an organic fluid rather than water. Organic working fluids typically have

low boiling points, making them suitable for low-grade (a low temperature source) waste heat

recovery. ORC-WHR is adaptable to the multiple heat sources available from engines, namely: the

2
tailpipe (TP) exhaust gas, the exhaust gas recirculation (EGR) circuit, the charge air cooler (CAC)

and the engine coolant [9-10, 16].

The main components of a typical ORC system are an evaporator, condenser, pump, expansion

tank and an expander. Figure 1.2 shows an ORC-WHR example implemented on a HDD engine

using TP gas as heat source.

Figure 1. 2. Schematic of ORC-WHR system coupled with Engine. (x – Vapor fraction)

As shown in Figure 1.2, the high pressure pump continuously circulates working fluid through

the WHR circuit. The process 1-2 represents the work consumed by the pump, 𝑊 . The working

fluid undergoes a phase change inside the evaporator (process 2-3) and exits in vapor phase (at

point 3) after extracting heat from the exhaust gases. The high pressure working fluid vapor then

expands inside the turbine (process 3-4), where the heat energy is converted to electrical energy to

be utilized as needed, 𝑊 . Finally, the working fluid flows from the turbine to a condenser where

it changes back to liquid phase (process 4-1). This cycle is then repeated for continuous operation.

Utilization of a turbine bypass valve ensures safe turbine operation by re-directing liquid working

3
fluid straight to the condenser. From T-S diagram of the process 1-4 as described above, it can be

concluded that work input into the cycle, 𝑊 is always less than work output, 𝑊 generating a

positive net work across the cycle, an important characteristic of Rankine cycle.

Figure 1. 3. T-S diagram for Rankine cycle

There are some conditions and requirements associated with the operation of the ORC system

as stated. 1) Superheating is required for safe turbine operation and thus dictates minimum

temperature requirement for the cycle. 2) The working fluid in a Rankine cycle follows a closed

loop and is reused constantly, so it’s important to maintain its integrity and thermal capacity for as

long as possible. 3) It should be noted that the working fluid deteriorates at high temperature and

therefore dictates maximum temperature limit for the ORC.

ORC systems have been explored as early as 1970’s mostly for the small scale applications

with overall system efficiency less than 10% [13-15]. Since then ORC systems have been applied

to multiple stationary applications such as biomass heat and power recovery, geo-thermal heat

recovery, solar-energy collector and waste heat recovery from industrial processes. While the

4
Rankine cycle has a long history in stationary applications, its implementation in the automotive

field has additional barriers to adaptation. The transient nature of heat sources, tight packaging

constraints, system weight, safety and environmental issues, limited (and varying) cooling

capacity, and the cost-conscious nature of the HDD industry has created challenges for ORC-WHR

implementation. Advanced control design is a technology enabler for efficient and safe ORC

system operation in an automotive application.

One of the biggest challenges of HDD engine ORC system implementation is the transient

nature of the various heat sources. As the engine encounters different speed/load transients during

real-world operation, there are abrupt yet disparate mass flow rate and temperature changes in both

the TP and EGR gas streams. This transient behavior exposes many control challenges, which are

difficult to address with traditional feedback control while simultaneously attempting to minimize

system expense and component complexity. Most advanced control strategies require component

and system models, emphasizing the importance of accurate ORC component modeling.

1.2 Open Loop Experimental analysis

It is important to understand the open loop system dynamics before designing a control

solution. An open loop experimental evaluation is performed in this section to understand the

different time constants involved for the ORC system in consideration. For an ORC-WHR system,

the engine exhaust gas mass flow rate and temperature are uncontrollable inputs whereas pump

speed is the controllable input that maintains the working fluid flow through the ORC system. The

effects of step changes to controllable and uncontrollable inputs on working fluid temperature are

analyzed herein to elucidate the critical system dynamics. The experiments are conducted on a 13

L Heavy Duty Diesel engine in AVL’s engine dynamometer. AVL’s PUMA software is used to

5
record temperature and mass flow data of the ORC system and of engine exhaust conditions. The

experimental setup is shown in Figure 1.2. For more information refer to section 5.3. The TP

evaporator is placed downstream of the after-treatment system for two reasons; 1) to capitalize on

thermal energy from the reactions within the after-treatment system, and 2) to avoid interfering

with the after-treatment system functionality. Thermocouples and pressure transducer are placed

at each of the locations 1, 2, 3 and 4 as shown in Figure 1.2. Coriolis mass flow meters are located

at junction 2 and are used to measure the working fluid flow going into the evaporator.

1.2.1 Step Change in Uncontrollable Input – Engine Conditions

(a)

6
 (b)
Figure 1. 4. (a) Engine condition for a step change in torque and (b) corresponding exhaust conditions at
the evaporator inlet exhaust power at the evaporator inlet, and working fluid temperature response at the
evaporator outlet.

In this test, a step change in engine torque is evaluated at constant engine speed. Figure 1.4(a)

shows the engine conditions and the top plot of Figure 1.4(b) shows the corresponding exhaust

conditions. As seen from the Figure 1.4, the change in engine exhaust gas mass flow occurs rapidly

whereas the exhaust temperature change is temporally delayed due to the thermal inertia of the

after-treatment system located between the engine and the evaporator. After every change in

engine conditions, the thermal inertia of the after-treatment system takes a finite amount of time

to arrive at steady state.

7
 The bottom graph of Figure 1.4 (b) shows the corresponding change in working fluid

temperature at the evaporator outlet with a settling time of 650 sec. This slow settling time of the

working fluid temperature at evaporator outlet is a function of the after-treatment systems thermal

inertia. Exhaust power is a function of exhaust flow and temperature. At 150 s, when the step

change in engine torque is initiated, the exhaust power at the evaporator inlet changes

instantaneously due to the fast dynamics of the exhaust mass flow rate, represented by region A in

figure 1.4 (b). The change in exhaust gas mass flow rate shows up as a linear trend in the working

fluid outlet temperature during this region. Region A persist for 100 s after which the evaporator

inlet exhaust temperature begins to rise, increasing the exhaust power in region B.

Correspondingly, the rate of working fluid temperature rise increases.

The heat flow in the evaporator is a two-step process, first the heat source transfers energy to

the tube wall and then, from the tube wall, the heat is transferred to the working fluid. In the first

step of the heat transfer process, thermal inertia, i.e. the wall thickness, evaporator design and

selected wall material, along with the exhaust conditions affects the rate of heat transfer process.

For a given evaporator design, this rate will change based on the engine operating condition. The

heat transfer rate in the second heat transfer step, from the wall to the working fluid, is dictated by

the working fluid type, mass flow rate, and phase. Figure 1.4 (b) shows that the working fluid

temperature dynamics are a function of both after-treatment thermal inertia and the evaporator

thermal inertia. The rate of heat transfer in the evaporator can be manipulated by adjusting the

pump speed as shown in the following section.

1.2.2 Step Change in Controllable Input – Working Fluid Pump speed

In this test, engine conditions were kept constant while the working fluid pump actuator

position was subjected to step changes as seen from Figure 1.5. For each 3% step change in

8
actuator position, the pump speed change by 100 rpm, directly altering the working fluid mass

flow rate. During this experiment, the turbine bypass valve is completely closed and turbine valve

is completely opened. These fixed valve position allowed the evaporation pressure to change with

the working fluid mass flow rate.

Figure 1. 5. Working fluid evaporator outlet temperature response to the step-change in pump-actuator
position.

The initial working fluid step change at 100s stimulated a slow working fluid temperature

response, steady state at around 400 s as shown in Figure 1.5. The second 3% high speed pump

step change created a faster working fluid temperature response, eventually resulting in working

fluid saturation. The fixed valve position operation sets in motion a chain of events, which causes

this disparate response. The evaporation pressure varies proportionally with the working fluid mass

flow which in turn alters the evaporation temperature proportionally. These compounding effects

9
create the disparate severity in response to the two working fluid mass flow changes. Thus to avoid

saturation, control systems must take care in defining both the operating region and the rate of

control input change. This behavior also illustrates the non-linear nature of the ORC system, as

the system response varies with operating condition and the working fluid sate.

From Figure 1.4 and 1.5, it can be also concluded that the input dynamics (pump speed/working

fluid flow and evaporation pressure) are faster than the output dynamics (working fluid

temperature).

1.3 Problem statement

Rankine cycles are very well established and in use for numerous stationary application where

there are little concerns of any transients interrupting the WHR system operation. However, for

any WHR-ORC system in an automotive application, the heat source is highly dynamic and

therefore the control of the ORC system should be carefully structured. The control strategy should

not only be able to deal with the highly dynamic heat source but should also consider the actuator

physical constraints and working fluid temperature constraints without inhibiting turbine

operation. Finally, the overarching goal is to maximize the power output, i.e. waste heat recovery,

over a typical drive cycle. The aforementioned controls requirement clearly falls within the scope

of Model Predictive Control. In comparison to classical PID strategy, utilizing an MPC strategy

has following advantages:

1) It can save significant calibration and tuning resources especially if one or more evaporator

are utilized in the ORC system. In such cases, MPC needs minor updates to objective

function and constraints.

2) It is perfectly capable of handling multivariable problems

3) It explicitly considers constraints in the optimization process

10
 4) It is also capable of rejecting disturbance or modeling error by utilizing state feedback at

each step.

5) Lastly, the preview feature of MPC takes into account the future inputs and optimize the

control sequence for the entire prediction horizon.

Many studies have compared the classical PID controller which is the state of the art in the

automotive industry and MPC, an advanced model based control technique [17-22]. In all these

studies, it was shown that MPC outperforms PID controller in maintaining the vapor state at the

evaporator outlet in a transient driving conditions. An opportunity for further improvement is

anticipated in the context of vehicle connectivity and look-ahead strategies that rely on a known

horizon. Thus improving the ORC systems power generation capability.

The development of MPC can be found as early as 1970s. Since then multiple MPC techniques

have being developed that uses a model based approach to predict the future control inputs. Some

authors have exploited EPSAC [17-18] as an MPC technique to improvise the power production

from an ORC system. The advantage of using EPSAC is that it allows the use of input-output

models in form of transfer function thus avoiding the need of state estimators. This also limits the

usage of that transfer function within the identified range of the model. While others [19, 23] have

used linear MPC where the non-linear evaporator model is linearized over a predefine set of points.

The drawbacks of linear MPC is that it requires a different linear model for different operating

conditions. Tuning, each linearized MPC would be additional challenges for smooth switching

between multiple models. However, very few studies have used nonlinear MPC with state

estimators.

In summary, the main goal of this research is to investigate a Non-linear MPC (NMPC) with

augmented EKF (Extended Kalman Filter) as state estimator. The scope of the sub-tasks is derived

11
from the main objective, i.e. the additional chapters will address development of enabling

modeling capabilities, error analysis based on model/measurement inaccuracies, as well as

experimental validation of the proposed control approach. The EKF is augmented with a

disturbance model to reject both modeling and measurement errors. Since MPC is model based

approach, first an enhanced finite volume modeling methodology is presented that enhances phase

length estimation inside the evaporator. This model can be then used to calibrate and tune control

oriented model.

1.3 Objectives

The main objectives of this dissertation are to develop:

1. Enhanced evaporator model for estimating phase lengths - A methodology is described and

validated to predict phase lengths inside the evaporator. (Chapter 2)

2. Augmented EKF model - The estimator is augmented with disturbance rejection model to

take into account modeling errors and instrumentation inaccuracies (Chapter 4). The

performance of which is experimentally validated in Chapter 5.

3. MPC with preview - The benefits of MPC with preview capability is explored where future

exhaust conditions are known. MPC with preview showed benefits in reducing the control

effort and in some instances better superheat tracking ability. (Chapter 4)

1.4 Dissertation Outline

Chapter 2, explains the evaporator modeling and introduces methodology to enhance the phase

length estimation undergoing beneath the surface of the evaporator. This methodology uses PSO

(Particle Swarm Optimization) technique for identification of additional multipliers to improve the

heat transfer coefficient. The identified multipliers and methodology is validated using transient

thermal image data.

12
 In Chapter 3, following the experimental open loop study, efforts have been made to

understand the open loop characteristics of the actuators, inputs-output correlation and the thermal

inertia affecting the evaporator dynamics in a simulation environment.

In Chapter 4, MPC control strategy is explained with the problem formulation. For model

based controls approach, it’s very crucial to model the system as accurate as possible. Sometimes

it’s extremely hard to model the actual system and in such cases, a disturbance model can be

incorporated in an MPC controls strategy. This is also explored and explained in Chapter 4. One

other important aspect which has not being investigated in literature for a WHR-ORC is its preview

capability. With known future inputs MPC can provide much better results from an optimization

stand-point.

Chapter 5, experimentally validates the developed MPC strategy for a TP evaporator. Step

inputs and drive cycles are used to tune and evaluate the performance of the MPC strategy.

Sinusoidal drive cycle with varying time period suggest that the thermal inertia plays a vital role

in buffering some of the high frequency exhaust conditions. Finally, the thesis ends with

conclusions, and future work in Chapter 6.

13
 CHAPTER 2. EVAPORATOR MODELING

2.1 Introduction

This Chapter presents the modeling and validation of a counter-crossflow heat exchanger used

to extract thermal energy from Heavy-Duty Diesel (HDD) engine exhaust. The finite volume

evaporator modeling methodology is enhanced for both accurate working fluid temperature and

phase length estimation, facilitating improved offline waste heat recovery simulation and accurate

control-oriented model development.

Transient model calibration and validation experiments were performed on a stand-alone flow

bench. Heated gas was passed through the evaporator, replicating different engine exhaust gas

conditions. In contrast to other studies, thermal imaging data served to identify the working fluid

liquid, mixed and vapor phase lengths within the evaporator. The FVM modeling methodology

was enhanced based on the thermal imaging data to accurately predict the working fluid phase

lengths. Once calibrated, working fluid phase lengths predicted by the proposed model were

validated against thermal imaging from additional transient experimental flow bench data sets.

2.2 Background

2.2.1 Heat Exchanger Modeling

Literature studies reveal two approaches for heat exchanger modelling, 1) a Finite Volume

Model (FVM) and 2) a Moving Boundary Model (MBM). The FVM is a discretized model,

dividing the evaporator into multiple uniform cells and solving the energy balance equations in

each cell. Such an approach yields high accuracy at the expense of computation speed [24-26].

FVM discretization depends on the amount of accuracy and computation time required for a given

application. Numerous published studies [20, 25-27] have utilized FVM for both stationary and

on-road applications.

14
 MBM are generally implemented for control purposes because of their low computation cost

[28-30]. Unlike an FVM, the MBM solves the energy balance equations in only three cells, one

for each working fluid phase: pure liquid, mixed phase (liquid and vapor) and pure vapor. The size

of each computational cell is free to vary with the lengths of each working fluid phase during

transients. Unfortunately, MBM calibration currently relies on only sensible parameters, such as

the entry and exit temperatures of both the working fluid and the heat source, because no studies

have examined the physical working fluid phase lengths. The utilization of tuned empirical

correlations describing the heat transfer in each phase presents a case where the MBM could

experience substantial deviation in the relative working fluid phase lengths and still arrive at the

correct working fluid exit temperature. While this inaccuracy can be acceptable in some instances,

this flaw becomes problematic when MBM are utilized for control, as the phase lengths are often

the states of interest during implementation [28, 31].

2.2.2 Importance of Phase length estimation

ORC WHR control strategies often critically rely on working fluid evaporator outlet

temperature prediction [20, 28-30], which is especially important when utilizing a turbine

expansion device. Due to the high rotational speeds of turbine expanders (10,000-120,000 rpm),

induction of working fluid droplets damages the turbine blades. Thus, during normal operation of

the turbine expander working fluid must be in vapor condition only. This constraint challenges

modern ORC WHR control strategies to accurately predict and control the evaporator working

fluid exit conditions during vehicle transients to avoid bypassing multi-phase working fluid around

the turbine, sacrificing power generation.

Redeveloping predictive ORC-WHR control strategies to also consider the working fluid phase

boundary locations within the evaporator provides an avenue for enhanced ORC-WHR power

15
generation during real-world operation. Knowledge of the mixed-phase to vapor boundary location

within the evaporator can provide an estimate of the real-time vapor ‘buffer’ before working fluid

evaporator exit conditions reach saturation. When utilized in conjunction with working fluid mass

flow and models of the compressible pipe volumes, an enhanced predictive control could alter

ORC system operation well in advance of detrimental working fluid saturation at the evaporator

outlet. Unfortunately, no validated physics-based models exist for prediction of the working fluid

phase change location. Error in the phase boundary estimation could lead to saturated exit

conditions during highly transient engine operation, which reduces the expander machine power

production duration, leading to Rankine cycle efficiency reduction. These considerations create

the motivation for research in this chapter.

2.2.3 Sensitivity of ORC-WHR Energy Recovery to Working Fluid Phase Length

To illustrate the importance of phase boundary layer estimation accuracy, a feedback plus

feedforward control is designed to regulate the vapor phase length as shown in Fig. 2.1. The heat

exchanger model utilized is a FVM and exhaust conditions are collected offline from a 1-D GT-

POWER engine model developed in [32]. The controller outputs the working fluid pump speed,

which is inversely related to working fluid vapor phase length.

Figure 2. 1. Vapor phase length control diagram.

16
 The target vapor phase length is held constant at 10% of the total evaporator length.

Hypothetical vapor phase length estimation model errors of 7% and 2% (both overestimating the

vapor phase length within the evaporator) are then simulated during a step down in engine speed

and torque. In the example, a simultaneous step down of engine speed and torque is considered

(namely, from 1500rpm, 1250Nm to 1400rpm, 1000Nm). Fig 2.2 shows the results of vapor phase

length, turbine power, and accumulative energy from a waste heat recovery system with the

aforementioned errors in phase length prediction. In Fig 2.2(a), the simulation with insertion of

the 7% phase length estimation error undergoes a period of saturation (i.e. vapor phase length falls

to zero), whereas the simulation with 2% phase length estimation error maintains vapor phase

along the entire 1000s test case. Based on the results at 200s in Fig 2.2(a), we can deduce that if

the vapor phase length within the evaporator can be predicted within 6%, the saturation period

could be avoided during this transient. In the Fig 2.2(b), when the phase estimation error is 7%,

the turbine is bypassed for the duration of the saturation period, resulting in a 9 % reduction in

cumulative energy collected over the transient. Minimizing the phase length prediction error can

maintain turbine power production without interruption and improve the effectiveness of waste

heat recovery.

17
Figure 2. 2. Organic Rankine cycle waste heat recovery system power comparison between two working
fluid boundary layer estimation accuracies: (a) vapor phase length over the total boiler length as a
percentage, and (b) normalized turbine power and normalized cumulative energy. In the legends, 7%
model error means the results from a simulation subject to a vapor phase length estimation error of 7%
prior to the feedback control. In contrast, 2% model error means the results from a simulation with a 2%
vapor phase length estimation error prior to the feedback control.

2.3 System Configuration

Figure 2.3 illustrates the evaporator design utilized in this study. The Figure schematically

represents a simplified version of the actual evaporator, due to intellectual property constraints

surrounding the novel evaporator design. The exhaust gas enters on the right and leaves the

evaporator on the left. Ethanol, the working fluid selected for this study, enters the evaporator core

on left top and splits into vertical tubes, which are arranged in counter-cross flow direction with

respect to exhaust gas.

18
 Figure 2. 3. Evaporator design schematic.

The evaporator is connected to a flow bench, which is supplied with hot exhaust from a gas

burner. A FLIR A300 thermal camera is utilized to capture temperature changes along the

evaporator’s skin. An example of captured thermal image is shown in Figure 2.4.

Figure 2. 4. An example of the Thermal Image from FLIR A300 camera utilizing an 18mm lens displaying
the line chosen for detailed temperature analysis from the thermal images. In consideration of the
sponsor’s intellectual property rights, some features of the evaporator have been intentionally blurred.

Figure 2.4 is purposely blurred to hide proprietary design features of this evaporator. However,

the reader can still distinguish the color stratification across the top of the evaporator, which

indicates the working fluid temperature variation during transitions from liquid (blue) to mixed-

19
phase (green-yellow) and then to vapor (red). Thermal video is captured during transient flow-

bench tests. With the availability of this thermal data at different conditions, the next task is to

process the images and extract the working fluid phase lengths along the evaporator core.

