Design Tools and Mechanisms for Progressive

Cavity Pumps
by
Kevin P. Simon
B.S., Franklin W. Olin College of Engineering (2012)
S.M., Massachusetts Institute of Technology (2015)

Submitted to the Department of Mechanical Engineering

in partial fulfillment of the requirements for the degree of

Ph.D. in Mechanical Engineering

at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY

February 2019
Massachusetts Institute of Technology 2019. All rights reserved.

Author ............
Signature redacted
Department of Mechanical Engineering
November 20, 2018

Certified by.........
Signature redacted_____,
Alexander H. Slocum
Walter M. May & A. Hazel May Professor of Mechanical Engineering
Thesis Supervisor

Signature redacted
Accepted by. ............... .
.

Nicolas Hadjiconstantinou
OFTEHNOLY
Chairman, Department Committee on Graduate Theses
FEB' 25 2019
LIBRARIES
ARCHIVES
2
Design Tools and Mechanisms for Progressive Cavity Pumps

by
Kevin P. Simon

Submitted to the Department of Mechanical Engineering

on November 20, 2018, in partial fulfillment of the
requirements for the degree of
Ph.D. in Mechanical Engineering

Abstract
This thesis presents tools to design progressive cavity pumps (PCPs), with an em-
phasis on low-viscosity fluids. These models indicate that high speed operation can
increase sealing performance, decrease pump size, and eliminate gear-reductions. New
models for estimating both laminar and turbulent internal flow and shear losses in
these pumps are presented. The new models are capable of estimating pump perfor-
mance 1000x faster than traditional simulation methods, and do not require empirical
calibration, making them 'designer-ready'. A proof-of-concept turbulent PCP was de-
signed using these models. Its volumetric efficiency is within 20% of predicted values.
This thesis also presents a novel one degree-of-freedom hypocycloidal bearing to
constrain the motion of the rotor for increased performance and control. 3 different
bearing topologies have been developed: roller, rail, and flexural. An experimental
PCP concept with integrated hypocycloidal rail bearings was developed and tested
with efficiencies as high as 45%. Experimental data are compared with a new lubrica-
tion theory model which accounts for rotor motion, rotor geometric error, and stator
geometric error. The experimental and theoretical results show strong agreement,
proving that low-order lubrication theory models are accurate simulation tools.
Additionally, performance results from the rail bearing pump and first-order anal-
ysis inspire new scaling laws for connecting the volumetric and mechanical efficiency
of PCPs. These scaling laws show strong agreement in both turbulent and laminar
flows.
A new generation of PCPs has the potential to transform irrigation, water purifica-
tion, oil-sand extraction, among other applications. The new tools required to create
these PCPs also have strong implications for how traditional PCPs are designed.

Thesis Supervisor: Alexander H. Slocum

Title: Walter M. May & A. Hazel May Professor of Mechanical Engineering

3
4
Acknowledgments
This thesis was made possible by support from the MIT Tata Center, the Tata Trusts,
the Martin Family, and the Abdul Latif Jameel Water & Food Systems Lab. I also
want to acknowledge the Ann Goss Foundation for providing me the Jackson W. Goss
Fellowship to help build Khethworks in the Martin Trust Center's Global Founders'
Skills Accelerator during the summer of 2015. I particularly want to thank the MIT
Tata Center. Not only did they sponsor 4 of my years in grad school. The 12 trips to
India, connections, and faith in the potential of my pump work was essential to this
project.
Thank you Prof. Slocum for your faith in my skills while constantly injecting
non-linear genius to my process. You've been a world-class advisor.
Mom, Julia, and Dad: thank you for everything that you have done to support,
care for, and teach me. You were my first champions, and I would not be 'here'
without you.
I am deeply grateful to my partner, Hilary White. Your love, support, and patience
in the last 5 years mean the world to me.
One of the most essential ingredients for good research is a thriving intellectual
community. While the nature of a thesis is individual, this work would have been
impossible without the wisdom, knowledge, and support of my colleagues at MIT
and elsewhere. Thank you David, Hilary, Abe, Marcel, Thanh, Tyler, and Lei. It is
impossible for me to thank and account for every person who has shaped this thesis.
You all in particular have critically influenced this work. PERG and Makerworks
have been cornerstones of my MIT experience, and I am deeply grateful for those
groups of wonderful individuals.
The data presented in this thesis would not have been possible without the help
of Aaron Johnson's CT stator scanning at Amphenol TCS; Nick Sondej's 3D rotor
scanning; David Taylor's data collection code; or Aaron Ramirez's help instrumenting
my torque sensor. Thank you so much for your help.

5
6
Contents

1 Introduction 27
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.2 PCP Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.3 Why PCPs? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4 Goals of this Research . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5 PCP Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5.1 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5.2 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5.3 D esign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.5.4 Water Applications . . . . . . . . . . . . . . . . . . . . . . . . 37

2 Lubrication Theory and Geometric Errors 39

2.1 Lubrication Theory Model . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1.1 Derivation and Clarifications . . . . . . . . . . . . . . . . . . 39
2.1.2 Numerical Implementation . . . . . . . . . . . . . . . . . . . . 46
2.2 Modeling Geometric Error in PCPs . . . . . . . . . . . . . . . . . . . 48
2.2.1 Imperfect Rotor and Stator Model . . . . . . . . . . . . . . . 50
2.2.2 Rotor Surface Error . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.3 Rotor Error Motion . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.4 Stator Surface Error . . . . . . . . . . . . . . . . . . . . . . . 58
2.3 Internal Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.1 Torque Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.2 Direct Integration from the Lubrication Theory Model . . . . 63

7
 2.4 Implications for Qth . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

.
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

.
3 Turbulence in Progressive Cavity Pumps 75
3.1 Network M odels . . . . . . . . . . . . . . . . . . 76

.
3.1.1 Background . . . . . . . . . . . . . . . . 76

.
3.1.2 Extension of Network Model . . . . . . . 80

.
3.1.3 Model Validation . . . . . . . . . . . . . 86

.
3.1.4 Combined Turbulent and Lubrication Flow Resistance 89
3.2 RANS Turbulence Models . . . . . . . . . . . . 91

.
3.2.1 Turbulence Model Application . . . . . . 91

.
3.2.2 Inclusion of Cell-Averaged Inertial Terms 95

.
3.3 Conclusions . . . . . . . . . . . 96
.

4 PCP Design 97
4.1 1-DOF Cycloidal Bearing Design 97
4.1.1 Orbiting Strategy..... 108
4.1.2 Hirth Coupling Assembly. 109
4.2 Rolling Bearing Experiment . . 110
.

4.2.1 M otor . . . . . . . . . . 111

.

4.2.2 Sensors . . . . . . . . . . 112

.

4.2.3 Total Design . . . . . . . 112

.

4.2.4 Cavitation . . . . . . . . 116

.

4.2.5 Results . . . . . . . . . . 118

.

4.3 Rail Bearing Experiment..... 119

4.3.1 Structure Design..... 120
4.3.2 M otor . . . . . . . . . . 121
.

4.3.3 Sensors . . . . . . . . . . 121

.

4.3.4 Total Design . . . . . . . 125

.

4.3.5 Results . . . . . . . . . . 129

.

4.4 Water PCP Concept . . . . . . 142

.

8
5 Conclusion 145
5.1 Rem arks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.2 Future W ork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

9
10
List of Figures

1-1 PCP rotor (left), stator (right), and cavities (bottom). . . . . . . . . 28

1-2 A depiction of typical PCP geometry from the side (left) and in cross-
section (right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1-3 The clearance gap within a progressive cavity pump. The rotor has
six pitches over its 0.36 m length with a rotor radius of 0.020 m. The
yellow regions correspond to cavities. . . . . . . . . . . . . . . . . . . 31
1-4 The ranges of flow applications where pumping technologies are cost
effective, compared to the requirements of irrigation and water purifi-
cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2-1 Different zones of the stator relative to the center of the rotor, adapted
from Andrade et al. [3]. . . . . . . . . . . . . . . . . . . . . . . . . . 41
2-2 The critical dimensions of a PCP viewed from the side (left) and outlet
(right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2-3 A representation of the numerical grid used to evaluate the Poisson
pressure equation. Each node (O,z) corresponds to a point where the
pressure is evaluated. . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2-4 Net flow rate at different sections as a measure of conservation of mass
for a centered grid (left) and a staggered grid (right) . . . . . . . . . 48
2-5 The pressure difference between Andrade's results and the model in
this paper on a grid where n2 = 500 and no = 250. . . . . . . . . . . 49
2-6 An example of sinusoidal surface error on the rotor. (Ae = 185 pm,
n = m = 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

11
2-7 Geometric definitions of the two motion-driven error modes. . . . . . 52
2-8 Time averaged flow rates for different surface error amplitudes and
spatial frequencies (n=m=1 to n=m=10). The mesh was created with
1400 nodes in z and 400 nodes in 0 to resolve the spatial frequencies. 53
2-9 A two dimensional depiction of rotor error motion. . . . . . . . . . . 53
2-10 An example of rotor radius changes corresponding to small angle rotor
motion. (A 1 = 83 pm, A 2 = 83 im, '1 = 0, '72 = 7r/2) . . . . . . . . 54

2-11 Experimental results from Gamboa with 0.042 Pa-s oil compared with
Andrade's centered rotor lubrication theory model and the rotor error
motion uncertainty presented in this thesis [18, 3]. . . . . . . . . . . . 55
2-12 Experimental results from Gamboa with 0.134 Pa-s oil compared with
Andrade's centered rotor lubrication theory model and the rotor error
motion uncertainty presented in this thesis [18, 3]. . . . . . . . . . . . 56
2-13 Experimental results from Gamboa with 0.481 Pa-s oil [18]. The theo-
retical flow-rate for this pump, Qth, is 5.39 m 3 /s. Both the experimen-
tal data and skewed-rotor model have flow-rates greater than Qth. . . 57

2-14 Rail bearing experimental results and lubrication theory simulation

based on rotor and stator scans. ...... .................... 58
2-15 Gamboa's experimental results compared with CFD simulation by Pal-
adino [19, 44]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2-16 The sensitivity of 2D lubrication theory and 3D CFD simulations to
grid size. 2-D lubrication theory has a much faster convergence rate
than conventional CFD. . . . . . . . . . . . . . . . . . . . . . . . . . 60
2-17 Comparison between Vetter's experiment and lubrication theory with
a corresponding turbulent viscosity. . . . . . . . . . . . . . . . . . . . 70
2-18 Lubrication theory results compared with the Paladino CFD for the
same geometry (botton) [44. Simulated (left) and experimental (right)
torques from this work. . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3-1 A depiction of a single flow resistor. . . . . . . . . . . . . . . . . . . . 77

12
3-2 Illustration of the cavity nodes and connectivity in a PCP network
model. The arrows represent slip flow between cavities. . . . . . . . . 80
3-3 Pressure distribution along the rotor of a representative 2 cavity pump.
Transverse sealing lines connect cavities that are two nodes apart, and
are horizontal lines above. Longitudinal sealing lines connect neigh-
boring cavities, and appear at an angle above. . . . . . . . . . . . . . 81
3-4 The effect of different geometric variables on the longitudinal pres-
sure loss coefficient. Higher values of K/K indicate better geometry-
independent sealing and delayed transition to turbulence. . . . . . . . 87

3-5 Results from the orifice network model compared with experimental
data from Gamboa and Vetter [19, 57] . . . . . . . . . . . . ..... 88
3-6 Results from the water prototype plotted against predictions from the
orifice network model. Black lines connect experimental data and sim-
ulation results at the same speed and pressure. . . . . . . . . . . . . . 88
3-7 A viscous channel followed by a mixing dominated sudden expansion. 89

3-8 Results from the long orifice network model compared with experimen-
tal data from Gamboa and Vetter [19, 57]. . . . . . . . . . . . . . . . 90
3-9 Results from the water prototype plotted against predictions from the
long orifice network model. Black lines connect experimental data and
simulation results at the same speed and pressure. . . . . . . . . . . . 91

4-1 Example pressure distribution over a water hydrodynamic bearing,

computed using lubrication theory. In this case, L = 20 mm, Rj =
15 mm. c = 20 pm, e = 10 pm, and Q = 419 rad/s . . . . . . . . . . 100
4-2 Front view of the CAD for the vertical (left) and horizontal (left) flex-
ural hypocycloidal crankshaft bearing stages. . . . . . . . . . . . . . . 103
4-3 Energy stored in the othrogonal flexure system. . . . . . . . . . . . . 104

4-4 Angled view of a flexural hypocycloidal crankshaft bearing. . . . . . . 105

4-5 The dynamic response of the under-constrained floating stage . . . . 107

4-6 Sketch of the orbiting bearing topology. . . . . . . . . . . . . . . . . . 108

13
4-7 Image of the water pump crank assembly with one of the Hirth cou-
plings visible from the side. . . . . . . . . . . . . . . . . . . . . . . . 109
4-8 Sectioned view of the water pump hypocycloid bearing and crank shaft
rotor.. ...... ....... . ............... . . . . ....... 111
4-9 Sectioned and whole views of the water pump assembly CAD. . . . . 114
4-10 Picture of the full water pump assembly. . . . . . . . . . . . . . . . . 115
4-11 Image of the crankshaft rotor (top). Close up of the inlet side (left) and
outlet side (right) of the crank with roller bearings and axial support
ball. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4-12 Cavitation on the inlet side of the water prototype rotor due to incor-
rectly designed PCP inlet conditions. . . . . . . . . . . . . . . . . . . 116
4-13 Section view of the PCP inlet and inlet bearings with the side inlet
(top), and the axial inlet (bottom). . . . . . . . . . . . . . . . . . . . 118
4-14 Results from the water prototype plotted against predictions from the
orifice network model. Black lines connect experimental data and sim-
ulation results at the same speed and pressure. . . . . . . . . . . . . . 119
4-15 Sectioned isomeric view of the rail bearing pump outlet with ball gear
joint, outlet bearings, and eddy current probe CAD (left). Assembled
ball gear joint (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4-16 The residual from fluid accelerations for the highest flow-rate experi-
ment. This residual at most is 3% of the nominal pressure. . . . . . . 123
4-17 The pressure loss coefficient computed for various needle-valve posi-
tions in the flow loop . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4-18 CAD cross-section (top left) alongside a picture (top right) of the drive-
shaft end of the rotor inside the stator with an eddy-current target
(yellow in the CAD, and aluminum in the picture) attached. CAD
cross-section of the inlet end of the rotor inside the stator with an
eddy-current target attached (bottom left)alongside a picture of the
eddy-current probe placement on one end of the stator (bottom right). 126
4-19 Whole and sectioned views of the rail bearing pump assembly CAD. . 127

14

._'_11
.......... - - -I-- J
4-20 Image of the assembled rail-bearing pump with flow loop. . . . . . . . 128
4-21 Close up of the rotor inside of the clear stator. . . . . . . . . . . . . . 128
4-22 Sectioned isomeric view of the rail bearing pump inlet with ball drive,
outlet bearings, and eddy current probe CAD (left). Assembled rail
bearing prototype (right). . . . . . . . . . . . . . . . . . . . . . . . . 128
4-23 lubrication theory simulation results alongside flow rates from the rail
bearing prototype experiment. . . . . . . . . . . . . . . . . . . . . . . 129
4-24 Efficiencies from the rail bearing prototype experiment. Volumetric
efficiency, mechanical efficiency, and total efficiency are o, o, and x
respectively . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4-25 Measured shaft torque from the rail bearing prototype experiment
(left) and lubrication simulation torques (right). . . . . . . . . . . . . 131
4-26 Scaling model for PCP efficiency compared with experimental data for
volumetric efficiency (o), mechanical efficiency (o), and total efficiency
(x). ......... .................................... 132
4-27 Scaling model for PCP efficiency compared with experimental data
from the water prototype. . . . . . . . . . . . . . . . . . . . . . . . . 133
4-28 The simple single quadratic resistor slip flow relationship used in the
turbulent scaling laws compared with experimental results and the ori-
fice network model. Black lines connect experimental data and simu-
lation results at the same speed and pressure. . . . . . . . . . . . . . 134
4-29 Estimates of motor torque from the water (+) prototype experiment
and scaling law fit (*). Black lines connect experimental data and
simulation results at the same speed and pressure. . . . . . . . . . . . 135
4-30 A comparison of similar turbulent and laminar pumps to illustrate the
effect of turbulent sealing. . . . . . . . . . . . . . . . . . . . . . . . . 136
4-31 The flow (left) and pressure (right) signal overlaid in the same time
window with the valve 1.5 turns open. . . . . . . . . . . . . . . . . . 139
4-32 The torque (left) and probe (right) signal overlaid in the same time
window with the valve 1.5 turns open. . . . . . . . . . . . . . . . . . 139

15
4-33 Torque (left) and pressure (right) measurements for the rail bearing
prototype running at 200 rpm with mineral oil at various valve posi-
tions. The grey line at the bottom of the torque measurements is a
kinematically repeatable parasitic torque that was measured by run-
ning the pump with only enough working fluid to lubricate the rotor. 140
4-34 Circumferential distribution of deviations from the stator's internal
designed geometry (left), and the rotor radius (right). . . . . . . . . . 141
4-35 The effect of stator and rotor error on the pressure fluctuations assum-
ing a laminar flow loop pressure drop. . . . . . . . . . . . . . . . . . . 142
4-36 Estimated flow rates for the water PCP concept proposed (left) and
the hypothetical efficiency curve for the water PCP concept (right). 143

-1 The torque speed curve for the 34Y214 motor wired in parallel. . . . 157

16
List of Tables

2.1 R, values in different regions of the PCP. These are updated to correct
for some typos in Andrade's original work [3]. . . . . . . . . . . . . . 42
2.2 OR 0 values in different regions of the PCP. These are updated to correct
for some typos in Andrade's original work [3]. . . . . . . . . . . . . . 43
2.3 Simulation runs to illustrate flow and rotor dynamics. . . . . . . . . . 72

3.1 The pressure profiles of different PCP models for the 4 sealing cavity
PCP. The linear models were calibrated against a two cavity PCP
lubrication simulation, and were simulated with Re ~ 50. . . . . . . 86

4.1 Critical design dimensions of the water prototype . . . . . . . . . . . 113

4.2 Critical design dimensions for the rail bearing prototype. . . . . . . . 120

17
18
 Nomenclature

cI Angle defining stator region I [rad]

&2 Angle defining stator region II [rad]

Tcy1 Fluid shear stress tensor in cylindrical coordinates [Pa]

7 Total fluid shear stress in cartesian coordinates [Pa]

Tf Reference frame shear stress [Pa]

1 Cumulative flow [m 3 ]

A Pressure matrix [Pa]

R Rotation matrix [-]

AP Static pressure rise [Pa]

APL Laminar sealing line static pressure drop [Pa]

APT Transverse sealing line static pressure drop [Pa]

APTL Long orifice static pressure drop [Pa]

A Driven stage displacement [im]

6 Flexure displacement [im]

6 Surface roughness [m]

7 Total PCP efficiency [-]

TIL Total PCP efficiency for laminar operation (from scaling laws) [-]

r/M Mechanical efficiency [-]

W7T total PCP efficiency for turbulent operation (from scaling laws) [-]

19
 r'v Volumetric efficiency [-]

rqm, Motor controller efficiency [-]

71 Meridional angle of the rotor error translation [rad]

72 Meridional angle of the rotor error pitch [rad]

w Nominal clearance gap [m]

P Coefficient of friction for the Hirth coupling (Section 4.1.2) [-1

P Dynamic viscosity [Pa-s]

Pt Effective turbulent eddy viscosity [Pa-s]

v Kinematic viscosity [m 2 /s]

v Poisson ratio (Chapter 4) [-]

Q Rotational velocity [rad/s]

Q* Dimensionless rotational velocity [rad]

we Critical speed for shaft whip [rad/s]

p Fluid density [kg/M 3]

Ub Flexure bending stress [Pa]

T Torque [N-m]

TK Flow meter time constant [pulses/s]

TO Parasitic torque [N-m]

0 Circumferential coordinate [rad]

Es Angle between the stator section centerline and the horizontal plane [rad]

n Rotor surface normal [-]

V Rotor surface velocity [m/s]

A Angle between stator regions I and III (Chapter 2) [rad]

A Fluid resistor matrix coefficient [im3 /Pa-s]

A' Angle between stator regions I and IV [rad]

20

- T- -1---l- -- , mlw---"1111 1 -, - , - 1- 11--711-11-1-11 - -1 1 1 . 1 - -1.11-1-1,1-1,1-1- - "I - - 1 -- 111-11 -. 111,11, '-.- I-

*11"M
A1 Amplitude of the rotor error translation [m]

A2 Amplitude of the rotor error pitch [m]

Ae Surface error amplitude [m]

Af Driven stage displacement amplitude [m]

B Angle between stator regions II and III [rad]

b Sealing line width [m]

B' Angle between stator regions II and IV [rad]

biong Longitudinal sealing line width [m]

btrans Transverse sealing line width [m]

c Nominal journal bearing clearance height [m]
C1 Circumferential coefficient of integration [m 2 /Pa-s]

C1 Coefficient of integration (Chapter 4) [m]

C2 Axial coefficient of integration [m 4 /Pa-s]

C2 Coefficient of integration (Chapter 4) [m]

Cd PCP modulus of drag [N-m s/rad]

Cf PCP coefficient of friction [-]

C, Pressure loss coefficient [-]

CQ PCP flow modulus [m 3 /rad]

CR Flow resistance modulus [Pa-s/m 3l

Cow Coefficient of integration [m 2 /s]

c~r PCP torque modulus [m 3]

CpL Laminar pressure loss coefficient [-]

CpT Turbulent pressure loss coefficient [-]

CsL PCP laminar slip modulus [m 3/Pa-s]

CsT PCP turbulent slip modulus [m 3 /s Pal

21
D Plate stiffness [N-m]

Df Hydraulic diameter [im]

Dh Hydraulic diameter [m]

dcsR Distance between the center of the stator and the center of the rotor [m]

dA Surface area differential [m 2]

E Young's modulus (Chapter 4) [Pa]

e Journal bearing eccentricity [im]

Ef Flexural energy [J]

Eotat Total Hirth coupling stiffness [N/m]

Ek Ekman number [-]

f Friction factor [-]

Ff Applied flexure force [N]

Fp Hirth coupling preload force [N]

FT Tangential Hirth coupling force [N]

Fc, Critical flexural buckling load [N]

g Gravitational acceleration at the surface of the earth [m/s 2]

h Journal bearing clearance height [m]