2.4 Thermal Image Processing

This section overviews the methodology for extracting thermal image temperature information

at each point along a line spanning the length of the evaporator between the working fluid inlet

and outlet, as shown in Figure 2.4.

2.4.1 Still Image Processing

The in-house thermal image processing methodology includes three aspects:

1. Converting pixel to pixel distances along the isometrically-oriented evaporator images to

a physical distance;

2. Extracting quantified temperature information along the line of interest from the images;

3. Detecting the working fluid phases from the extracted temperature information.

The Forward Looking Infrared (FLIR) thermal imaging camera used for capturing the thermal

images and videos provided an isometric view of the boiler as shown in the Figure 2.4. Projection

of a 3D object to a 2D plane implies the loss of information about the depth of the object.

Consequently, to accurately determine the distance to each point on the boiler from a given

reference point a quadratic fit was used based on the measured physical distance between the two

different points from the reference. This function converted the distance in terms of pixels to a

physical distance.

FLIR Tools software determined the temperature at each pixel of the thermal images. The

location for each point along the line was converted to pixel coordinates. The temperature at each

20
point was then determined using the temperature data from FLIR tools by matching the pixel

coordinates.

2.4.2 Video Processing

The transient FLIR videos, on the other hand, lacked the metadata required for FLIR Tools to

provide any temperature information. The data obtained from the project industrial sponsor had

undergone post-processing to reduce the file size. As such, an in-house processing methodology

was developed to extract the temperature information.

The key strategy for extracting temperature information from the videos was to match the color

of each of point on the line of interest to the video’s legend. The colors on the legend remain

constant for the duration of each transient test. However, based on the actual temperatures achieved

in each transient test case, the upper and lower limit of the temperature scale changed between

videos. Since the highest and lowest temperature for each video was known, a simple linear fit was

created between these two temperatures. Initial attempts sought the translation of each video still

frame to grayscale for temperature identification. However, this approach led to ambiguous results

due to the identical grayscale intensities for two or more different colors from the legend, and it

was thus abandoned.

21
 Legend Color Intensity Map
250

200

150

100

50

0
0 50 100 150 200 250 300 350 400 450 500
Array Index
Figure 2. 5. Comparison of RGB intensities on each point on the temperature detection line.

The procedure for analyzing the color videos as follows: the red, green and blue intensities

must be compared at each point from the line of interest to the image’s color scale legend. The

color intensities from each point of the temperature legend were plotted, as shown in Figure 2.5.

The red intensity component was then utilized to split the temperature identification space into

three zones with the array index increasing in the direction of increasing temperature. Until an

index of 150, the red component intensity increases linearly at an approximately constant rate.

Between 150 and 220, the red intensity continues to be linear, but at a much faster rate of increase.

The linear trend is lost only beyond an array index of 220. These three ranges were then utilized

within the in-house software to narrow the search space when identifying the temperature of any

pixel from the image. Once the search space was narrowed, the difference in color between the

point along the line of interest and legend entries in that sector was then found by the following

equation:

22
 Δ= 𝑅 −𝑅 + 𝐺 −𝐺 + 𝐵 −𝐵

The legend array index that yielded the minimum difference relative to the point of interest

was then fed into the linear function to convert the pixel location to a temperature value. For

validation, this procedure was applied to still images where full temperature discretization was

available via the FLIR Tools. The trend in temperatures for any still image was preserved,

providing confidence for identification of the working fluid phase change locations during the

transient tests. The offset noted in Figure 2.6 could be due to the line of interest not referring to

the exact same locations in the video and the image used.

Once the temperature data is extracted from any still image, the pixels along the line of interest

must be properly assigned to a physical length along the evaporator herein referred to as working

fluid ‘tube distance’. A sample result of the temperature mapping and subsequent length

assignment is provided in Figure 2.7.

Since temperature data was extracted only from the top surface of the evaporator, the length

assignment process has a length scale discretization corresponding to twice the evaporator height,

owing to the evaporator design illustrated in Figure 2.3.

23
 Temperature, ° C

Figure 2. 6. Temperature comparison data between FLIR tools and Matlab code operating on the same
still thermal image.

Figure 2. 7. Extracted temperatures along the top surface of the evaporator (along the defined line of
interest shown in Figure 2.4) with respect to normalized working fluid tube length in the evaporator. Also
shown are the three regions of interest (A-C), corresponding to the different working fluid phases.

24
2.4.3 Phase Change Detection

As shown in Figure 8, there are three possible regions of interest inside the evaporator,

corresponding to the three phase regions of the working fluid: (i) Region A is the pure liquid phase;

(ii) Region B represents the mixed-phase region (saturation) where both liquid and vapor working

fluid exist; and (iii) Region C represents the pure vapor phase. In region A, where the working

fluid temperature is below the saturation temperature, heat transfer is initially single phase

convection. As the temperature rises through the subcooled region towards point the ‘mix point’,

‘M’, vapor begins appearing from nucleation sites. This mechanism of heat transfer is termed as

subcooled nucleate boiling [33].

The transition between regions A and B, is shown as point ‘M’ where the subcooled nucleate

boiling heat transfer is converted to saturated nucleate boiling. Heat transfer in region B is

characterized by the working fluid quality. Temperature variation is noted in region B due to local

differences in pressure and vapor formation. As the quality of the vapor increases through region

B, nucleation is suppressed and heat transfer is a result of two-phase forced convection. This two-

phase region is characterized by very high heat transfer coefficients, which will be discussed later

in this paper.

At point the vapor transition point, ‘V’, of Figure 2.7, the working fluid quality reaches unity.

Heat transfer in the vapor region is due to convection within the vapor film. To determine points

‘M’ and ‘V’ from the thermal image results, saturation temperature is calculated based on the

evaporator pressure for each test. Point M is assigned as the first point along the evaporator length

where observed fluid temperature reaches the saturation temperature. The first instance where

observed temperature surpasses the saturation temperature is then identified as point ‘V’.

25
 The aforementioned length scale discretization due to the use of thermal data from only the top

evaporator surface plays an important role in detection of the working fluid phase change locations.

The thermal images only provide temperature data at the end or the beginning of each vertical tube

within the evaporator. Thus, the actual phase change location could be at any point along the

vertical tube length between adjacent top surface locations. The actual mix point can be between

point M and the immediately following top surface temperature point whereas the actual vapor

point can be between point V and the immediate preceding point. Thus, the certainty of

experimentally determined working fluid phase changes is limited by twice the evaporator height.

2.5 Evaporator Modeling

In an ORC-WHR system, there are two types of heat exchangers: evaporators which accept

thermal energy from heat sources, and condensers which reject thermal energy from the ORC loop.

ORC system efficiency depends critically on evaporator design and performance. As such, the

accuracy ORC system models heavily relies on the evaporator model. Two assumptions are made

for evaporator modeling described herein: (i) axial heat conduction (along the working fluid tube

length) in working fluid, wall and exhaust gas are neglected, and (ii) the wall temperature is

uniform across the tube wall thickness. In addition, experimental results have shown minimal

pressure drop across the evaporator, and hence pressure drop is neglected in this modeling work.

Mass balance, energy balance and momentum balance principles are all considered in evaporator

modeling. The modeling work contained herein is adopted from Bin Xu et al. [32, 34].

Transference of the main modeling methodology is first confirmed by applying said methodology

to the current evaporator, which is a substantially different physical design from the previously

published work. Subsequent sections will then enhance the modeling methodology to obtain

additional phase length prediction accuracy. Only the relevant equations of the underlying FVM

26
methodology are outlined in this work. For a complete overview of the modelling methodology,

readers are guided to prior works [32, 34].

The working fluid mass balance is presented in Eq. (2.1).

, ̇
+ =0 (2.1)

where subscript 𝑓 denotes fluid, 𝐴 is the cross-sectional area, 𝜌 is density, 𝑚̇ is mass flow

rate, and z is axial position along the evaporator tube length.

The energy balance for both the working fluid and the exhaust gas share the same general form,

which is presented in Eq. (2.2)

( ) ̇
+ = 𝜋𝑑𝑈∆𝑇 (2.2)

where 𝑝 is pressure, ℎ is enthalpy, 𝑑 is the effective flow path diameter for either the working fluid

or exhaust gas, 𝑈 is the heat transfer coefficient, and Δ𝑇 is the temperature difference between the

fluid (working fluid or exhaust gas) and the wall.

The tube wall energy balance is shown in Eq. (2.3)

𝐴 , 𝑐 , 𝜌 𝐿 =𝐴 , 𝑈 , ∆𝑇 , + 𝜂 𝐴 , 𝑈 , ∆𝑇 , (2.3)

where subscript 𝑤 denotes wall, 𝑐 is heat capacity, 𝐿 is the length in axial direction, 𝐴 ,

represents the heat transfer area between working fluid and wall, 𝑈 , is the heat transfer

coefficient between working fluid and wall, 𝜂 is the heat exchanger efficiency multiplier that

accounts for heat loss to the environment, 𝐴 , is the heat transfer area between exhaust gas and

wall, and 𝑈 , is the heat transfer coefficient between exhaust gas and wall.

Ordinary differential equations of Eqs. 2.1 and 2.2 are represented as follows:

= 𝑚̇ , − 𝑚̇ , (2.4)

( ̇ )
= 𝑚̇ ℎ − 𝑚̇ ℎ + 𝐴𝑈∆𝑇 (2.5)

27
where subscripts 𝑖𝑛 and 𝑜𝑢𝑡 denote spatial context in the axial direction, and 𝑣 is the working fluid

volume in one discretized cell.

Eqs. 2.3–2.5 can be solved by utilizing 1D Euler differentiation as shown by Eqs. 2.6-2.8:

, , ∆ , , , ∆ ,
𝑇 , ( ) = 𝑇 , ( ) + 𝜂 ∆𝑡 (2.6)
, ,

𝑚 , ( ) = 𝑚 , ( ) + (𝑚̇ , − 𝑚̇ , )∆𝑡 (2.7)

𝑑(𝑣𝑝)
(𝑚ℎ)𝑡(𝑘+1) = (𝑚ℎ)𝑡(𝑘) + + (𝑚̇ 𝑖𝑛 ℎ𝑖𝑛 − 𝑚̇ 𝑜𝑢𝑡 ℎ𝑜𝑢𝑡 + 𝐴𝑈∆𝑇)∆𝑡 (2.8)
𝑑𝑡

where 𝜂 is again the efficiency multiplier that accounts for the heat losses from exhaust gases to

environment, k is the time step indices, and ∆t is the time step length.

Figure 2.8 illustrates the finite volume modeling methodology. The heat exchanger is theorized

to be a long, straight tube which is divided into N cells. Within each cell, heat flows from the

exhaust to the wall and finally to the working fluid. Exhaust gas flows from the right to the left

and temperature gradually decreases, whereas working fluid flows from left to right and

temperature gradually increases. Meanwhile, the working fluid enters as liquid phase and exits as

vapor phase. For each discretized cell, the FVM solves Eqs. 2.6, 2.7 and 2.8 simultaneously.

Figure 2. 8. Schematic representation of Finite Volume Method used for evaporator modeling.

28
2.5.1 Heat Transfer Coefficients

The evaporator in [32, 34] utilized a helical coil type tube structure whereas the evaporator

tubes in the current device are arranged vertically. Hence a new set of equations for the respective

heat transfer coefficients are described in this section. In an evaporator, heat is transferred from

high temperature, high flow exhaust gases to the working fluid. The heat transfer coefficient for

each working fluid phase is calculated utilizing the average Nusselt number for that phase and the

evaporator tube geometry. The average Nusselt number, 𝑁𝑢 is calculated as follows,

∗
𝑁𝑢 = (2.9)

where 𝛼 is the heat transfer coefficient, 𝑑 is the hydraulic diameter, and 𝜆 is thermal conductivity.

Flow is defined to be a turbulent or laminar depending on the Reynold’s number, which is

calculated using Eq. (2.10) where 𝜐 is defined as kinematic viscosity and 𝜈 is defined as flow

velocity and is calculated using Eq 2.11.

∗
𝑅𝑒 = (2.10)

̇
𝜈 = (2.11)
∗

Since exhaust gas dynamics are fast relative to changes in working fluid enthalpy, a single

value of heat transfer coefficient between the exhaust gas and the tube wall is utilized for the entire

evaporator. As the exhaust gas traverses through the evaporator, the gas velocity gradually

decreases. Hence, an average Nusselt number is used for the exhaust gas to tube wall heat transfer

calculation. The heat transfer coefficient between exhaust gases and the wall is given as:

𝑈 , = 𝑚 (2.12)

where 𝑑 is the streamed length of a single tube, and 𝑚 is the heat transfer coefficient multiplier

that accounts for differences between the actual evaporator construction compared to the vertical

29
tube type structure. 𝑁𝑢 is the average exhaust gas Nusselt number based on the evaporator

geometry given by the expressions (2.13 – 2.16) [33]:

2 2 0.5
𝑁𝑢 = 0.3 + (𝑁𝑢 , + 𝑁𝑢 , ) (2.13)

/
𝑁𝑢 , = 0.664 ∗ 𝑅𝑒 , ∗ 𝑃𝑟 (2.14)

.
. ∗ , ∗
𝑁𝑢 , = (2.15)
.
. ∗ , ∗( )

𝑅𝑒 , = (2.16)

where 𝑑 = (𝜋/2)𝑑 is the streamed length of a single tube (i.e., the length of the flow path

traversed over a single tube) with inner diameter do. Void fraction, 𝜓 = 1 − 𝜋/4𝑎𝑏, where the

transverse pitch ratio is expressed as 𝑎 = 𝑠 /𝑑 and the longitudinal pitch ratio is calculated as

𝑏 = 𝑠 /𝑑 . For these expressions, 𝑠 and 𝑠 are the lateral and longitudinal spacing between tubes,

respectively.

The exhaust gas thermal conductivity, 𝑘 , , required in Eq 2.12, is expressed as:

.
𝑘 = 1.07 + − ( )
(2.17)

For the working fluid, individual heat transfer coefficients are assigned based on the fluid

phase. The vertical orientation of the working fluid tubes and the presence of a fully turbulent

working fluid flow (as calculated via the Reynolds number) results in utilization of the Gnielinski

correlations corresponding to flow in vertical tubes [33]. The pure liquid and pure vapor phase

heat transfer coefficients are generated with equations (2.18) and (2.19)
𝜉𝑓,𝑇𝑃,𝑖
𝑅𝑒𝑓,𝑖 𝑃𝑟𝑓,𝑖 𝑘𝑓,𝑖
.
𝑈𝑓,𝑖 = 8
(1 + (𝑑/𝑙) ) (2.18)
𝜉𝑓,𝑇𝑃,𝑖
0.667 𝑑𝑓,𝑖
1+12.7 𝑃𝑟𝑓,𝑖 −1
8

𝜉 , = (1.8 log (𝑅𝑒 ∗ ) − 1.5) (2.19)

where i represents the chosen phase (i.e. either fluid or vapor).

30
 For two phase working fluid flow, the heat transfer coefficient between the fluid and the tube

wall is also based on empirical correlations for flow through a vertical pipe [33] and is expressed

as follows,

. .
.
⎧(1 − 𝑥) . (1 − 𝑥) .
+ 1.9𝑥 . ,
+⎫
⎪ , ⎪
𝑈 , = 𝑈 , .
(2.20)
⎨ . , . , ⎬
⎪ 𝑥 1 + 8(1 − 𝑥) ⎪
⎩ , , ⎭
.
In the Eq. (2.20), 𝑈 , and 𝑈 , are calculated using Eq. (2.18). The terms (1 − 𝑥) and

. ,
𝑥 ensure a value of unity for 𝑥 = 0 and a value of for 𝑥 = 1. Further discussion on this
,

implementation will follow in section 2.7.

2.6 Sensitivity Analysis: FVM Discretization

To solve the governing equations (2.1-2.3), a finite volume modelling method is used. The

FVM discretizes the heat exchanger uniformly and solves the governing equations in each cell. To

access the accuracy and viability of the discretized evaporator model, a sensitivity analysis is

performed discern the impact of discretization on model accuracy.

Figure 2.9 illustrates the impact of model discretization on the working fluid exit temperature

and the phase change location predictions within the evaporator. Figure 2.9 is evaluated at the

steady state condition stated in Table 2.1.

Gas flow [kg/h] 358.9

Gas inlet temperature [ºC] 328.3
Gas outlet temperature [ºC] 188.7
Ethanol flow [kg/h] 39.7
Ethanol inlet temperature [ºC] 77.4
Ethanol inlet pressure [bar] 19.9
Ethanol outlet temperature [ºC] 282.4
Table 2. 1. Steady state experimental conditions for evaluating FVM discretization.

31
 Figure 2. 9. Working Fluid temperature across the length of evaporator tube for different FVM
discretization.

Table 2.2 provides a quantitative error summary at the different FVM discretizations

represented in Figure 2.9. In Table 2.2, temperature error is calculated as the difference between

simulated and experimental outlet temperatures. Phase length errors are calculated as a difference

between the model prediction and the post processed thermal image data. Overall, while the model

provides adequate working fluid and exhaust gas outlet temperature predictions with FVM

discretizations as low as 10 cells, the trend in the phase length prediction error forces further FVM

discretization.

32
 Evaporator outlet Phase Length
temperature error error
Working Exhaust Mix Vapor
Cells Fluid gas point point
% % % %
3 5.94 3.58 4.23 20.63
10 0.26 0.32 14.94 16.67
100 0.06 0.04 0.57 10.71
500 0.00 0.07 0.57 10.00
Table 2. 2. Error summary for different levels of discretization of the FVM model.

Figure 2.10 compares working fluid temperature predictions from the 500 cell FVM with the

experimental thermal image data. With the 500 cell discretization, the FVM predicts the liquid

transition to two-phase (mixing point) location with 0.57 % error, while the vapor point prediction

is still subject to a 10% error. The vapor phase location error does not decrease substantially with

further FVM discretization. Thus, there is room for FVM phase length prediction improvement,

which will be discussed in detail in section 2.7.

300
Thermal Image
FVM
250

200

150

100

50
0 0.2 0.4 0.6 0.8 1
Normalized length of tube along evaporator
Figure 2. 10. Comparison of simulated working fluid temperature from FVM model 500 cells
discretization and corresponding thermal image.

33
 The FVM is utilized as a plant model for offline simulation and hence these models are not

limited by their computation time. Since accuracy is prioritized in this study over computation

time, the 500 cell FVM is utilized for the remainder of this work.

2.7 Enhancing FVM Phase Prediction Accuracy

Utilization of empirical correlations from literature with multipliers reflecting the loss of heat

from the exhaust gases to the ambient within a FVM resulted in a best case 10% vapor phase length

prediction error, regardless of further discretization. The empirical heat transfer coefficients in Eq.

(2.18) and (2.20) may not directly apply to the heat exchanger geometry utilized herein, leading to

inaccuracy of the correlations for specific scenarios. Thus, this section outlines additional

calibration parameters necessary for enhanced FVM phase length prediction accuracy. The steady

state point stated in Table 2.1 (and represented in Figure 2.10) is still utilized, and a sensitivity

analysis is performed on the additional multipliers proposed to improve the vapor phase point

detection without affecting the evaporator outlet temperatures.

Multipliers were introduced to the heat transfer coefficient equations (2.18, 2.20), and the

resulting equations are stated below as (2.21) and (2.22),

,
, , ,
.
𝑈 , = 𝑼𝒍𝒊𝒒 (1 + (𝑑/𝑙) ) (2.21)
, . ,
. ,

,
, , ,
.
𝑈 , = 𝑼𝒗𝒂𝒑 (1 + (𝑑/𝑙) ) (2.22)
, . ,
. ,

. .
. . . ,
𝑈 , = 𝑈 , (1 − 𝑥) (1 − 𝑥) + 𝑼𝒎𝒊𝒙,𝟏 ∗ 1.9𝑥 +
,

.
.
. , . ,
𝑥 1 + 𝑼𝒎𝒊𝒙,𝟐 ∗ 8(1 − 𝑥) (2.23)
, ,

34
 Definitions of the newly introduced multipliers are as follows: 𝑼𝒍𝒊𝒒 is the tuning parameter

added to the liquid phase heat transfer coefficient equation. 𝑼𝒗𝒂𝒑 is the tuning parameter added to

the vapor phase heat transfer coefficient equation, and 𝑼𝒎𝒊𝒙𝟏,𝟐 are two heat transfer multipliers

added to the mixed-phase flow inside the boiler. In the case of two phase flow, Umix1 and Umix2 are
. .
applied within Eq. 2.23 such that the terms (1 − 𝑥) and 𝑥 still preserve heat transfer

coefficient continuity.