I Current [A]

I Flexure second moment of area [m 4 ]

I,, No-load current [A]

K Laminar viscous coefficient [-]

K Simplifying coefficient (Chapter 2) [-]

k Flexure stage stiffness [N/m]

Kb Buckling coefficient [-]

kh Horizontal flexure stiffness [N/m]

22
KL Longitudinal laminar viscous coefficient [-)

Km Flow meter K-factor [m 3 /pulse]

KT Transverse laminar viscous coefficient [-3

Kt Motor torque constant [V-s/rad]

K, Motor voltage constant [rad/V-s]

kv Vertical flexure stiffness [N/m]

L Longitudinal cavity fluid conductivity [m 3 /Pa-s]

L PCP rotor length (used for non-dimensionalization in figures) [m]

I Length of the rotor [ml

L* Dimensionless longitudinal cavity fluid conductivity [-3
if Flexure length [m]

is Effective sealing line length [m]

M Flexure bending moment [N-m]

M Floating stage mass [kg]

m Rotor mass (Chapter 4) [kg]

m Surface error spatial frequency in z (Chapter 2) [-]

n Surface error spatial frequency in 0 [-]

No Number of grid cells in 0 [-]

np Number of parallel flexure stages [-]

N2 Number of grid cells in z [-]

P Motor power [W]

p Static pressure [Pa]

P1 First cavity relative static pressure [Pa]

P2 Second cavity relative static pressure [Pa]

Pc Power of convergence [-]

23
P Relative static outlet pressure [Pa]

Pr Nominal rotor pitch [m/rev]

P, Nominal stator pitch [m/rev]

Prot Rotor input power [W]

Qm Lubrication theory model predicted flow rate [m 3 /s]

Qexp Experimentally measured flow rate [m 3 /s]

Qthr Maximum flow rate, computed from the rotor geometry [m 3 /s]

Qth, Maximum flow rate, computed from the stator geometry [m 3 /s]

Qth Theoretical maximum flow rate (could use Qths or Qthr) [m3 /S]

R Resistance (Chapter 4) [Ohm]

r Radial coordinate [m]

RI Journal bearing radius [m]

RL Longitudinal cavity sealing resistance [Pa-s/m 3]

Ro Radial distance between the center of the rotor and the wall of the stator [im]

Rr Nominal rotor radius [im]

Rs Nominal stator radius [m]

RT Transverse cavity sealing resistance [Pa-s/m3 ]

R,li Total pump slip flow resistance [Pa-s/m 3]

Ro Rossby number [-]

S Rotor surface coordinate (Chapter 2) [m]

S Slip flow rate [m 3 /s]

SL Laminar slip flow rate [m 3 /s]

ST Turbulent slip flow rate [m 3 /s]

Scr Rotor center coordinate [i]

T Transverse cavity fluid conductivity [m 3 /Pa-s]

24
t Time [s]

T* Dimensionless transverse cavity fluid conductivity [-]

U Slip flow speed [m/s]

u Axial fluid velocity [m/s]

V Voltage [V]

v Radial fluid velocity [m/s]

Vf Reference frame velocity [m/sj

w Tangential fluid velocity [m/s]

Wf Flexure width [m]

z Axial coordinate [im]

AR Area ratio [-]

E Rotor eccentricity [im]

errm Rotor error motion magnitude [m]

errs Surface error magnitude [m]

Re Reynolds number

25
26
Chapter 1

Introduction

There are many types of pumps for moving fluids and gases, and societies of any type
critically depend on them. The use of pumps for irrigation and drinking water is
among the oldest applications of engineering. One of the first pumps, the shadoof,
was documented in the city of Uruk, in what is now Iraq, around 3,300 BC [58]. Denis
Papin invented the world's first true centrifugal pump in 1687 [62]. In the meantime,
the role of pumps in irrigation has grown significantly. In 2016 the agricultural
sector in India consumed 173 TWh of energy [31]. Much of that energy was used
for irrigation. In spite of the maturity and importance of this field, there is still a
gap in the use of electrical power for irrigation or the production of drinking water.
This problem is acutely relevant in developing countries. For example, East India has
among the highest 'fraction of rural persons going hungry' in India [12]. Tragically,
this is a region where many agricultural fields lay un-used for over half of the year
because of a lack of access to affordable irrigation and market linkages. Ending hunger
and providing access to clean water are both UN Sustainable Development Goals [17].
Providing access to clean water, and making solar energy economical are both NAE
Grand Challenges [46]. Creating more efficient low-energy water pumps is directly
relevant to these broad societal goals.

27
 I _4

Figure 1-1: PCP rotor (left), stator (right), and cavities (bottom).

1.1 Motivation

This research was initially motivated by the author's work with small plot farmers
in East India as part of a Tata-trusts sponsored fellowship [49]. Agriculture is a
major source of livelihood for over 300 million farmers in India. In the summer,
farmers are able to rely on the monsoon rains to irrigate their crops, but winter
and dry-season farming require irrigation. Some farmers have the benefit of canals,
but the majority of farmers in India depend on groundwater for irrigation. Even
though there is considerable growth in groundwater usage, the cost of pumping water
and access to energy often makes year-round irrigation unfeasible for these farmers.
Many small plot farmers use costly and polluting diesel powered pumps to irrigate
their fields. Current irrigation technologies are not accessible for the majority of
Indian farmers. Over 50% of all cropped land in India is unirrigated [1]. Where there
is access to electricity, the subsidized agricultural load on India's electricity grid is
costly and decreases the hours of available energy for other uses. In the 2015-2016
season, agricultural irrigation accounted for 17.3% of electrical power consumption in
India [31].
Increasing the efficiency of groundwater pumps has the ability to save government
money on energy subsidy programs; increase the quality of grid operations; enable
year-round irrigation; and enable solar irrigation to compete with diesel pumps. This
improvement could be realized by creating better positive displacement pumps, which
has the potential to be more efficient than centrifugal pumps for the pressure rise
(100-350 kPa) and flow rate (1-2 lps) requirements of these farmers.

28
1.2 PCP Principles

The progressive cavity pump (PCP) was invented in 1930 by Moineau and presented in
his doctoral thesis [36]. Some authors refer to them as progressing cavity pumps, but
this thesis will exclusively use 'progressive cavity pumps'. PCPs are rotary positive
displacement pumps with a single helix rotor eccentrically rotating in a hypocycloidal
trajectory inside of a double helix stator. Multiple helix PCPs do exist, but that ge-
ometry is most commonly used in 'mud-motors.' This work focuses exclusively on
single helix PCPs. An example of the geometry, and the respective cavities, contain-
ing discrete volumes of fluid, are shown in Fig. 1-1. Those cavities spiral along the
axis of the pump. PCPs are commonly used in chemical processing and oil and gas
applications because of their efficiency and ability to handle high viscosity fluids and
mixed phase flows. Even though the pumps have no valves and few components, cur-
rent PCPs are too expensive for use in irrigation due to the manufacturing tolerances
of the rotor and stator. Beyond price concerns, the efficiency of PCPs is driven by slip
flow along sealing clearance lines. This slip flow rate is proportional to the viscosity
of the fluid, which makes designing PCPs for water particularly challenging.

PS
4E

R.

Pr-

Figure 1-2: A depiction of typical PCP geometry from the side (left) and in cross-
section (right).

Each cavity progresses along the axis of the PCP at a rate that is set by the rotor
speed and pitch. The product of axial cavity velocity and the cross-section of the
pump are the basis of Eq. 1.1, an estimate for the ideal flow-rate if there is no slip
flow [47, 57, 39]. E is the eccentricity of the spirals. R, is the radius of the rotor. R,

29
is the radius of the stator. P, is the pitch of the stator, typically expressed in units
of [m/rev]. P, is the pitch of the rotor, which is half of the pitch of the stator. Q is
the rotational speed of the rotor. Unless otherwise indicated, Q is expressed in S.I.
units, [rad/s]. These geometric terms are shown in detail in Fig. 1-2.

(1.1
Qths = [8ER, + r(R2 - R ,2)]
27r

Another, less common, estimate for the maximum theoretical flow-rate of a PCP,
Eq. 1.2, is based on the area that the rotor sweeps out [4].

Qthr = - ERrPQ (1.2)

7r

The clearance can be visualized for an example pump geometry (6 pitches, 0.36 m
long, rotor radius of 0.02 m, stator radius of 0.020185 m, eccentricity of 0.004039 m)
in Fig. 1-3. This is the geometry of the pump used in experiments by Gamboa et
al., which are commonly used to validate PCP models [19, 43, 3, 11, 47]. While the
rotor-stator clearance in rigid stator PCPs gets to be small between the cavities, it
is never zero, as seen in Fig. 1-3. It follows that the pressure difference between the
cavities creates back-flow along those clearance lines, visualized in Fig. 1-2, which
reduces the pump's volumetric efficiency. The shaft-to-flow efficiency of a PCP is the
product of volumetric efficiency and mechanical efficiency.

1.3 Why PCPs?

This thesis focuses on PCPs for two reasons. Progressive cavity pumps have ex-
ceptional flow characteristics: they have the advantages that positive displacement
pumps have over centrifugal pumps at high pressure differences and low flow rates (i.e.
low specific speeds) without the challenges that most positive displacement pumps
have with valving, flow pulsations, handling solids, or complexity.
Centrifugal pumps reduce in efficiency with specific speed and with flow rate [28].
One highly optimized centrifugal pump, with a still greater specific speed than the

30
 Rotor-Stator Clearance [ml x 10-3
0 .3 5
14

0.3 .12

0.25 10

0.2 8
N
0.15 6

0.14

0.05 2

0
1 2 3 4 5 6
Theta [rad]

Figure 1-3: The clearance gap within a progressive cavity pump. The rotor has six
pitches over its 0.36 m length with a rotor radius of 0.020 m. The yellow regions
correspond to cavities.

applications considered in this thesis, reaches a maximum efficiency of 60%. However,

the designer of that pump notes that with increasing pressure., the efficiency of the
pump would be reduced considerably [10].
In spite of these properties, PCPs are mainly used in oil and gas, sewage., and
chemical processing because of the high cost of precision manufacturing for the com-
plex shapes. As such, there is limited literature on applications outside of oil and gas.
When this thesis was written, there was not enough literature for an engineer skilled
in the art of fluid machinery design to quickly and easily do early stage design or
evaluation the feasibility of a water-moving PCP. Most low-order models are tailored
to high viscosity fluids; all models assume that manufacturing is arbitrarily precise;
and many models have ambiguities that make it difficult for a designer to quickly
implement them.
It is this author's belief that there are, as yet, un-designed PCPs with the potential
to improve water delivery and purification. Figure 1-4 qualitatively shows what this
design space could look like. There are likely other existing applications for PCPs
beyond those as well. The lack of literature on this subject; the magnitude of current
water-related challenges; and the fundamental advantages of PCPs justify a PCP-

31
centric research effort to expand the world of possibilities for PCPs working with
water.

Cost Effectiveness

New PCP Tech

Irrigation

+
Purification

nrifugal
Flow(Q)
Figure 1-4: The ranges of flow applications where pumping technologies are cost
effective, compared to the requirements of irrigation and water purification.

1.4 Goals of this Research

The goal of this research is to develop tools to design a PCP for water which can
ideally be molded from plastic for low cost. There are obvious challenges with such a
technology. Plastics are softer and more compliant than metal, making them less well
suited to supporting the large pressure loads or resisting abrasion. Furthermore, plas-
tic molding processes are generally less precise than metal forming processes. Since
flow resistance in a thin channel is highly sensitive to the channel width, these varia-
tions in clearance can have substantial effects on slip flow performance. In addition to
challenges with plastic, water has a low dynamic viscosity of p ~ 0.001 Pa-s, resulting
in a large slip flow that subtracts from total pump efficiency. Given these challenges,
several aspects of PCP design need to be better understood before a low-cost water
moving PCPs can be properly designed.
To alleviate contact and abrasion between the rotor and stator, supports need to be
developed that constrain the rotor to the one degree of freedom (DOF) hypocycloidal

32
motion of PCPs. These supports may not be necessary, but they make possible
considerable improvements in pump life and behavior. By preventing contact between
the rotor and stator, pump life can be extended to the life of the rotor supports.
Clearances up to the largest particle size can be sustained for a large portion of the
pump's life, even for relatively soft materials. This would reduce the need to use
exotic materials for pump components, while even increasing the lifetime efficiency
of a PCP. Preventing rotor-stator contact can also increase the mechanical efficiency
of a PCP and enable higher speed operation. It has already been noted in literature
that higher-speed operation results in greater volumetric efficiencies at the expense
of mechanical efficiency and pump life [57].

Tools for understanding the trade-off between mechanical and volumetric efficiency
in PCP design are critical for optimization. Mechanical efficiency monotonically de-
creases with speed, and volumetric efficiency monotonically increases with speed [57].
For any design point, there is an optimal speed and geometry which balances these
effects. With the exception of experimental work by Vetter et al. and CFD work by
Paladino et al., there has been limited work to study the role of mechanical efficiency
in rigid stator and rotor PCP designs [57, 44]. This research aims to develop tools for
better understanding this tradeoff, starting with rigid rotor and stator PCPs. While
compliant stator pumps also have promise for pumping water, tools for rigid stator
pumps are still necessary to fully explore the design space. Many of the techniques de-
veloped and improved upon in this thesis are building blocks for modeling compliant
PCPs.

Significantly reducing the cost of rotors and stators will likely require switch-
ing to molded parts. Molded plastic components typically have looser tolerances
than machined metal. To understand the tradeoffs between lower cost manufactur-
ing processes and PCP performance, better modeling tools need to be developed for
predicting the connection between manufacturing error and pump performance.

The low viscosity of water makes is particularly challenging to design and model
PCPs. Clearance gaps are often on the order of 0.1 mm, and cavity diameters are
on the order of 10 mm. Modeling the generation and convection of turbulence across

33
these varying length-scales on a moving domain is challenging and computationally
expensive. The first CFD model for a water moving PCP that had any agreement
with experimental data was presented in 2009 [42]. It took 100 hours to run. CFD
simulations for PCPs are faster now, but efficient design still needs faster tools for
predicting pump performance. Furthermore, this new model also needs to be able to
account for small variations in rotor and stator geometry due to the manufacturing
tolerances mentioned above.

1.5 PCP Literature Review

1.5.1 Experiments

The data collected by Gamboa et al. is one of the most detailed experimental studies
of the performance of rigid stator PCPs in literature, and is commonly used as a
benchmark for PCP simulations. Almost every rigid-stator PCP model validates
their results against Gamboa's data [20].
Berton conducted rigid stator PCP experiments with water and non-Newtonian
fluids to validate their CFD model [5]. Their experiment appears to pay great at-
tention to detail, and their experimental results show good agreement with their
simulation. However, they only share a limited amount of experimental data, making
it hard to extract many insights about PCP performance and design from their work.
Chandel et al. has performed experimental studies comparing compliant stator
PCPs with centrifugal pumps for handling fly ash slurry [7]. Ward et al. have also
evaluated the efficiency of elastomeric stator PCPs for pumping water with solar
power, and found promisingly efficient performance [60].

1.5.2 Models

The three existing ways to model PCPs are with: equivalent fluid resistors; lubrication
theory; and full 3 dimensional (3D) numerical computational fluid analysis (CFD)
such as FLUENT or ANSYS.

34
 Poiseuille flow is a simple model for PCP back-flow, and was the first and most
common approach for modeling flow slippage within PCPs [36, 18]. Pessoa et al.
presents a Poiseuille inspired model which requires calibration to manage the length
and width of the clearance gaps. This makes the model better suited to on-line
control than pump design [47]. Zheng presents an interesting and promising method
for computing the cavity-to-cavity flow resistance for laminar flows directly from PCP
geometry using a few simple assumptions [63].

Lubrication theory provides the most geometrically-driven reduced order model

of back-flow leakage in PCPs. Andrade et al. developed a 2D numerical model for
flow within PCPs [3]. By assuming that the Reynolds number is small, the inertial
and transient terms are neglected, leaving the equations for viscosity dominated flow
in an annular channel, where the channel height is small compared to its radius. It
is important to note that models in cartesian coordinates, which assume that the
curvature of the rotor is negligible, were unable to match experimental data [3].
Lubrication-theory based models in cylindrical coordinates account for the curvature
of the rotor and stator, matching experimental data well. However, this model cannot
capture the behavior of low-viscosity fluids such as water. This type of model has
also been implemented by Li et al. for single screw extruders [35].

Numerical simulation with CFD and moving meshes are also able to predict the
volumetric performance of rigid and compliant rotor-stator combinations, but are at
least three orders of magnitude more computationally expensive than the cylindrical
lubrication-theory model mentioned above. Numerical simulation is particularly pop-
ular for analyzing PCPs with elastomeric stators. The compliant stators are used to
provide a slight interference fit, which seals the pump cavities. However, these sta-
tors eventually deform under the outlet pressure, providing more clearance for more
slip. This interaction creates a non-linear effect which rapidly reduces the efficien-
cies of this type of pump beyond a certain pressure. Therefore, numerical simulation
is particularly important for predicting the pressure rise that a design is capable of
delivering. Chen et al. have implemented a fluid-solid interaction model that uses
ABAQUS and FLUENT for the elastic and fluid models respectively through coupled

35
iterations [9]. Zhou et al. also present numerical simulation of compliant stator PCPs
for both circular and even thickness stators [64].

Other authors investigate a fully 3D transient model of flow within rigid stator
and rotor PCPs. Fully transient models with three spatial dimensions have been
implemented in CFD with it-E [44] and large-eddy simulation [5] with results that
match experimental data [20] from the literature well. Mrinal et al. have also run
a CFD simulation against Gamboa's data with 21% error [37]. Mrinal claims that
inaccuracies in reporting the rotor and stator geometry is the primary source of error.
Although these models provide accurate results, they are computationally expensive.
This drawback makes them ill-suited to developing new designs. Full CFD models
must address moving meshes and fully discretize all three flow velocity vectors in
three dimensions.

Given that PCPs have advantages with handling mixed-phase flows, authors have
also developed CFD simulations to model the performance of PCPs for pumping both
fluids and gases. de Azevedo has successfully implemented a mixed-phase model which
is able to describe experimental data very well [11].

1.5.3 Design

Saveth put forward design equations for what he defines as optimal PCP design [48]. It
appears that Saveth's model minimizes fluid velocity for a desired flow rate and fixed
well diameter and rotational speed. This is a useful metric, particularly because the
velocity of entrained particles dictates pump wear rates, but there are a broad range
of design parameters to balance in PCP design, such as mechanical and volumetric
efficiency. Nelik puts forward a more detailed discussion of how pump geometry and
operation drive pump life [38]. Vetter also provides a detailed discussion of PCP wear
mechanisms, and performs accelerated life testing for predicting pump life [55].

36
1.5.4 Water Applications

At the time of this writing, there were at least 2 different PCP solutions for solar
powered irrigation. According to Holthaus, Kunen et al., Sunculture sold a 500 USD
PCP called the 'RainMaker', and Davis & Shirtliff sold a 1,500 USD PCP called the
D3 Solar, made by DAYLIFF [23, 30]. Given the non-quadratic shape of the D3 Solar
pump curve, it is likely that it uses an elastomeric stator. Holthaus presented two
case studies showing the long-term economic feasibility of solar powered irrigation.
Both authors still note that the high capital costs are the main reason why farmers
have not adopted the technology. Loans are often difficult to acquire for small plot
farmers. For small plot farmers who do have access to financing options, the financial
risks associated with large loans still deter adoption.
Ward et al. also noted that PCPs have considerable potential for efficiently lifting
water with solar power in 1987 [60, 59]. The pump, motor, and electronics that they
tested cost between $900 and $4,000 in 1987.
The Life Pump by Design Outreach is another interesting application of PCPs to
rural water delivery. The Life Pump is a manually operated PCP with an elastomeric
stator [6]. According to Bixler, PCPs have an inherent advantage over other hand-
pump technology. The simplicity of the rotor and stator mean that maintenance are
requirements are low. Additionally, PCPs deliver water continuously, whereas other
pumps deliver water in pulses. With continuous flow, less strong users have an easier
time pumping water. One full installation of their system costs 10,000 USD, including
spare parts and training.
Vetter performed a tribological and systems-level study to evaluate the feasibility
of PCPs for solar powered water pumping from deep wells [54]. They particularly
noted the advantage that PCPs have when operating across a wide range of pressures.

37
38
Chapter 2

Lubrication Theory and Geometric

Errors

Stokes flow for Newtonian fluids is a well understood and repeatable behavior. Given
that many PCPs operate with high viscosity fluids, and low Re flows, laminar lubri-
cation theory is a promising tool for understanding internal PCP flows. Especially in
the context of assessing how manufacturing errors and wear can affect performance.
This would enable more cost effective design and manufacturing choices and the ra-
tional setting of tolerances. Andrade et al. developed a powerful 2D simulation for
these types of flow, with only minor errors from experimental data [3]. This type of
model has also been implemented by Li et al. for single screw extruders [35]. This
chapter extends their work to account to variations in the surface of the rotor and
stator; rotor error motion; and uses these new modeling capabilities to better explain
internal PCP behavior.

2.1 Lubrication Theory Model

2.1.1 Derivation and Clarifications

Before discussing modifications to the original theory, this section presents an abbre-
viated form of Andrade's derivation below for completeness [3]. Beginning with the

39
full Navier-Stokes and continuity equations in cylindrical coordinates:

l Orv 1w OU0
(2.1)
r Or r 0 Oz
Du Du Ou w ou' OP +i[1 a rOU) 1 02U 02U-
p +u- +v- + =p
at az Or r 0 Oz r Or Or r2 002 OZ2

(2.2)
(Dv Dv Dv wov w2 '\ Op a(I Orv"\ 1 0 2v 2 Ow 0 2V
p- +U +V- +---=
at Oz Or rOO r0 Or Or r Or r2 02 r2 00 Oz 2

(2.3)
(Ow Ow ow wow vw 1 p [ 0 (1Orw" 1 92 w 2 0v D2 W
t Oz Or r 0 J rO Or r Or r2 002 r20 Oz2

(2.4)

where u is the flow in the axial direction, v is the flow in the radial direction, and
w is the flow in the tangential direction. In lubrication flows, the fluid velocity in
the radial direction is much less than in the axial or tangential direction (v << w,
u). Similarly, velocity gradients in the radial direction are much greater than the
velocity gradients in the axial or tangential directions (i.e. >> 2, 2). Andrade
further assumes that the material derivative terms are small, since Re is small for the
flows investigated in their work. With these assumptions, the Navier-Stokes equations
simplify to:

Op [10 Ou'~
0= + P1a (ra (2.5)
Oz r Or Or
Op
0 = (2.6)
Or
10=
0
1 +IO(rw) r _ (2.7)
rO00 .Or Lr OTrJ

These equations can be integrated to find analytical solutions for u and w. Equa-
tion 2.9 is updated based on a typo in Andrade's original work [3].