2.7.1 Manual Parameter Tuning

With the insertion of additional tuning parameters, it was critical to understand their influence

on prediction of the working fluid phase lengths and outlet temperature. Therefore, for the

conditions listed in Table 1, 𝑚 (the general heat transfer coefficient multiplier in Eq. (2.6)) and

𝜂 (the efficiency multiplier in equation Eq. (2.12)) were identified first and held constant.

Subsequently, each of the newly proposed multiplier [𝑈 , 𝑈 , , 𝑈 , 𝑈 , ] values were

changed ± 50% as a sensitivity analysis, as shown in Table 2.3.

In Table 2.3, the rows with U=1 represent the baseline FVM phase length and outlet

temperature with just the general heat transfer coefficient and efficiency multipliers utilized. The

percent change [C, %] in phase length and outlet temperatures between subsequent parameter

perturbations and the baseline was then calculated as

𝐶, % = 𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒 − 𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑖𝑒𝑟 ∗ 100/𝑡𝑎𝑟𝑔𝑒𝑡 (2.24)

where the 𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒 is the parameter value obtained while holding the new heat transfer multiplier

at unity, 𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑖𝑒𝑟 is the parameter value obtained when the new multipliers are

perturbed (as 0.5, 0.1, 1.0, 10), and 𝑡𝑎𝑟𝑔𝑒𝑡 is the parameter value obtained from the thermal image

data as shown in Figure 2.10.

35
 % CHANGE, C
Change in Phase Change in Evaporator
Length outlet temperature
multiplier Mix point Vapor Working Exhaust gas
value point Fluid
1 0.57 10.00 0.00 -0.02
𝑼𝒗𝒂𝒑 1.5 6.32 12.86 -1.59 0.78
0.5 -17.8 2.38 5.35 -2.77

1 0.57 10.00 0.00 -0.02

𝑼𝒍𝒊𝒒 1.5 -0.57 9.05 -0.99 0.30
0.5 2.87 12.14 2.76 -1.13

1 0.57 10.00 0.00 -0.02

𝑼𝒎𝒊𝒙,𝟏 1.5 -0.57 10.71 -0.39 0.19
0.5 2.30 7.86 1.24 -0.76

1 0.57 10.00 0.00 -0.02

1.5 0.57 10.00 -0.04 -0.07
𝑼𝒎𝒊𝒙,𝟐 0.5 0.57 10.00 0.07 -0.12
10 2.87 12.38 2.62 -1.08
0.1 5.17 9.76 4.11 -1.93

𝑼𝒎𝒊𝒙,𝟏 1.5 -0.57 10.95 -0.39 0.19

= 𝑼𝒎𝒊𝒙,𝟐 0.5 5.17 9.29 4.36 -2.08
Table 2. 3. Analysis the effects of additional multipliers on phase lengths for test conditions in Table 1.
[Negative error means under-estimation and positive error is over-estimation]

Figures 2.11 – 2.15 display the results of Table 2.3. Figure 2.11 shows the effects of changing

the 𝑼𝒗𝒂𝒑 value. As expected, when 𝑼𝒗𝒂𝒑 is increased by 50 % (=1.5), the heat transfer coefficients

of both the pure vapor and the mixed phases increase, resulting in increased heat transfer between

exhaust gas and working fluid in both regions of the evaporator. Thus, the vapor phase change

occurs earlier than baseline. This increase in heat transfer also results in higher working fluid outlet

temperature. Decreasing 𝑼𝒗𝒂𝒑 by 50% (=0.5) resulted in exactly the opposite behavior.

36
 Working Fluid Temperature, ° C

Figure 2. 11. Effects of varying Uvap values on phase lengths and outlet temperature.

Figure 2.12 illustrates the influence of 𝑼𝒍𝒊𝒒 on the working fluid phase lengths. The trend of

𝑼𝒍𝒊𝒒 is similar to that of 𝑼𝒗𝒂𝒑. However, the magnitude of phase length change is reduced. 𝑼𝒍𝒊𝒒

indirectly influences 𝑼𝒇,𝒎𝒊𝒙 by changing the value of 𝑼𝒇,𝒔𝒂𝒕 . Since 𝑼𝒇,𝒔𝒂𝒕 can be found in both the

numerator and denominator of Eq. (2.23), the overall impact of 𝑼𝒇,𝒔𝒂𝒕 on 𝑼𝒇,𝒎𝒊𝒙 is minimal. A

fifty percent increase in 𝑼𝒍𝒊𝒒 from 1.0 to 1.5 produces no discernable change in phase lengths

from the baseline FVM. However, the trend can be verified by increasing Uliq by a factor of ten,

which shows the phase changes occurring earlier and a higher outlet temperature. Predicted phase

lengths displayed a weak inverse relationship with Uliq manipulation, requiring an order of

magnitude parameter changes prior to phase length response.

37
 Working Fluid Temperature, ° C

Figure 2. 12. Effects of varying Uliq values on phase lengths and outlet temperature.

The effects of 𝑼𝒎𝒊𝒙,𝟏 and 𝑼𝒎𝒊𝒙,𝟐 on working fluid temperature and heat transfer coefficient are

shown in Figures 2.13 and 2.14, respectively. Comparing Figures 2.13 and 2.14 reveal the stronger

influence of 𝑼𝒎𝒊𝒙,𝟏 on the overall heat transfer coefficient, 𝑼𝒇,𝒎𝒊𝒙 relative to 𝑼𝒎𝒊𝒙,𝟐 in affecting

the vapor point and mix point occurrence The weaker influence of 𝑼𝒎𝒊𝒙,𝟐 is the result of the

,
normalization term ahead of 𝑼𝒎𝒊𝒙,𝟐 .
,

Figure 2.14 [2.14(b) is a zoomed in version of 2.14(a) for a clearer representation of the phase

length changes] shows the heat transfer coefficient evolution along the evaporator tube length for

various values of 𝑼𝒎𝒊𝒙,𝟏 and 𝑼𝒎𝒊𝒙,𝟐 . Increasing the 𝑼𝒎𝒊𝒙,𝟏 by a factor of 10 affects the magnitude

of overall heat transfer coefficient, but it does not influence the phase length significantly.

Meanwhile, reducing the 𝑼𝒎𝒊𝒙,𝟏 by a factor of 10 not only reduces the magnitude of the heat

transfer coefficient but also increases the mix phase zone as seen from Figure 2.14 (a).

38
 (a)
Working Fluid Temperature, ° C

(b)
Figure 2. 13. Effects of varying Umix,1 and Umix,2 values on phase lengths and outlet temperature.

39
 (a)

(b)

Figure 2. 14. Effects of varying Umix,1 and Umix,2 values on heat transfer coefficient.

Figure 2.14 (a) shows that, in the single phase convective heat transfer region A, the heat

transfer coefficient is relatively constant despite changes to the Umix multipliers. In the mixed phase

40
region B, as the quality of the fluid increases across the evaporator, the heat transfer coefficient

increases rapidly. This is a physically correct phenomenon of bubble formation and transport

resulting from higher fluid turbulence and mixing. The empirical correlations approximate the high

heat transfer in the mix phase region caused by latent heat transfer from the formation of bubbles

and the corresponding increase in turbulence. As the liquid film thickness reduces along the tube

in the mixed phase forced convective region, the corresponding heat transfer coefficient increases

with the vapor quality. Eventually, as liquid film thickness continues to decrease, it reduces the

effect of forced convection and there is a rapid drop in heat transfer coefficient just before the

onset of region C. Hence the heat transfer coefficient of the single phase vapor region C is always

lower than the multiphase convection and evaporation heat transfer of region B. For reference,

Appendix A [35], shows the typical flow patterns occurring inside the vertical tube heated

uniformly over its length.

This sensitivity analysis of the newly proposed parameters provided useful insights regarding

their relative impacts. These insights were utilized to configure the particle swarm optimization

(PSO) bounds for the automated parameter identification discussed in the next section. For the

PSO, 𝑼𝒎𝒊𝒙,𝟐 was eliminated and only 𝑼𝒎𝒊𝒙,𝟏 was utilized in the mix phase zone.

2.7.2 Particle Swarm Optimization Parameter Identification

Once the relative impacts of each multiplier on the phase length and outlet temperature were

understood, a PSO was utilized to identify the optimal values of all the model parameters for peak

outlet temperature and vapor phase length accuracy. The five parameters that were simultaneously

identified were [𝜂 , 𝑚 , 𝑈 , 𝑈 , ,𝑈 ].

The main criteria for modeling an evaporator is to accurately estimate the outlet temperatures

and the PSO cost function weights were adjusted accordingly, i.e. higher weights were used for

41
temperature error compared to phase length error. The relative weights can be altered if the goals

of future model implementations shift toward phase length prediction accuracy. The implemented

PSO cost function is shown as Eq 2.25

𝑇, − 𝑇, 𝑇, − 𝑇,
𝐽 = 10 ∗ + 10 ∗ +
200 200

, , , ,
+ (2.25)

where 𝑇 , denotes the working fluid outlet temperature, 𝑇 , is the exhaust gas outlet temperature,

𝑙 , is the mix phase point predicted by the FVM model, 𝑙 , is the mix phase point

determined from the thermal images, 𝑙 , is the vapor point from FVM model, and 𝑙 , is

the vapor point determined from the thermal images.

The cost function in Eq 2.25 involves two different variable types, working fluid

temperatures and phase lengths, which have disparate orders of magnitude owing to their

respective units. In order to compare these on equal footing, the temperature and phase length

errors were normalized by 200 and 5000, respectively.

Table 2.4 shows the sensitivity study of different PSO targets for the operating conditions

introduced in Figure 2.7 and Table 2.1. The varying PSO targets in Table 2.4 physically represent

experimental uncertainty in 𝑙 , and 𝑙 , . As stated previously, the evaporator design

relative to the thermal information available provides a level of uncertainty in the thermal imaging

based phase length data on the order of twice the evaporator height. The objective behind this

sensitivity study was to examine any improvement in phase length prediction while still

maintaining the outlet temperature within acceptable margins of ±10℃. In this study, importance

was given to improving vapor phase length as compared to mix point. If the complete ORC system

42
is considered, the turbine is located after the evaporator, and safe operation of turbine requires

confident prediction of the working fluid vapor quality exiting the evaporator.

Case# Phase lengths Evaporator Outlet

Temperature (˚C)
𝑙 𝑙 Working Exhaust gas
Fluid
Thermal Image 0.36 +/- 0.04 0.84 +/-0.04 282.4 188.7
1 Baseline FVM results 0.348 0.758 281.7 188.7
% Error in FVM -3.33 -9.76 -0.21 -0.37
results

2 PSO target 0.36 0.84 282.4 188.7

% Error in FVM 0 -2.62 -4.74 -2.8
results

3 PSO target 0.38 0.84 282.4 188.7

% Error in FVM 7.77 -2.85 -3.25 -2.06
results

4 PSO target 0.34 0.84 282.4 188.7

% Error in FVM -3.89 -3.33 -4.5 -1.32
results

5 PSO target 0.36 0.8 282.4 188.7

% Error in FVM 3.89 -5.95 -0.74 0.26
results

6 PSO target 0.38 0.8 282.4 188.7

% Error in FVM 4.44 -6.43 -2.38 -0.28
results

7 PSO target 0.34 0.8 282.4 188.7

% Error in FVM -9.44 -3.81 -3.72 -0.95
results
Table 2. 4. Summary of Sensitivity analysis of different PSO targets. Negative error in the above Table
indicates under-estimation whereas positive error corresponds to over-estimation.

Case 5 provided the least error in the working fluid outlet temperature estimation while

providing improvement in the vapor point prediction. Note that if the constraints for working fluid

outlet temperature accuracy were relaxed, the parameters identified in Case 2 could provide

addition phase length prediction accuracy. The PSO identified parameters from case 5 are

compared with the baseline parameters in Table 2.5.

43
 Parameters Baseline PSO
𝑚 0.4517 0.4168
𝜂 0.9646 0.9539
𝑈 1 0.9346
𝑈 , 1 2.7335
𝑈 1 2.6927
Table 2. 5. PSO identified variables for Case #5 in Table 2.4.
Working Fluid Heat Transfer Coefficient, W/m 2K

Figure 2. 15. Comparison of working fluid heat transfer coefficient from the FVM baseline model and the
FVM with PSO identified values.

The effects of the multipliers 𝑼𝒎𝒊𝒙,𝟏 and 𝑼𝒗𝒂𝒑 on the heat transfer coefficient is shown in

Figure 2.15. 𝑼𝒎𝒊𝒙,𝟏 and 𝑼𝒗𝒂𝒑 were increased by 2.8 and 2.7 times, respectively, leading to

enhanced heat transfer prediction in the mixed-phase and vapor regions. The increase in heat

transfer in each region is proportional to the increase in applied multiplier.

On interpreting PSO identified values, a value of 0.93 for 𝑼𝒍𝒊𝒒 does not substantially affect

either the outlet temperature as observed during the sensitivity analysis shown in Figure 2.12.

However, it can be concluded from Figures 2.11 and 2.13(a) that increasing 𝑼𝒗𝒂𝒑 and 𝑼𝒎𝒊𝒙,𝟏

44
increases the evaporator outlet temperature, which in turn is counteracted by the lower overall heat

transfer coefficient multiplier mU. It can be also observed from Figure 2.13 (a) and later confirmed

from Figure 2.14 that increasing 𝑼𝒎𝒊𝒙,𝟏 also shifts the complete mix phase towards the evaporator

outlet, the desired direction.

In order to match the evaporator outlet temperature and the working fluid phase lengths, the

additional multipliers predict a higher heat transfer in the mixed phase region, as shown in Figure

2.15. This implies that the introduction of the generic empirical correlations (Eq. 2.12, 2.18, and

2.20) failed to capture the intensity of the real heat transfer phenomena within current the

evaporator design. The introduction of additional multipliers compensates for lost physics in the

mix phase region.

Figure 2.16 shows the working fluid temperature throughout the evaporator length for the

Table 2.1 test conditions with both the baseline FVM and FVM utilizing the proposed PSO

identified parameters. The FVM traces are plotted with the experimental thermal imaging data for

comparison. The enhancement to FVM phase length prediction performance when utilizing the

newly added PSO identified parameters is clear. While there is still room for further improvement,

no further iterations were made within the scope of this study.

45
Figure 2. 16. Comparison of working fluid temperature along the length of the evaporator from the FVM
baseline model and the FVM with PSO identified values.

2.8 Results

In this section, the performance of the enhanced PSO identified FVM, utilizing the parameters

listed in Table 2.5, is evaluated relative to thermal imaging results collected from two different

transient experimental runs. All of the experiments were performed on a thermal-flow bench,

where the flow (hot gas) and temperature can be both controlled independently.

2.8.1 Transient Test Sequence #1

The first test sequence imitates a cold start. Figure 2.17 shows the test conditions for this

experiment. Note that working fluid mass flow was initiated prior to the exhaust temperature rise.

From 0s to 20s both the exhaust and working fluid are cold. Once the exhaust burner is initiated

both the exhaust mass flow (360 kg/hr) and the exhaust temperature (327 ˚C) were kept constant

for the remainder of the test. Conversely, working fluid pressure was kept constant at 20 bar and

the outlet temperature was maintained at 280˚C by altering the working fluid mass flow rate.

46
 Flow, kg/hr
Temperature ° C Temperature ° C

Figure 2. 17. Exhaust and Working Fluid conditions for Test 1.

Outputs of the enhanced FVM utilizing a 500 cell discretization are shown in Figure 2.18.

Despite the initial offset in simulated exhaust outlet temperature (only observed in the cold region,

which was outside the model calibration range), the FVM model precisely tracked the outlet

temperatures. The mean error for the exhaust temperature was 1.45% whereas the working fluid

the mean outlet temperature error was 1.68%. The model still exhibits a slight discrepancy at the

onset of vapor phase, indicative of a slight disparity in wall thermal inertia.

47
 Temperature, ° C
Temperature, ° C

Figure 2. 18. Baseline FVM simulation results for the evaporator outlet temperatures for Test 1.

Figure 2. 19. Phase length comparison between the baseline FVM model and experimental thermal video
for Test sequence #1.

Phase length predictions of the baseline FVM when utilizing test sequence #1 data as inputs

are shown in Figure 2.19. Note that the high frequency error on the thermal imaging results is

produced by the spatial resolution restrictions of utilizing the top evaporator surface for phase

length prediction with an evaporator of this design. After 75s, as the working fluid temperature

48
increases, the FVM model prematurely predicts mix point. This may indicate a mismatch between

the selected heat transfer correlations and the thermal behavior of the evaporator during harsh

transients. However, as the system reaches equilibrium at the warmer working fluid temperature

(after 300s), the mix phase length prediction accuracy increases, yet the length of the vapor phase

is over-estimated (with respect to the evaporator outlet). These results correlate with Figure 2.10,

which also displayed over-estimation of the vapor phase length relative to the thermal image

results during another fully-warmed condition.

The next step was to evaluate the PSO identified values (Table 2.5) in the enhanced FVM over

the complete transient. Application of additional multipliers showed a great improvement in the

vapor phase length prediction with some penalty in mix phase length and working outlet

temperature estimation [Figure 2.20]. Refer to Table 2.6 for the quantitative error comparison

between base line and enhanced FVMs. Overall, the enhanced FVM exhibited a 43% improvement

in mean vapor point prediction error.

Baseline FVM Calibrated FVM

Mix pt Vapor pt Mix pt Vapor pt
Max % error 28 7.7 28.2 6.8
Mean % error 3.37 2.4 3.73 1.36
Table 2. 6. Comparison of the phase length estimation between baseline FVM model and calibrated FVM
model.

49
 Figure 2. 20. Phase length comparison between the enhanced FVM and the experimental thermal video
for Test #1.

Figure 2.21 displays the time varying phase length error comparisons between the baseline and

enhanced FVMs. Both the mix point and vapor point predictions were shifted towards the

evaporator outlet. Thus, the entire saturation zone was shifted towards the evaporator outlet.

However, it is important to note that the change in vapor point position was greater than the change

in mixing point prediction.

50
 Normalized Error
Normalized Error

Figure 2. 21. Phase length error comparison for the baseline and enhanced FVMs relative to the thermal
imaging results during transient Test 1.

Simulated exhaust and the working fluid outlet temperature error from the enhanced FVM

remain representative of the experimental values, as shown in Figure 2.22. Although parameter

identification for the enhanced FVM was limited to a single experimental condition, the new model

accurately estimated the vapor phase length. Utilization of a broader data set during paramter

identification will provide even better results during transient validation.

51
Figure 2. 22. Simulated working fluid and exhaust gas outlet temperature results from the enhanced FVM
plotted relative to the experimental values for transient Test #1.

2.8.2 Transient Test Sequence #2

The second transient experiment maintained a constant ethanol flow rate (82 kg/hr) and exhaust

temperature (350°C) while exhaust mass flow rate was ramped up and down, as shown in Figure

2.23. The phase length predictions of the enhanced FVM over transient test #2 are displayed in

Figure 2.24. During portions of the test where all three working fluid phases are present within the

evaporator (~100-500s), the enhanced FVM was able to predict the mix point accurately. However,

during instances when the working fluid was saturated at the evaporator outlet, the model under

predicted the mix phase length. This may be due to the utilization of turbulent equation (5.18),

which increases the heat transfer coefficient and triggers the occurrence of mix phase prematurely.

Note that the enhanced FVM was not calibrated with data displaying saturated working fluid outlet

conditions.