40
 R (G) 0

IV

Figure 2-1: Different zones of the stator relative to the center of the rotor, adapted
from Andrade et al. [3].

u= (9 - pg) + c lnr+ c2 (2.8)

(az 4y
r ap /
=-lnr--
1)
+
c3 r c4
(2.9)
2 r
-

2p62

A general depiction of the critical geometric variables for a PCP is shown in Fig. 2-
2. The geometry of the stator is defined by R0 , which is the distance between the
center of the rotor and the stator wall. For any cross-section that is orthogonal to the
z-axis, there are four different regimes for R,. Those regions are defined in Table 2.1
by terms computed in Eqs. 2.10 through 2.17, and are visualized in Fig. 2-1.

41
 Region 0 Limits Ro(0, z)
I A' < 0 < A (2E - dcsR) cos(0 + e)) + 2 - (2E - dcsR) 2 sin(O + e 8 )2
II B < 0 < B' -(2E + dcsR) cos(0 + E)) + R2 - (2E dcsR) 2 sin(0 + E8 ,)2
III A <0< B R
sin(0+6,)

IV B'< 0 < A' -+s

Table 2.1: RO values in different regions of the PCP. These are updated to correct for
some typos in Andrade's original work [3].

s (2.10)

dcsR = 2E cos(Qt - 19,) (2.11)

Rs
a, = arctan (2E R) (2.12)

a2 = arctan (2 (2.13)
2E+ dcSR

A = a - , (2.14)

A' = 27r - ee - ES (2.15)

B = 7r - a2 - E3 (2.16)

B' = 7r + a2 - 93 (2.17)

E8 is the angle between the stator section's line of symmetry and the horizontal
plane. Alternatively, E) is the twist angle of the stator helix at an axial coordinate,
z. P, is the pitch of the rotor. dcsR is the distance between the centers of the rotor
and stator cross-sections, R, is the radius of the stator, and ai and a2 are half the
angular span of the two cavities.
is also needed to solve the final Poisson equation, Eq. 2.30, and can be ana-
lytically evaluated for accuracy and efficiency. -, is analytically expressed in Table
2.2.
The velocity of the fluid on the surface of the stator and stator is subject to the

42
 PS
4E

Cl W RS

Pr

Figure 2-2: The critical dimensions of a PCP viewed from the side (left) and outlet
(right).

Region 0 Limits ,(0, z)

2
I A' < 0 < A -(2E - dcSR) sin(O + s) - (2E-dCSR) sin(O+E) cos(e+e,)
2
\R8-(2E-dcsR) 2 sin(O+E),)

II B < 0 < B' (2E + dcsR) sin(O + E) 2

_ (2E+dcsR) sin(O+E),) cos(+Es)
2
VR2-(2E+dcsR)2 sin(6+E),)
-R, cos(O+e,
III A < 0 < B
)

sin 2 (0
)

R, cos(0+80)
IV B'< 0 < A' smi(O+E)

Table 2.2: %1Ro values in different regions of the PCP. These are updated to correct
for some typos in Andrade's original work [3].

43
no-slip condition. Therefore, the boundary condition for the fluid velocity at the
stator wall is the relative velocity of the stator to the center of the rotor, as can be
seen in Eqs. 2.18 through 2.20. The velocity at the exterior of the rotor is the relative
velocity of the rotor surface compared to the center of the rotor cross-section, as can
be seen in Eqs. 2.21 through 2.23. Those boundary conditions are combined with
Eqs. 2.8 and 2.9 to find Eqs. 2.24 and 2.25. Equation 2.25 is updated from Andrade's
original work to address a typo in bracket placement [3].

u(Ro) = 0 (2.18)

v(Ro) 2EQ sin(Qt - E,) cos(O + 8 ,) (2.19)

w(R,) -2EQ sin(Qt - E,) sin(6 + ,) (2.20)

u(Rr) 0 (2.21)

v(Rr) 0 (2.22)

w(Rr) QRr (2.23)

S2 (R_
2
1
/11\R{ 2 I R
-

S= g- -- )+ ln (2.24)
Oz 4p Rr K~RrJ In Rr,

W OP Rr ln(r) - -+ K - ln(R) - + K]
00 2p R, 2 r L 2 . (2.25)
[w(Ro)Ro - QR2 r -Rr

0 ~ r +2 R 7

R2(ln Rr - 1/2) - R 2(ln Ro - 1/2) (2.26)

R 2- R2

With flow velocities calculated, they can be substituted into the integral form
of the continuity equation for the region between the rotor and stator to give the
following expression:

44
 Ow+ O(ru)
0 = dr (2.27)
JR 0
I Or V) 00 O
Rr Ro
D(TV) dr OW d
0-z] (2.28)
JRRo

+
Or 00 r Oz d
r Rr

Combined with the boundary conditions in Eqs. 2.18 through 2.23, the continuity
equation is:

fR,,
rudr + a Ro wdr -w(Ro)
w( - Rov(Ro) (2.29)
OZR o 0JRR~( 0

By evaluating the integrals in Eq. 2.29 with Eqs. 2.8 and 2.9 Andrade et al. derived
Eq. 2.30 a Poisson equation of similar structure to the classic Reynolds lubrication
equation.

00 (C1l OCow O(PzC 2

+ w(Ro) - Rov(Ro) (2.30)
)

C2 OP) +Oz 0
Oz 00 00

The coefficients for Eq. 2.30 are Eqs. 2.31 through 2.33. These coefficients are
presented as updates from Andrade's original paper with typo corrections [3].

1R
2Rr 0r
2
InRo - R lnR - (R - R 2)(1 - K)] - R,(ln Rr - 1/2
r 0 rr + K) In(Ro/ Rr) }
C, R (2.31)
2 -- 2 t
C
R2- Rr
R4 -R
2
4
+ RoRr )2
-1
ln( Ro/ Rr)
[ ln(Ro/Rr)
R2 - R ]
02 r
-

}
(2.32)
w(Ro)Ro - QR2 R2 - R R ln(Ro/Rr) + QR ln(Ro/Rr) (2.33)
Cow = R 20 - R2r 2
R(ln R. - 1/2) - R 2(ln Ro - 1/2)
(2.34)

Once the static pressure, p, is computed along the rotor, the axial velocity can be

45
integrated across the outlet to determine the flow-rate of the working fluid.

27r RO
Q() = ru(r, 0, z)drdO (2.35)
0 R

,
Which can be computed efficiently and accurately in Eq. (2.36) using C2 from the
above analysis [3].

Q(z) =j C2 ( -pg) dO (2.36)

2.1.2 Numerical Implementation

The numerical implementation of the lubrication theory model in this thesis has a
more compact handling of boundary conditions than Andrade's original work, and
accelerates matrix creation in MatLab through vectorization [3]. Due to the structure
of the Poisson equation, the integration coefficients can be computed on a staggered
grid, illustrated in Fig. 2-3, for a second order numerical method to conserve mass.
Figure 2-4 compares the mass conservation properties for a centered and staggered
grid. The superior mass conservation properties of the staggered grid verifies the
necessity of the method.

Because C1, K, and CO, are differentiated with respect to 0, they are evaluated
at 0 + dO/2. C2, being differentiated with respect to z, is evaluated at z + dz/2.
This holds for the backwards difference estimation of the first derivative. If forward
differences are used, then the dO/2 and dz/2 should be negative to keep the derivative
centered on [0, z]. This results in three different staggered grids.

The following finite difference estimations were used in generating the numerical
solution to Eq. 2.30.

46
 (e,z+dz)
k A

4 ...

dz/2
A 19

(e,z) (e+dE),z)
dO/2

Figure 2-3: A representation of the numerical grid used to evaluate the Poisson
pressure equation. Each node (O,z) corresponds to a point where the pressure is
evaluated.

a
/((
COP)
=p
[[
[C1 ][RDo][p] (2.37)

a
C2 = [FDz] [C21][BDz] [p] (2.38)
)

Oz
aCow
= [BDo] [Cow] (2.39)
&(pgC2
= [BDz] [C 2 ] (2.40)
)

49Z

[FDo], [BDo], [FDz], and [BDz] are forward or backwards finite difference matri-
ces for their respective coordinate direction. As mentioned above, the constants C1,
C 2 , and Cow need to be evaluated on a grid that is staggered from the pressure vari-
able to ensure that the finite differences for pressure and the coefficients are centered
on the same point, conserving mass.

The results from this model closely match the results found by Andrade. In Fig. 2-

47
 Centered Grid Staggered Grid
0.8- 0.7508_

0.79
0.7506
0.78-
0.77- 0.7504-

0.76
0.7502
CS

-
0.75
0.74- 0.75-

0.73
-

0.7498
0.72
-

n 71 , , , i 0.7496 II . I 1I

-
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
z/L [-] z/L [-]

Figure 2-4: Net flow rate at different sections as a measure of conservation of mass
for a centered grid (left) and a staggered grid (right)

5, the flow rate on a grid with 500 nodes in z and 250 nodes in 0 at t = 0 show a
1% difference in flow-rate between the models. This model predicts 3.18 m 3 /s and
Andrade's model predicts 3.13 m3 /s. This difference, and the difference in pressure
fields can be caused by small differences in how the flow velocity is integrated, the
numerical scheme implemented, or how the grid is set up. Due to the vectorization of
matrix creation and narrower bandwidth finite difference matrix fron a more compact
handling of the boundary conditions, the system solves considerably faster than the
original implementation by Andrade et al. Implementation details are included in the
accompanying code.
Evaluating the flow-rate on 3 grids: [N,,No] = [200,40], [400, 80], and [800,160]
and applying Eq. 2.41, finds that the order of convergence, pc, is approximately 1.3
[33]. Note that the lubrication model implemented for this thesis shows a nonlinear
convergence behavior in Fig. 2-16.

Pc ~ In U2Z 17 /ln(2) (2.41)

2.2 Modeling Geometric Error in PCPs

Wear, manufacturing tolerances, deformation, and bounded rigid-body motion (i.e.

backlash) are present in any real machine. Many authors have noted the presence

48
 1 AP AndradeAPSimon [Pal
1 6000

0.8
4000

0.6 2000

0
0.4
-2000

0.2 -4000

-6000
0
1 2 3 4 5 6
9 [rad]

Figure 2-5: The pressure difference between Andrade's results and the model in this
paper on a grid where n, = 500 and no = 250.

of these effects in PCPs, and a few have attempted to study them. Vetter at al.

have done experimental work to better understand the effect of clearance gap size on

the performance of a PCP. Their experimental work was coupled with a tribological
study to understand how wear can affect the life-time performance of a PCP [54, 57].
Due to a lack of more sophisticated numerical modeling tools at the time, Vetter's
model uses the simplifying assumption that the wear results in a uniform increase in
clearance gap size, w, even though the tribological study indicated that rotor wear

was non-uniform [54, 57].

This section presents a new way to use lubrication theory to analyze the effect of
non-uniform wear, manufacturing tolerances, deformation, and bounded rigid-body
motion on PCP performance. A new version of the PCP lubrication equation is

presented which can account for spatial variations in the rotor radius, R,. This new
Poisson equation is then implemented with different deviations from 'ideal' rotor and

stator geometry.

49
2.2.1 Imperfect Rotor and Stator Model

No surface is perfect, and the lubrication theory model derived by Andrade is extended
in this thesis to account for imperfections in the surface of the rotor or stator. Notably,
spatial variations in R, require the conservation equation to be modified to include
a -w(R,) ! on the right hand side. This gives Eq 2.42, the Poisson equation for
pressure within a PCP that can handle spatial variations in Rr and R,.

&/Op N& _ aC0 w &(pgC2 ) ___ ___

C1) + C2- - + +w(Ro) 0 - Rov(Ro) -w(Rr)

(2.42)

Because the location of the rotor surface, RO, is already treated as variable in
6 and z, there is no need to modify the Poisson equation further to include stator
surface error. Instead, the deviations just need to be directly added to the theoretical
Ro values and its derivatives.

2.2.2 Rotor Surface Error

This new model can simulate the effect of rotor surface imperfections on the per-
formance of the pump. Figure 2-6 shows one example of sinusoidal surface error,
expressed in Eq. 2.43.

n6 rmz
err, (0, z) = Ae sin(---) sin( ) (2.43)
2 1

Where err,(6, z) is the surface error perturbation on R,, A, is the amplitude of

the error, and n and m are the spatial frequencies of the error in 0 and z respectively.
Figure 2-8 shows that not only does the amplitude of the surface error affect the
performance of the pump, but the spatial distribution of that error also has an effect
on performance. This is especially the case when the error amplitude is comparable
to the sealing gap size. This means that, even when solving for lubrication flow in
transient high viscosity fluids, there is non-trivial dependence, not just on the average
size of the clearance gap, but also on how that gap varies spatially.

50
 Rotor Error Distribution (n=m=2) [im) Surface Error [m]
0.35
150

0.3
100

0.25
50

- 020
0
N

-
0.15
-50

0.1
-100
0.05
-150

0 104
1 2 3 4 5 6
Theta [rad]

Figure 2-6: An example of sinusoidal surface error on the rotor. (A, = 185 pm.
n = m= 1)

2.2.3 Rotor Error Motion

This new surface error model can also predict the effects of solid body rotor error
motion on PCP performance. Given the forces acting on the rotor, it is unlikely that
an unconstrained rotor would sit perfectly centered in the stator cavity. Given that
the pressure-driven slip flow approximately scales with the clearance to the third
power, this rotor motion can significantly affect PCP performance. Berton et al.
have made note of this fact and implemented a transient CFD model which allows
for the rotor to move in response to fluid forces [5]. They claim that they were
able to accurately reproduce results from experimental data, but do not present a
quantitatively repeatable procedure.

Equation 2.44 maps solid body rotor surface error corresponds to rotor motion.
A 1 and A 2 both represent the error magnitude from displacement in the x-y plane
and skew respectively. 71 and 2 both represent the direction of the displacement and
skew in the x-y plane.

errm(O, z) = A 1 cos(6 - 71) + A 2 cos(O - '72) 2z- 1 (2.44)

51
 I PITCH ...A

ou r

TRANSLATION

Figure 2-7: Geometric definitions of the two motion-driven error modes.

Under this formulation. the skew is constrained by contact between the rotor and
stator through Eq. 2.45.

w > [A1 + A 2 Icos(7Y1 - _y2)1]2 + [A 2 sin(71 - y2)] 2 (2.45)

The range of possible PCP performances can now be computed by minimizing and
maximizing the total flow rate with rigid rotor motion as the constraint. Equation
2.46 is the maximization problem, and Eq. 2.47 is the minimization problem. Both
are readily solved using a non-linear optimization function, such as a pattern search
or a multi-start gradient method [53, 21], to find the rotor positions corresponding to
maximum and minimum flow-rates. Under many operating conditions, the maximum
flow rate corresponds to the rotor sitting in the center of the cavity. However, in
low pressure rise, high viscosity applications, the maximum flow-rate can occur with
an uncentered rotor. In this case, the lubrication-theory model predicts that it is
possible for the flow-rate of a PCP to exceed the 'theoretical flow-rate,' as it does
in Fig. 2-13. Figure 2-11 show how these maximum and minimum flow pump-curves
encapsulate the results for Gamboa's experiments with oil [18]. These results provide
new insights into how PCPs work, and are designed.

52
 Effect of Rotor Surface Error on Flow
0.75

-
---1 mode
--2 modes
3 modes
4 modes
' s --- 5 modes
-- 6 modes
---7 modes
-- 8 modes
-- 9 modes
0.7 _
,10 modes,

n ar-

0 0.05 0.1 0.15

Surface Error Wave Amplitude [mm]

Figure 2-8: Time averaged flow rates for different surface error amplitudes and spatial
frequencies (n-m=1 to n=m=10). The mesh was created with 1400 nodes in z and
400 nodes in 0 to resolve the spatial frequencies.

Figure 2-9: A two dimensional depiction of rotor error motion.

max Q
(2.46)
s.t. w > V[A1 + A 2 1cos(7Y - )]2 + [A 2 sin(71 -_72)]2

And the mirrored minimization problem

min Q
(2.47)
s.t. w > l[A 1 + A 2 1cos('yi - Y2) ]2 + [A 2 sin(71 -_72)]2

53
 Rotor Error Distribution (n=m=2) [pm] Rotor Radius (Rr) [m] 0.02004
0.35

-
100 00.02002
0.3 -- 0.02
0.01 998
50 0.25
0.01996
0.2- 0.01994
0.01992
0 0.15
0.0199
-50 0.1 0.01988

0.01986
0.05
-100 0.01984

0
1 2 3 4 5 6
Theta [rad]

Figure 2-10: An example of rotor radius changes corresponding to small angle rotor
motion. (A 1 = 83 pm, A 2 = 83 pm, Yi = 0, 72 = ir/2)

The rotor sits in an off-center position inside of the stator cavity. This small

displacement and skew can have a significant effect on the performance of the pump.
The only other model that explicitly accounts for 'undesired' motion of the rotor
within the stator involved full 3D CFD fluid-structure interaction simulation [5]. It
was an industry paper which lacked all of the information necessary to recreate the
author's results. Motion on the order of 100 pm seems minuscule in a ~0.1 m radius
pump. Prior to this work, resolving the fluid-solid interaction, changing mesh, and
even smaller length-scales (1-10 pm where the rotor is close/in contact with the stator
wall) would significantly complicate the problem. The fast, semi-analytical, structure
of this lubrication theory model has enabled this deeper exploration of the PCP flow

problem.

Because deviation between experimental data and the lubrication theory model
are captured by rotor motion or surface error, it is plausible that this model captures
all relevant flow phenomena in the laminar regime. This an important finding because

flow within PCPs is not simple: it is three dimensional, transient, rotating, and occurs
across orders of magnitude in length-scales. It is not obvious that the lubrication

theory assumptions are sufficient to capture all of the relevant physics in this problem.

As seen in Figs. 2-11 and 2-12, rotor motion and error can account for >10%

54
 Performance of Model PCP with 0.042 Pa-s Oil
S100 rpm
0.5 200 rpm
0 0 0 300 rpm
0 0 0 400 rpm
0.4 - 0 -+ centered rotor
0.4 00 motion uncertainty

.3 -
0 .0.2 0 .00 0

0 200 400 600 800

AP [kPa]
Figure 2-11: Experimental results from Gamboa with 0.042 Pa-s oil compared with
Andrade's centered rotor lubrication theory model and the rotor error motion uncer-
tainty presented in this thesis [18, 3].

variations in flow rate, and as much as a 2x variation in slip-flow. This means that
PCP designers need to seriously consider the effect of rotor motion on performance.
The precision of a manufacturing process, and corresponding clearance gap, now
has a clear and quantifiable effect on pump performance, and the range of possible
performances. This source of variation needs to be considered in the design process.
When that variation is unacceptable, constraints can reduce the rotor's range of

motion within the stator. The design of those constraints is discussed later in this
thesis.

This analysis also provides motivation to re-think the meaning of the maximum
theoretical flow rate, Qth in Eq. 2.48, that is commonly used in literature. Note that
pitch is commonly presented in units of [m/rev], so a factor of 27r is included to make

units match with , which is in [rad/s]. Both experimental results and higher-order
modeling demonstrate that Eq. 2.48 is not a true limiting case.

55
I I

0.6- Performance of Model PCP with 0.134 Pa-s Oil

S100 rpm
S200 rpm
0.5
0 o0 0 -- centered rotor
0.4 - 0- q
0 0
F~~I 0 00n 0 .0
L--0.3 - * *- *

*
0. 0 0 00
0 0
0.2
-

0.1
-

0
0 200 400 600 800
AP [kPa]
Figure 2-12: Experimental results from Gamboa with 0.134 Pa-s oil compared with
Andrade's centered rotor lubrication theory model and the rotor error motion uncer-
tainty presented in this thesis [18, 3].

Qths = [8ERs + 7r (Ri - R )] (2.48)

-R2]27r

Comparison with Plastic Rotor and Stator Experiment

It has now been shown that the experimental data collected by Gamboa et al. [19]
is within the bounds of uncertainty on the rotor location. While this is strong evi-
dence that this lubrication theory model is applicable to low Re PCPs, an experiment
showing direct agreement would provide stronger evidence still. To reduce the uncer-
tainty on the motion of the rotor, I constructed a PCP constrained by hypocycloidal
bearings, as described in Section 4.3. Both the rotor and stator were measured, and
4 eddy current probes were used to track the position of each end of the rotor. The
surface error data from the scans were combined with the rotor trajectory to model
the rail bearing pump performance.

56
 0.481 Pa-s Oil Experiment Comparison
1.1

-
1.05-

-- - - - - - - - - - - - -

-
5 0.95

0.9
* Gamboa Data
0.85 o Skew Rotor Model
- Qthr
- -- ths
na I I I I I I . I I I I I I I I
0 200 400 600 800
A P [kPa]

Figure 2-13: Experimental results from Gamboa with 0.481 Pa-s oil [18]. The theo-
retical flow-rate for this pump, Qth, is 5.39 m 3 /s. Both the experimental data and
skewed-rotor model have flow-rates greater than Qth*

The experimental and lubrication theory results are shown together in Fig. 2-14.

The model shows excellent agreement with the experimental data. This result further

validates the efficacy of the lubrication-flow model for PCPs.

CFD Comparison Note

If rotor position is the cause of the error between the centered lubrication theory
model and the experimental results by Gamboa et al., then traditional CFD should
have the same error. However, Paladino's CFD study, shown in Fig. 2-15, shows
excellent agreement with the experimental results [44]. Because the rotor is centered
in the Paladino CFD model, this result contradicts the hypothesis, unless there is
another explanation.

It appears that the discrepancy between the CFD and lubrication theory models

is due to insufficient convergence in their CFD work. Figure 2-16 shows the conver-
gence of the lubrication theory and CFD models alongside the experimental flow rate.
Instead of refining the mesh until it reached an asymptote, it appears that Paladino

57
 Measured Pump Flow Rates
0.35- + 50 rpm
x 10 rpm
* 15( rpm
0.3- 0 20( rpm
+ 25( rpm
X 30 rpm
0.25-

0.2-

+
0

+
U. 0.15- 0
0

0.1
-

0.05-

U
7 IIII[I 111l~l~ 11liii 1111 1
0 100 200 300 400 500 600
A P [kPa]

Figure 2-14: Rail bearing experimental results and lubrication theory simulation
based on rotor and stator scans.

et al. refined the mesh until it agreed with experimental results [44]. They justified
this approach because the purpose of the study was to demonstrate the capability of
the CFD method to qualitatively describe the behavior of PCPs.
Given that shape of the CFD and lubrication theory convergence curves, it appears
that the Paladino CFD simulation is actually approaching the lubrication theory
model. The lubrication theory model is simply converging three orders of magnitude
faster than the CFD. This makes sense: with lubrication theory, one two dimensional
PDE need solving. With conventional CFD, a three dimensional PDE needs solving
for each of the three velocity directions.

2.2.4 Stator Surface Error

This surface error framework can also be applied to the stator surface. Because the
basic lubrication theory model already has a spatially variable R, adding error or
deformation-driven variations to R, has a simple effect on the computation of R,:
recalculate R, with the new distribution of R,. This spatial variations in the stator

58
 Paladino CFD PCP with 0.042 Pa-s 01
0.5

0.45 .*

*
0.4

*
0.35

0.3

0.251

*
0.2
*4

*
-
0.15

*
**
- **

*
0.1
**

0.05
*

0
0 200 400 600 800 1000 1200
Pressure [kPa]

Figure 2-15: Gamboa's experimental results compared with CFD simulation by Pal-
adino [19, 44].

must also be incorporated in OR,/0, and are included below, in table 2.2.4.

Region 0 Limits %(0, z)

R - 2 -(2E-dcSR )2 sin(O+E))cos(O S)
I A' < 0 < A

2
II B < 0 < B' (2E + dcsR) sin(0 + Os) + 9 -(2E+dcsR) si(n+s)cos(O+G 8

)
R R1-(2E+dcsR )2 sin(9+E s) 2
*

III A <0 < B aRs -1 R, cos(O+eE)

600 sin(O+e.) sin2_(_+e,,)

IV B' < 0 < A' aR, 1 R, cos(+e)

aO sin(O+8E) smn( +E)

Compliant Stator Note

-

By coupling the pressure field with the mechanics of the elastomer, it should be
possible to make a computationally efficient fluid-solid interaction model. Such a
model would have immediate applicability to elastomeric PCP design. Elastomeric
PCPs require a minimum amount of slip (stator expansion) to lubricate the sealing

59
*
 Convergence Plot
1.15
1.1 Of

1.05
1

0.95-
X0.9
.

0 0.85
-

0.8
0.75
-Paladino CFD
0.7 ' - : I~bcto~erlP=0.042
- - - Lubrication Theory 200MM-
Pa-s

0.65 AP =275.79 kPa

10 3
10 4 105 106
nodes

Figure 2-16: The sensitivity of 2D lubrication theory and 3D CFD simulations to grid
size. 2-D lubrication theory has a much faster convergence rate than conventional
CFD.

lines. Otherwise, the pump life will be reduced [38, 13]. Such a model would enable

PCP designers to balance volumetric efficiency with lubrication of the elastomer.

2.3 Internal Forces

2.3.1 Torque Scaling

With models for the flow and pressure field inside of PCPs, it is possible to approxi-
mate the forces acting on the rotor and stator. This is important for understanding

how shear losses scale with pump size and speed, and for understanding the expected
loading on any supports. Equation 2.51 is the simplest expression for the torque re-
quired to drive the rotor. This relationship can be found by starting with conservation
of energy, Eq. 2.49, and substituting Qthr7V for Q in Eq. 2.50.

60
 r QXP (2.49)

7 QtjVAP (2.50)

Qth= P
(2.51)
Q?7M

ij is the total efficiency, qv is the volumetric efficiency, and 7M is the mechanical

efficiency. Since Qth is a function of geometric parameters and rotor speed, as de-
scribed in Eq. 2.92, rotor torque can be expressed as a function of pressure, geometry,
and secondary effects (such as volumetric losses or friction) in Eq. (2.52).

T- 4ERrPsAP (2.52)

Flow rate and volumetric efficiency (,v) are notably missing from Eq. (2.52). This
allows for direct computation of mechanical efficiency from pressure rise and torque
alone. Even more information can be gleaned about the mechanical behavior of the
pump by expanding Eq. 2.52 to include the drag on the rotor and drive system, cAd;
and any parasitic torque associated with seals, linkages, or bearings, T,. The linearity
of these relationships can be seen experimentally and analytically in Fig. 2-18. In
this work, it is assumed that the force of drag scales linearly with surface speed
by a dimensional coefficient of drag, Cd, and that the parasitic torque is invariant
with speed or pressure. The remaining efficiency term, (1 - Cf), in the pressure
rise proportional power term, is referred to as the pump coefficient of friction. It is
described as the coefficient of friction because it is analogous to the role that friction
plays in computing pulley efficiency.

4ERrPsIXP
r + CdQ + o (2.53)
7r(1 - Cf)

The torque coefficients in Eq. 2.53 map onto terms in the shear tensor in Eq. 2.75.
There are viscosity and surface-speed independent terms, which are only driven by

61
 pressure rise and geometry. These terms determine the theoretical minimum value for
the pump coefficient of friction, Cf. That theoretical minimum value is added to any
mechanical losses that also scale with pressure. The terms in Eq. 2.75 also include
pressure rise independent shear forces, which are linearly proportional to speed and
viscosity, setting the theoretical minimum value for the dimensional pump coefficient
of drag, Cd. Given this relationship, it is likely that Cd is linearly proportional to
dynamic viscosity, p, but that assumption is excluded from Eq. 2.53 because there
may be more complicated mechanical interactions elsewhere in the PCP structural
loop.

For the bearing pump presented in Section 4.3, the pump coefficient of friction,
Cf, is 0.0514, the dimensional coefficient of drag, Cd, is 0.0286 [N-in s/rad], and the
parasitic torque, T 0 , is 0.125 N-m. Although the equations appear complicated, they
provide both accurate and understandable estimates of PCP performance for how
simple they actually are.

4ERrPs(
CQ ~ (2.54)
ir

CsL SL (2.55)
AP
CsT =ST (2.56)
AP
(cQQ c 8LZOP)L$P
-
r/L ( QQ P
CdQ T)Q (2.57)
(1C AP + CdQ + T,
)

7T = _ (2.58)
- QAP +c C+o T,

(2.59)

The volumetric efficiency in the numerator of Eqs. 2.57 and 2.58 is a well-known
and documented scaling law for PCPs. However, the introduction of a mechanical ef-
ficiency term offers more complete insight into PCP behavior, analogous to an affinity
law.

62

..........................
---,,,,,,,,,,"I
2.3.2 Direct Integration from the Lubrication Theory Model

Now that flow velocity can be analytically expressed in terms of known pressure
gradients, the velocity gradients can also be analytically computed for evaluating
strain-rate in the flow. The velocity gradients used to compute the shear stresses are:

9n )2 r Roa' 2)n R

)
9z Oz ) .4
(2.60)

+ Z2 4,
RJ 1 - (n ) [&-)

R,
In
00
aO Pg
op R2
Oz )4pi
- in
r
R
R

-_
_ 2
1 n
2(R
Rn,-
__2I

In (
}} (2.61)

+ p R2 1
OO~z 4pi Rr kn RrJ
Rr

( I-2r R,
(2.62)
-

pg Oz J4/1 r rIn( )R,

OW 02 pRr +OpRK (r RJ
Inkr)-I+K R2 n - 1-+K]
Oz &OOz 2/- R,2 20 0 2p oz Rr r

+ (r2 R
(R - R2
2
~
Ro + w(Ro) + 2Ro [w(Ro)Ro - QR 2]
-Rr(R -R5)2

(2.63)

(W

ao0
2p[n(r)
0022/,i R
- + K] -- [ln(R) - +K + (# -%Rr

+ (r2 - R { (R 2i - R 2) aw(o) Ro +
)r(
0w(Ro) + 2Ro%- o[w(Ro)Ro - QR 2]
R -R) 2

(2.64)

63
 Ow
-w = _OpR
- ,r, 1 - [-+
- -ln(r) 1
K + ln(R,) -1 +K
R
Or 00 2[ f Rr 2 r2L1 2
(2.65)
Row(Ro) - QRr 1+ (R)2 R) 2

+
0 r

rOr
a w) =-- 9P 1 2Rr ln(Rr) + -]
1+

+
Or (r) 8o 2

-
r2 I 2 (2.66)

-
w(Ro)Ro - QR Rr 2
- 2()
R0 - R2 r r

)
OK _ORO ~2Ro log(Ro) 2log(R,) - 1]
(2.67)
(90 ao R2 R+R 2Ro _
OK ORO 2Ro log(Ro) 2 log(Ro) - 1
R2- 2 (2.68)
Oz Oz 0 +Rr RO

Region 0 Limits (0, z)

a9z

I A' < 0 < A - (2E - dcsR) sin(0 + 8) - g sin(Qt - E)) cos(O + 8

)
+ ir sin(0+e8 )(2E-dCSR) [2E sin(o+E ) sin(t-Gs)-(2E-dcSR) cos(O+E),)]
Pr VR2-(2E-dCSR)2 sin(O+E8
)

II B < 0 < B' 7(r2E + dCSR) sin(0 + 8 s) - 2E sin(Qt - Theta) cos(6 + 8 ,)

7r sin(0+eS)(2E+dcSR)[2E sin(O+E 8,) sin(t-eS )+(2E+dcSR)cos(O+Gs)j

Pr VR2-(2E+dcsR )2 sin(0+,)

-rR, cos(O+GE)
III A <0<B Pr sin2 (0+E
)

7rR, cos(O+9E)
IV B' < 0 < A' Pr sin2 (O+E
)

These equations can be further simplified for the purpose of calculating shear
forces on the rotor. Op/Or is negligible in lubrication flows. The spatial velocity
gradients are evaluated at r = R, giving the following terms:

OUr=R, = 0 (2.69)
Oz
64

. ........
..
 an l=R, = 0 (2.70)

Ou Op Rr R -R2 2

or |r 9p - -2
o0l( (2.71)
0 z 4/p R2 In (-

Ow
z=, -- =0 (2.72)
am
O r=Rr = 0 (2.73)
00

0 (w R Op K + log (Rr) Row(Ro) - QR2

+2
)=- r- -2Q
(2.74) r- =
Or r 0 p R 2- R2

In cylindrical coordinates, the shear stress tensor on the surface of the rotor re-
duces to Eq. (2.75). One interesting note is that multiple matrix entries in Eq. 2.75
contain terms that are inversely proportional to dynamic viscosity, canceling out the
dynamic viscosity term external to the matrix in Eq. 2.75. This has an interesting
theoretical consequence: there is a minimum, viscosity-independent, shear force on
the pump walls that is driven by pressure gradients. These viscosity-independent
shear forces are analogous to the wall forces in Poiseuille, pressure-driven pipe, flow.
In Poiseuille flow, a parabolic pressure profile results from a pressure gradient which
is opposed by viscous forces. Greater viscosity decreases the size of the parabolic flow
profile, decreasing the shear strain-rate at the wall. The decrease in shear strain-
rate is proportional to an increase in dynamic viscosity, which ensures that pressure-
gradients set the wall force instead of viscosity. Of course, viscosities still influence
pressure gradients in these flows, which in turn have an effect on the wall stresses.

0 rjQ()
-

TcyI= ry (E) 0 0 (2.75)

1-- 0 0

65
 The rotation matrix for converting from cylindrical to cartesian coordinates, R,
is Eq. 2.76.

cos(-O - Qt) sin(-O - Qt) 0

- sin(-O - Qt) cos(-6 - Qt) 0 (2.76)

0 0 1

At each axial coordinate, z, along the length of the rotor, there is a reference
frame velocity which is used to simplify the computation of the pressure field. The
z-gradient in reference frame velocity contributes to the viscous shear stresses in the
fluid. From Andrade et al., the velocity of the reference frames are expressed in
Eq. 2.77.

Vf = [-2EQ sin(Qt - E),) cos(E),)]& + [2EQ sin(Qt - 6,) sin(E),)]Q + 0 (2.77)

aVf 2EQw'r
O = Pr [cos(Qt 8) cos(O,) + sin(Qt - 19,) sin(19,)]2
%
-

(2.78)
( 2EQ7r
[Cos (Qt - 18 ) sin(O,) - sin(Qt - 8 ) cos(1 8 )]Q + 0
-P)

The resulting shear stresses induced by the relative velocities of the reference
frames is Eq. 2.79.

0 0 a&z
Vk

Tf = P 0 0 (2.79)

L. f v/ L
z 0

The viscous shear tensor, 7, in cartesian coordinates is then Eq. 2.80.

T = Rig1 RT + Tf (2.80)

The surface normal vectors to the surface are calculated as the cross product of

66
the tangent and bi-tangent [32]. The surface of the PCP rotor is parameterized in
both 0 and z. If the vector pointing to the surface coordinate is:

r= S(z, 0)
= [2E cos(Qt - E)) cos(O,) + Rr cos(6 + Qt)] (2.81)
+ [-2E cos(Qt - 9,) sin(E 8 ) + Rr sin(0 + Qt)] 9 + z

aS 4EKw
az = E~ r[sin (Q t - E)) co s( O ) cos(Qt - E) sin(ES )] 2
PS

-
(2.82)
4EKr
+ cos(Qt - OS) cos(ES )] p
Es[s i n (Qt - E ,)s i n ( ) ,)

+
-
S
as
Rr cos(0 + Qt)p + o0 (2.83)
00 = Rr sin(0 +F Qt)i +

Thus, the surface normal is specified as:

as as (2.84)
ao az

dA = dOdzRr (2.85)

With an explicit expression for the traction vector, the instantaneous surface
velocity of the rotor is needed to compute the power. Equation 2.86 computes the
velocity of the rotor surface in the absolute reference frame, V'r. The original derivation
of the kinematics are presented by Andrade et al. [3].

r = -[2EQ sin(Qt - 6,) cos(E8 + RrQ sin(0)]i (2.86)

)

+ [2EQ sin(Qt - E)) sin(E + RQ cos(0)] + Oi

)

The power input to the system through the rotor, Prot, can now be determined
with Eq. 2.88.

T = Tvisc - PI (2.87)

Prot = -1 dA (2.88)
)

67
 Torque can be computed about the rotor's center of rotation, Sc, with Eq. 2.90

Scr = 2E cos(Qt - - 2E sin(Qt - e,) + 02 (2.89)

T= (S-Sc) x I() dA (2.90)

Effective Turbulent Viscosity

Effective turbulent viscosity, or eddy viscosity, is a concept that comes from the
Reynolds Averaged Navier Stokes (RANS) equations [33]. Turbulent viscosity is used
as a proxy for convective transport of momentum that results from spatial velocity
fluctuations. Eq. 2.91, can be used to make an alternative estimate of mechani-
cal losses with turbulent flow to the laminar lubrication model alone. The effective
turbulent viscosity is computed from the ratio of the predicted slip velocity in the
laminar model, SL, and the actual turbulent slip flow, ST, which can be estimated
with the network model in Section 3, or directly from experimentation. Turbulence is
a complicated phenomenon, which behaves differently at different length-scales. It is
inaccurate to assume that turbulent diffusion of momentum is homogenous through-
out the flow. However, by conservation of momentum, the shear force transferred
between layers of flow should be very similar. This is an analogous argument to the
Poiseuille flow comparison with Eq. 2.75. This means that effective turbulent viscos-
ity should give a better order-of-magnitude for viscous losses than normal dynamic
viscosity alone.

At= SL (2.91)
ST

Compliant Rotor Note

It is possible to couple these rotor forces with a structural model for the rotor to
efficiently solve fluid-solid interactions. This type of simulation is important for de-
signing rotors out of new materials, such as injection molded plastics, for use with

68
constraints, such as the bearings described in Chapter 4. A supported plastic rotor
can eliminate rotor-stator contact in an unpressurized environment. However, rotor
deflection under pressure will affect clearance gaps, and may bring the rotor and sta-
tor into contact. Furthermore, this fluid-solid interaction simulation would be able
to model the compression of an elastomeric rotor[29].

Experiments

To the author's understanding, Vetter et al. have presented the only experimental
data in a peer-reviewed journal for rigid rotor and stator PCP mechanical losses prior
to this work [57]. The direct shear loss calculations are compared with their mechan-
ical losses using an effective turbulent viscosity computed from Vetter's volumetric
efficiency. Effective turbulent viscosity is a good proxy for the wall shear stresses
that would result in mechanical losses. The numerical and experimental results are
compared in Fig. 2-17. There is a clear difference between the experimental and the-
oretical mechanical efficiencies. This is likely due to a combination of two effects:
inaccuracies stemming from the use of an effective turbulent viscosity, or mechanical
losses between the rotor and stator. Due to the large magnitude of this difference, the
authors believe that most of that difference is due to contact between the rotor and
stator. Effective turbulent viscosity should give a close estimate, and the observed
difference is much larger than that, indicating the presence of another mechanism for
mechanical losses.
Paladino's torque calculations for water and oil indicate that there are greater
viscous shear losses in water than oil, as can be seen in Fig. 2-18 alongside results
from the lubrication theory model, which does draw their results into question [44].
This has also been noted by Berton et al. [5].

2.4 Implications for Qth

Section 2.2.3 raised concerns about the validity of Eq. 2.48, the equation for Qth
that is commonly used. Figure 2-13 shows that the experimental results by Gamboa,

69
 1 Estimated PCP Efficiency
1 ------ --------
------------

0.8 X-

0.6-

a)

= .4
u.5
-

Total
0.2- --- Volumetric
Mechanical
- * Vetter Exp. Volumetric
X Vetter Exp. Mechanical
0 ' , I I I I I I I I I I I I I I
0 200 400 600 800 1000 1200
Rotor Speed [rad/s]

Figure 2-17: Comparison between Vetter's experiment and lubrication theory with a
corresponding turbulent viscosity.

which exceed Qth, are not necessarily anomalous, and have a plausible explanation.
Table 2.3 shows Gamboa's experimental results alongside centered rotor and skewed
rotor lubrication theory model results.

4
Qthr = -ERrPsQ (2.92)
7r

In both the 0.042 Pa-s minimal flow condition, and the 0.481 Pa-s maximal flow
condition, the rotor is off center. Additionally, the minimal flow rotor position for
the 0.042 Pa-s experiment under-predicts the experimental flow rate. This means
that the position of the rotor inside of the stator is not dictated by maximizing or

minimizing flow. It is more plausible that the forces on the rotor sets the position of
the rotor, which may not correspond to a maximal or minimal flow condition.

There is considerable difference between Eq. 2.48, the commonly used equation for
theoretical maximum flow-rate, and the flow rate at high viscosities and low flow rates

(i.e. case 1, 4, and 5). Qthr, predicted by Eq. 2.92 by contrast, matches the maximum
flow rate of the lubrication theory model to within 1%. Since the lubrication-theory
model for a perfectly centered rotor with high viscosity fluid has closer agreement to

70
 '1

Simulation Torque Experimental Bearing Pump Torque

3- 45-
+ 50rm
-- 50 rpm * 100 rpm
2.5- - 100 rpm * 150 rpm
-150 rpm 3.5- 0 200 rpm
- 200 rpm + 250 rpm
2-
- 250 rpm
3- x 300 rpm
- 300 rpm,
E
z
-

1.5- Z2-
12.5-
0.5-

0.5-
0.5-

0 50 100 150 200 250 0 50 100 150 200 250

AP [kPa AP [kPa]
Torque Comparison
14
A Paladino CFD (0.042 Pa-s)
A Paladino CFD (0.001 Pa-s)
12 Bounding Case (0.042 pa-s)
Bounding Case (0 Pa-s)

10
-

0
0 100 200 300 400 500 600 700 800 900
P [kPa]

Figure 2-18: Lubrication theory results compared with the Paladino CFD for the
same geometry (botton)[44]. Simulated (left) and experimental (right) torques from
this work.

Eq. 2.92 than Eq. 2.48, the author believes that Eq. 2.92 should be used to estimate the
theoretical flow-rate for PCPs. Since neither Qth, or Qth, establish a true maximum

possible flow rate, the fact that Eq. 2.92 is more parsimonious than Eq. 2.48 is another
mark in its favor. A physical argument for this preference is that the area swept by
the rotor is a better measure of flow-rate than the volume of the cavities. While much
of contemporary literature uses Eq. 2.48, Nelik also uses Eq. 2.92 [38].

While this delineation may seem inconsequential given the 2.5% difference, it has
been shown experimentally and theoretically that greater than Qth, flow rates are
possible. It is important to have consistent and as specific as possible definitions for

71
 1 2 3 4 5
w [rpm] 400 400 400 400 400
p [Pa-s] 1 0.042 0.042 0.481 0.481
Rotor Pos. Centered Centered Min Q Centered Max Q
AP [kPa] 213 213 213 31 31
Qth,[l/s] 0.539 0.539 0.539 0.539 0.539
Qthr[l/S] 0.515 0.515 0.515 0.515 0.515

Qm [l/s] 0.510 0.470 0.417 0.511 0.543

Qexp[l/s] - 0.4368 0.4368 0.570 0.570
Pout[W 109 100 89.2 15.9 16.8
Pin[W] 180 108 117 52.7 55.0

Table 2.3: Simulation runs to illustrate flow and rotor dynamics.

a term like Qth, which is critically important to early-stage design and linear-network
models. Since these greater than Qth, flow rates are seen at high ratios of pQ/AP,
it is likely a consequence of the rotor wobble sweeping an extra-large cavity area,

without losing too much back-flow from the wider sealing lines.

2.5 Conclusions

Expanding lubrication theory tools for simulating PCP performance has important
implications for design insights and quantifying their embodiments. The three most
important insights from the model presented in this chapter are: the role of rotor
speed; the sensitivity of PCPs to geometric imperfections; and the role and presence
of rotor motion.