52
 Mass Flow
1000 Exhaust gas

Flow, kg/hr
Working fluid

500

0
0 100 200 300 400 500 600 700 800
Exhaust Gas Temperature
Temperature ° C Temperature ° C

400

200
Inlet
Outlet
0
0 100 200 300 400 500 600 700 800
Working Fluid Temperature
400 Inlet
Outlet
200

0
0 100 200 300 400 500 600 700 800
time, sec
Figure 2. 23. Exhaust and Working Fluid conditions for Test 2.
Normalized tube length

Figure 2. 24. Phase lengths comparison between the enhanced FVM model and the thermal videos for
Transient Test #2.

53
 The discrepancy of the vapor length during saturated outlet conditions can be improved by

conducting parameter identification over a broader data set. No efforts were made in this work to

improve the phase length predictions during saturated outlet conditions. Instead, focus was placed

on accurate vapor phase length prediction. Even with the aforementioned shortcomings, the

enhanced FVM still accurately predicts the working fluid and exhaust outlet temperatures in all

the conditions, as shown in Figure 2.25, which is the most significant requirement of an evaporator

model. The mean error for the entire transient test was 0.2% for the exhaust temperature and 1.55%

for the working fluid.

Temperature, ° C
Temperature, ° C

Figure 2. 25. Simulated working fluid and exhaust outlet temperatures using the enhanced FVM plotted
relative to the experimental values during transient Test #2.

2.9 Conclusions

In this research work a high fidelity finite volume evaporator model was developed and

validated for accurate estimation of working fluid phase lengths and evaporator outlet temperatures

for a vertical tube cross-flow evaporator. A model discretization analysis showed that increased

discretization increases the model’s phase length prediction capability. A discretization of 10 cells

54
is enough to accurately model the evaporator outlet temperatures, while a 500 cell discretization

provided the best phase length prediction capability.

A vertical tube type evaporator was experimentally tested on a flow bench. Thermal imaging

data obtained from the corresponding transient tests was analyzed for experimental

characterization of the working fluid phase lengths inside the evaporator. The baseline FVM

estimation for mixed phase length is very accurate when all the three phases exist in the evaporator.

However, the baseline FVM over-predicted the vapor phase region.

To improve the accuracy of the vapor phase length prediction, additional calibration

parameters were included within the heat transfer coefficients for the mixed and vapor phases. A

PSO was used to identify the optimal values for these multipliers utilizing the thermal imaging

data. The enhanced FVM was then validated over two separate transient experimental data sets.

The model adaptations included in the enhanced FVM resulted in a 43% improvement in vapor

phase length prediction relative to the baseline FVM.

The current model formulation was not calibrated for instances where the working fluid is

saturated at the evaporator outlet. The enhanced FVM still yielded accurate working fluid and

exhaust gas outlet temperature results during these conditions, but the mixed phase length

prediction accuracy suffered. Further calibration of the enhanced model will be necessary for

accurate mixed phase length prediction during saturated working fluid outlet conditions.

Overall, a high fidelity FVM evaporator model was expanded to enhance the working fluid

phase length predictions during transient operation. The methodology developed herein can bolster

the evaporator design process and serve as basis for improving the moving boundary model for

advanced ORC system control development.

55
 CHAPTER 3. DYNAMIC CHARACTERIZATION OF
EVAPORATOR

3.1 Introduction

This Chapter overviews the dynamics of the ORC system. First, the importance of

characterizing the thermal inertia of the evaporator and working fluid properties is explained.

Second, open loop analysis is performed on the ORC inputs and outputs that explains the plant

dynamics and requirements of the controller design for an ORC-WHR system.

3.2 Background

The thermal inertia of the evaporator affects the transient performance of ORC-WHR system.

Under fluctuating loads, ORC system will respond in certain amount of time based on the thermal

inertia. For optimal control of the set-point temperature or superheat tracking, the response time

of the evaporators play a vital role [19, 24]. Evaporator dynamics is the slowest among other

components of the ORC system, and therefore, the system inertia can be very well represented by

the evaporator [36].

In this study, the ORC-WHR system contains two variants of the evaporator, the TP evaporator

and the EGR evaporator. As the name suggests, the TP evaporator extracts heat from the TP

exhaust and is located after the after-treatment system of the engine. Whereas the EGR evaporator

extracts heat from the EGR exhaust flow and is located before the turbocharger and is in series

with the EGR cooler to provide additional cooling. Because of their inherent locations both the

evaporators are exposed to different dynamics of the exhaust heat source. Below is an example of

a typical heat source for each evaporator.

56
 EGR

Temperature, ° C
500 TP

400

300

200
0 500 1000 1500 2000 2500 3000 3500 4000
time, s
EGR
Mass Flow, kg/hr

1000 TP

500

0
0 500 1000 1500 2000 2500 3000 3500 4000
time, s
Figure 3. 1. Typical TP and EGR exhaust conditions

As shown in Figure 3.1, the frequency component of the exhaust mass flow dynamics for the

both evaporators are similar, whereas the magnitude comparison shows that flow going through

EGR evaporator is much less than the TP, because of which, in comparison to the TP evaporator,

the EGR evaporators are much smaller in size. The exhaust temperature for EGR evaporator is as

dynamic as the mass flow, but for the tailpipe evaporator the exhaust temperature is much

dampened because of the after-treatment system. This shows that both the evaporators are exposed

to different time-scale of exhaust conditions. Moreover the dynamic response of evaporator is a

function of, 1) thermal inertia, 2) working fluid properties, 3) exhaust conditions and, 4) working

fluid flow rate. A sensitivity study has being conducted for these parameters in the following

sections.

3.3 Evaporator Thermal Inertia

The main aspects that affect the dynamic response of the evaporator can be grouped in the

following areas: wall material, boundary conditions and geometry [36]. In this study, wall material

and geometry of the evaporators are already fixed. The present work aims at investigating the

57
varying boundary conditions for two evaporators of different size, TP and EGR. Figure 3.2 shows

the relative size comparison of the two evaporators under consideration.

(a) TP Evaporator (b) EGR Evaporator

Figure 3. 2. Evaporator size comparison

Figure 3.2 (a) shows a single core TP evaporator. In this study, there are two single core TP

evaporators connected in series. Therefore, effectively, the actual length of the TP evaporator is

2*L whereas the width, W and height, H remains the same for this series connection. Figure 3.2

(b) shows the dimensions of the EGR evaporator with approximately half the width of the TP

evaporator and is also half of the TP evaporator’s effective length.

For this analysis, fluctuating loads with varying frequency are passed through the evaporator

and the corresponding dynamic response of the evaporators is analyzed. This sinusoidal heat

source can be viewed as alternating uphill and downhill road sections that can be encountered by

the trucks in real-world operations. Understanding the effects of this fluctuating heat source is

crucial for optimized ORC system operation. The analysis can aid as a tool for effective controller

design for optimized dynamic operation in real world scenarios.

58
 % change

(a) T=30s
% change

% change

(b)T= 60s (c) T=120 s

% change

(d) T=240 s (e) T=600s

Figure 3. 3. Thermal response of TP and EGR evaporators to sinusoidal exhaust conditions with varying
time period, T

59
 15

TP response
EGR response

10

% change

0
0 100 200 300 400 500 600
Time period, s
Figure 3. 4. Summary of thermal response of TP and EGR evaporator.

Figure 3.3 illustrates a systematic investigation of evaporator thermal response to different

exhaust power frequencies. The thermal response of the evaporator is measured by the change in

working fluid temperature at the evaporator outlet. The time-period of the sinusoidal exhaust

power waveform is varied from 30 s - 600 s. For shorter time-periods, as seen from Figure 3.3 (a),

the heat source fluctuations are filtered out by the thermal inertia of the evaporators. This is

because heat transfer process inside the evaporator is a two-step phenomenon. In Step 1, heat

energy flows from the exhaust gas to tube walls and depends on the tube material and thickness.

These factors dictate the time necessary for the wall temperature respond. In Step 2, once the wall

temperature responds, the heat then flows from the wall to the working fluid. Short period exhaust

power fluctuations cannot fully influence the working as the power fluctuations cannot fully

overcome the tube wall thermal inertia. Therefore, at high frequencies, only minimal changes in

the working fluid temperature are observed. As the time-period of the heat signals is increased, the

working fluid experiences increased fluctuation, as seen from Figure 3.3 (b) to 3.3 (e).

60
 Relative to the TP evaporator, the dynamic response of the EGR evaporator is larger in

magnitude and dynamically faster, i.e. the working fluid more closely tracks the exhaust power

fluctuations. In other word, the smaller EGR evaporator dampens incoming exhaust power

fluctuations less. Note that the higher thermal inertia TP evaporator produces a pronounced phase

shift in the working fluid temperature response. This phase shift corresponds to the time taken by

the thermal mass to complete the heat transfer process. Figure 3.4 summarizes the amplitude

response to varying time-periods of the heat source for both the evaporators.

For optimal operation of the ORC system, the superheat tracking error should be as low as

possible. If the temperature/enthalpy change of the working fluid at the outlet of the evaporator is

to be remained unchecked than it’s possible that the state of the working fluid will go below the

saturation point inhibiting the turbine operation. In the following Chapters, it will be cleared on

how to control this magnitude variation to the fluctuating heat source with a well-tuned model

predictive control.

3.4 Working fluid properties

(a) (b)
Figure 3. 5. (a) ORC system pressure as a linear function of Working fluid mass flow (b) Saturation
temperature vs system pressure correlation for ethanol as working fluid

61
 The correlation between working fluid flow rate and system pressure is a function of

component characteristics i.e. the valves, turbine design, and tubing size in the ORC system. For

a fixed valve position and a given turbine and tube design, the system pressure is a linear function

of working fluid mass flow rate as seen in Figure 3.5(a). On the other hand, Figure 3.5(b) shows

that increasing pressure causes the saturation temperature to increase non-linearly. This behavior

is working fluid dependent. The curves shown in Figure 3.5(b) correspond to ethanol, the working

fluid this study. A different working fluid will provide a different characteristic saturation

temperature and system pressure correlation.

The slope of the saturation temperature shows that for pressures less than 18 bar, a small

change in pressure will cause the saturation temperature to vary rapidly. This makes system

operation difficult for pressures below 18 bar. Therefore, a control strategy should operate above

pressures of 18 bar where slight pressure changes do not drastically affect the saturation

temperature. Proper bounding of system operation will aid in maintaining superheat within

acceptable limits.

3.5 ORC System dynamics evaluation

Once the evaporator dynamics are quantified, the next task would be to characterize the ORC

systems inputs and outputs. For this analysis, an open loop evaporator and turbine model was

created as shown in Figure 3.6. ORC system dynamics is a function of following parameters

1. Disturbances/Uncontrollable Inputs, 𝑚̇ ,𝑇

2. Control Input, 𝑚̇

62
 Figure 3. 6. Open loop plant model with inputs, outputs and control objective.

The input to the evaporator model is the working fluid mass flow rate which is controlled by

the pump. The known disturbances that act on the evaporator model are the exhaust gas

conditions, 𝑚𝑒𝑥ℎ and 𝑇 . The model output to be controlled is the working fluid temperature,

𝑇 /or superheat (∆𝑯𝒔𝒉 = 𝑇 -𝑇 ). The control objective is then to maximize the power

production. In this simulation study, turbine is modelled as a map based function of evaporation

pressure and saturation temperature. For a given ORC system operating with fixed valve position,

evaporation pressure, 𝑃 is a linear function of working fluid. 𝑃 is estimated by formulating

the mass flow and pressure relation from the experimental turbine map.

Table 3.1 summarizes the control objective of the ORC-WHR plant model. Superheating is

required to be as low as possible that in order to increase the mass flow through turbine increasing

its power output. From Figure 3.5 it can also be inferred that operating at higher pressure will lead

to better controllability characteristic which will indirectly influence higher saturation temperature.

For fixed valve openings and turbine speed, higher pressure can be achieved by increasing working

fluid flow which in turn increase turbine power output. Ideally, condenser saturation temperature

should be as low as possible to ensure complete cool-down of the working fluid before its

circulated back in the evaporator. However, for on-board vehicle operation, cooling is limited to

63
radiator size. It is also dependent on ambient temperature and engine coolant temperature set-point.

For simplicity, in this study the condenser saturation temperature was assumed to be constant.

Parameter Objective Constraints

Minimum Maximum
Superheating, ∆𝐻 Minimize Safe turbine operation WF deterioration
limit
Evaporation Saturation Maximize None None
temperature, 𝑇
Condenser saturation Minimize Ambient temperature Available cooling
temperature, 𝑇𝑐𝑜𝑛𝑑 capacity
Table 3. 1. Summary of the control objective of the ORC-WHR system.

3.5.1 Uncontrollable Inputs: Exhaust Conditions

For an automotive application, the ORC-WHR system is passive in nature. Therefore, the

exhaust conditions are considered known (measured) disturbances or uncontrollable inputs. For

this study, the engine conditions are changed in stepwise fashion while working fluid flow remains

constant at 𝑚̇ = 0.025 kg/s. The test conditions are shown in Table 3.2 and the corresponding

results are shown in Figure 3.7.

Test A Test B

𝑚̇ , kg/s 0.3 0.3 - 0.35 - 0.4

𝑇 , ˚C 300 – 350 - 400 350

Table 3. 2. Exhaust conditions for sensitivity analysis on the uncontrollable inputs

Figure 3.7 (a), shows the exhaust temperature sensitivity. It can be observed that after the

warmup (at 350 s), fixed magnitude (50 ˚C) step changes in exhaust temperature produce working

fluid temperature changes of varying magnitude at the evaporator outlet. This non-constant

behavior is a function of operating condition and results in a non-linear behavior.

64
 (a)

(b)
Figure 3. 7. Open loop simulation to characterize ORC system time constant for step changes in (a)
exhaust mass flow (Test A) and (b) exhaust temperature (Test B). [τ is the time constant for the response
of the working fluid (WF) temperature].

It can be also inferred from the diminishing returns in turbine power with increased working

fluid temperature that, at a constant working fluid flow there is an optimum working fluid

65
superheat level. Note that turbine power is a function of working fluid mass flow and the enthalpy

change across the turbine as shown in equation (3.1) below,

𝑃 = 𝑚̇ ∗ (ℎ − ℎ ) (3.1)

Where, 𝑚̇ is the working fluid mass flow rate, ℎ is the enthalpy of the working fluid going

into the turbine and ℎ is the enthalpy of the working fluid out of the turbine.

As the temperature of the working fluid rises, its heat carrying capacity reduces saturating

enthalpy, ℎ and hence further increase in the exhaust power do not produce substantial gains in

turbine power. Therefore 𝑃 is nearly constant for the last step change as seen from Figure

3.7 (a).

Similarly, step changes in exhaust flow expose nonlinear evaporator dynamics, as shown in

Figure 3.7 (b). However, the working fluid temperature changes are temporally damped. This can

be attributed to the evaporator design and the reduced time available for heat transfer as the exhaust

flow is increased. In summary, the evaporator non-linearity or the evaporator time constant is a

function of engine operating conditions.

3.5.2 Controllable Inputs: Working fluid mass flow

The control input for an ORC system is the pump speed which manipulates the working fluid

flow through the system to maintain vapor state at the evaporator outlet. In this case study, the

working fluid flow is subjected to step changes to examine the system dynamics. The results of

the working fluid step changes correlates with the experimental open loop analysis shown in

chapter 1. For every step increase in working fluid flow rate the evaporator outlet temperature

decreases. Correspondingly, the increase in working fluid mass flow increases the turbine power

output, which correlates with Equation 3.1.

66
Figure 3. 8. Open loop simulation to characterize ORC system time constant for step changes in working
fluid flow.

As shown in Figure 3.8, at 1800 sec (red dotted line) further step change in working fluid flow

results in saturation. This loss in vapor quality arises for two reasons, 1) increasing the mass flow

increases saturation pressure resulting in higher saturation temperature (bottom plot of Figure 3.8),

reducing the superheat magnitude; 2) increasing the working fluid mass flow rate reduces the time

available for heat transfer. Thus, for a given exhaust condition there is only one optimum working

fluid flow for maximum turbine power generation and further increase of working fluid flow will

lead to saturation, inhibiting turbine operation.

Another important conclusion is that the given ORC system is open loop stable since for a

finite change in model inputs (𝑚̇ ,𝑇 , 𝑚̇ ), the model outputs (𝑇 ) reaches a stable steady

state values, an important fact to be considered when designing an appropriate closed loop control

law.

67
3.6 Conclusion

In this Chapter, an overview of different parameters influencing the dynamics of ORC-WHR

system was presented. It was shown that the thermal inertia of the evaporator can be characterized

by means of sinusoidal heat source with varying time period. Large size evaporators adds buffering

towards high frequency heat source whereas the small size evaporators will respond faster to such

high frequency inputs. This discrepancy in response time creates additional control challenge.

It was also demonstrated that for a fixed valve position, working fluid flow linearly changes

the system pressure. This in turn proportionally changes the saturation temperature. It can be thus

concluded that the ideal operating condition would be where a slight change in pressure would not

affect saturation temperature substantially.

Open loop analysis showed the nonlinear nature of the evaporator, which is a function of engine

operating conditions. The system time constant varies with exhaust mass flow, exhaust

temperature and working fluid mass flow rate. From the open loop simulation it can be inferred

that the ORC system is open loop stable. Since for each step changes in the input conditions the

system arrived at a stable steady state condition. Further, it was also shown that for each exhaust

condition only one optimal working fluid flow exist that provides the maximum turbine power

output. An ideal controller will try to achieve this optimal point at transient engine conditions

without inhibiting turbine operation nor violating the working fluid temperature constraints.

68
 CHAPTER 4. CONTROL DESIGN

4.1 Introduction

In this Chapter, a model predictive controller is designed for optimal control input based on

the cost function objective. It is shown that for optimal turbine power generation low superheat is

desired, consequently a superheat tracking controller is designed and the performance of designed

controller is simulated over step inputs, drive cycle and a real-world sinusoidal time-varying

exhaust conditions.

4.2 Background

Under highly fluctuating exhaust conditions, the main control requirement for uninterrupted

turbine expander operation is to maintain the working fluid in a superheated state. Prior works

have shown superior ORC-WHR performance through utilization of advanced control strategies

like linear/nonlinear model predictive control [18-20, 23, 28], supervisory predictive control [21],

and extended prediction self-adaptive control [22, 29] compared to traditional PID control.

Feru et al. [19] developed a switching MPC control strategy linearized over three operating

conditions. Their proposed control strategy produced 3% increase in average output power relative

to a classical PI strategy. In their simulation study, Esposito al. [20] proposed a nonlinear MPC

where the receding horizon optimization problem was solved using a Particle Swarm Optimization

(PSO) algorithm.

Hou et al. [21] developed a supervisory predictive control with constrained regulatory and

economy targets within the optimization procedure. Their proposed strategy demonstrated less

overshoot and shorter settling time under process and measurement disturbances in comparison to

a traditional optimized PID control. Hernandez et al [22] evaluated the Extended Prediction Self-

Adaptive Control (EPSAC) for multivariable systems (in this case, an ORC system). EPSAC used

69
input/output models for prediction, avoiding the need for state estimators. Simulation results

suggested that for a steady heat source, PID control and EPSAC produced similar performance.

However, for an unsteady heat source, EPSAC outperformed the PID controller in regulating the

superheat temperature.

The next frontier is enhancing the MPC performance through a look-ahead approach.

4.2.1 Look-ahead ORC control

All of the aforementioned advanced control strategies have considered exhaust mass flow and

temperature as disturbance inputs. The controller designs only consider the current exhaust

condition for computing the control inputs for the complete prediction horizon. A key MPC

capability is explicitly accounting for future disturbances, which automatically results in a

feedforward control action [37]. Hence, better knowledge of disturbances over an extended

horizon which will ensure more optimal control. Many research works [38-45] have exploited the

potential of look-ahead control strategies utilizing future road topography and traffic conditions

for fuel economy improvements.

Ganji, and Kouzani [38] have shown hybrid electric vehicle (HEV) performance and efficiency

improvements by optimizing the split ratio between engine and battery power utilizing future road

conditions. Gong et al. [39] evaluated trip based optimal power management of a plug-in hybrid

electric vehicle (PHEV) using a Dynamic Programming (DP) algorithm with and without historic

traffic data. In comparison to a simple trip model without historic data, simulation results showed

an improvement of 18% in fuel economy due to usage of historic traffic data. Similar energy

management optimal control strategies have been evaluated in a PHEV application by multiple

researchers [40-42].