Rotor speed, Q, is clearly an important design variable for sizing PCPs. Q also has
a clear, but less obvious effect on pump efficiency. First it is important to recognize
that a PCP's resistance to slip flow, Q, is very insensitive to Q for a given PCP
design, but the forward flow, Qth, is directly proportional to Q. This means that for
rigid stator PCPs, as the rotor speed is increased, the forward flow will increase at a

72

.......
.....
.
faster rate than the slip flow, increasing the pump's volumetric efficiency. This has
already been experimentally demonstrated in the laminar and turbulent flow regimes
[18, 57]. In rigid stator PCPs, this increase in volumetric efficiency is opposed by a
decrease in mechanical efficiency as the surface speed of the rotor increases the shear
forces in the clearance. These two competing effects mean that there is an optimal
speed, for a given PCP, where the mechanical and volumetric losses are balanced.
This trade-off is seen in Figs. 2-17 and 4-23; and has been described by Vetter [57].
Given the spatially concentrated wear on the rotor described by Vetter [57]; me-
chanical forces discussed in Fig. 2-17; and effect of rotor motion on flow in Section
2.2.3, it is likely that there is contact between the rotor and stator, governed by the
static pressure and viscous forces on the rotor. The rotor motion is important for
calculating the mechanical and volumetric efficiencies (10% or more for both cases).
Therefore, further work on conventional, meaning: unconstrained, rotor PCPs should
be capable of modeling the fluid-solid interactions on the rotor. Berton et al. have
noted this importance [5]. They claim to have implemented this capability, but do
not present their results in detail. It should be possible to use the work in this thesis
to extend the lubrication theory model to include fluid-solid interactions.

73
74
Chapter 3

Turbulence in Progressive Cavity

Pumps

To develop new PCPs for use with low-viscosity fluids, such as water, this chap-
ter presents newly adapted methods for estimating the performance of a PCP with
turbulent flows. These methods are important because low-viscosity fluids are a ma-
jor application of artificial lift systems. Water and high-temperature hydrocarbons
from oil sands extraction (SAGD) are two noteworthy examples. Without accessible
models for turbulent operation, the design engineer's analytical toolbox will be lim-
ited. This chapter investigates a method for directly and efficiently calculating the
critical geometric parameters for a network model. This method lends itself to effi-
cient pairing with lubrication-theory analysis of PCPs. The model is then compared
with experimental data from literature and experiments from this work to confirm its
accuracy over a range of geometries, in both laminar and turbulent conditions.
Formulations of the asymptotic Navier Stokes equations are also presented with
eddy viscosity included to enable use with Reynolds Averaged Navier Stokes (RANS)
equation closures [33]. This formulation is a PCP specific implementation of the
general turbulent lubrication theory equations [25].
The linear network models presented by Pessoa et al. have shown promise for un-
derstanding the performance of pumps that have already been modeled or extensively
tested [47]. Unfortunately, due to the extent of manual calibration that they require,

75
their methods are unable to inform design decisions in the early stage of pump con-
ceptualization. This section includes a refinement of their method, and a process for
integrating the lubrication theory model into the linear network calibration process
to rapidly get a physical understanding of PCP mechanics.

3.1 Network Models

3.1.1 Background

PCPs can be modeled as devices that deliver fluid at a known rate, Qth, minus a
slip flow, S, as expressed in Eq. 3.1. Because Qth is straightforward to compute,
see Eqs. 1.1 and 1.2, the challenge of this approach is in modeling the slip flow. The
simplest way to model the slip flow is with a fluidic resistor analog. One way to model
the resistance of a fluid element is with the Darcy-Weisbach equation, which assumes
that the pressure difference across a channel is proportional to the dynamic pressure
of the fluid and the channel aspect ratio (hydraulic diameter to length) Eq. 3.2. Here,
U is the area averaged fluid velocity, i, is the channel length, and Dh is the hydraulic
diameter. The Darcy friction factor, f, is that coefficient of proportionality. The
term famar is the laminar correlation for f in Pessoa's model, and is calculated
with Eq. 3.3 [47].

Q = Qth - S (3-1)

Fluid network models for PCPs assume that each cavity has uniform pressure.
These network models are robust, fast, and easy to understand. However, most
require calibration to experimental data or numerical simulation, although Zheng
has recently developed a promising technique for estimating laminar flow resistance
directly from geometry [63]. This section discusses the different strategies that can
be used to estimate the resistance of the sealing lines.

76
 ........

.
AP - f 1, pU2 (3.2)
Dh 2

flaminar = - (3.3)
Re
S =Uwb (3.4)

Dh =2w (3.5)

Re UDh _ 2S (3.6)
v ub

A P =CP P2 (3.7)

Pi P0

Figure 3-1: A depiction of a single flow resistor.

Pessoa et al., whose work this is largely based upon, uses the friction factor corre-
lation for straight ducts, Eq. 3.2, to estimate the flow resistance of PCP sealing lines
[47]. Since 1, is used as a calibration factor in their work, it can obscure the nature of
the model. Furthermore, the equivalent pipe resistor model requires non-physically
large surface roughness to hydraulic diameter ratios on the order of 0.1 to match ex-
perimental data. An alternative pressure loss modeling approach is used by Edwards
et al., represented in Eq. 3.7 [14]. There are two advantages to this formulation. It
is more transparent about the nature of the modeling approach, and it allows for a
more nuanced discussion of the flow resistance mechanisms.

77
 K
Re (3.8)

APL K (3-9)
2bw2
blong =rRs (3.10)

OX= - 2E cos(208 ) (3.11)

&Yr
ae =- 2E sin(2E8 ,) (3.12)
=ZrS (3.13)
08 27r
7r
" Xr 2x__2
2 gr 22 + 8Zr 2

btrans =
S
r
OS,
+
aes +r zr(3.14)
aes

Most fluid fittings and connections are modeled by a static pressure loss coefficient,

C,, which is inversely proportional to Re in the laminar regime and constant with
respect to Re in the turbulent regime. Equation 3.7 is consistent with that convention,

and allows for better articulated comparisons between the flow characteristics in PCP
sealing lines and other flow restricting devices. The pipe analogy used to formulate
Eq. 3.2 for PCP seal line modeling introduces an extra hydraulic diameter, Dh, term
along with the l calibration term. Given that the static pressure loss coefficient
formulation is more commonly used to model flows in complex geometries, the author

encourages the use of Eq. 3.7 over Eq. 3.2 [14, 15, 22]. As such, the rest of the network
modeling work discussed in this thesis will also center around Eq. 3.7.

Equations 3.7 and 3.6 connect the sealing line geometry to the laminar flow re-
sistance, Eq. 3.9. A detailed description of how to compute the longitudinal and

transverse sealing line widths, blong and birans respectively, is in Pessoa's original pa-
per [47]. w is the sealing line height, p is the dynamic viscosity, and U is the average
fluid velocity in the channel. For many flow components, the pressure drop coefficient
is independent of Re in the turbulent regime [14]. While this is a large assumption,
it is well supported for such a complex flow structure; it provides a surprisingly good
predictor; and it is an excellent starting point for discussing the role of turbulence

78
in the mechanics of PCP flow. This results in Eq. 3.15 as an estimate for nonlinear
turbulent flow resistance between cavities.

APT = S 2 CpTP
2 2
(3.15)
2b w

CPT ~ ( A (3.16)
AR

Equation 3.16 is a commonly accepted estimate for the pressure loss coefficient of
a mixing jet [22]. In this case, AR is the area ratio between the vena contracta of
the orifice jet and the surrounding cavity. For a PCP, AR is approximately w/(4E).
Because w << 4E for all applications discussed in this thesis, Cp will be treated as
1 for turbulence modeling applications, unless otherwise mentioned. However, this
assumption should be revisited if a PCP with large clearances or small cavities is ever
designed. For the orifice model, it is assumed that the flow separates immediately
downstream of the narrowest point. This assumption implies that the seals act as
perfectly imperfect diffusers where none of the kinetic energy in the dynamic pressure
created by acceleration through the seal is recovered.
The results from Edwards' work can also be used to estimate when flow through
the sealing line transitions from laminar to turbulent [14]. In all of Edwards' ex-
periments, the flows in the fittings studied transition to turbulent when the laminar
coefficient of friction becomes approximately equal to the turbulent coefficient of fric-
tion. This conclusion was also put forward by Fester et al. [15]. For the progressive
cavity pump sealing lines examined in this thesis, and likely others as well, Cp is ap-
proximately 1. Therefore, it is hypothesized that the transition to turbulence in PCP
seals likely occurs when Re = K. While this might seem obvious to some readers,
it is very different from pipe flow, where K = 64, but the transition to turbulence
occurs at approximately Re=1,500, where Cp jtransitions to its turbulent value. This
is a useful design hypothesis, because it allows for pump designers to begin to pre-
dict what slip flow magnitude will induce turbulent flow for a given pump geometry.
Additionally, it often predicts the transition an order of magnitude earlier than most

79
PCP research papers in the current literature.

3.1.2 Extension of Network Model

Before discussing new deterministic calibration procedures for the sealing line resis-
tances, K, the structure of the fluid network model needs to be refined. This section
presents clarifications and modifications on Pessoa's original model [47]. The general
structure of a single lobe PCP resistor network is depicted in Fig 3-2. Since Pessoa's
model is already calibrated by a single coefficient, sealing length: 1s, it may seem
unnecessary to focus in detail on terms which may simply be calibrated out [47].
However, it will be demonstrated that small changes in the model structure has a
significant impact on the predictive capabilities of the tool, and the intuition for PCP
performance that it builds.

Central Nodes

Inlet Boundary Nodes Outlet Boundary Nodes

Figure 3-2: Illustration of the cavity nodes and connectivity in a PCP network model.
The arrows represent slip flow between cavities.

Boundary Conditions and the Right Hand Side

Pessoa et al. previously presented a powerful application of linear networks to model-

ing PCP performance [47]. A few elements of their work were unclear to the author.
As such, this section begins with specific definitions for those elements as a starting
block for this work. Pessoa's paper does not include a specific formulation for the
right hand side of the resistor network system of equations. It is inferred, based
on Eq. 3.17 for slip flow from Pessoa's paper, that the pump outlet is connected to

80
the last node by one transverse sealing line, and to the second-to-last node by one
longitudinal sealing line.

S = n-4n 2 + A~-n1(3.17)
RL RT

Relative Pressure Distribution [-Relative

Rotor Pressure Distribution (AP/AP)

0.9 0.9

0.8 0.8

0.7 0.7

0.6 - 0.6

3 0.5 0.5

0.3 0.3

0.2 0.2

0.1 0.1

0 0
1 2 3 4 5 6
Theta [rad]

Figure 3-3: Pressure distribution along the rotor of a representative 2 cavity pump.
Transverse sealing lines connect cavities that are two nodes apart, and are horizontal
lines above. Longitudinal sealing lines connect neighboring cavities, and appear at
an angle above.

The new slip flow formulation proposed in this work, Eq. 3.18, assumes that both
longitudinal and transverse slippage occurs between the pump outlet and the final
cavity. This assumption is reasonable given the general geometry of PCPs shown
in Fig. 3-3, generated using lubrication theory [3]. Further inspection of equivalent
ghost points for the network outlet also reveals that there are indeed three leakage
paths to the outlet. Equation 3.19 is the equation for the network with boundary

conditions. T = 1/RT, L = 1/RL, and A = -2(L + T).

++
S-=pm (3.18)
RT RL RL

81
 0

0
Aj- 2 T_ 1 Li 0 0 Pi- 2
0
T- 2 Ai_ 1 T Li+1 0 Pi- 1
0
Lj- T_ Ai TI+1 Lj+2 Pi (3.19)
2 1
0
0 Li_ 1 T Aj+ 1 T+2 Pi+1
0
0 0 Li T+ 1 Ai+2

Po(Tn + Ln)

Differences in Pressure Drop Coefficients

Given the differences in rotor and stator curvature on the longitudinal and transverse
sealing lines, it is assumed that T and L have different values, respectively. All values
for T are assumed to be identical, and the same assumption applies to all values of L.
For the reader who is skeptical of this delineation, a simple proof of the importance of
treating the two sealing line types differently is presented below. A two cavity pump
can be represented by the mass balance for a linear network with Eq 3.20, where
T=1/RTand L=1RL.

82
 0

L T -2(T + L) T L 0l P1 0
(3.20)
0 L T -2(T + L) T L P2 0

PO

[Po]

After normalizing the equation such that T* = T/[2(T+L)] and L* = L/[2(T+L)]

and simplifying, Eq. 3.20 simplifies to Eq. 3.21.

[1 -T* P1 L*
=[ PO (3.21)
1P2 0.5

The solution is:

Pi = PO L* + T* L*T* + r5 (3.22)
1 (1 -T*2

( L*T* + 0.5\ (3.23)

P2 = P I - T*2
)

In the simplest of cases, when L* = T* = 0.25, P 0.4PO and P2 = 0.6PO.

This result indicates that the ratio of the sealing values plays an important role in
determining the pressures in each cavity. The results from the lubrication theory
model, shown in Fig. 3-3, predict that P = 0.34PO and P2 = 0.65PO. This matches
qualitative descriptions of linear cavity pressure distributions in literature, and indi-
cates that longitudinal sealing lines are the driving leak path, connecting neighboring
cavities.

83
Calibration Algorithm

The network formulation presented here can quickly determine L and T with one
result from the lubrication theory model presented in Chapter 2, which is a first-
principles model that has shown strong agreement with experimental data. The
calculation begins with a two cavity PCP in a fluid where Re < 50, and then compute
s, P1 , P2 , and P, with lubrication theory. Those values are input to Eq. 3.25. The first
two rows of Eq. 3.25 come from Eq. 3.20; and the bottom row is based on Eq. 3.18,
the equation for the outlet slip flow, s. Equations 3.24 and 3.25 represent the same
system of equations.

-2(L+T) T L P1 0

T -2(L+T) T+L P2 0 (3.24)

-L -(L+T) 2L+T Po S

P2 -2P1 Po-2P
1 [ 1 0
T
P1 -2P 2 +Po PO-2P2 0 (3.25)
L
PO-P2 2PO-P1 -P 2 S

Since the 3x2 matrix, (A), in Eq. 3.25 is not square, the Moore-Penrose inverse
directly solves for L and T in Eq. 3.26.

. . 0
T
= (AT A)-lAT 0 (3.26)
L
S

After computing the total resistance for the transverse and longitudinal sealing
lines, the known geometric parameters specify the laminar viscous coefficient, K, with
Eqs. 3.27 and 3.28.

84
 KT 2btransW2 (3.27)
Ty
2biongw2 (3.28)
KL= L-2u
Lp

The two pressure loss coefficient, single sealing length, and the single sealing length
with old boundary conditions (see above) were all calibrated against a two cavity
PCP using Eq. 3.26 and then used to predict the performance of a 4 cavity PCP
with the other geometric parameters held constant. Table 3.1 shows the distribution
of pressures inside of a four cavity PCP. In accordance with the methods described
by Pessoa et al., the single sealing length models were calibrated manually, iterating
until the model matched the data [47]. The two pressure loss coefficient model was
calibrated with Eq. 3.26.
The two pressure loss coefficient model shows better agreement with the lubri-
cation theory model than either of the single sealing line length models. The first
and last cavities of the single sealing length model are off by 5% of the outlet pres-
sure, while the single sealing length model with simple boundary conditions has a
non-physical pressure distribution, see the last column of Table 3.1.
The data from Gamboa et al. were also included for the 4 cavity PCP for reference
[18]. The specific Gamboa data are for 0% gas volume fraction operation with 0.042
Pa-s oil running at 400 rpm. The pressure measurements are from just before the
pressure spikes at 180'. PO for the Gamboa data was specified assuming that there is
a 826.13 kPa pressure differential across the pump. These data match the two sealing
length and lubrication theory model to within 2%.

Independence of KL and KT

This section presents the sensitivity of K to different PCP design variables. With
the new orifice network formulation, it is possible to separate sealing line length, KL
and KT, from the number of cavities. It is valuable to know what factors drive the
pressure loss coefficient, and what factors are independent of K. The design variables

85
 Value Gamboa Lubrication 2 seals 1 sealing length Simple BC
P1 /Po 0.210 0.211 0.211 0.251 0.187
P2 /Po 0.398 0.405 0.403 0.400 0.277
P3 /Po 0.595 0.600 0.597 0.600 0.4749
P4 /Po 0.812 0.792 0.788 0.750 0.443

Table 3.1: The pressure profiles of different PCP models for the 4 sealing cavity PCP.
The linear models were calibrated against a two cavity PCP lubrication simulation,
and were simulated with Re ~ 50.

that could affect the viscous pressure loss coefficient are listed below. Some of these
design variables have a direct influence on sealing line performance, but they all have
additional nonlinear influence through cross coupling with K. It has been shown in
Table 3.1 that K is independent of the number of pitches. Figure 3-4 is an example
of how K varies with geometric variables. In this case, the base design is the pump
from Gamboa's experiments [191. Note that varying multiple geometric variables at
the same time may result in nonlinear cross terms not portrayed in Fig. 3-4.

* w: The clearance gap appears separately from K in the flow resistance equa-
tions, PCP performance is very sensitive to w, it makes sense that w and K
have a co-dependence.

" P,: The rotor pitch, and, by kinematics, the stator pitch directly affects the
curvature of the sealing line, so K is dependent on it.

" E: The pump eccentricity should mostly affect sealing line width, and have a
minimal impact on K.

" R,: The rotor radius directly affects the curvature of the sealing lines, so it
plays a role in determining K.

3.1.3 Model Validation

The orifice network model shows agreement with experimental data from literature.
Figure 3-5 shows both the turbulent data from Gamboa and Vetter alongside the

86
 Pressure Coefficient Dependence

1.6

1.4-

1.2-

0 1

0.8-

0.6 E - KL
-- w -KL
0.4- Rr-K
+-P - K
0.2
-

0.5 1 1.5 2 2.5

Figure 3-4: The effect of different geometric variables on the longitudinal pressure
loss coefficient. Higher values of K/KO indicate better geometry-independent sealing
and delayed transition to turbulence.

model's predictions [18, 57]. The orifice network model has similar agreement with
the data as Pessoa's equivalent pipe resistor model [47]. This makes sense because
both functions have the same quadratic pressure and flow relationship. Gamboa's
experimental results have been extensively studied, and other linear networks have
shown the same degree of agreement after manual calibration. The Reynolds numbers
for flow inside the seals of the water Gamboa experiment range from 1,000 to 9,000.
The pump used in Vetter's experiments has not been tested at low Reynolds numbers,
making manual calibration of a network model unfeasible. The orifice network model
was able to accurately and quickly capture their performance results without any
calibration beyond setting C, to 1.

A proof-of-concept small water pump, discussed in more detail in section 4.2, was
designed using the orifice network model. The experimental results are compared

87
 Gamboa Water Comparison Vetter Experiment Comparison
0.7- 0.6-
: 0 o Linear Network 300 rpm Linear Network
+ Gamboa Data 300 rpm +Vetter Data
0.6-
o Linear Network 400 rpm 0.5-

+
+ Gamboa Data 400 rpm
0.5 o Linear Network 600 rpm
0.4-
0+ Gamboa Data 600 rpm
0.4- 0 0
+ 0
0 :9* 0.3-
0.3- + o+
IL E 0.2-
0.2
00
0.1-
0.1
0*

0 0-

-n 1 -0 .1 ! I I II I I I I . I . I I I I I I f II I I II
0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200
AP [kPa] AP [kPa]

Figure 3-5: Results from the orifice network model compared with experimental data
from Gamboa and Vetter [19, 57].

with the orifice network model in Fig. 3-6. The experiment was run between 600 and
2300 rpm, and between 10 and 150 kPa static pressure differences.

Experiment and Orifice Network Total Flow Error

0.8- 0.05
-

+ Experim
o Model
0.7 I 0

-0.05
0.6-
-0.1
0.5-
II
S0.4 I -0.15
0 I -0.2
0.3
I I
*

2-0.25
0.2-
-0.3-
0.1- I -0.35-

n- ........... - f4A
. ..i ; : i :
;..............
i : i :
0 50 100 150 0 50 100 150
AP [kPa] AP [kPa]

Figure 3-6: Results from the water prototype plotted against predictions from the
orifice network model. Black lines connect experimental data and simulation results
at the same speed and pressure.

Given these differences, it is likely that there are flow mechanisms which are
not accounted for by Eqs. 3.7 and 3.15. Idel'chik has noted that diverging curved
channels have a greater turbulent and friction resistances than equivalent straight
channels [27]. Pessoa et al. also noted that 'one of the strongest hypotheses in these
approaches is the simplification of the channel geometry' [47]. While experimental
comparison shows the orifice network's impressive ability to model flow performance,

88
the turbulent pressure loss correlation still leaves room for improvement in the model.

Implications for Compliant PCPs

Since this model is capable of efficiently estimating turbulent flow through PCP chan-
nels, the orifice network model could be coupled with a deformation model for a
compliant stator or rotor.

3.1.4 Combined Turbulent and Lubrication Flow Resistance

If the earlier hypothesis that turbulent jet mixing is the primary mechanism for
turbulent flow resistance, then it stands to reason that the viscous momentum transfer
mechanisms, such as the ones observed in the laminar resistor model, are still present
in the flow. Therefore a combined turbulent and laminar resistor, analogous to the
channel and orifice shown in Fig. 3-7, as described in Eq. 3.29 can provide more
predictive accuracy.

Figure 3-7: A viscous channel followed by a mixing dominated sudden expansion.

APTL = S2 CpP + S pK (3.29)

2b2W2 2bW2

K was computed using the laminar methodology described above. With the ad-
dition of a new flow resistance mechanism, Cp needed to be reduced to 0.8 to match
the data from Gamboa, Vetter, and in this thesis. A C, value of slightly less than 1
is analogous to flow separation occurring downstream of the flow restriction.
As Figs. 3-8 and 3-9 show, the empirical loss coefficient of 0.8 and the geometrically
driven K value show excellent agreement with experimental data for three different

89
 Gamboa Water Comparison with Laminar & Turbulent C Vetter Slip Flow Comparison with Laminar &Turbulent C,
0.8- p 0.8 --
o Linear Network 300 rpm
0.7 0 + Gamboa Data 300 rpm
,0 0 Linear Network 400 rpm
0.6 + + Gamboa Data 400 rpm
o Unear Network 600 rpm
0.6-
0.5 - + Gamboa Data 600 rpm

0.4 - 0

03 0.4-
-

U_ 0 0

0.1-
0.2 %+ o ++

+
0.3

-
0.1- +

+
0.2
0- 0 *
0 +
0 +0
09
0

-0.1-
0 100 200 300 400 500 600 700 0 200 400 600 800 1000 1200

AP [kPa] AP [kPa]

Figure 3-8: Results from the long orifice network model compared with experimental
data from Gamboa and Vetter [19, 57].

pumps operated across a wide range of speeds, pressure, and length-scales. It must
be admitted that the addition of complexity to a model can increase predictive power
without necessarily using better physics. However, the structure of this model makes
physical sense: it is possible for momentum to be dissipated by both viscosity in
a channel and jet mixing downstream of the channel. This is different from the
previous laminar/turbulent equivalent pipe friction factor. In that formulation, it
would not make sense to combine viscous and turbulent resistances in series. Since
K is computed from first principles using lubrication theory, only one term, C,,
is necessary to fit the model. A C, of 0.8 is physically plausible, especially when
compared to the surface roughness to diameter ratio, c/D, of approximately 0.1 is
necessary to fit the equivalent turbulent pipe model to data.