70
 Huang and Bevly [43], investigated setting optimal vehicle speeds for a Heavy Duty truck to

save fuel and trip time utilizing roadway geometry and a constrained Nonlinear Programming

solver. Simulation results showed that the truck driving at optimal speed was able to significantly

reduce fuel consumption with small penalty in travel time, when compared to constant speed cruise

driving. Whereas Turri [44], exploited platooning and look-ahead control for reducing fuel

consumption in a MPC framework. In this work, the optimal speed profile was established for the

whole platoon rather than the first vehicle. This was done not only to reduce fuel consumption but

to also maintain safe inter-vehicular distances. Additionally, Gáspár and Németh [45] evaluated a

multi-objective optimization problem to predict an adaptive energy-efficient cruise control

strategy based on the motion of surrounding vehicles. This paper proposed a look-ahead control

utilizing several factors for calculating the optimal vehicle speed, such as: energy reduction, road

slopes, traveling time, and speed limits.

4.2.2 Opportunity and Uniqueness of the current work

While many studies [18-20, 23, 28] have exploited the benefits of complex model based control

strategies, all the published works utilize current exhaust conditions for prediction of the optimal

control input. The approach pursued in this research study, differs in two ways.

Firstly, in this study, the potential benefits of look-ahead control strategy for an ORC system

are investigated. It is assumed that the future vehicle speed can be predicted utilizing road

topography and V2V connectivity [43-45]. This vehicle speed is then used to predict the engine

speed and torque, culminating in the estimation of future exhaust conditions. The predicted exhaust

conditions are then used to estimate the optimal control inputs in this MPC framework.

Secondly, MPC is inherently a state feedback control scheme using the current state and a

system model for prediction. In literature, an extended Kalman filter (EKF) is widely used for

71
nonlinear state estimation by successive linearization of the nonlinear model at current predicted

states [46]. In this study, the EKF is augmented with a disturbance model for offset free MPC

tracking [47-48]. The performance of the proposed controller is then compared to the baseline

NMPC control strategy, which lacks the look-ahead feature.

This Chapter is organized as follows: first the system under consideration is described,

followed by control oriented modeling methodology. Then the control strategy is described with

the augmented state estimator. The augmented state estimator performance is evaluated for

disturbance rejection. Then MPC’s preview capability is exploited in the next section. For the look-

ahead control strategy, future exhaust conditions are considered when calculating the optimal

control input for the given prediction horizon. Performance of the look ahead NMPC strategy is

then evaluated relative to baseline NMPC performance for a step sequence exhaust conditions and

for a drive cycle.

4.3 Moving Boundary Control Model

For control design, it is desired to have a low-computation model which is very efficient in

capturing all the system dynamics of the evaporator. The finite volume model presented in Chapter

2 with 100 cells results in 300 states and is not acceptable for online implementation of MPC

strategy. Therefore, in this section a reduced order control oriented model is presented.

During normal operation, i.e. typical heat exchange process, the evaporator contains working

fluid in three different phases: liquid, mix and vapor. There are two boundaries that subdivide the

evaporator into these three phases. The main function of the MBM is to track the length of each

working fluid phase inside the evaporator by determining the boundary locations. Figure 4.1

illustrates the three working fluid phases for the MBM approach.

72
 Figure 4. 1. Schematic representation of Moving Boundary Model used for evaporator control-oriented
modeling (counter flow).

Applying energy balance equations to the working fluid and wall in each phase results in a 6-

state MBM [28, 49]. Dynamics of the exhaust gases are neglected due to their fast transient

characteristics. The derivation of the MBM assumes the existence of all three phases. The system

of differential equations are summarized below:

Liquid region:

Phase length dynamics in the liquid phase (𝐿 ):

dL1
 f ,1  h f ,1  h f ,l  
dt
1   f ,1  dh

2
AL1   f ,1 
h
 
h f ,1  h f ,l  in 
  dt (4.1a)
m f ,in  h f ,in  h f ,l    dtube L1U fw,1 Tw,1  T f ,1 

Wall temperature dynamics (𝑇 , ):

dTw,1 dL1
Ac p  w L1  Ac p  w Tw,1  Tw,l  
dt dt
 dtube L1U fw,1 T f ,1  Tw,1  
(4.1b)
 d shelleqv L1mHTCU g , w TTP ,1  Tw,1 

Two-phase region:

Phase length dynamics for the mixed phase (𝐿 ):

73
 dL1 dL
A   f ,1  h fl  h fg    A 1      f ,l  h f ,l  h fg   2 
dt dt
1  f ,1 dhin
 AL1
2 h dt
 h fl  h fg  
m f ,in  h f ,l  h f , g   (4.2a)

 d tube L2U fw,2 Tw,2  T f ,2 

Wall temperature dynamics (𝑇 , ):

dTw,2 dL1
Ac p  w L2  Ac p  w Tw, l  Tw , g  
dt dt
dL2
Ac p  w Tw ,2  Tw, g    d tube L2U fw ,1 Tsat  Tw ,2  
dt (4.2b)
 d shelleqv L2 mHTCU g , w TTP ,2  Tw,2 

Vapor region:

Evaporator outlet enthalpy dynamics (ℎ , ):

dL
 
A   f ,3 h f , out  h f ,3   f ,1  h f , g  h f ,out   1 
  dt
dL
 
A   f ,3 h f , out  h f ,3   1     f ,l   f , g  h f , g  h f ,out   2 
  dt
1   f ,3  dh f , out
2
AL3   f ,3 
h

h f ,out  h f ,3   
  dt
1   f ,1  dh
 AL1   h fg  h f , out   in  (4.3a)
2  h  dt
m f ,in  h f , g  h f , out  
 dtube L3U fw,3 Tw,3  T f ,3 

Wall temperature dynamics (𝑇 , ):

74
 Tw,3 dL1
Ac p  w L3  Ac p  w Tw, g  Tw,3  
t dt
dL2
Ac p  w Tw, g  Tw,3  
dt
 d tube L3U fw,3 T f ,3  Tw,3   (4.3b)

 d shelleqv L3mHTCU g , w TTP ,3  Tw,3 

where 𝐿 = 𝐿 − (𝐿 + 𝐿 ); 𝑑 and 𝑑 are the heat exchanger hydraulic diameters on the

working fluid and exhaust gas sides, respectively; 𝑚 is a calibration parameter for the gas side

heat transfer coefficient, and 𝜂 is a calibration parameter accounting for heat loss between the

exhaust gas and the ambient. The subscripts 𝑖 = 1,2, 3 stand for liquid, two-phase, and vapor

regions, respectively.

Assuming the exhaust gas temporal dynamics are ignored, the exhaust gas temperature

evolution is predicted by the following algebraic equations:

1 g , wTw,1  mg C pg 2Tg ,2  2Tg ,3  Tg , in 

 dtube LU  
Tg ,1   (4.4a)
m g C pg   d tube LU
1 g,w

 dtube L2U g , wTw,2  m g C pg 2Tg ,3  Tg ,in 

Tg ,2   (4.4b)
m g C pg   dtube L2U g , w

 d tube L3U g , wTw ,3  m g C pgTg , in 

Tg ,3   (4.4c)
m g C pg   d tube L3U g , w
Thus in this model based design the evaporator system consists of 6 states,

𝐿 ,𝑇 , ,𝐿 ,𝑇 , ,𝑇 , ,ℎ , .

75
4.4 NMPC Problem Formulation

For the ORC system under consideration, the primary control input is a working fluid mass

flow, which is actuated via pump speed manipulation. The NMPC is then formulated as a finite

horizon optimal control problem subject to system dynamics, as well as state and input constraints.

For control formulation, the resulting differential and algebraic equations in (4.1 – 4.4) can be

written in the standard nonlinear state space form:

𝑥̇ = 𝑓(𝑥, 𝑧, 𝑤, 𝑢)

𝑧 = 𝑔(𝑥, 𝑤) (4.5)

𝑇
where 𝑥 = 𝐿 , 𝑇 , ,𝐿 ,𝑇 , ,𝑇 , ,ℎ , is the dynamic state vector; 𝑧 = 𝑇 , , 𝑇 , , 𝑇 , is the

algebraic state vector; 𝑢 = 𝑚̇ , is the control input; 𝑤 = 𝑚̇ , 𝑇 , is an exogenous known

disturbance vector (input from the engine to the ORC system) containing future predictions.

Then the NMPC formulation is written as follows:

𝑚𝑖𝑛 𝐽(𝑥(𝑡), 𝑢(. ))

(∙),𝒙(∙)

𝑥̅̇ (𝜏) = 𝑓 𝑥̅ (𝜏), 𝑧̅(𝜏), 𝑢(𝜏), 𝑤(𝜏)

𝑦(𝜏) = ℎ 𝑥̅ (𝜏), 𝑧̅(𝜏), 𝑢(𝜏), 𝑤(𝜏) + 𝑑(𝑡)

𝑥̅ (𝑡) = 𝑥 (𝑡)

𝑑(𝜏) = 𝑑(𝑡) ∀𝜏 ∈ 𝑡, 𝑡 + 𝑇
(4.6)
S.t. :
𝑦 ≤ 𝑦(𝜏) ≤ 𝑦 , ∀𝜏 ∈ 𝑡, 𝑡 + 𝑇

𝑢 ≤ 𝑢(𝜏) ≤ 𝑢 , ∀𝜏 ∈ [𝑡, 𝑡 + 𝑇 ]

𝑢(𝜏) = 𝑢(𝑡 + 𝑇 ), ∀𝜏 ∈ 𝑡 + 𝑇 , 𝑡 + 𝑇

𝑥 ≤ 𝑥̅ (𝜏) ≤ 𝑥

76
where 𝐽: ℝn × ℝm → ℝ is the performance index for optimization; 𝑛 and 𝑚 are dimensions of

state and input; 𝑇 and 𝑇 denote the prediction and control horizon, respectively with 𝑇 ≤ 𝑇 ,

and 𝑑(𝑡) is the disturbance state added for offset free tracking. The superscripts 𝑙𝑏 and 𝑢𝑏 indicate

the lower and upper bounds of the constrained variables, respectively. The bar (∙̅) denotes predicted

variables based on the control model using the estimated state feedback, 𝑥(𝑡), and predicted input

𝑢.

The NMPC computes the control input at each time instant 𝑡 by minimizing the performance

index, J.

(4.7)
𝐽= 𝑊 (𝑦(𝜏) − 𝑦 (𝜏)) + 𝑊 Δ𝑢

where 𝑦 is the reference working fluid superheat to be tracked by the output 𝑦, and Δ𝑢 is rate of

change of working fluid mass flow. Weights, 𝑊 and 𝑊 are tuned to attain the desired control

performance.

4.5.1 ACADO Implementation

For the current NMPC simulation study, ACADO [50-52] an open-source software

environment, has been used to implement the proposed controller scheme. ACADO implements

real-time iteration scheme to solve optimal control problems. The ACADO toolkit is equipped

with a collection of algorithms for solving various automatic control and dynamic optimization

problems. ACADO is also capable of automatically generating a highly efficient s-function code

for fast MPC applications using MATLAB/Simulink while performing a robust optimization for

medium to large scale non-linear problems.

77
4.6 Disturbance Model and Observer Design

NMPC is inherently a state-feedback algorithm and assumes the availability of full-state

information through measurement or estimation at the initial condition. In order to achieve offset-

free tracking in this formulation, the nominal system dynamics (Equation 4.5) are augmented with

additional integrating states or disturbances [53-54]. The general augmented linearized system is

given as follows:

𝑥 (𝑡 + 1) = 𝐴 𝑥 (𝑡) + 𝐵 𝑢(𝑡) (4.8)

𝑦(𝑡) = 𝐶 𝑥 (𝑡)

In which:

𝑥
𝑥 = is the augmented state matrix, where 𝑑 is the disturbance and is assumed to stay constant
𝑑

over the prediction horizon.

𝐴 𝐵 𝐵
𝐴 = , 𝐵 = , 𝐶 = [𝐶 𝐶 ]
0 𝐼 0

In this study, the values 𝐵 = 0 and 𝐶 = 1 are utilized making the augmented system as an

“output integrator” [55]. The augmented states in equation (4.8) is estimated at each time step

given the output measurement by means of a posteriori observer, containing a prediction phase

and correction phase.

Prediction Phase: For forward prediction of the state trajectories 𝑥 (𝑡 + 1), an implicit numerical

integration method called Rosenbrock-Wanner (ROW) is used.

𝑥 (𝑡 + 1) = 𝐴𝑥 (𝑡) + 𝐵𝑢(𝑡) (4.9)

For the ORC system dynamics as stated in equations (4.1 – 4.4), a two-stage ROW method

described below [28] is implemented to estimate 𝑥 (𝑡 + 1):

78
 𝑥 = 𝑥 + 𝑑𝑡 𝑏𝑘

(4.10)
𝑊 𝑘 = 𝑓 𝑥 + 𝑑𝑡 𝑎 𝑘 + 𝑑𝑡𝐽 𝑑 𝑘 , 𝑖 = 1,2

𝜕𝑓(𝑥 )
𝑊 = 𝐼 − 𝑑𝑡𝑑 𝐽 , 𝐽 =
𝜕𝑥

The coefficients for the two-stage ROM method are: 𝑎 = ,𝑑 =1+ ,𝑑 = − ,𝑏 =

√

& 𝑏 = 3/4. For detailed derivation of the ROM method, one can refer to [56].

Correction Phase: In the correction stage, an Extended Kalman Filter (EKF) is utilized,

𝑥 (𝑡 + 1) = 𝑥 (𝑡 + 1) + 𝐾 [𝑦(𝑡 + 1) − 𝑦 (𝑡 + 1)] (4.11)

𝐾
where 𝐾 = is the augmented state matrix, 𝐾 is the Kalman gain, and 𝐾 is the disturbance
𝐾

gain that is tuned to achieve the required performance for overshoot and rise time for a unit step

change.

Note that the augmented state formulation (4.8) is only used to obtain and define the states (𝑥, 𝑑 )

but is not used in the NMPC formulation (4.5).

4.7 Simulation Results

In this section simulation results are presented to demonstrate the performance of the proposed

controller over step inputs in exhaust conditions and for a drive cycle. For this simulation study,

all the working fluid flow is passed through turbine by keeping the bypass valve closed and turbine

valve completely opened. The evaporator pressure is then a function of working fluid flow rate,

more flow will result in higher evaporation pressure and vice-versa.

79
 The NMPC designed in this section utilizes 100 prediction steps (N) with model time step (Ts)

of 0.6 seconds, making the prediction horizon (Hp) 60 seconds. Note that in simulation studies

prediction horizon of less than 60 sec resulted in short-sightedness for the MPC to solve the optimal

problem. The control horizon is set equal to the prediction horizon.

4.7.1 NMPC Cost function formulation

The main objective of the ORC-WHR system is to recover as much heat energy as possible

through the turbine shaft. For this case, the NMPC cost function was formulated to track a constant

fraction of exhaust power to be recoverable. A quadratic cost function was formulated as stated

below,

( , ) ( )
𝐽=𝑊 +𝑊 (4.12)

A superheat constraint is imposed, 30 < 𝑆𝐻 < 150 and the other parameters are as follows

𝑃 = 100 kW, 𝑃 = 15 𝑘𝑊, 𝑑𝑢 = 0.0025, 𝑑𝑢 = 0, 𝑊 = 1 , 𝑊 = 5.

Figure 4. 2. Generated sinusoidal exhaust mass flow with varying time periods to test the NMPC
controller

80
 To test the controller and the cost function, a sinusoidal exhaust condition was generated as

shown in Figure 4.2. In a transient condition, exhaust mass flow changes frequently with RPM and

requested torque, whereas the change in exhaust temperature is very slow due to the thermal inertia

of the after-treatment system. Hence in the generated sinusoidal heat source signal, exhaust

temperature was kept constant and the time period of the mass flow was varied from 4s to 240 s.

The performance of the formulated MPC with objective function for maximizing the turbine

power is shown in Figure 4.3. It can be seen from Figure 4.3 that during the initialization section

from 0 to 2000 sec, MPC brings the superheat as low as the constraints allow in order to maximize

the turbine power. Once the oscillatory heat signals are initiated at 2000 sec, the lower time period

signals don’t show significant response from the TP evaporator, which was an expected behavior

as explained in section 3.3 of Chapter 3. As the time period keeps increasing, there is enough time

for the heat transfer to occur from exhaust gas to working fluid and thus the working fluid

temperature at the evaporator outlet becomes more responsive to changes in heat source signal.

The controller response to changes in the working fluid temperature can be seen in the top plot of

Figure 4.3 (b). As the time period keeps on increasing the amplitude of the working fluid flow

increases. Now, since all the valve positions are fixed, a change in working fluid causes the

saturation pressure to change accordingly. This can be seen in the bottom plot of Figure 4.3 (b).

This causes the saturation temperature to change and since pressure dynamics is faster compared

to temperature dynamics, at the on-set of 120 sec heat signal, working fluid saturates.

81
 (a)

(b)
Figure 4. 3. Power maximization MPC performance for the exhaust condition shown in Figure 4.2. Plot
(a) shows the response of the working fluid temperature at the evaporator outlet to different sinusoidal

82
inputs with the superheat plot shown in the bottom graph. Plot (b) shows the controller generated control
input and corresponding response of the saturation pressure.

The nature of the cost function, drives the MPC solution to keep the superheat as low as

possible in order to maximize the turbine power. Although, the controller is not violating the

minimum superheat constraint but its formulation is not helping ORC-WHR dynamics. For

maximum turbine power output, minimum superheat is required which was verified from the open

loop analysis and the turbine power as cost function. This suggests that tracking the power might

not be the ideal cost function objective. Hence the cost function was modified to track superheat

instead, as shown in equation 4.13.

( ) ( )
𝐽=𝑊 +𝑊 (4.13)

With 𝑆𝐻 = 200, 𝑆𝐻 = 20°𝐶, 𝑑𝑢 = 0.005, 𝑑𝑢 = 0, 𝑊 = 1 and 𝑊 = 5.

It was found that for the TP evaporator alone, the minimum superheat that could be tracked was

20 ˚C and the results of which are shown in Figure 4.4, for input heat conditions of Figure 4.2.
Superheat, ° C

(a)

83
 Control Input, kg/s
Pressure, bar

(b)
Figure 4. 4. Plot (a) Superheat tracking MPC performance for the exhaust condition shown in Figure 4.2.
Plot (b) shows the controller generated control input and corresponding response of the saturation
pressure.
Two important observations can be made from the results of superheat tracking shown in

Figure 4.3. First, the control effort for the lower time period doesn’t change at all as the thermal

inertia of the evaporator is dampening the fast changing heat signals. Second, with the increase in

time period of the heat signals, the dynamic response increases and significant deviation in the

control input is observed. Both these observations correlate with what was observed in the open

loop simulation study shown in Chapter 3.

4.7.2 Disturbance Rejection

In a WHR-ORC system, the two main sources of disturbances are, 1) a mismatch between the

plant and control models, and 2) measurement noise. Under such circumstances it is crucial for

any controller to have robust performance and track the desire temperature as accurately as

possible avoiding any steady state tracking error. The performance of the NMPC when utilizing

84
the disturbance augmented state estimator is evaluated in this section to evaluate the capability of

the augmented system.

A. Plant and Control Model Mismatch

In this study, the FVM ORC plant model is more accurate than the 3 cell MBM control oriented

model due to discretization disparities. In this section, we show how the disturbance augmented

state estimator enhances the temperature tracking performance of the NMPC by addressing the

mismatch in FVM and MBM evaporator model output predictions. Figure 4.5 below shows the

step sequence of the exhaust test condition utilized for this comparison.
Exhaust Mass Flow, kg/hr

Exhaust Temperature, ° C

Figure 4. 5. Step sequence of the exhaust test conditions for plant and control model mismatch
comparison

Figure 4.6 shows the NMPC performance comparison with and without the disturbance model

for the test conditions shown in Figure 4.5. It can be clearly observed that, depending on the

exhaust condition, there is an offset of 3-10˚C without the disturbance model. Augmenting with

the disturbance model decreases this offset to less than 1˚C.

85
 Superheat Tracking Error, ° C

(a)
Control Input, kg/s
Pressure, bar

(b)
Figure 4. 6. NMPC performance comparison with and without the augmented disturbance model during
a period of mismatch between the plant and control models.

The performance of the augmented model can be seen as equivalent to integral action, which

assists in the elimination of steady state error. The mean absolute error using the baseline NMPC

control design was found to be 7.26˚C whereas inclusion of the disturbance model reduces the

86
mean absolute error to 0.46˚C. Thus, the inclusion of disturbance model reduces the offset in

tracking as expected.

B. Measurement Noise

Exhaust gas mass flow rate and exhaust temperature determine the waste heat power available

for recovery in the evaporator. On the vehicle, exhaust mass flow rate is available via the ECU

whereas the exhaust temperature is measured via thermocouples. In this section, to demonstrate

effectiveness of the NMPC and augmented estimator in the face of measurement noise/error, the

control oriented model is purposely sent faulty thermocouple temperature values. Precisely, a

constant temperature difference/offset (-10˚C, + 40˚C) is added to the control model input and the

NMPC performance is evaluated as shown in Figure 4.7. The plant model is sent the

corrected/actual values (+10˚C, - 40˚C) without any offsets and no changes were made to the

exhaust mass flow measurements.

On perturbing the offsets, it was found that a maximum of 10 ˚C could be applied to the control

model without affecting the controller performance whereas a maximum of -40 ˚C on the opposite

end of perturbation was successfully simulated. As seen from Figure 4.7, even with a -40˚C inform

erroneous exhaust temperature sensing, the disturbance augmented NMPC performance is

admirable. It is interesting to note that, for negative offsets overshoot are larger. This can be

attributed to the slower response of the control input arising due to the integral action of the

disturbance rejection. Setting a value for 𝐾 in equation (4.11), is often a trade-off between

decreasing overshoot and increasing settling time. In this study, 𝐾 was chosen to be a constant

for all offset conditions and, as a result, the overall response for the -40˚C offset case produced

more overshoot and increased settling time. It is also worth noting that the overshoot and settling

87
time characteristics vary with the operating conditions, which further emphasizes the nonlinearity

of the heat exchanger model.

Figure 4. 7. Disturbance augmented NMPC performance comparison for various constant temperature
offsets between plant and control model.

4.7.3 Look-ahead NMPC

In this section, a look-ahead non-linear model predictive controller (NMPC) for tracking the

working fluid evaporator outlet temperature is described. Specifically, the proposed control

scheme consists of an NMPC to regulate the evaporator outlet working fluid temperature and a

state estimator, which is augmented with a disturbance model. The stated approach addresses and

explores two distinct features. First, the proposed MPC approach has a preview capability – i.e.

the MPC incorporates future reference information into the control problem to improve the

controller performance. This addresses and enhances the predictability of the nonlinear evaporator

88
dynamics. Second, the control oriented evaporator model is based on empirical correlations which

are bound to differ from the actual component design and other disturbances like measurement

noise. Hence, the inclusion of a disturbance model in the estimator design contributes to off-set

free tracking by considering the model and plant mismatch. A schematic of the proposed closed-

loop control law is shown in Figure 4.8.

Figure 4. 8. Proposed control law for the ORC-WHR system.

In this study, it is assumed that the exhaust conditions are known for the complete drive cycle.

These conditions are stored in the Future exhaust conditions block in Figure 4.8. The Future

exhaust conditions block then sends out the current and future exhaust conditions based on the

MPC prediction horizon. In future studies, this block will be replaced by an engine model with

engine speed and torque as inputs, and exhaust temperature and mass flow as outputs.

In the following section simulation results are presented to demonstrate the performance of the

proposed controller over step inputs in exhaust conditions and for a drive cycle. For this simulation

study, the evaporator bypass valve is completely closed and turbine valve is completely opened,

allowing the evaporation pressure to vary proportionally with working fluid mass flow rate.

89
A. Drive Cycle

Figure 4.9 shows an aggressive heavy duty diesel drive cycle run on the Clemson University’s

state of the art heavy duty engine dynamometer. ETAS INCA was used to record ECU parameters

to calculate real time exhaust mass flow rate. Thermocouple measurements of exhaust gas

temperature were recorded in the tailpipe just before the evaporator . This drive cycle was

developed by our project sponsor to represent a typical heavy duty truck during highway driving.

The drive cycle was initiated once the engine was completely warmed.

1400 1500
Speed, rpm

Torque, Nm
1300
Engine

Engine
1200 1000

1100

1000 500
0 200 400 600 800 1000 1200
time, s
900 450
mass flow, kg/hr

Temperature, ° C
800
Exhaust

Exhaust
400
700
350
600

500 300
0 200 400 600 800 1000 1200
time, s

Figure 4. 9. HDDE Drive Cycle exhaust test conditions for NMPC analysis

The bottom plot of Figure 4.9 shows that the rate of change of exhaust temperature doesn’t

correpsonds to that of exhaust mass flow. As explained before, this is attributed to the thermal

inertia of the aftertreatment system, which is located upstream of the evaporator and reduces the

high frequency thermal components of the varying exhaust gases. This recorded exhaust conditions

are used for the following simulation studies.

90
 Without preview
With 60s preview
30 Reference

20

10

0
0 200 400 600 800 1000 1200

Without preview
14 With 60s preview

12

10

6
0 200 400 600 800 1000 1200
time, s
(a)
Control Input, kg/s
Pressure, bar

(b)
Figure 4. 10. Look-ahead NMPC performance (‘with preview) compared to the baseline NMPC (‘without
preview’) during a drive cycle.

It can be observed from Figure 4.10, that the performance of both control strategies is

acceptable in keeping the temperature error within ±10℃ during this heavily transient drive cycle.

A quantitative summary of temperature error for both strategies is shown in Table 4.1.

91
 While the impact of the look-ahead NMPC on temperature error is minimal, there is a

substantial reduction in control action with preview capability. This reduction in control action

will be beneficial by increasing pump durability and reducing pump power consumption. The

controlled working fluid mass flow change will also inhibit localized abrupt pressure changes

inside the evaporator, resulting in uniform temperature distribution across the evaporator and better

estimation of working fluid phase lengths inside the evaporator.

Mean Absolute Max Absolute Total Turbine

Temperature Error, Temperature Error, Energy
˚C ˚C produced, kJ
Without Preview 3.14 12.32 13.216
Look-ahead NMPC 4 14.1 13.184
Table 4. 1. Quantified NMPC temperature error comparison with and without preview capability for the
drive cycle in Figure 4.9.

I. NMPC Preview time sweep

The look-ahead NMPC preview time window was altered to assess its impact on NMPC

performance. For the drive cycle shown in Figure 4.9, three different scenarios were evaluated.

The minimum prediction horizon (Hp) for stable NMPC performance was established as 60s.

Subsequently, preview time windows of 10s, 30s, and 60s were simulated. More specifically, for

a preview time of 10s, only 10s of future exhaust conditions were provided to the NMPC and, for

the remimaing 50s of the prediction horizon, the exhaust conditions were kept constant. A

summary of these conditions with results is shown in Table 4.2.

Figure 4.11 compares all three preview time scenarios. Similar conclusions to Figure 4.10

results can be made. No significant difference in temperature error prediction was observed

whereas less control action overshoot was observed as the preview time increased. The 10 sec

preview time case is tending towards the case of without preview from Figure 4.10. Figure 4.11

again exhibits unexpected results in terms of temperature tracking performance. No clear trend in

92
evaporator temperature error reduction emerged from the preview window length alterations. In

an ORC system, the evaporator experiences a slower thermodynamic response because of its

thermal inertia. This slow response is due to the heat capacity of the heat exchanger tube wall as

well as the flow regime dependent heat transfer characteristics of the working fluid. These heat

transfer characteristics are a function of the evaporator tube design, material, and working fluid

properties. The evaporator design in this study was capable of filtering out large amplitude exhaust

condition fluctuations, resulting in a more robust system and a more effective dynamic control.

Therefore for a simplified ORC system operation, which in this case is a single evaporator

operating under a constant pressure working fluid flow, knowledge of future exhaust stream

conditions does not necessarily enhance the temperature tracking ability of the controller.
Superheat, ° C
Control Input, kg/s

Figure 4. 11. Look-ahead NMPC performance comparison with different preview window sizes (total
preview time).

93
 Preview time Prediction Mean absolute Max absolute
Horizon Temperature error, temperature error,
time ˚C ˚C
10s 60s 4.71 16.16
30s 60s 3.96 14.07
60s 60s 4.01 14.06
Table 4. 2. NMPC temperature error for different preview times of Figure 4.10.

B. Sinusoidal Input:

To further investigate the look-ahead NMPC effectiveness under a highly fluctuating heat

loads, a sinusoidal heat exhaust source with varying frequency mass flow oscillations was designed

and the controller performance was evaluated. Due to the substantial temperature trend dampening

of the after-treatment system, high frequency oscillations in exhaust temperature at the evaporator

inlet are not possible. Thus, for this exercise, the exhaust temperature was kept constant at 350˚C

while the mass flow frequency was varied as shown in Figure 4.12. For the sake of simplicity, a

sinusoidal heat source was considered where half of the sine wave can be assumed to be a

representative of acceleration and the other half a deceleration action. The time period of sine wave

varied from 4 sec to 240 sec and was applied for 600 sec to give enough time for the controller to

achieve steady state. Figure 4.12 shows the same sinusoidal signals as shown in Figure 4.2, except

the initialization time period.

94
 Figure 4. 12. Sinusoidal exhaust test conditions for comparison of the look ahead and baseline NMPC.

Figure 4.13 shows the performance of both the look-ahead NMPC (‘with preview’) and

baseline NMPC (‘without preview’) NMPC controllers. The look-ahead preview time (Hp) was

chosen to be 60 sec. Like previous tests results, the top plot from Figure 4.12 (a) shows that the

temperature error from both NMPC controllers are comparable. The summary of mean and max

temperature error for each of the sine wave frequency is quantified in Table 4.3. For heat source

fluctuations with smaller time periods, the temperature error is small and the error increases with

the period of the heat source oscillation. For smaller heat source oscillation time periods (4 sec to

20 sec), the thermal inertia inherent to the evaporator design dampens the heat source oscillations,

resulting in only minimal changes in the working fluid mass flow actuation from the controller.

95
 Superheat, ° C
Turbine Power, kW

(a)
Control Input, kg/s
Pressure, bar

(b)
Figure 4. 13. Performance comparison for the look-ahead NMPC (‘with preveiw’) and the baseline
NMPC (‘without preview’) while subjected to the sinusoidal heat source. Hp=60 s. SH Reference = 20˚C

96
 From Table 4.3 and Figure 4.13, it can be concluded that the mean and max error for both the

controllers are comparable for all the sine waves with the exception of 60s, 90s and 120s time

periods. The maximum gain was seen from the 90s time period, the look-ahead NMPC strategy

showed 40% improvement in mean superheat error reduction and 37% in the max superheat error

reduction. While at higher time periods look-ahead NMPC strategy generated slower control

response, affecting the temperature error mitigation, as seen from 240s time period response.

Time period Mean Superheat error, ˚C Max Superheat error, ˚C

Without Look-ahead Without Look-ahead
Preview NMPC Preview NMPC
4s 0.22 0.09 0.36 0.33
10s 0.24 0.26 0.55 0.66
20s 0.52 0.67 1.42 2.1
60s 3.77 3.29 7.62 5.9
90s 6.93 4.17 13 8.2
120s 5.66 4.9 11.46 10.97
240s 3.25 5.46 8.3 8.98
Table 4. 3. NMPC temperature error comparison for different time periods of the sinusoidal input of
Figure 4.11.

The main advantage of look-ahead NMPC is shown in the control action plot of Figure 4.13

(b). The control action of the look-ahead NMPC (‘with preview’) is smooth and well regulated as

compared to the baseline NMPC. The reduction in amplitude of the control action is summarized

in Table 4.4 using equation (4.14).

%𝐴 = (𝐴 , − 𝐴 , ) ∗ 100⁄𝐴 , (4.14)

where A is the amplitude of the control signal, 𝑤𝑜 = without and 𝑤 = with.

97
 Sinusoidal Time % reduction in control
period input amplitude Look-
ahead NMPC
4s 95.38
10s 95.16
20s 85.38
60s 56
90s 33.55
120s 25.91
240s 20.25
Table 4. 4. Percentage reduction in control input amplitude for the look-ahead NMPC (‘with preview’)
for different time periods of the sinusoidal input in Figure 4.11.

From the results of Table 4.3 it was concluded that for certain frequency having the preview

capability helps maintain superheat within tight limits. To prove this, a further reduction in

superheat reference was initiated and the simulated results are shown in Figure 4.13. As expected

at lower time periods there was no change in superheat prediction. But at time period of 90 sec

which showed the maximum error in Table 4.3, the control strategy without preview saturates

inhibiting turbine operation and in turn reducing the power producing capability.

It is important to note that from Figure 4.13 and Table 4.3, a reduction in 5 ˚C superheat

tracking should have been attainable by without preview control strategy. However, a reduction in

5 ˚C superheat causes the working fluid mass flow to increase which leads to increase in saturation

pressure and therefore the saturation temperature. This increase in saturation temperature reduces

the operability window for the controls strategy to track the superheat. This further emphasizes

that the non-linearity of the ORC system as it changes the acceptable operability limits depending

on the operating point of the system.

98
 Superheat, ° C
Turbine Power, kW

(a)
Control Input, kg/s
Pressure, bar

(b)
Figure 4. 14. Performance comparison for the look-ahead NMPC (‘with preveiw’) and the baseline
NMPC (‘without preview’) while subjected to the sinusoidal heat source. Hp=60 s. SH Reference = 20˚C

For a HDD truck that follows an aggressive drive cycle either through traffic or road load

conditions, the corresponding exhaust conditions at the ORC evaporator can vary on 10s to 90s

time intervals. In such scenarios, a look-ahead NMPC would be beneficial in reducing the ORC

system control action, reducing the pump power consumption and improving system efficiency.
99
In addition, for control action that demands rapid variations in pump speed, the current drawn to

honor the controller request would be detrimental to pump longevity. Obviously, robustness in

control action is improved with the introduction of preview feature.

This case study shows the evaporator design’s effectiveness at mitigating some frequency

components of the fluctuating exhaust thermal load irrespective of the controller utilized. Also

with the knowledge future exhaust condition dynamics, a look-ahead NMPC controller can reduce

the required ORC system control action and also, to smaller degree, reduce evaporator outlet

temperature error.

4.8 Conclusions

In this chapter, a disturbance augmented, look-ahead NMPC strategy is proposed for superheat

tracking of a single evaporator WHR-ORC system. The NMPC is designed based on a 3-cell

control oriented moving boundary model. For state estimation, an EKF was designed, which was

augmented with a disturbance model for offset free MPC tracking.

Simulations were utilized to evaluate the performance of the NMPC over step changes in

exhaust conditions and over a drive cycle. With the inclusion of disturbance rejection functionality,

the steady state offset in evaporator outlet temperature decreases significantly. The disturbance

rejection capability was further validated by adding 10 - 40˚C modeling error during the

simulation, which still produced minimal steady state temperature tracking errors. The

performance comparison between a look-ahead NMPC which benefited from a preview of

upcoming exhaust conditions and a baseline NMPC without a preview of exhaust conditions

revealed comparable temperature tracking performance. The thermal inertia of the evaporator

played a vital role in attenuating the fluctuating frequency components of the exhaust conditions.

Hence, knowing the future exhaust conditions gained only minimal benefits. However, the look-

100
ahead NMPC generated a smoother control action. The controlled action can be helpful towards

maintaining pump durability and reducing pump power consumption.

101
 CHAPTER 5. NMPC EXPERIMENTAL VALIDATION

5.1 Introduction

In this Chapter, the controller designed in Chapter 4 is experimentally validated. First, the

model was calibrated online to counter the performance shifts due to system aging. Second, NMPC

weights were calibrated to give satisfactory control response for both a step rise and a step down

in engine conditions. Drive cycle verification showed that the thermal inertia of the evaporator

helped to dampen the high frequency components of the heat source and the tuned controller

limited working fluid superheat overshoots/undershoots within acceptable limits. Finally, the

generated sinusoidal cycle was utilized to understand the thermal response of the evaporator for

different time period of the engine conditions.

5.2 Background

While there are multiple studies conducted to determine the best optimal control strategies for

WHR-ORC system but they are all simulation based studies, ranging from a traditional PID to

more advanced model based approaches. There are only few studies that have validated their

control design experimentally [58-60].

In their study, Peralez et al. experimentally compared traditional PID control strategy with a

PID augmented with dynamic inverse model in feedforward control to track the superheat [58].

The ORC system utilized R245fa as working fluid which had a higher mass flow due to their

limited heat carrying capacity. This high mass flow results in higher operating pressure. Therefore,

for safe operation the exhaust gas was bypassed from the evaporator to limit the pressure rise inside

the ORC system. This approach is not an ideal solution has it reduces the ability of the ORC system

to recover heat energy.

102
 Hernandez et al. compared MPC and PID control strategy for ORC-WHR in their experimental

investigation for tracking the optimized working fluid temperature [59]. In this study, an optimal

evaporating temperature is computed for tracking. The computed temperature which is a function

of exhaust flow and temperature is used to track by MPC and PID. MPC outperformed PID strategy

in tracking the temperature and generated 17% more electrical power output. However, this study

utilized a lower order model which is identified over a specific range of exhaust conditions and for

it to be implemented real time more identification needs to be done which might change the

required tuning efforts and controls structure of the proposed MPC.

Seitz et al. investigated six different control approaches for possible real-time implementation

strategy [60]. Following strategies were evaluated experimentally, 1) Static feedforward with PI

feedback, 2) MB (Moving Boundary) feedforward with robustified PI feedback, 3) MB

feedforward with LQR GS feedback (proposed controller). 4) MB feedforward with PI feedback,

where P gain dependent on the sign of the control error (sign(𝑇 , , − 15 K)), and 5) IMC

feedback controller, based on the simplified MB model. The proposed controller MB feedforward

with LQR GS feedback (proposed controller) was found to be showing superior performance in

terms of thermal to electric efficiency. However, the identified model lacks accuracy and it’s valid

only around the identified points. The superheat tracking of their proposed controller is poor and

has maximum overshoots of 60 ˚C in some instances.

To address some of this concerns and gaps, this Chapter focuses on testing the designed NMPC

controller from Chapter 4. This section systematically address the issues and their remedies, as

faced during online implementation of the NMPC control algorithm. Three test scenarios are

evaluated, ramp inputs for tuning the NMPC algorithm, drive cycle to verify the tuned weights of

103
the NMPC and sinusoidal heat signals to understand the frequency rejection capability of the

thermal inertia of the evaporator.

5.3 Experimental Set-up

Figure 5. 1. Experimental set-up showing the Engine, after-treatment system and the ORC rig in a
transient capable heavy duty engine dynamometer at Clemson University’s Automotive research facility.

All of the following tests are conducted in a state of the art dynamometer facility at Clemson

University’s Automotive Engineering department. The NMPC algorithm is implemented via

DSPACE MicroAutoBox to run in parallel to the engine operation. ORC system being the passive

system doesn’t provide any feedback to the ECU. AVL’s PUMA is used to collect temperature

and pressure data of the ORC system and initiate a heavy transient drive cycle. DSPCA

ControlDesk software is used to collect NMPC related parameters.

All of the testing was done without the turbine in the ORC system and hence the turbine power

output numbers are not available. The primary focus of this study was to investigate the superheat

104
tracking ability of the controller under varying exhaust conditions. In the absence of the turbine,

the superheated working fluid flowed straight to condenser, which utilized the facilities water

maintained at 25˚C and represented infinite cooling capacity for this study. The system’s lack of

turbine operation did not alter the NMPC controller performance presented in this study. The

working fluid temperature at the condenser outlet was maintained at 35 ˚C for all experiments.

(a) (b)
Figure 5. 2. (a) AVL’s PUMA and INCA data collection interface (b) DSPACE MicroAutoBox for
implementing the NMPC algorithm
For the NMPC validation, the prediction horizon was kept at 60 s, and equal to control horizon.