As such, the author recommends this long orifice network model as the turbulent
model for PCP design. It provides remarkable turbulent predictive capability; a
prediction for the transition to turbulence; and has a minimal increased computational
cost compared to the C, = 1 model because of the efficiency of the lubrication theory
model.

90
 0.8- Experiment and Orifice Network 0.15- Total Flow Error
-+ Experiment
0.7- 0.7 --
Model
0.1
0.6
: 0.05-
0.5
-

0.40
-

0.3 - -0.05 -
0

,
0 .2 9 *.

0.1
-

0 -0.15

-0.1- -0.2 I
0 50 100 150 0 50 100 150
AP [kPa] AP [kPa]

Figure 3-9: Results from the water prototype plotted against predictions from the
long orifice network model. Black lines connect experimental data and simulation
results at the same speed and pressure.

3.2 RANS Turbulence Models

While the orifice network model presented in the first half of this section is a powerful
tool for modeling turbulent flow inside of PCPs, there still is a need for a higher order
turbulence model. Predicting pump efficiency, for example, requires the computation
of wall stresses to estimate mechanical efficiency. Even though it may be possible
to develop empirical correlations for mechanical and hydraulic power losses, a higher
order model will still be valuable tool for better understanding flow mechanisms.
The most common method for simulating turbulence is with the Reynolds Aver-
aged Navier-Stokes (RANS) closure. The RANS equations depend on methods for
approximating the correlation between the velocity components. The most famous
approximation for that correlation is the Boussinesq hypothesis which assumes that
the turbulent correlation term in RANS has a similar mathematical structure to vis-
cous momentum diffusion, and can be treated as an effective eddy viscosity, itt.

3.2.1 Turbulence Model Application

This section presents a few strategies for understanding the role of turbulence in
high Re PCP behavior. Before examining the models, the effective eddy viscosity
ratio itt/ip from Gamboa et al.'s experiments with water in a metallic PCP needs

91
to be estimated [20]. From inspection, a non-dimensionalized pressure reveals the
connection between viscosity and the no-flow pressure rise.

AP*=Z PIQO (3.30)

PIQ

The bulk effective eddy viscosity for the lubrication model can be found by com-
paring the pressure where Andrade et al.'s model has zero net flow with the pressure
where the pump actually stops delivering fluid. The terms come from the data in
Fig. 3-5.

SL
A't = S (2.91)

Pt _ APQ=O, turb - 1 (3.31)

A APQ=O, lam

The calculated pty/ from the Gamboa experiment and Andrade model, predicts
a ratio of 5.4 at 300 pm and 5.7 at 400 rpm. These values indicate an important
nonlinear effect which is not captured by low Re lubrication flow.

Turbulent PCP Lubrication Equations

This section presents a general formulation for incorporating eddy viscosity in the
PCP specific lubrication theory equations. The additional eddy viscosity term changes
the structure of the lubrication theory equations. After assuming that the inertial
terms only contribute to the effective eddy viscosity, an assumption that should be
investigated further, this new radially variable term changes the solution strategy for
the coefficients in the 2D Poisson pressure equation. The derivation below shows how
the velocity and viscosity fields can be integrated to compute new coefficients for the
pressure equation. First, a radially variable eddy viscosity, pt, is added to the original
momentum equations, Eqs. 3.32 through 3.34:

92
 Ou
0= + t, [a r (3.32)
Oz r Or Or)

0= Op
Or (3.33)
1 op (O ~1O(rw)~1
+ yl a [ Or (3.34)
r 00 Or _r Or_ }

which then become Eqs. 3.35 through 3.37:

OU
0= O+ (t +Otr (3.35)
0z rOr I
Op
0 = O (3.36)
Or
rOOap + It(rW) (3.37)
r8
00 r r Or_

The eddy viscosity pt is determined by a separate RANS closure; such as the

Spalart-Allmaras, Wray-Aggarwal, or r-c closures [61, 51, 34]. Solving these equations
for u and w, the first derivatives, which will later be useful in shear stress calculations,
are:

OU Op r C1
= - (3.38)
Or Oz 2(p + pt) r(p + pt)
+

r) = Ow Op log(r)r rc 3
(3.39)
Or Or 0 /-t + pt P + Pt

Integrating a second time, the velocities include integrals that require numerical
evaluation.

lop jr r* fr dr*
U = dr* + ci J r( + C2 (3.40)
2 Oz R, P + Pt , r*(P + -ct)
p 1[ log(r*)r* r C3 fr r *C4
(3.41)
+

00 r JF. /I A*it Rr,/ -it r

Inputting the boundary conditions from Andrade et al., Eqs. 2.18 through 2.23,

93
 to find the constants of integration to be:

= r dr* (3.42)
Ro dr*
J Rr r* (tt+pt)

C2 = 0 (3.43)
Row(R0 ) - QR - - f1R lo(r) dr*
C3 = (3.44)
fRo r*tdr*
r, P+pt
c4 = QR2 (3.45)

The resulting equations for u and w are then:

U -
__ R, * d,* fr 'r p+At
r* dr* f r dr*

)
Pprdr* (3.46)
~Oz ft [, +/It
2fS2ROr r*I+pt
dr* Ro dr*
R~r p-t

Op 1
[fr log(r*)r* * r* dr
R, log(r*)r*
I
*
ftr _+pt
-

W R
0 r JR A + Pt Jft r A + 1-t f "0 * dr*
0
(3.47)
QR2 Row(Ro)
r
- R2 r ' dr*
~ AA ++
r r f( r* dr*
Jr A+pt

Following the lubrication theory approach from Andrade at al., integrating the
continuity equation in cylindrical coordinates and applying the same boundary con-
ditions, the resulting conservation of mass formulation is [3]:

a'Rr rudr + wdr - w(Ro) -- Rov(Ro) = 0 (3.48)

azfR, r6 oR, ao
Next, u and w are substituted into this equation, and integrated to obtain a 2-
dimensional Poisson equation for P in z and 0. The terms are taken from Eqn. 3.46
and Eqn. 3.47.

C2 = -(Cow) + a(pgC 2 ) +w(Ro) a + Rov(Ro) (3.49)

00 0 Oz 0 az ao
)

Oz
)

The corresponding constants are in Eqs. 3.50 through 3.52. They are evaluated on

94
the appropriate grid points, see Section 2.1.2, based on the geometry and turbulent
viscosity. Then the Poisson equation can be numerically solved for the pressure field.

R Er log (* T* r r* f Rolog(r*)r* dr*

C1 = 'dr* - dr* RrRI+pt dr (3.50)
Rr LIR0 1t A +1At
/Ro Ro
A
r
JR0

r* dr*
fRo r*
r dr*
dr]

C2 =
Rr Rr A ++t
dr* j2fSR ro + r* /I
dr*
_ R r /+
Ro dr*
/ dr (3.51)
r /p+t Rr p+pt
2logRo Row(Ro) - QR 2
Ro /r r* dr*
Cow R log + /+/t +r dr (3.52)
r RrI R
JR, p * dr*
r*ft J r

This approach decouples the lubrication model for momentum inside of the pump
from the turbulent viscosity closure. This reduces the problem to fixed point iter-
ation between Eqn. 3.49: a linear 2 dimensional PDE for pressure and flow; and a
RANS closure for the eddy viscosity, Mt. This form has a similar structure to the
implementation of RANS turbulent lubrication theory discussed by Hori [25].

Due to the scope of this thesis, the numerical scheme was not implemented. Future
work should numerically implement this model with a RANS closure.

3.2.2 Inclusion of Cell-Averaged Inertial Terms

The contribution of the turbulent viscosity to the momentum equation needs to be

compared with the inertial terms from the spatial velocity gradients in Section 2.3.2.
These terms can be analytically computed in the first inner loop iteration of the
simulation because the eddy viscosity is zero at all points. Evaluating the mean
inertial terms, from later inner loop iterations, will require numerical differentiation.
Given the accuracy of the long orifice network model in section 3.1.4, future work on
the turbulence model should include these averaged inertial terms in the analysis.

95
3.3 Conclusions
This chapter presents a new network model for estimating high Re flows in PCP seal-
ing lines, and proposes basic elements for creating a high Re version of the lubrication
theory model discussed in Chapter 2. The new network models rely on a pressure
loss coefficient that can be directly computed from the lubrication theory model and
a turbulent mixing pressure-loss coefficient. This work has increased the physicality
of fluid resistor network models in PCPs; improved calibration methods; increased
the accuracy of the turbulent network model; and validated its use as a design tool
with an experimental water-moving prototype PCP.
This chapter also presents a PCP specific formulation of general turbulent lubri-
cation, which has not yet numerically implemented.
Future work should focus on refining and further validating the long orifice network
model; and numerically implementing the general turbulent lubrication model.

96
Chapter 4

PCP Design

4.1 1-DOF Cycloidal Bearing Design

This chapter presents the design of a hypocycloidal bearing which is capable of con-
straining the rotor motion in a PCP such that it never contacts the stator. This new
type of bearing adds some complexity to a PCP assembly, but it comes with distinct
advantages, such as the ability to manufacture PCPs from molded plastic parts. The
kinematics hypocycloidal trajectory of the PCP rotor has been examined extensively
in literature, and the reader is directed there for an in-depth discussion of the subject
[38, 50, 40].
Preventing contact between the rotor and the stator increases the number of ma-
terial choices and means that pump life is considerably extended. In an unconstrained
PCP, the rotor wears against the stator until clearances are so large that the rotor
and/or stator need replacing [55, 57]. With a constrained rotor PCP, there may
be some initial contact between the rotor and stator, but they will eventually wear
together and cease contact when the structural loop then goes entirely through the
bearings. Thus, PCPs with bearings can be designed with an arbitrary surface rough-
ness, or from plastics. The clearance will be then set by the stiffness and accuracy of
the bearings and deflection of the rotor due to pressure. With constrained motion, it
may prove to be advantageous to have high surface roughness, a patterned surface,
or another exotic trait (such as hair or a pressure-compensating behavior).

97
 These bearings come with costs and challenges. They increase assembly complex-
ity, cost, and risk over-constraint as well as adding loss mechanisms to the system.
The author believes that it may be possible to offset those costs by using higher speed
rotors with low-cost materials, structured surfaces, and/or eliminating the losses form
rotor-stator contact

Sliding

Sliding contact bearings were not investigated given their high coefficient of friction.
Friction directly impacts efficiency, and a first order friction efficiency analysis showed
that sliding bearings were not able to achieve the desired mechanical efficiency without
lubrication.

Rolling Element

Rolling element bearings offer low friction, are easy to pre-load, and are very stiff.
For rolling element bearings operation in a wet, and potentially dirty, environment is
the major challenge.

Hydrodynamic

Hydrodynamic bearings rely on pressure that is generated by the flow of fluid through
a converging channel. One example hydrodynamic journal bearing pressure distribu-
tion is shown in Fig. 4-1. This effect can be used to create thrust and journal bearings.
For channels that are thin enough to provide reasonable lift, the Reynolds equation
Eq. 4.1 is the general reduced form of the Navier Stokes equations [25].

a (ph 3 p )+ 0 (ph 3 Qp 0 ( pa
h(ua + Ub ) + pa
h(va + Vb)
Ox 12p ax ay 12p ay Ox 2 / y 2 (4.1)
+ P(Wa - Wb) - PUa
A+
- PVa + hoat
Ox Oy 09t

h is the height of the channel. u, v, and w are the velocities of the two bodies in
x, y, and z respectively. The subscripts a and b are for the top and bottom bodies
respectively. When the channel height and body velocities are known, the equation

98

.. .....
. ........
becomes a 2D Poisson equation, identical in structure to the equation used to solve
for pressure fields inside of progressive cavity pumps derived by Andrade et al. in
the previous chapter [3]. This is no coincidence: that model is based on the same
assumptions as lubrication theory.

For a journal bearing of radius R rotating at some speed Q, under quasi-static

conditions, the Reynolds equation reduces to Eq. 4.2.

( V O + = (4.2)
00 6pR 2 O -0y 6/pLc9y M0

In the coordinates of the journal bearing, 0 is the circumferential coordinate and y

is the axial coordinate. While the one dimensional version of the Reynolds equation,
which assumes uniform pressure distribution in y, is considerably simpler than Eq. 4.2,
the one dimensional equation is only applicable for 'infinitely short journal bearings',
which are often referred to as ISJB in literature. The ambient pressure boundary
conditional at both sides of the journal bearing have a strong effect on the force
generated by the bearing. Pressure builds parabolically from the edges of the bearing
towards center, causing the lifting force to scale with L 3 instead of L.

h and its spatial derivative are computed with Eq. 4.3 and Eq. 4.4 respectively.
c is the clearance between the shaft and the bushing, and e is the eccentric distance
between the center of the bushing and the center of the shaft. Low eccentricities result
in poor lift, and high eccentricities result in stress concentrations in the center of the
gap which shorten bearing life. Best practices recommend designing for eccentricities
between 0.6c and 0.7c under expected loading to get the most out of a bearing design.

h = c + e cos(O) (4.3)
Oh
- - - e sin(0) (4.4)
00 Rj
c = R3 - R, (4.5)

The pressure field is then integrated to find the radial force on the bearing in

99
 1 240

0.9 220
-200
0.8
180
0.7
160
0.6 140
.. 0.5 120

0.4 100
80
60

0.1 20

0.5 1 1.5 2 2.5 3

0 [rad]

Figure 4-1: Example pressure distribution over a water hydrodynamic bearing, com-
puted using lubrication theory. In this case, L = 20 mm, Rj = 15 mm. c = 20 pm,
e = 10 pm, and Q = 419 rad/s

Eq. 4.6. The adverse pressure gradient in the expanding region of the channel often
causes separation and film break down, preventing pressure recovery, this effect is en-
capsulated in the Giimbel (or half-Sommerfeld) boundary condition [8, 16]. Therefore,
without pressure recovery from an expanding flow, hydrodynamic journal bearings are
often modeled only half-way around (0 = [0, 7r])

F= j pidA (4.6)
0 0
The drag forces on the shaft can be similarly computed by the shear forces from
the pressure field. Given the similarity of the hydrodynamic bearing and PCP models,
the equations in Section 2.3.2 can be directly used to compute the shear forces.
Since this model takes bearing geometry as an input and estimates forces, there
typically are significant normal and orthogonal components to the force acting on the
rotor. This is because, at equilibrium, the shaft sits off-center in the bearing relative
to the applied force. This is called the 'attitude angle' of the bearing, and shifts the
position of the shaft in the bearing.

100
 This model can also be used to estimate the effect of angular displacements induced
by misalignment or moments applied to the bearing. h can be linearly varied in y
without adding much complexity to the solver. This results in an additional moment
applied on the bearing, which should counteract the applied moment. Since this is
an inverse problem as well, the solver must iterate to reach a moment balance.

The same methodology can be applied to a linear hydrodynamic bearing. Pressure

fields are generated by converging flow, similarly to the journal bearings. Those
pressure forces provide lift, and apply a moment. Using conservation of momentum,
the position of the hydrodynamic bearing is iterated until the net moment is zero,
and the lift is equal and opposite to the applied force.

Flexural

The linear constraints can also be implemented with a one degree of freedom (DOF)
flexure with two or more parallel blade flexures [24]. Equation 4.11 relates the design
bending stress (Ob) with the range of motion (6), Young's modulus (E), stiffness (k),
flexure width (wf), length (if), and the number of parallel stages (n,). One particular
challenge with designing a precision flexure for operation over a 'wide' range of motion

is the parasitic shortening of the flexural blades. The most straightforward solution
to this parasitic shortening is to use two flexure stages in series with a floating stage.

When the lengths of both serial flexures are the same, then the floating stage has
off-axis motion relative to the ground stage and the driven stage, but there is no

parasitic off-axis motion of the driven stage in the ground reference frame due to

flexure shortening.

101
 F = kU (4.7)

k = 12 (4.8)
npJ3
'Pf

b Mwf (4.9)
2I
M = fI(4.10)
2
b 3E6wf (4.11)
nply
The minimum thickness of the flexures is set by buckling and off-axis stiffness.
Compressive loading of the flexures could buckle the flexure stage, which would result
in damage and inefficiency from contact between the rotor and stator. For most
cases, Euler buckling, Eq. 4.12, is the only concern, but plate buckling, Eq. 4.13,
should also be considered for particularly wide flexures. The wide geometry may be
necessary to limit the dynamic over-shoot of the floating stage, as discussed below.
Both Euler buckling, and post-buckling phenomena in the case of plates, have been
covered extensively in other material, and will not be discussed more in this thesis.

Fe, = Tr22 EII(4.12)

r (Kblf ) 2
wr 2 D
Ncr = k" 2D(4.13)

(mb ab) (4.14)

a mb
Eh3
Dh
= (4.15)
12(1 - v 2
)

Even if the flexures are not at risk of buckling, sufficient off-axis stiffness must be
maintained to prevent the rotor from coming in contact with the stator. The clearance
is often on the order of hundreds of micrometers, so stiffness is a critical design
constraint. As demonstrated in Section 2.3, loading on the rotor is complicated, and
both ends of the rotor generate different reaction forces in the radial and tangential

102
directions. To ensure the life of the flexure, and prevent contact, the loading of the
flexure in different positions, from the fluid model, need to be considered when loading
the flexural bearing in different deflected positions.

ae
0

7M1

Figure 4-2: Front view of the CAD for the vertical (left) and horizontal (left) flexural
hypocycloidal crankshaft bearing stages.

Flexure dynamics present two risks for use in the cycloidal crankshaft bearing.
The first is the possible presence of parasitic forces required to move the flexure
from one position to another. Just like a car driving over a hill, net force over
a cycle averages to zero, but this could create additional losses in the motor with
varying torque and un-recovered energy inputs. Additionally, the dynamic motion of
the floating intermediary stage is decoupled from the motion of the rotor, and can
overshoot, crashing into the side walls.
The most straightforward way to analyze the parasitic force and torque to actuate
the crankshaft flexure system is through the principle of virtual work, Eqs. 4.16 and
4.17. The terms Ff, Ef, and rf refer to the force required to deflect the flexures, the
energy stored in the flexures, and the torque required to rotate the rotor respectively.
Because the crank-shaft ensures that the two flexures are 90' out of phase, the energy
in the system can be described as a sum of trigonometric functions in Eq. 4.18.
This trigonometric identity indicates that the energy stored in the flexure system is
constant if the horizontal and vertical flexures have the same stiffness, k. This can
be seen in Fig. 4-3

103
 Elastic Potential Energy

1-

0.8

- 0.6

-
- Horizontal Flexure
N - -Vertical Flexure
0.4- -Total Energy

0.2

0 iii
0 1 2 3 4 5 6
Flexure Position [0]

Figure 4-3: Energy stored in the othrogonal flexure system.

Ff = M (4.16)
Ox
TY = Of (4.17)
00
Ef kvA 2 khA 2 kA 2 (4.18)
2 2 2
OIf 0
(4.19)
00

Thus, since Ef can be expressed independently of a linear translation of the rotor

and is constant in 0, there are theoretically no parasitic forces or torques required
to drive the flexure. In practice this is true for small deflections, and still accurate
for tip-rotation-free large deflections. The force to deflect the flexures is still a part
of the design process because they must be deflected in the first place to assemble
the flexure system, and because those forces are still exerted on the crank elbow
between the two flexures. Because flexures do not enable continuous rotation, roller
bearings are necessary between the flexures and rotor to maintain sufficient degrees
of freedom. Those bearings have friction, which does change the torque required to
actuate the crank flexure. Large deformation stiffening of flexures, as the actuation

104
 -. 4

force becomes more in line with the flexure's axis, also create minor energetic minima.
The most noticeable deviation from the linear theory in plastic flexures is the plastic
deformation of the stages when the system sits unused. One or both of the flexures
are always deformed. This means that one of the stages will stress relax to create a
new energetic minimum.

Figure 4-4: Angled view of a flexural hypocycloidal crankshaft bearing.

A = Af sin(Qt) (4.20)

M6 = -k6 + k(A - 6) (4.21)

(026) SkAf sin(Qt) - 2k (4.22)

at2

Af k sin (Qt) +Csi 2k

+C 2 cos t r) (4.23)
2k - MJQ2 M SVM

Since 6(t = 0) = 0, C2 = 0. The initial floating stage velocity initial condition is

also assumed to be zero, giving the condition 6(t 0) = 0, resulting in Eq. 4.25

105
 -AfkQ =C k (4 24)
2 k -MQ2 M
C -AfkQ (4.25)
2k - MQ 2 2k

The resulting dynamical equation is Eq. 4.26.

6(t) =2k -
2k
2
-MQ2
[sin(Qt) - Q
V 2k sin t
M
(4.26)

The first sine term is the driven frequency response and the second term is the
natural frequency of the under-constrained spring-mass component. Since, for any
non-resonating frequency, the two sine waves must be additive at some point in time,
the magnitude of displacement for the floating stage is bounded by Eq. 4.27.

6max= max Afk

2k - MQ2 (
(i ) (4.27)
V2k)
Equation 4.27 can be further simplified and, non-dimensionalized by a dimension-
less frequency, Q*, and displacement, 6*, expressed as Eq. 4.30.

* (4.28)
6
(4.29)
Af
1
* max (4.30)
2(1 - *)

The non-dimensional response of the floating stage is presented in Fig. 4-5. Min-
imizing 6* is critical for ensuring that the flexures are compact enough for down-well
applications, and for minimizing the flexure stresses. Values of Q* less than 0.5 and
greater than 1.5 appear to have the best deflection characteristics. High values of
Q* are particularly interesting because 6* can be less than 0.5 in that design region.

106
Since all intermediary speeds between 0 and Q* need to be passed through, the ro-
tor acceleration protocol needs to be carefully designed to prevent to flexure from
resonating. Fluid damping will decrease the displacement gain, but that is left for
future work, as this calculation provides a conservative estimate for design.

Floating Stage Response

7-

6-

5-

3-

2-

I I II I
0 0.5 1 1.5 2 2.5 3

Figure 4-5: The dynamic response of the under-constrained floating stage

This low frequency resonance occurs because the floating intermediary stage is
under-constrained. The driven stage is constrained by the rotor and other flexures
in the crankshaft flexure system, and the ground stage is rigidly attached to the
structure. One solution to the under-constrained dynamics was put forward by Panas
et al. [45]. They add a flexural linkage which forces the position of the ground stage
along the primary translating axis to be the average of the ground stage and driven
stage positions. This modification eliminates the first resonant mode, and significantly
improves the dynamic response of the flexure system.

107
4.1.1 Orbiting Strategy

A rolling element orbiting around a central rotating bearing could also enforce the
hypocycloidal constraint, similar to the hypocycloid mechanisms described by Hsu
[26]. This mechanism would be similar to a planetary gear drive with one key differ-
ence: for a single spiral PCP, the planet is 1/2 the diameter of the ring gear, and the
sun gear has zero diameter. This means that any radial force holding the orbiting
planet will not be aligned with the center of stiffness for the journal bearing, as seen
in Fig. 4-6. A secondary orbiting gear can be placed symmetrically about the journal
bearing, but this increases complexity. Due to the potential complexity of this design,
it was not selected for the prototyping round presented in this thesis. However, that
does not mean that the design should not be investigated in the further. As such, a
brief overview of the critical design concepts is presented here.
The crux of the orbiting design is that the rolling motion creates the hypocycloidal
trajectory that is necessary to prevent rotor-stator contact. If the planet ever slips
relative to the ring instead of rolling, the rotor and stator will come into contact. This
means that any frictional rolling element would require a particularly large pre-load
to ensure that it does not slip in a lubricated environment. As such, a spur-gear plus
ring-gear topology is recommended to enforce a radial constraint.

A A

=bib Od6 11016MOTOR

-

FREE ROTATING SHAFT

Figure 4-6: Sketch of the orbiting bearing topology.

To the author's knowledge, there are four main design considerations this orbiting
gear:

e Tooth separation forces

108
 * Tooth shear forces

* Hertz contact stresses

" Tooth crowning to prevent edge loading

4.1.2 Hirth Coupling Assembly

Alignment of the rotor, stator, and bearings is critical for the performance of the
eccentric bearing support. However, connecting this many components with precise
concentricity and angular alignment is challenging due to the loads and the length of
the tolerance loop. Hirth couplings are capable of delivering high alignment precision
and strength through elastic averaging. Elastic averaging is the logical reciprocal of
exact kinematic constraint [52]. It leverages the combined benefit of many alignment
features acting in parallel. Providing that there is enough compliance in the alignment
features to allow the parts to mate, the joint can combine high stiffness and precision.
By the law of large numbers, the errors average out, causing the precision to scale with
the square root of the number of elements, as shown in Eq. 4.31 [52]. Furthermore,
the stiffness and strength scales with the number of aligning elements, described in
Eq. 4.32, allowing higher strength than exact kinematic joints, such as kinematic
couplings. Hirth couplings provide precision alignment of the rotor crank, as seen in
Fig. 4-7, and the stator+bearing assembly, as seen in Fig. 4-9.

Figure 4-7: Image of the water pump crank assembly with one of the Hirth couplings
visible from the side.

e= eavg (4.31)

109
 Etota 1 (4.32)

Both Curvic couplings and Hirth couplings operate on the same principle. Curvic
couplings have the benefit of being easier to machine with high precision grinding
methods. This manufacturing method is largely responsible for the popularity of
Curvic couplings over Hirth couplings. That said, Hirth couplings were still selected
for three reasons: the large contact areas reduce contact stress; the flat faces can be
self-locking; and the geometry can easily be created via 3D printing and molding.

In a tapered joint, torsional forces can be resisted by friction, provided that the
taper angle is properly set. This is the same concept as the critical sliding angle for
a block with a given coefficient of friction. Equation 4.33 describes the taper locking
tooth root angle, 0, for a given coefficient of friction, p. In a wet environment, or with
low-fiction materials, the coefficient of friction is too small to self-lock with reasonably
tapered teeth. In this case, torsion can be resisted with an axial preload, as described
in Eq. 4.34. Fp is the preload force. FT is the twisting force applied at the coupling.
6 is still the tooth root angle. Large values of 0 indicate a steep tooth.

0 = tan--( - (4.33)

Fp FT cos(0) - p sin(0)
sin(0) + /Icos(0)

4.2 Rolling Bearing Experiment

The first prototype was developed to evaluate the performance of a high speed,
submersible PCP with a constrained rotor. The rotor was constrained with rotor-
mounted ball journal bearings rolling between two rails with a small clearance shown
in Fig. 4-8.

110
Figure 4-8: Sectioned view of the water pump hypocycloid bearing and crank shaft
rotor.

4.2.1 Motor

A custom wound Hyperion motor provided the mechanical power for this prototype.
The winding resistance was measured by applying a fixed current with a power supply
and recording the voltage drop. The no load current was calculated at a range of
speeds to find the internal friction losses. The motor constant was found by measuring
the stall torque and the no-load speed.

V = + IR (4.35)
Kv
Prlmc = IR (4.36)
2
Pr=mcI Q+ R (4.37)
Kv

- () +4PRTmc
I 2R (4.38)

(4.39)

P is the electrical power going into the motor controller. I, R, V, and Q are the

111
motor current, resistance, voltage, and rotational speed [rad/s] respectively. K, is
the motor constant in units of [rad/(V-s). Once the motor current is estimated, the
typical motor characteristics are straightforward to calculate.

V =P1 mc (4.40)

Q = (V - IR)K, (4.41)
1
Kt = 1(4.42)
K(
r = (I - I 1)K, (4.43)

Qmot (4.44)
VI
pump (4.45)

Kt is the motor torque constant. 1 mot and Tlpump are the motor (wire-to-shaft)
efficiency and the pump (shaft-to-water) efficiencies respectively.

4.2.2 Sensors

Motor speed was measured with a DT-2234A laser photo tachometer. The pump
outlet pressure was measured with a MG1-200-A-9V-R digital pressure gage made
by SSI Technologies inc. Flow was measured using an Omega FTB792 turbine flow
meter.

4.2.3 Total Design

The water PCP prototype was designed taking multiple constraints into account.
The pump was designed to deliver up to 1 liter per second of water against a pressure
of 100 kPa or greater. Even with a relatively large number of pitches, high-speed
operation (>2,000 rpm) is necessary to achieve reasonable volumetric efficiencies with
a clearance gap of 200 pum. A large clearance was selected to be representative of a
plastic molding process. The high speed operation required a short stator to avoid

112
 Term Symbol Value Unit
Clearance w 0.2 [mm]
Rotor Radius Rr 10 [mm]
Eccentricity E 5 [mm]
Stator Pitch PS 50 [mm/rev]
Rotor Length 1 200 [mm]
Table 4.1: Critical design dimensions of the water prototype

running into problems with whip, estimated by Eq. 4.46, or other dynamic problems
with the shaft. The rotor was constructed with a shallow pitch to fit the greatest
amount of pressure resistance in a small footprint. This represents a fundamental
trade-off between sealing benefits from speed and rotor length.

S 48E (4.46)
VmL3

The bearing design for the water PCP prototype has sealed ball bearings mounted
on the rotor which are designed to roll along guides shown in Fig. 4-8. To fit the bear-
ings onto the crank and connect the cranks to the rotor, insert and screw preloaded
Hirth couplings were used to align and hold the parts together. This overall design
was selected to simplify the assembly process. One end of the rotor is attached to a
crank assembly, inserted into the stator, and the other end of the crank assembly is
then attached to the rotor. Finally, the bearing guides are slid over the crank and
locked in place with the external Hirth couplings shown in Fig. 4-9. The axial con-
straint is provided by a ceramic ball on the end of the inlet side of the crank, which
rides on a stainless-steel puck. This style of constraint was chosen for low friction
and good wear resistance. The pressure forces on the pump outlet are designed to
pre-load the axial constraint.

To separate component alignment from sealing, an enclosing tube is slid over the
whole Hirth coupling assembly. 0-rings separate different pressure regions of the
pump, and tie rods hold the whole assembly together. A small window on the drive

113
side of the pump provides access to tighten and align the helical motor shaft couplings.
Helical couplings were used to accommodate the rotor eccentricity. However, those
couplings frequently broke, and were replaced with a ball gear drive in the rail bearing
prototype.

Outlet Saddle
Drive Shaft
I

Crank Supports

Rotor Stator

Hirth Joint
Crank Supports

Axial Support
.
.

Inlet
Figure 4-9: Sectioned and whole views of the water pump assembly CAD.

114
 Figure 4-10: Picture of the full water pump assembly.

Figure 4-11: Image of the crankshaft rotor (top). Close up of the inlet side (left) and
outlet side (right) of the crank with roller bearings and axial support ball.

115
4.2.4 Cavitation

With increased speed comes an increased risk of cavitation for flows with water. This
is more than a theoretical concern. Vetter et al. studied this in detail [56]. In this
work, the first version of the water pump prototype was unable to exceed 1,500 rpm.
Additional power input to the motor only resulted in increased current draw from
the motor, indicating an asymptotic increase in the torque required to turn the rotor
without an increase in rotor speed. Asymptotic power draw and the pitting shown in
Fig. 4-12 indicate that cavitation is the likely cause.

Figure 4-12: Cavitation on the inlet side of the water prototype rotor due to incor-
rectly designed PCP inlet conditions.

Inlet pressure drop cavitation in PCPs has been investigated by Vetter et al. in
their 1997 paper [56]. This cavitation mechanism is driven by the pressure drop
required for the fluid to fill the inlet cavities as they axially progress, growing in
volume. Vetter et al. characterize the pressure drop as a combination of laminar, A,
and turbulent, (, momentum dissipation terms

116
 2
AP ( +A ) sub pU (447)

( and A are experimentally determined pressure loss coefficients for the PCP inlet.
1 is the channel length, and d is the effective hydraulic diameter for the inlet.

The pressure reduction from the outside of a rotating body to the inside of the
rotating body could cause a drop in pressure, resulting in cavitation. This form of
vortex cavitation comes from Eq. 4.48, which reduces to Eq. 4.50 assuming that the
crank bearings impose solid body rotation at the inlet. Solid body rotation is a
conservative assumption, but since this model neglects the vortex stretching at the
inlet, it should give a reasonable scale of the phenomenon.

P -0(4.48) -
Or r
O= prQ2 (4.49)

pQ 2 (R 2 - r 2
) (4.50)
2

R is the ID of the enclosure; Q is the rotational speed of the fluid in solid body
rotation; and r is the radial coordinate of the fluid within the enclosure. For a water
pump with an ID of 32.3 mm, and a rotational speed of 1,500 rpm, the pressure
reduction at the core is approximately 160 kPa, which is over 60% greater than a
vacuum.

Given that a simple analytical model supported the hypothesis that vortex cavi-
tation was driving cavitation, the inlet of the water PCP prototype was redesigned
such that the inlet port is aligned with the center of the enclosing tube, as seen in
Fig. 4-13. This modification apparently eliminated the source of asymptotic power
draw. Because redesigning the inlet to be at the center of fluid rotation instead of
the outermost radial point, reduced pressure is the likely source of cavitation. Future
work should apply modern computational analysis techniques to better design around
the inlet pressure drop cavitation described by Vetter [56].

117
 Stator
Inlet

Bearing
Crank Supports Axial Support
Stator
Inlet
Rotor

Bearing
Crank Supports Axial Support
Figure 4-13: Section view of the PCP inlet and inlet bearings with the side inlet
(top), and the axial inlet (bottom).

4.2.5 Results

The results from the water PCP prototype experiment in Fig. 4-14 demonstrate ex-
cellent agreement with the long orifice model in Section 3.1.4. This verifies the effec-
tiveness of the long orifice model as a design tool. The validation of this model means
that it is possible to create small plastic pumps, with the associated tolerances, and
still achieve high volumetric efficiencies. Unfortunately, the estimated i-nput torque
indicates that this specific pump is not mechanically efficient enough to be a useful
water pump. This can be seen in Fig. 4-27.

118
 0.8_ Experiment and Orifice Network
+ Experiment
0.7- o Model

0.6

0.5-

y 0.4
-

0.3-

0.2-

0.1

0-

0 50 100 150
AP [kPa]

Figure 4-14: Results from the water prototype plotted against predictions from the
orifice network model. Black lines connect experimental data and simulation results
at the same speed and pressure.

4.3 Rail Bearing Experiment

To demonstrate the design and role of the cycloidal constraints in PCP design, the
nominal PCP geometry used by Gamboa et al. was tested with the addition of linear
roller crankshaft bearings described in Section 4.1 [18]. The position and phase of the
rotor cranks is set by precision adjustment mechanisms, aligned on a surface plate
before assembly, fine tuned with eddy current probes after assembly, and tracked
with an eddy current probe during operation. For reference and consistency, the
critical geometric variables from the Gamboa experiment are included in Table 4.2.
As a note, Andrade et al. rounded the rotor radius to 20 mm when testing their
lubrication theory model [3]. Their stator radius was increased by the same amount
to avoid changing the nominal gap width.

119
 Term Symbol Value Unit
Clearance w 0.185 [mm]
Rotor Radius Rr 19.94 [mm]
Eccentricity E 4.039 [mm]
Stator Pitch P, 119.99 [mm/rev]
Rotor Length 1 360 [mm]

Table 4.2: Critical design dimensions for the rail bearing prototype.

4.3.1 Structure Design

PCP rotors are subject to loading in all six degrees of freedom. The axial moment
is the design parameter of primary concern for any PCP designer because it governs
the transmission of power from the drive train into the fluid. Pitching moments are
generated on the rotor by the unequal pressure in cavities. The high pressure cavities
at the outlet provide a greater pitching moment about the center of rotation than the
lower pressure cavities at the inlet.
The axial torque is set by the pump geometry and mechanical efficiency, IM, with
Eq. (2.52).

T 4ERrPsP (2.52)
7F'17M

The critical design parameters of torque and pressure can now be connected by
a dimensional PCP torque coefficient, c, in Eq. 4.51, describing the torque necessary
to provide a certain outlet pressure rise.

T 4ERrPs
c'r = - = (4.51)
AP 7,1?M
This torque was used in combination with outlet pressure rise driven axial loads to
provide coarse estimations of rotor and bearing loading. The bearings for this pump
were designed assuming that the mechanical efficiency would be at least 50%. Once
the preliminary pump was designed, final torque and load calculations were computed
using the lubrication theory model. The structure was refined to accommodate any

120
differences between the coarse and high resolution models. The Hirth couplings were
sized such that the same torque values did not induce shear failure in the root of the
teeth. The ball gear drive, shown in Fig. 4-15, was also sized to ensure that the shaft
did not fail on torsion, and that the ball gear joint teeth did not fail in shear.

Figure 4-15: Sectioned isomeric view of the rail bearing pump outlet with ball gear
joint, outlet bearings, and eddy current probe CAD (left). Assembled ball gear joint
(right).

4.3.2 Motor

To ensure precise control of the pump speed, a 34Y214S-LW8 high-torque stepper

motor was used in the prototype with a MLA10641 stepper motor controller. Both
the stepper motor and motor controller were provided by Anaheim Automation. The
windings were connected in the 4-lead bipolar parallel configuration. The resulting
winding resistance is 0.205 Ohms, and the resulting winding inductance is 2.3 mH.

4.3.3 Sensors

All measurements were recorded with a National Instruments cDAQ-9174 with a NI

9211 module for recording and amplifying the temperature signal, and a NI 9205
module for converting sensor voltages to a digital signal.
A K-type thermocouple was used to record the temperature of the working fluid at
the pump. The same fluid was processed in a rheometer to verify the precise viscosity,
0.156 Pa-s, of the working fluid in the pump during operation.

121
 Pressures were recorded with a Dwyer 628-02-GH-P1-El-S5 pressure sensor. It is
designed for a pressure range of 0 to 50 psia, a signal voltage between OV and 1OV,
and is rated as accurate to 1.0% of the full-scale range.

A Macnaught MX series fuel and oil flow meter for a 1" NPT connection and a
range of 1.6 to 32 gpm was used to measure the flow rate of the working fluid. It is a
positive displacement rotary lobe flow meter. A scanned copy of the calibration sheet
is included in Appendix 5.2. The temporal resolution of this sensor is a function of
the ratio of the average flow rate and the flow constant, or K-factor, which is 36.141
pulses per liter, or 136.81 pulses per gallon. The time constant, described in Eq. 4.52,
is the temporal resolution that the flow meter can measure.

TK - KmQ (4.52)

A higher temporal resolution estimate of flow rate can be determined by combining

the data from the flow meter and the much higher bandwidth pressure sensor. Figure
4-17 shows that the pressure drop in the flow loop is hyperbolically proportional
to the flow rate. It appears that the pressure-loss coefficient, Cp, is beginning to
transition to a turbulent value, being independent of Re. Since the transition is only
just beginning at the highest Re encountered in this experiment, the laminar pressure
loss coefficient will be exclusively used to predict the flow rate at a high temporal
resolution.

Assuming that unsteady effects are small, the pressure drop across the flow loop
and the flow rate are related by a flow resistance, CR, by Eq. 4.53. The delineation
between the pressure-loss coefficient Cp, and the flow resistance, CR, expressed in
Eq. 4.54 might seem arbitrary, but they are treated as explicitly different terms to
maintain consistency with conventional definitions. It can be shown by applying
unsteady Bernoulli to the results of this calculation that inertial effects are small in
Fig. 4-16.

122
 Unsteady Pressure Residual
2000
I I

-
..11lid
1500

-
1000 --

500-

0
CO

T -500-

-1000

-1500-

-2000-
I I I

50 100 150
Time [s]

Figure 4-16: The residual from fluid accelerations for the highest flow-rate experiment.
This residual at most is 3% of the nominal pressure.

AP =CRQ (4.53)
CR (X Cp (4.54)

If CR is known, the instantaneous flow rate can be accurately estimated with

Eq. 4.55.

CR
(4.55)
AP

Because the flow meter is quite accurate over a longer timescale CR can be deter-
mined by comparing the integral of the computed Q in Eq. 4.55 with the totalized
volume of flow over the same window, V.

123
 Flow Loop Pressure Loss Coefficient
108 _

CL10 7
0 :

6
10

101 102
Re

Figure 4-17: The pressure loss coefficient computed for various needle-valve positions
in the flow loop.

V = Cf AP(t)dt (4.56)

V
CR = t P(t) (4.57)
0e'APt

With CR computed, it is possible to produce a high bandwidth estimation of the

pump flow rate with Eq. 4.55.

For a turbulent flow loop, where CR is independent of the flow rate, Eqs. 4.58
through 4.61 can be used instead of the above analysis.

124
 AP = CRQ2 (4.58)

Q= (4.59)

V = VA-P(t)dt
z- (4.60)
f

CR = (4.61)
(ff !2- _A(t)

Four Lion Precision ECL202 eddy current probes were used to track the rotor
position. They were calibrated to cylindrical aluminum targets. A scanned copy of
the calibration sheet is included in Appendix 5.2.

An in-line continuous rotation torque transducer recorded the input torque re-

quired to drive the rotor and bearings. Assuming a best-case mechanical efficiency of

90%, a worst case mechanical efficiency of 50%, Eq. 4.51 was used with Table 4.2 to
determine a torque coefficient range of 0.0143 [Nm/kPa to 0.0257 [Nm/kPa]. If the
pump should be testable between 50 kPa and 250 kPa. This means that the torque

meter should be able to measure torques varying between 0.715 Nm and 6.425 Nm.
Because the stepper motor is capable of 4.59 Nm of torque, the TQ513-100 rotating

shaft torque sensor from Omega was chosen with a full scale reading of 12 Nm to

have sufficient overhead torque for measurement.

4.3.4 Total Design

The purpose of the rail bearing PCP experiment was to evaluate the feasibility and
effect of using hypocycloidal bearings to constrain the rotor's motion. Eddy current

probes were used to track aluminum targets on each end of the rotor, shown in Fig. 4-

18. Pre-loaded micro-adjustment mechanisms positioned the bearing blocks using 100
thread-per-inch set screws with roller-ball tips, shown in Fig. 4-22. Once the bearing
block was positioned, four screws were tightened to hold it in place. This separated
the alignment functionality from the structural loop.

125
 -4

N
I
w w Wj

iI

i~rI

4 E~wi!~'
jii

Rur 04W /1

Figure 4-18: CAD cross-section (top left) alongside a picture (top right) of the drive-
shaft end of the rotor inside the stator with an eddy-current target (yellow in the
CAD, and aluminum in the picture) attached. CAD cross-section of the inlet end of
the rotor inside the stator with an eddy-current target attached (bottom left)alongside
a picture of the eddy-current probe placement on one end of the stator (bottom right).

To ensure modularity and adjustability, while maintaining precision alignment,

each module of the pump was designed with a Hirth coupling, as seen in Fig. 4-19.
The same Hirth coupling connects different stages of the crank and rotor such that
it can be precisely assembled with the bearing blocks.

126
 0@
LA!?P
Outlet Saddle

Drive Shaft

Crank Supports

-Stator

Rotor

-Hirth Joint
Crank Supports

Inlet

Figure 4-19: Whole and sectioned views of the rail bearing pump assembly CAD.

127
 II
'I lI
I II
I II,
II
'4

.9. __

Figure 4-20: Image of the assembled rail-bearing pump with flow loop.

Figure 4-21: Close up of the rotor inside of the clear stator. il

/ /1/ awillow,
/ i/I
/1/

~//
/ /

44
Figure 4-22: Sectioned isomeric view of the rail bearing pump inlet with ball drive,
outlet bearings, and eddy current probe CAD (left). Assembled rail bearing prototype
(right).

128
4.3.5 Results

The results from the rail bearing experiment appear to follow mechanical and volu-
metric characteristics curves. The parallel lines seen in Fig. 4-23 are typical of low Re
flows in PCPs. The close agreement between the actual flow rates, and the flow rates
predicted by lubrication theory is due to the geometric error modeling in Chapter
2. Given the fact that the measured geometric error from 3D printing and possibly
wear, as seen in Fig. 4.3.5, is multiple times larger than the nominal clearance of
185pm, this degree of agreement would not be possible without the geometric error
accounting.