For the NMPC validation, the prediction horizon and control horizon were set at 60s. The

measurement update rate and the control input refresh rate were each set to 0.2 s. Note that the

ORC evaporator has time constant of several minutes, making the 0.2s ample time for the

controller to take appropriate action in response to varying exhaust conditions. The turbine bypass

valve was fixed at 30% open and the turbine valve was closed for the entire test. This allows the

evaporation pressure to vary as per the working fluid flow rate. Ethanol was used as the working

fluid for this ORC system.

5.4 Online Model Calibration

The system under consideration consisted of prototype components and had accumulated a

significant run time hours over the course of this development project. Due to which at the

105
beginning of the NMPC testing a system level performance shift was observed because of system

aging. The exact cause of the aging is being investigated. The hypothesis for the potential causes

of this aging could be working fluid aging, internal working fluid leakage not captured by the

flowmeters or component damage/aging.

(a) (b)
1300 1100 data set1
data set2
data set1
data set2
1200 1000

1100 900
0 50 100 0 50 100

(c) (d) data set1

300 1100
data set2
data set1
data set2
290
1000

280
900
0 50 100 0 50 100

(e) (f)
50 24
data set1
data set1
data set2
data set2 22
40

20
30
18
0 50 100 0 50 100

(g) (h) data set1

250 0.025
data set2

0.02
240
0.015
data set1
data set2
230 0.01
0 50 100 0 50 100
time, s time, s
Figure 5. 3. System aging characterization for the same engine conditions with data set 1 being the green
ORC system and data set 2 representing the aged ORC system.

106
 Shown in Figure 5.3, is an example of performance comparison of the aged system with respect

to the green system. For identical engine conditions and working fluid superheat set-point, the

required working fluid flow was 40% lower. The heat transfer coefficient described in equations

(4.1 – 4.3) of chapter 4, was re-calibrated to match the performance of the aged system.

5.4.1 EKF state estimation error

A high heat transfer rate was observed for the green ORC system and thus a higher heat transfer

coefficients were identified during evaporator identification process for the control oriented model.

EKF also utilizes this control-oriented model for state estimation. Now, due to system aging, the

EKF generated erroneous states. The higher baseline heat transfer coefficients resulted in over-

predicting of the working fluid outlet temperature.

An example of behavior of the aged system relative to the estimates based on green system is

shown in Figure 5.4. The engine conditions are held constant at 1200 rpm 1000Nm. In the first

200 sec of Figure 5.4, it can be seen that the baseline heat transfer coefficients made the model

overestimate the evaporator outlet temperature due to aged system’s reduced mass flow at the same

working fluid superheat set-point and that resulted in over-prediction of working fluid evaporator

outlet temperature. This lead to the over-prediction of working fluid enthalpy and a larger vapor

phase region culminating in decreasing phase lengths L1 and L2 for pure liquid phase and mix

phase, respectively.

107
 Evaporator States L1 106
15 L2 1.8
Enthalpy

10 1.7

5 1.6
0 200 400 600 800 1000 1200
Heat Transfer Coefficients
1
htc-a
htc-b
0.5
htc-c

0
0 200 400 600 800 1000 1200
WF Evaporator Outlet temperature
250
Estimated
240 Measured

230

0 200 400 600 800 1000 1200

time, s
Figure 5. 4. Online tuning of heat transfer coefficients for MBM of TP evaporator resulted in stable state
estimation by the EKF

To reduce error in the EKF estimation, the heat transfer coefficients were reduced in Figure

5.4 until working fluid outlet temperature agreement was obtained. The recalibration of these

coefficients resulted in stable phase length estimations and working fluid enthalpy predictions. The

new identified heat transfer coefficients were used for the remainder of the testing. This exercise

of online tuning the heat transfer coefficients depending on the age of the system showed that it’s

vital to keep track of the system level performance shifts. For model based advanced controls

strategy modeling a parameter characterizing the system age would be beneficial.

The proposed control approach that was shown in Figure 4.8 can serve as a frame-work to

include a system aging adaptation variable, 𝜀(𝑡) that will manipulate the heat transfer coefficients

as shown in Figure 5.5.

108
Figure 5. 5. Proposed control law (Figure 4.8) incorporated with system aging adaptation variable, 𝜀(𝑡).

The system aging block will store the working flow mass flow required to maintain a fixed

amount of superheat at the evaporator outlet for a set of exhaust conditions. Then, in real time

implementation the working fluid flow from the NMPC block will be compared to the stored

working fluid mass flow values for that set of exhaust conditions. If the required working fluid

flow is less than the stored values than the 𝜀(𝑡) value will manipulate the heat transfer coefficient

for the MBM model in EKF block. In future experimental testing, a nominal operating would be

repeated at equal intervals to understand the behavior of the heat transfer coefficients as the system

ages. This will dictate the necessary change in 𝜀(𝑡) variable needed to accurately portray the age

of the system.

109
5.5 Ramp-up and Ramp-down Events

A conventional method of controller tuning is examination of its response to step changes. In

this section, two input ramps (with rise times of 30sec and 5sec) of engine speed and torque are

used to analyze the evaporator response and tune the NMPC weights towards tracking a constant

superheat of 60˚C. Note that a superheat of 60˚C was recommended by the project sponsor to

safeguard these early prototype components and to avoid constant switching between saturation

and vapor phase as the controller was tuned. In simulation, it was concluded that the best way to

optimize turbine power was to lower the superheat temperature. In future testing, this reference

will be optimized further with the turbine installed.

1400 RPM
Torque

1200

1000
0 50 100 150 200 250 300 350 400 450

70
Superheat
65 Reference
60

55

50
0 50 100 150 200 250 300 350 400 450

0.017 28

0.016
26
0.015

0.014 24
0 50 100 150 200 250 300 350 400 450
time, s

Figure 5. 6. Tuned NMPC performance for a 30s ramp in engine conditions.

The weight 𝑊 in the cost function shown in equation 4.13 in Chapter 4, was tuned to 20 to

give an acceptable performance in superheat tracking of +/- 5˚C as shown in Figure 5.6. In the

tuning process, it was found that to ensure superheat tracking within +/- 5˚C a larger weight on 𝑑𝑢

110
term was necessary. The slower evaporator response due to its thermal inertia required a slower

change in mass flow to maintain tracking performance for both the rise and fall in engine

conditions.

It should also be noted that the first 60 sec of Figure 5.6 shows zero steady state error which

confirms that the disturbance model introduced in the NMPC formulation is performing as

anticipated. Since the both the turbine bypass valve and the turbine valve were fixed, the change

in working fluid flow changes the evaporation pressure as seen from the bottom plot of Figure 5.6.

This change in evaporation pressure changes the evaporation temperature which in turn affects the

superheat tracking condition.

1600
RPM
1400 Torque

1200

1000
0 100 200 300 400 500 600 700 800

70
Superheat
65
Reference
60

55

50
0 100 200 300 400 500 600 700 800

0.018 24

23
0.016
22

0.014 21
0 100 200 300 400 500 600 700 800
time, s
Figure 5. 7. Tuned NMPC performance for a ramp input of rise time 5 sec

Next, the controller performance in response to a more aggressive 5s ramp in engine conditions

is shown in Figure 5.7. With the same cost function weights, the tracking performance is outside

the +/- 5˚C target range, however the overshoot is still below 10 ˚C and is easily recovered as seen

111
after 400 sec in Figure 5.7. Again the controller performance remarkable in reducing the steady

state error.

5.6 Drive cycle

A true test of the controller would be to verify its performance on heavy transient conditions

representative of real-world driving. For this an engine dyno test cycle was developed as shown in

Figure 5.8 (a). In real world operations, majority of the life cycle of a heavy duty diesel engines is

spent on highways and accordingly a near constant engine speed and variable load cycle was

developed. The bottom plot of Figure 5.8 (a), is the corresponding exhaust mass flow and

temperature conditions recorded at the evaporator inlet location.

Speed, rpm
1600 Torque, Nm

1400

1200

1000
0 200 400 600 800 1000 1200

1600 290

1400
280
1200
270
1000

800 260
0 200 400 600 800 1000 1200
time, s
(a)

112
 70

Superheat, ° C
65

60
Reference
55 Measured

50
0 200 400 600 800 1000 1200

0.018 30
Working Fluid

Working Fluid
pressure, bar
flow, kg/s

0.016 25

0.014 20

0.012 15
0 200 400 600 800 1000 1200
time, s
(b)
Figure 5. 8. (a) Engine conditions for the constant speed drive cycle (b) Tuned NMPC superheat tracking
performance and generated working fluid flow rate

As seen from the exhaust condition plot, the mass flow dynamics are fast, changing rapidly

with the drive cycle. However, temperature dynamics are relatively slow and dampened by the

after-treatment system. The NMPC superheat tracking performance over this highly-transient

engine cycle is provided in Figure 5.8(b). The maximum overshoot in working fluid superheat is

7.9˚C with a mean error of 2.9˚C. The working fluid flow predominantly mimics the exhaust

temperature trend without suffering variation from the highly transient exhaust flow changes. This

can be attributed to the high frequency of the exhaust gas flow and the large thermal inertia of the

evaporator.

113
 87.25

NMPC Computation time, ms 87.2

87.15

87.1

87.05

87

86.95

86.9

86.85
0 200 400 600 800 1000 1200
time, s
Figure 5. 9. NMPC Computation time for the constant speed drive cycle

An inhibiting factor for real time MPC implementation is the incurred computation cost. The

designed controller and model computation time averaged 87ms over the entire drive cycle as

shown in Figure 5.9. With an update rate of 0.2s, the controller has ample time to complete the

calculations and provide optimal control inputs to the system.

In Chapter 4, it was found that the main advantage of knowing the future exhaust conditions is

reduced control effort, however in the experimental validation no significant variation in working

fluid mass flow was observed as seen from the bottom plot of the Figure 5.8 (b). Therefore, in the

drive cycle case, the preview capability was not validated experimentally. Instead, a series of

sinusoidal exhaust conditions were experimentally tested to evaluate the controller performance.

5.7 Sinusoidal Inputs

Since no oscillatory behavior was seen during the drive cycle tests. Sinusoidal tests were

conducted at different frequency to understand the variation of the heat transfer process and best

114
approach to tune the given ORC system under different frequency of heat signals. This set of tests

were prepared to be in correspondence to the open loop simulation done in Chapter 3. For this

analysis, controller performance is experimentally evaluated over a series of sinusoidal heat signals

with time periods of 20 secs, 60 secs, and 120 sec.

5.7.1 Time period of 20 s

From the perspective of real world applicability a sinusoidal waveform can be treated as

alternate acceleration and deceleration/coast-down. To generate the sinusoidal exhaust conditions

at the evaporator inlet and recognizing the real world applicability, amplitude of the engine speed

was kept at 100 rpm whereas amplitude of the torque was kept at 400 rpm. Figure 5.10 (a) shows

the engine conditions for 20 sec time period. As stated before the dynamics of exhaust mass flow

is in cohesion with the engine speed and torque. Whereas the exhaust temperature is seen steadily

increasing with a dampened frequency component.

Engine Conditions
mass flow, kg/hr

temperature, oC
Exhaust gas

Exhaust gas

(a)

115
 70
Reference
65 Measured

60

55

50
0 50 100 150 200 250 300 350

0.0168 21

20.8
0.0166
20.6
0.0164
20.4

0.0162 20.2
0 50 100 150 200 250 300 350
time, s
(b)
Figure 5. 10. (a) Engine conditions for the sinusoidal cycle with time period of 20s (b) Tuned NMPC
superheat tracking performance and generated working fluid flow rate, 𝑊 =10

With very high frequency of heat source, there is no time for heat transfer and in this case the

change in control effort needed to maintain superheat tracking is very low. This was expected as

described in the open loop simulation in Chapter 3.3 and in Chapter 4.7.3 B. As a result, the

required working fluid flow steadily increases over time with increase in exhaust temperature. The

maximum absolute superheat tracking error was found to be 2.1 ˚C and mean error was 0.87 ˚C.

5.7.2 Time period of 60 s

Engine conditions for a 60 s heat signal is shown in Figure 5.11 (a) and was run for a 600 sec.

With the same weights, 𝑊 =20 as for the 20 s time period, slight oscillations can be seen in the

superheat tracking performance.

116
 RPM
Torque
1400

1200

1000

800
0 100 200 300 400 500 600
Time, s
1600 276

1400 275

1200 274

1000 273
0 100 200 300 400 500 600
time, s
(a)
70
Reference
Superheat, ° C

65 Measured

60

55

50
0 100 200 300 400 500 600
Time, s
Evaporation Pressure, bar

0.016 20.5
mass flow, kg/s

Working Fluid
Working fluid

0.0155

0.015 20

0.0145

0.014 19.5
0 100 200 300 400 500 600
time, s
(b)
Figure 5. 11. (a) Engine conditions for the sinusoidal cycle with time period of 60s (b) Tuned NMPC
superheat tracking performance and generated working fluid flow rate, 𝑊 =10

117
 However, the overshoot and undershoot is still maintain within +/- 5˚C which is quite

remarkable of the controller. The maximum absolute superheat tracking error was found to be 4.3

˚C and mean error was 1.66 ˚C.

5.7.3 Time period of 120 s

Similar to 20s and 60s time period, 120s time period heat signal was also tested with same

amplitude of 100 rpm and 400 Nm. As expected with increase in time period, oscillation increased.

Note that same weights of 𝑊 =10 were used initially resulting in a maximum absolute superheat

tracking error of 5.23 ˚C and mean error of 13.06 ˚C. It was only with a larger weight that the

oscillations were brought back within the +/- 5˚C range as seen from the Figure 5.12 (b).
Exhaust mass
flow, kg/hr
Temperature, ° C
Exhaust

(a)

118
 (b)
Figure 5. 12. (a) Engine conditions for the sinusoidal cycle with time period of 120s (b) Tuned NMPC
superheat tracking performance comparison with weights, 𝑊 =10 and 𝑊 =60 with generated working
fluid flow rate

At slower frequency of heat source, there is plenty of time available for heat transfer and the

corresponding changes can be seen at the working fluid evaporator outlet temperature. Comparing

the three time periods, it can be concluded that 120s is the worst case. In such situations a slower

working fluid flow produces better tracking performance as shown in the Figure 5.12 (b). In

Chapter 1, it was experimentally shown that the temperature dynamics is slower than the working

fluid mass flow change. This is because the heat transfer process is slow as the exhaust gas first

heats up the tube walls before transferring the energy to the working fluid. It is because of this

process that the rate of change of working fluid had be slowed down to match the temperature

119
dynamics time constant. A direct realization of this approach can be seen in Figure 5.12 (b), where

the slower working fluid flow generated less oscillations.

Table 5.1 summarizes the results of the sinusoidal study. Even in this study, the mean

computation time was found to be 0.0871s. This boost confidence in real time NMPC

implementation.

Time Weight, Mean Error, Max Error, NMPC Computation time, s

period, s 𝑊 ˚C ˚C
20 20 0.95 1.88 0.0871
60 20 1.75 3.74 0.0871
120 10 5.23 13.06 0.0871
40 1.74 4.44 0.0871
Table 5. 1. NMPC superheat tracking error for different time period sinusoidal heat signal.

5.8 Conclusions

In this Chapter, an experimental investigation was conducted to evaluate the NMPC controller

performance under different engine conditions. During the initial experimental implementation,

there was a performance shift observed due to system aging. This caused the EKF to estimate

faulty state values and therefore the control oriented model had to be tuned online to match the

new heat transfer efficiency of the aged ORC system.

Once the state estimation outputs were stabilized, ramp inputs in heat source with rise times of

30s and 5s were introduced as disturbance. NMPC was then tuned to this ramp inputs to maintain

the superheat tracking within +/-5 ˚C range. With the tuned NMPC weights, a heavy transient drive

cycle with near constant engine speed and fast varying load was run. The mean superheat tracking

error was found to be 2.9 ˚C for the drive cycle. Although the drive cycle was transient in nature,

and the superheat tracking was exceptional, but the working fluid flow rate didn’t fluctuated

120
according to the heat. In fact the working fluid flow steadily increased with the rise of exhaust gas

temperature at the evaporator inlet.

To further investigate the dampened response of the evaporator a sinusoidal heat signals with

varying time periods were generated. As expected high frequency (time period =20s) heat signals

caused no changed to the control input and superheat tracking error. As the time period of the heat

signal was increased, subsequently more time was available for heat transfer and hence more

oscillatory behavior was observed, both for the working fluid temperature and for the working

fluid flow. It was also showed that as the heat signal time period is increased, to reduce the

corresponding oscillations in superheat, a higher penalty to the rate of change of working fluid

flow is needed to maintain the tracking performance within +/- 5 ˚C.

121
 Chapter 6. Conclusion, Contributions and Future Work

This Chapter provides concise highlights of the outcome of preceding Chapters. This section

also explains the limitations and future work of this study. Finally, the original contributions of

the presented research work and its impact are accentuated.

6.1 Conclusions and Contributions

This thesis contributes towards development of an NMPC controller for WHR-ORC system in

an HHDE application with enhanced capabilities under highly transient conditions. The main

contribution is the development of the NMPC control strategy with preview capability, followed

by characterization of its benefits. To this end, 1) an enhanced evaporator model is developed for

accurate estimation of changing phase lengths inside the evaporator and 2) an augmented EKF

with disturbance rejection is proposed.

Enhanced evaporator modeling and proposed improvement of the mixed-phase heat

transfer coefficients. First, discretization of the FVM showed that increased discretization

increases the model’s phase length prediction capability. However, the working fluid temperature

at the evaporator outlet doesn’t necessarily benefit by increasing the discretization. It was found

that while 10 cells are enough to accurately model the evaporator outlet temperatures, but a 500

cell discretization provided the best phase length prediction capability. Second, additional

multipliers are introduced to the existing empirical equations to improve the phase length

estimation accuracy. This additional multipliers were identified using Particle Swarm

Optimization (PSO) technique against thermal imagery data. The newly identified multipliers for

the empirical equations were validated against transient experimental data and showed 43%

improvement in vapor phase length prediction in comparison to the baseline FVM.

122
 The identification set was limited to the available thermal data, and is valid within the

identification range. However, this study serves as a proof-of-concept methodology exhibiting that

the phase length estimation can be improved by using thermal imaging data. The methodology

developed herein can provide additional insight to the evaporator design process and serve as basis

for improving moving boundary models for advanced ORC system control development.

NMPC Controller design. For a WHR-ORC system in an automotive application the highly

dynamic nature of the heat sources provide a significant control challenge. Moreover, the non-

linear behavior of the evaporator dynamics changes with the engine operating conditions and

possess additional control challenges. In order to address these challenges a NMPC controller was

designed. First, an augmented disturbance rejection estimator in an EKF framework was

designed to reject modeling errors and measurement inaccuracies, thus improving steady state

performance. Simulation study demonstrated the elimination of steady state error after the

implementation of disturbance rejection in EKF. Further, it was shown that the augmented

estimator also helps reject measurement inaccuracies.

Second, during the initial formulation of objective function for NMPC controller it was found

that the best way to optimize for turbine power without violating any constraints was to track the

superheat. Then, the superheat tracking performance was compared between NMPC with preview

capability and NMPC without this feature for a drive cycle. It was found that knowing the future

helps improve the control effort and therefore improved robustness. However, it doesn’t provide

any significant improvement to the superheat tracking performance for the drive cycle.

Additionally, this comparison study was extended to sinusoidal heat signals with varying time

periods of 4s to 240s. This case study showed that the minimum superheat that could be tracked

without using the preview feature was 20 ˚C. NMPC with preview capability helped the controller

123
in maintaining a tight tolerance. It was found that the minimum superheat that could be tracked

with preview feature was 15 ˚C. Thus providing higher turbine power in comparison to the NMPC

without the preview feature.

Third, experimental validation of the designed NMPC controller without preview feature was

conducted. During online implementation it was concluded that the control oriented model of the

evaporator was susceptible to system aging and had to be re-calibrated online to compute stable

state estimates for MPC.

Then the weights of the NMPC were tuned to the ramp inputs of heat source such that the

superheat tracking was within +/ 5 ˚C. The performance of the tuned NMPC was then evaluated

for a drive cycle and the mean superheat tracking error was found to be 2.9 ˚C. The required

working fluid for this heavy transient drive cycle didn’t varied dynamically in cohesion with the

heat source and therefore with preview capability of the NMPC was not tested online since the

only advantage of with preview capability was reduction in control effort. Therefore, to further

investigate the evaporator dynamics sinusoidal heat source was generated and validated

experimentally. In this study, highly dynamic heat source with time period of 20s and 60s didn’t

affect the performance of the controller as the thermal inertia of the evaporator helped buffer out

the high frequency components of the heat source. However, when the time period of the exhaust

conditions was increased to 120s, the evaporator outlet temperature dynamically varied with the

heat source and a maximum overshoot of 13.5 ˚C was observed. In this particular instance, it was

found that a heavier penalty to the rate of change of control input aided to bring the superheat

overshoot within +/- 5 ˚C limit. Thus, it was recommended that a high penalty on working fluid

change rate should be utilized to maintain superheat tracking performance within acceptable limits.

124
6.2 Future Work

The conclusion and developments presented in this thesis leads to further exploration in areas

of both modeling and control improvement.

Extending the enhanced evaporator model identification and validation. A high fidelity

FVM evaporator model was expanded to enhance the working fluid phase length predictions

during transient operation in this thesis. However, due to limited availability of the thermal data

the identification of the additional multipliers were valid within a given range of operation. In

future studies a more systematic design of experiements will be conducted to cover the entire range

of the ORC-WHR operation. Both identification and validation data sets will be expanded to

represent the complete operating range of the engine and the author hopes to developed a

correlation of the proposed multipliers that would be a function of the exhaust heat power.

Control strategy improvement. In this study, a disturbance augmented, look-ahead NMPC

strategy is proposed for superheat control of a single evaporator WHR-ORC system. In future

works, this study will be extended to ORC system with two evaporators and further investigate the

effectiveness of both the disturbance rejection and the preview capability. Due to time constraint,

the control strategy was only implemented on a TP evaporator. It would be interesting to exploit

the benefits of the NMPC’s preview capability on EGR evaporator which has less thermal inertia.

It was shown in Chapter 3 that the dynamics of the EGR evaporator is faster compare to TP

evaporator. Therefore, the reduce control effort with preview capability would be beneficial to

EGR evaporator since it will fail to buffer the high frequency components of the exhaust.

Moreover, the current experimental validation showed that system aging affects the

performance of the model based approach. For future testing, it is planned to characterize this

125
behavior at regular run time hours and generate deeper understanding required to accurately model

this type of aging.

126
 APPENDIX

Figure a. Heat Transfer and Flow Regimes in a vertical tube [33].

127
 REFERENCES

1. Schoettle, Brandon, Michael Sivak, and Michael Tunnell. "A survey of fuel economy and fuel

usage by heavyduty truck fleets." No. SWT-2016-12, http://atri-online.

org/wpcontent/uploads/2016/10/2016. ATRI-UMTRI. FuelEconomyReport. Final_.

pdf (2016).

2. UPDATE, POLICY. "UNITED STATES EFFICIENCY AND GREENHOUSE GAS

EMISSION REGULATIONS FOR MODEL YEAR 2018-2027 HEAVY-DUTY

VEHICLES, ENGINES, AND TRAILERS." Policy (2015).

3. D. Oeberlein, "Cummins SuperTruck Program Technology and System Level Demonstation

of Highly Efficient and Clean, Diesel Powered Class 8 Trucks," presenttion at US Department

of Energy Merit Review, 2013.

4. M. Allain, D. Atherton, I. Gruden, S. Singh, and K. Sisken, "Daimler’s Super Truck Program;

50% Brake Thermal Efficiency," in presentation at US DOE Directions in Engine-Efficiency

and Emissions Research (DEER) Conference, 2012.

5. V. Grelet, T. Reiche, L. Guillaume, and V. Lemort, "Optimal waste heat recovery Rankine

based for heavy duty applications," 2014

6. Chen, Haoxiang, Weilin Zhuge, Yangjun Zhang, Tao Chen, and Lei Zhang. "Performance

Simulation of an Integrated Organic Rankine Cycle and Air Inter-Cooling System for Heavy-

Duty Diesel Truck Engines." In ASME Turbo Expo 2017: Turbomachinery Technical

Conference and Exposition, pp. V003T28A004-V003T28A004. American Society of

Mechanical Engineers, 2017.

128
7. Teng, Ho, Gerhard Regner, and Chris Cowland. Achieving high engine efficiency for heavy-

duty diesel engines by waste heat recovery using supercritical organic-fluid Rankine cycle.

No. 2006-01-3522. SAE Technical Paper, 2006.

8. Legros Arnaud, Guillaume Ludovic, Diny Mouad, Zaïdi Hamid and Lemort Vincent.

Comparison and Impact of Waste Heat Recovery Technologies on Passenger Car Fuel

Consumption in a Normalized Driving Cycle. Energies ISSN 1996-1073

9. F. Yang, X. Dong, H. Zhang, Z. Wang, K. Yang, J. Zhang, et al., "Performance analysis of

waste heat recovery with a dual loop organic Rankine cycle (ORC) system for diesel engine

under various operating conditions," Energy Conversion and Management, vol. 80, pp. 243-

255, 2014.

10. G. Shu, G. Yu, H. Tian, H. Wei, X. Liang, and Z. Huang, "Multi-approach evaluations of a

cascade-Organic Rankine Cycle (C-ORC) system driven by diesel engine waste heat: Part A–

Thermodynamic evaluations," Energy Conversion and Management, vol. 108, pp. 579-595,

2016.

11. I. Arsie, A. Cricchio, C. Pianese, M. D. Cesare, and W. Nesci, "A Comprehensive Powertrain

Model to Evaluate the Benefits of Electric Turbo Compound (ETC) in Reducing CO2

Emissions from Small Diesel Passenger Cars," SAE Technical Paper 2014-01-1650, 2014,

DOI: 10.4271/ 2014-01-1650.

12. C. Zhang, G.-Q. Shu, H. Tian, H. Wei, G. Yu, and Y. Liang, "Theoretical Analysis of a

Combined Thermoelectric Generator (TEG) and Dual-loop Organic Rankine Cycle (DORC)

System Using for Engines' Exhaust Waste Heat Recovery," SAE Technical Paper 2014-01-

0670, 2014, DOI: 10.4271/ 2014-01-0670.

129
13. Davidson, Thomas A. "Design and analysis of a 1 kw Rankine power cycle, employing a

multi-vane expander, for use with a low temperature solar collector." PhD diss.,

Massachusetts Institute of Technology, 1977.

14. Monahan, J. and McKenna, R., 1976, September. Development of a 1-kW, organic Rankine

cycle power plant for remote applications. In Proceedings of the Intersociety Energy

Conversion Engineering Conference, New York, NY, USA(Vol. 1217).

15. Probert, S. D., Mohey Hussein, P. W. O'Callaghan, and Eli Bala. "Design optimisation of a

solar-energy harnessing system for stimulating an irrigation pump." Applied Energy 15, no. 4

(1983): 299-321.

16. Arias, Diego A., Timothy A. Shedd, and Ryan K. Jester. Theoretical analysis of waste heat

recovery from an internal combustion engine in a hybrid vehicle. No. 2006-01-1605. SAE

Technical Paper, 2006.

17. Hernandez, Andres, Adriano Desideri, Clara Ionescu, Sylvain Quoilin, Vincent Lemort, and

Robin De Keyser. "Experimental study of Predictive Control strategies for optimal operation

of Organic Rankine Cycle systems." In 2015 European Control Conference (ECC), pp. 2254-

2259. IEEE, 2015.

18. Grelet, Vincent, Pascal Dufour, Madiha Nadri, Thomas Reiche, and Vincent Lemort.

"Modeling and control of Rankine based waste heat recovery systems for heavy duty

trucks." IFAC-PapersOnLine 48, no. 8 (2015): 568-573.

19. Feru, Emanuel, Frank Willems, Bram de Jager, and Maarten Steinbuch. "Model predictive

control of a waste heat recovery system for automotive diesel engines." In System Theory,

Control and Computing (ICSTCC), 2014 18th International Conference, pp. 658-663. IEEE,

2014.

130
20. Esposito, Marco Crialesi, Nicola Pompini, Agostino Gambarotta, Vetrivel Chandrasekaran,

Junqiang Zhou, and Marcello Canova. "Nonlinear model predictive control of an organic

rankine cycle for exhaust waste heat recovery in automotive engines." IFAC-

PapersOnLine 48, no. 15 (2015): 411-418.

21. Hou, Guolian, Rui Sun, Guoqiang Hu, and Jianhua Zhang. "Supervisory predictive control of

evaporator in Organic Rankine Cycle (ORC) system for waste heat recovery." In Advanced

Mechatronic Systems (ICAMechS), 2011 International Conference on, pp. 306-311. IEEE,

2011.

22. Hernandez Naranjo, Jairo Andres, Adriano Desideri, Clara-Mihaela Ionescu, Sylvain Quoilin,

Vincent Lemort, and Robain De Keyser. "Increasing the efficiency of organic rankine cycle

technology by means of multivariable predictive control." In 19th World Congress of the

International-Federation-of-Automatic-Control (IFAC), vol. 47, no. 3, pp. 2195-2200. 2014.

23. Liu, Xiaobing, Adamu Yebi, Paul Anschel, John Shutty, Bin Xu, Mark Hoffman, and Simona

Onori. "Model predictive control of an organic Rankine cycle system." Energy Procedia 129

(2017): 184-191.

24. Feru, Emanuel, Bram de Jager, Frank Willems, and Maarten Steinbuch. "Two-phase plate-fin

heat exchanger modeling for waste heat recovery systems in diesel engines." Applied

energy 133 (2014): 183-196.

25. Quoilin, Sylvain, Richard Aumann, Andreas Grill, Andreas Schuster, Vincent Lemort, and

Hartmut Spliethoff. "Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles." Applied energy 88, no. 6 (2011): 2183-2190.

131
26. J. M. Jensen, "Dynamic Modeling of Thermo-Fluid Systems with Focus on Evaporators for

Regrigeration," PhD, Department of Mechanical Engineering, Technical University of

Denmark, 2003.

27. Benato, A., Martin Ryhl Kærn, Leonardo Pierobon, A. Stoppato, and Fredrik Haglind.

"Analysis of hot spots in boilers of organic Rankine cycle units during transient

operation." Applied Energy 151 (2015): 119-131.

28. Yebi, Adamu, Bin Xu, Xiaobing Liu, John Shutty, Paul Anschel, Simona Onori, Zoran Filipi,

and Mark Hoffman. "Nonlinear model predictive control strategies for a parallel evaporator

diesel engine waste heat recovery system." In ASME 2016 Dynamic Systems and Control

Conference, pp. V002T19A003-V002T19A003. American Society of Mechanical Engineers,

2016.

29. Peralez, Johan, Paolino Tona, Antonio Sciarretta, Pascal Dufour, and Madiha Nadri.

"Towards model-based control of a steam rankine process for engine waste heat recovery."

In Vehicle Power and Propulsion Conference (VPPC), 2012 IEEE, pp. 289-294. IEEE, 2012.

30. Zhang, Jianhua, Song Gao, Yannan Chen, and Guolian Hou. "Multivariable robust control for

organic rankine cycle based waste heat recovery systems." In Industrial Electronics and

Applications (ICIEA), 2013 8th IEEE Conference on, pp. 85-89. IEEE, 2013.

31. Zhang, H. G., E. H. Wang, and B. Y. Fan. "Heat transfer analysis of a finned-tube evaporator

for engine exhaust heat recovery." Energy Conversion and Management 65 (2013): 438-447.

32. Xu, Bin, Dhruvang Rathod, Shreyas Kulkarni, Adamu Yebi, Zoran Filipi, Simona Onori, and

Mark Hoffman. "Transient dynamic modeling and validation of an organic Rankine cycle

waste heat recovery system for heavy duty diesel engine applications." Applied Energy 205

(2017): 260-279

132
33. VDI Heat Atlas, DOI 10.1007/978-3-540-77877-6_23, © Springer-Verlag Berlin Heidelberg

2010

34. Xu, Bin, Xiaobing Liu, John Shutty, Paul Anschel, Simona Onori, Zoran Filipi, and Mark

Hoffman. Physics-based modeling and transient validation of an organic Rankine cycle waste

heat recovery system for a heavy-duty diesel engine. No. 2016-01-0199. SAE Technical

Paper, 2016.

35. Collier, John G., and John R. Thome. Convective boiling and condensation. Clarendon Press,

1994.

36. Jiménez-Arreola, Manuel, Roberto Pili, Christoph Wieland, and Alessandro Romagnoli.

"Analysis and comparison of dynamic behavior of heat exchangers for direct evaporation in

ORC waste heat recovery applications from fluctuating sources." Applied Energy 216 (2018):

724-740.

37. Bemporad, Alberto. "Model predictive control design: New trends and tools." In Decision and

Control, 2006 45th IEEE Conference on, pp. 6678-6683. IEEE, 2006.

38. Ganji, Behnam, and Abbas Z. Kouzani. "A study on look-ahead control and energy

management strategies in hybrid electric vehicles." In Control and Automation (ICCA), 2010

8th IEEE International Conference on, pp. 388-392. IEEE, 2010.

39. Gong, Qiuming, Yaoyu Li, and Zhong-Ren Peng. "Trip-based optimal power management of

plug-in hybrid electric vehicles." IEEE Transactions on vehicular technology 57, no. 6

(2008): 3393-3401.

40. Rajagopalan, Arun, and Gregory Washington. Intelligent control of hybrid electric vehicles

using GPS information. No. 2002-01-1936. SAE Technical Paper, 2002.

133
41. Boehme, Thomas Juergen, Florian Held, Christoph Rollinger, Heiko Rabba, Matthias

Schultalbers, and Bernhard Lampe. "Application of an optimal control problem to a trip-based

energy management for electric vehicles." SAE International Journal of Alternative

Powertrains 2, no. 1 (2013): 115-126.

42. Smith, R., M. Morison, D. Capelle, C. Christie, and D. Blair. "GPS-based optimization of

plug-in hybrid electric vehicles’ power demands in a cold weather city." Transportation

Research Part D: Transport and Environment 16, no. 8 (2011): 614-618.

43. Huang, Wei, and David M. Bevly. "Set terrain-based optimal speed limits for heavy trucks

energy saving." International Journal of Powertrains 1, no. 4 (2012): 335-350.

44. Turri, Valerio, Bart Besselink, Jonas Mårtensson, and Karl H. Johansson. "Fuel-efficient

heavy-duty vehicle platooning by look-ahead control." In Decision and Control (CDC), 2014

IEEE 53rd Annual Conference on, pp. 654-660. IEEE, 2014.

45. Gáspár, Péter, and Balázs Németh. "Design of look-ahead control for road vehicles using

traffc information." In Control and Automation (MED), 2014 22nd Mediterranean

Conference of, pp. 201-206. IEEE, 2014.

46. Jazwinski, A. H. "Stochastic Process and Filtering Theory, Academic Press." A subsidiary of

Harcourt Brace Jovanovich Publishers (1970).

47. Maeder, Urban, and Manfred Morari. "Offset-free reference tracking for predictive

controllers." In Decision and Control, 2007 46th IEEE Conference on, pp. 5252-5257. IEEE,

2007.

48. Borrelli, Francesco, and Manfred Morari. "Offset free model predictive control." In Decision

and Control, 2007 46th IEEE Conference on, pp. 1245-1250. IEEE, 2007.

134
49. J. M. Jensen. "Dynamic Modeling of Thermo-Fluid Systems with Focus on Evaporators for

Refrigeration," Energy Engineering, Department of Mechanical Engineering, Technical

University of Denmark, 2003.

50. Houska, Boris, Hans Joachim Ferreau, and Moritz Diehl. "ACADO toolkit—An open‐source

framework for automatic control and dynamic optimization." Optimal Control Applications

and Methods 32, no. 3 (2011): 298-312.

51. Quirynen, Rien, Sébastien Gros, and Moritz Diehl. "Efficient NMPC for nonlinear models

with linear subsystems." In 52nd IEEE Conference on Decision and Control, pp. 5101-5106.

IEEE, 2013.

52. Gros, Sébastien, Mario Zanon, Rien Quirynen, Alberto Bemporad, and Moritz Diehl. "From

linear to nonlinear MPC: bridging the gap via the real-time iteration." International Journal

of Control (2016): 1-19.

53. Maeder, Urban, and Manfred Morari. "Offset-free reference tracking for predictive

controllers." In Decision and Control, 2007 46th IEEE Conference on, pp. 5252-5257. IEEE,

2007.

54. Borrelli, Francesco, and Manfred Morari. "Offset free model predictive control." In Decision

and Control, 2007 46th IEEE Conference on, pp. 1245-1250. IEEE, 2007.

55. Bemporad, Alberto. "Model predictive control design: New trends and tools." In Decision and

Control, 2006 45th IEEE Conference on, pp. 6678-6683. IEEE, 2006.

56. Zedan, H. "An AN-stable Rosenbrock-type method for solving stiff differential

equations." Computers & Mathematics with Applications 13, no. 7 (1987): 611-615.

135
57. Peralez, Johan, Madiha Nadri, Pascal Dufour, Paolino Tona, and Antonio Sciarretta. "Organic

Rankine cycle for vehicles: Control design and experimental results." IEEE Transactions on

Control Systems Technology 25, no. 3 (2017): 952-965.

58. Hernandez, Andres, Adriano Desideri, Sergei Gusev, Clara M. Ionescu, Martijn Van Den

Broek, Sylvain Quoilin, Vincent Lemort, and Robin De Keyser. "Design and experimental

validation of an adaptive control law to maximize the power generation of a small-scale waste

heat recovery system." Applied energy 203 (2017): 549-559.

59. Seitz, D., O. Gehring, C. Bunz, M. Brunschier, and O. Sawodny. "Model-based control of

exhaust heat recovery in a heavy-duty vehicle." Control Engineering Practice 70 (2018): 15-

28.

PUBLICATIONS

60. Rathod, Dhruvang, Ujjwal Belwariar, Bin Xu, and Mark Hoffman. "An enhanced evaporator

model for working fluid phase length prediction, validated with experimental thermal imaging

data." International Journal of Heat and Mass Transfer 132 (2019): 194-208.

61. Rathod, Dhruvang, Bin Xu, Adamu Yebi, Ardalan Vahidi, Zoran Filipi, and Mark Hoffman. A

Look-ahead Model Predictive Control Strategy for an Organic Rankine Cycle-Waste Heat

Recovery System in a Heavy Duty Diesel Engine Application. No. 2019-01-1130. SAE

Technical Paper, 2019.

62. Xu, Bin, Dhruvang Rathod, Adamu Yebi, Zoran Filipi, Simona Onori, and Mark Hoffman.

"A comprehensive review of organic rankine cycle waste heat recovery systems in heavy-

duty diesel engine applications." Renewable and Sustainable Energy Reviews 107 (2019):

145-170.

136
63. Xu, Bin, Dhruvang Rathod, Shreyas Kulkarni, Adamu Yebi, Zoran Filipi, Simona Onori, and

Mark Hoffman. "Transient dynamic modeling and validation of an organic Rankine cycle

waste heat recovery system for heavy duty diesel engine applications." Applied Energy 205

(2017): 260-279.

64. Xu, Bin, Dhruvang Rathod, Adamu Yebi, Zoran Filipi, Simona Onori, and Mark Hoffman.

"A comprehensive review of organic rankine cycle waste heat recovery systems in heavy-

duty diesel engine applications." Renewable and Sustainable Energy Reviews 107 (2019):

145-170.

65. Rathod, Dhruvang, Simona Onori, Zoran Filipi, and Mark Hoffman. "Experimental

Investigation of Soot Accumulation and Regeneration in a Catalyzed Gasoline Particulate

Filter Utilizing Particulate Quantification and Gas Speciation Measurements." In ASME 2018

Internal Combustion Engine Division Fall Technical Conference, pp. V002T04A002-

V002T04A002. American Society of Mechanical Engineers, 2018.

66. Rathod, Dhruvang, Mark A. Hoffman, and Simona Onori. "Determining three-way catalyst

age using differential lambda signal response." SAE International Journal of Engines 10, no.

3 (2017): 1305-1312.

137