Measured Pump Flow Rates

0.35 + 50 rpm
-

- 10 rpm
* 15 rpm
0.3- 0 20 rpm
+ 25 rpm
x 30 rpm
0.25-

0.2
+
-

0
0.15- 00
X
0.1-
0
0.05-

U -. I 1 1- 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
0 100 200 300 400 500 600
AP [kPa]

Figure 4-23: lubrication theory simulation results alongside flow rates from the rail
bearing prototype experiment.

The volumetric efficiencies for the rail bearing experiment, seen in Fig. 4-24 also
follow the characteristic linear trend. The small deviations from linear are likely due
to measurement error. It should be noted that volumetric efficiency in this plot is not
computed with either Qths or Qthr. Not only do the nominal geometric deviations
shift the actual 'theoretical flow rate,' but the zero pressure rise flow rate extrapolated
from the experimental data is less than either Qth, or Qths. The Qth value used to

129
compute volumetric efficiency in this PCP is 0.0116Q l/s, while Qth, is 0.0123Q 1/s.
This provides a more real estimate of mechanical efficiency because this pump should
not exceed 0.0116Q 1/s even at zero pressure rise.
The mechanical efficiency curves in Fig. 4-24 also appear to follow characteristic
trends. The paring of volumetric efficiency and mechanical efficiency trends has
an interesting consequence: while the AP associated with the best efficiency point
depends strongly on rotational speed, the magnitude of the best efficiency point has
a weak dependence on rotational speed. This behavior has interesting consequences
for the design of PCPs that are discussed in more detail in Section 4.3.5.

1 Measured Pump Efficiencies

- 50 rpm
0.9 - 100 rpm
G G. ~.--150 rpm
0.8 - -200 rpm
- .- 250 rpm
0.7 -o y- -300 rpm

S0.5-
-

0.3
0.2
0.1
-

0-
0 50 100 150 200
AP [kPa]

Figure 4-24: Efficiencies from the rail bearing prototype experiment. Volumetric
efficiency, mechanical efficiency, and total efficiency are o, <>, and x respectively

The torque, speed, flow, and pressure data from the rail bearing experiment are
able to provide insight into PCP performance. Clear scaling laws emerge when the
data are compared. High temporal resolution data demonstrates how repeatable and
deterministic PCP behavior is at low Re. Detailed rotor and stator measurements,
combined with a constrained rotor trajectory, validate the accuracy of the error lu-
brication model in Section 2.2.
Torque measurements from running the pump at different speeds without a work-

130
 4.5 Experimental Bearing Pump Torque 3 Simulation Torque
+ 505m
50 rpm

-
4 ,100 rpm -
150 rpm 2.5- - 100 rpm
3.5- 0 200 rpm 2-15
5-20 : -- 150 rpm
: + 250 rpm - -- 200 rpm
rpm
300 rpm
rpm300 2- - 250 rpm
- - 300 rpm
rpm]
22.5 - z
1.5
00

1.5 -1-

0.5
0.5 0

0-0- 1 T
0 50 100 150 200 250 0 50 100 150 200 250
AP [kPa] AP [kPa]

Figure 4-25: Measured shaft torque from the rail bearing prototype experiment (left)
and lubrication simulation torques (right).

ing fluid in Fig. 4-33 show a kinematically repeatable parasitic torque. Given the
kinematic repeatability of this torque at a range of speeds, the best explanation for
this phenomenon is over-constraint of the rotor. Both ends of the rotor are attached
to bearings with the 1 DOF cycloidal constraint. This parasitic torque could be a
combination of interaction between the bearing units at each end of the rotor, or
interactions between the two opposed linear bearings in each cycloidal bearing unit.
The data in Fig. 4-25 are linear, and the slopes appear to asymptote towards a
common value as speed increases, likely as the bearings become properly pre-loaded.
The deviation appears to be a speed and pressure cross-term, which was not included
in the scaling laws to avoid unnecessary complexity. That noted, a pressure-speed
cross term might be necessary to properly model bearing PCPs over a range of op-
erating conditions. y-intercepts also have a linear relationship with rotational speed,
indicating a surface-speed proportional drag torque.

PCP Affinity Laws

The scaling laws developed in Section 2.3.1, Eqs. 2.57 and 2.58, were inspired by a
first-order analysis of the fluid stress tensor on the surface of the rotor, and they follow
similar trends to the characteristic efficiency curves from the rail bearing experiment
in Fig. 4-24.

131
 cLQQ - CSLAP)AP (2.57)
LAP + C( P +7 ) Q

TT (CQQ - CsT AP)AP (2.58)

1
7Tf AP + cdQ + o)Q

For the rail bearing pump presented in this section, the pump coefficient of friction,
Cf, is 0.0514, the dimensional coefficient of drag, Cd, is 0.0286 [N-m s/rad], and the
parasitic torque, 0 , is 0.125 N-m. cQ also has a slight deviation from theoretical ideal
value. The theoretical value for cQ is 1.23x10- 5 m3 /rad, but it was experimentally
found to be 1.16x10- 5 by examining the y intercepts in Fig. 4-23. The laminar slip flow
coefficient, CsL, is 5.65x10- 10 [m3 /Pa-s]. Although the equations appear complicated,
they provide accurate and straighforward estimates of the mechanical performance of
a PCP as seen in Fig. 4-26.

Scaling Validation

0.9-
0.8 --. -- ~

0.7 -

-
0.2 0__

0.3
0.2 - 100 rpm o0
-- 200 rpm
0.1 --- 300 rpm

0 50 100 150 200 250

AP [kPa]

Figure 4-26: Scaling model for PCP efficiency compared with experimental data for
volumetric efficiency (o), mechanical efficiency (o), and total efficiency (x).

This model also has good agreement with the water prototype from Section 4.2.
Torque was estimated from the power input to the motor controller using the method

132
described in Section 4.2.1, which is reflected in the relatively large error on torque and
mechanical efficiency estimations in Fig. 4-29. In spite of that complicating factor,
the PCP efficiency scaling laws still provide a good overall fit for pump performance,
as can be seen in Fig. 4-27. Figure 4-28 shows the strong agreement between the
quadratic slip flow scaling correlation, the long orifice network model from Section
3.1.4, and experimental data.

1 -Water Prototype Efficiencies

+ Exp Mech
0.9-7- x Exp Vol
S ?* Exp Total
0.8 0 Scale Mech
0 0 Scale Vol
0.7 - o Scale Total

0.6
0.5
0.4
0.3-
0.2 U
0.1
0- 1 1 liII I III
0 20 40 60 80 100 120 140
AP [kPa]

Figure 4-27: Scaling model for PCP efficiency compared with experimental data from
the water prototype.

For the water prototype pump, the pump coefficient of friction, Cf, is 0.147. The
dimensional coefficient of drag, Cd is 0.00205 [N-m s/rad], and the parasitic torque
is 0.075 N-m. The flow coefficient is cQ is 3.18x10- 6 m 3 /rad, and the turbulent slip
coefficient, CsT, is 5x10- 7 [m3 /vPa-s].
The author would like to emphasize that, due to the extensive use of experi-
mentally determined coefficients, this model is not currently intended for use as a
predictive tool for designing new pumps. This model is currently useful in three
ways.
It allows for pump designers to understand how efficiently a pump would operate at
different speeds and pressures. One interesting result of this model is that, for laminar

133
 0.8_ Experiment, Model, and Scaling
+ Experiment
0.7 0 Model
- * Scaling
0.6-

0.5

0.4

0.3

0.2
-

0.1

0~

-0 .1 4~ , I, I I I I I I, I
0 50 100 150
AP [kPa]

Figure 4-28: The simple single quadratic resistor slip flow relationship used in the
turbulent scaling laws compared with experimental results and the orifice network
model. Black lines connect experimental data and simulation results at the same
speed and pressure.

operation of this pump, increased speed has a small effect on the magnitude of the
maximum efficiency point, but it does have a significant effect on the pressure rise
where the maximum 7 occurs. This result has significant implications for the design
of a turbulent water pump. The efficiency scaling rules imply high-speed operation
is a better strategy for turbulent flow than it is for laminar flow. Since turbulent slip
flow is proportional to v/A P, instead of AP, the flow penalty of operating at higher
relative outlet pressures is diminished for the same mechanical losses and similar seal
characteristics. Figure 4-30 is an oversimplified illustration of this phenomenon. The
two theoretical pumps being compared have the same mechanical characteristics, and
the same slip flow performance at AP = 160 kPa. In the context of Eqs. 2.57 and
2.58, those two pumps are almost identical. Figure 4-30 shows a higher maximum
efficiency at a higher pressure for the turbulent pump, and shows a weaker slip flow
efficiency penalty at higher AP. Of course, this is a simplified comparison. It is
unlikely that two such different pumps would have such similar characteristics, but

134
 Measured and Scaling Torques
0.8- J
0.7

-
0.6-

0.5
z-
0.4-

0.3-
0.2-

0.1

II-IIIII 0 II I I I I I
0 20 40 60 80 100 120
AP [kPa]

Figure 4-29: Estimates of motor torque from the water (+) prototype experiment
and scaling law fit (*). Black lines connect experimental data and simulation results
at the same speed and pressure.

the analysis tools are illustrative of broader themes in PCP design.

As seen in the turbulent and laminar comparison, distilling PCP performance to

specific coefficients allows for a more rigorous discussion of how mechanical and hy-
draulic energy are transformed inside of PCPs. The curves in Fig. 4-26 can appear
complicated at first glance, but these scaling laws cast light on the underlying sim-
plicity of PCPs. Given simple bench-top testing, Eqs. 2.57 and 2.58, can provide
powerful insights into entire design strategies.

Furthermore, these scaling laws can inform how a PCP will operate in the field.
Some applications require high pressures which are challenging to test on a bench-top,
and operating conditions vary for just about any real application. All of the scaling
coefficients can be computed using discrete forms of Eqs. 4.62 through 4.67. Theoreti-
cally, only three performance points are needed before the model can be extrapolated,
but overfitting with additional data points can add to predictive confidence.

135
 Laminar and Turbulent Efficien Sc lmina
1 -- :s --- olaminar
-- M laminar
0.--- 17V lamninar
turbulent
- Mturbulent
0.7-- -- V turbulent

0.6

-
0. 5
0.4-
0.3-
0.2
0.1
0
-

0 100 200 300 400 500

A P [kPa]

Figure 4-30: A comparison of similar turbulent and laminar pumps to illustrate the
effect of turbulent sealing.

CQ
(4.62)
1 - Cf O(AP)

Cd = 0(4.63)
CQdc
To = AP QCQ
Q I - Cf
-cdQ (4.64)

While cQ and c, can both be predicted using existing methods, including some
in this thesis, it is helpful to verify the values experimentally once a PCP has been
built. Given that all pipes have a pressure drop, and that flow instrumentation can
have even higher flow resistances, it may be necessary to extrapolate a constant speed
pump curve back to the y-intercept.

136
 CsL = (4.65)
O(AP)
CsT = (4.66)
a(/AP)
cQ Q(P=o) (4.67)

Coth, = ERrPs (4.68)

7r

Future work should focus on determining how the coefficients scale with geometry
and fluid properties.

Dimensionless Scaling Laws

These scaling laws can also be presented in nondimensionalized form. To capture both
turbulent and laminar behavior, two different dimensionless pressures are needed. The
laminar correlation in Eq. 4.69 is a straightforward result from dividing the numerator
and denominator in Eq. 2.57 by cQ and Q. By nondimensionalizing this expression, it
can be seen that the only direct effect of speed on the magnitude of the best efficiency
point or a laminar and rigid PCP is through the r*/Q term. This explains the weak
effect that speed has on the best efficiency point magnitude in Fig. 4-24.

=L- 1 -CSAP (4.69)

TIL = 1 + Cd +'7-_
1-Cf AP* QAP*

The dominant interaction which dictates pump efficiency is the interaction be-
tween C, and Cd. Large values of C, require high speeds to be volumetrically efficient,
and Cd requires low speeds to be mechanically efficient.

T =
1 -+ CCd /AP+(4.70)
+ (4-*
1-Cf 'AP* QAP*

The nondimensional efficiency scaling for turbulent PCPs in Eq. 4.70 follows a

similar convention, but the pressure and slip flow correlation is based on the dynamic

pressure in the seal, creating a quadratic relationship. This means that there are two

137
dimensionless pressures in Eq. 4.70: AP* from Eq. 4.69, and the inertial dimensionless
pressure, AP+ in Eq. 4.72.
This inertial dimensionless pressure relationship matches experimental results for
slip flow quite well, as Fig. 4-28. The denominator still uses the laminar correlation
with AP*, because the torque data from the water PCP experiment was not high
quality enough to justify a new drag correlation. The geometric variables A, and V
describe the cross-sectional area of the sealing line and the cavity volume, m 3 /rev,
respectively.

AP* = (4.71)
P~Q

AP+ = (2wA, (4.72)

pQ2 yC

A= bw (4.73)

V = 8ERrPr (4.74)

Future work on these scaling laws should clarify the role of drag in turbulent PCP
operation. It is plausible that AP+ also drives the drag torque.

Signal Repeatability

Given the repeatability of Stokes flow, the performance of a low Re PCP should be
deterministic and repeatable for a given geometry and rotor trajectory. Results from
the rail bearing experiment support this hypothesis. For a given nominal pressure,
operating speed, and bearing position, the rotor trajectory and pressure signal are
repeatable within the time window, often to within the rated accuracy of the sensors.
Figures 4-31, 4-32, and 4-33 all show this repeatability. To the extent that the rotor
trajectory drifts within a time-window, the pressure signal does as well.
These experimental results are different from the very low seal line Re (0.133
Pa-s and 0.481 Pa-s oils with Re<65 and Re<18 respectively) flow results obtained
by Gamboa which show a small, but non-trivial deviation from the perfectly linear

138
 Flow Repeatability 300 rpm Pressure Signal Repeatability 300 rpm
0.294-
129

0.292
128-

0.29
C. 127-

0.288
U)126-

0.286-
125-

0.284-
124-

0.282-
123-

0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

Time in Window [s] Time in Window [s]

Figure 4-31: The flow (left) and pressure (right) signal overlaid in the same time
window with the valve 1.5 turns open.

pressure-flow relationship, as seen in Fig. 2-12 [19]. These deviations might be ex-
plained by experimental error, but given that Gamboa's results for 0.042 Pa-s oil in
the same pump displays the expectedly linear behavior, there is room for alternative
explanations for the deviations from linear behavior at low Re.

Torque Signal Repeatability 300 rpm Position Repeatability 300 rpm

2.8- 10
- 12
2.75- 9 -- - D1
-D2
2.7- 8

2.65
E
2. 2.6 ca 6

05
2.55
0
2.5
32-
2.45-
2-
2.4

2.35

0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

Time in Window [s] Time in Window [s]

Figure 4-32: The torque (left) and probe (right) signal overlaid in the same time
window with the valve 1.5 turns open.

The rail-bearing experiment demonstrated very linear pressure-flow behavior with

0.156 Pa-s oil, and a correspondingly similar seal line Re of < 55. Given that the
rail-bearing PCP has the same nominal geometry as Gamboa's PCP, albeit with
more manufacturing error and larger 'real' seal clearances; but has a constrained

139
trajectory, there may be rotor-trajectory-based explanations for the deviations in
Gamboa's experiments. One probable phenomenon is the shaft whirl that is observed
in hydrodynamic bearings [41, 2]. Whirl occurs in hydrodynamic bearings when
hydrodynamic forces couple with the inertia of the shaft/rotor, causing it to oscillate
in the cavity. There is not enough information to make a definitive statement about
Gamboa's results, but this is a plausible statement, and, if it were accurate, would
have real implications for pump life.

Windowed Torque Signal 200rpm Windowed Pressure Signal 200rpm

4

3.5 200-
3-

2.5 150

32 -
0100
51.5
50
0.50

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Time in Window [s] Time in Window

-
Figure 4-33: Torque (left) and pressure (right) measurements for the rail bearing
prototype running at 200 rpm with mineral oil at various valve positions. The grey
line at the bottom of the torque measurements is a kinematically repeatable parasitic
torque that was measured by running the pump with only enough working fluid to
lubricate the rotor.

PCP Error Analysis

The rail bearing PCP has significantly greater slip flow than would be predicted by the
nominal geometry sent to the 3D printer for fabrication. The analytical tools for rotor
and stator surface error described in Section were combined with detailed 3D scan
data for the rotor and stator geometries. The stator geometry was CT scanned by
the Amphenol TCS office in Nashua, NH. The rotor was scanned at the MarkForged
engineering development office. Both scans were performed in multiple parts and
were then registered together using the MatLab computer vision toolbox. Once the
scans were merged and registered against the CAD, they were sliced into 1000 axial

140
 sections. In each section, the error was fit to a 7th order polynomial function of 0.
Note that the registration process is imperfect, and some error on the order of 10's of
pm is likely. Figure 4-34 shows examples of the stator and rotor error distributions at
different slices. The stator error is computed as the actual R, minus the theoretical
RO value at that section of the geometry.

R. Error at Slice 400/1000 [mm]

90
120 0.3 600.1 Rr Error at Slice 24/1000
0.15

+
1SO 0.2 +0.05
S

+
. .
1* M6
-+. - * - +, . +

+
- -0.1 +++
-lo -0.05 + +
180

+
''0 +4

+
*I~ ~-0.25 -4.
t
4.+
+
+++
++4

-0.35
240 300
270 -0.4
0 1 2 3 4 5 6 7
0 [rad]

Figure 4-34: Circumferential distribution of deviations from the stator's internal de-
signed geometry (left), and the rotor radius (right).

The lubrication theory model was used to compute the slip flow resistance at
different rotor positions inside of the stator. Since the lubrication theory simulation
works with constant pressure boundary conditions, a fixed pressure was imposed, and
the resulting flow rate was used to estimate the slip flow resistance, R, [m3 /Pa-s].
This provides a pump curve estimate described in Eq. 4.75. Assuming that transient
effects are small, the actual outlet pressure rise of the pump can be estimated by
intersecting the pump curve with the low Re system curve, Eq. 4.76, resulting in
Eq. 4.77. According to the surface error lubrication model and the system curve
intersection, the estimated pressure fluctuations for the rail bearing pump operating
at 300 rpm with an approximately 1.5 turn open valve are shown in Fig. 4-35

141
 Qpump(AP) = Qth - RslpAP (4.75)
Qsystem(AP) CRAP (4.76)

A P Qth(4.77)
CR + Rslip

114- Simulated Pressure Fluctuations

112-

110

.
-

2 108
-

106-

104

102 I
0 0.2 0.4 0.6 0.8 1
Time in Window -]

Figure 4-35: The effect of stator and rotor error on the pressure fluctuations assuming
a laminar flow loop pressure drop.

While there is a clear difference in the experimental, Fig. 4-31, and theoretical
pressure fluctuations, Fig. 4-35, both data show that there is a deterministic connec-
tion between pump geometry, rotor position, and pump behavior. Additionally, both
simulations show similar magnitudes of pressure fluctuations.

4.4 Water PCP Concept

The question at the heart of this work is 'What will it take to make a cost-effective
water-moving PCP?'. In this thesis it has been shown that mechanisms to allow high
speed operation and low-cost manufacturing methods are plausible. Furthermore, the

142
 Water PCP Simulation at 3,600 rpm 1 Water Concept Efficiencies

0.9-
0.8 ---- :.

0.7
0.8

-
00.6

0.6 - 0.5-
LA?

.- 0.41
0.3-
0.2 - ..--- Eff
0.2 --- EffV.
- 0.1 -- Eff t

0 100 200 300 400 500 0 100 200 300 400 500
A P [kPa] AP [kPa]

Figure 4-36: Estimated flow rates for the water PCP concept proposed (left) and the
hypothetical efficiency curve for the water PCP concept (right).

water-moving PCP experiment demonstrated that the desired slip-flow performance

is possible. Unfortunately, neither the water PCP prototype or the rail bearing PCP
prototype achieved the mechanical efficiency necessary for cost-effective water pump-
ing. However, water-moving PCPs in literature have demonstrated high measured
mechanical efficiency in water, even at high speed [57, 60]. Furthermore, the me-
chanical efficiencies predicted by lubrication theory are also much higher than the
experimental values in this work, as seen in Fig. 4-25.

This means that, to the author's understanding, much higher mechanical efficiency
PCP designs are possible. The scaling laws in Section 4.3.5 still limit the maximum
efficiency, but improved bearing design will further increase pump efficiency. Figure
4-36 shows the efficiency curve of a hypothetical pump with increased mechanical
efficiency. This pump is predicted to have a volumetric efficiency of 83% at 200 kPa.
The scaling coefficients are: Cd = 3.5 x 104 [N-m-s/rad], CT = 4.6 x 10-7 [m 3 / _Pa-

s], Cf = 0.05, -r = 0.05 N-m, r(AP=200 kPa) = 62%. Given the low torque losses
analytically predicted in Section 2.3.2, there does not appear to be a fundamental
reason why this pump could not be designed.

The water prototype PCP has sufficient sealing characteristics. Thus, there is
currently no need to redesign the rotor and stator. Instead, more efficient bearings

143
and lower-cost assemblies are needed. Given the inherent benefits of flexures and hy-
drodynamic bearings, those are the recommended technologies for refinement. Both
bearing types are very simple, very efficient, and have long life when designed cor-
rectly. Such flexure stages could be molded with Hirth couplings and stacked, as they
are in the water and rail bearing prototypes. Costs could be reduced even further if
the stator is made from hydro-formed stainless steel or blow-molded plastic. Radial
ridges could be imprinted to align bearings with a radial elastically averaged coupling.

144
Chapter 5

Conclusion

5.1 Remarks

This thesis contains substantial advancements in the fields of PCP modeling and
mechanical design. These contributions make it easier to design water PCPs, and act
as a foundation for a future research agenda to develop new PCPs.
Multiple clarifications in the PCP modeling literature have been presented in
Chapters 2 and 3. The typos in Andrade's lubrication theory work have been identi-
fied and corrected forms presented [3]. The sources of error in the lubrication theory
model have been described, mathematically quantified, and experimentally verified
with techniques put forward in this thesis. Errors in the convergence of Paladino's
CFD study have been noted in Section 2.2.3. By studying the numerical conver-
gence of the PCP problem, one challenging issue with false overshoot in simulation
convergence has been identified.
The network model developed by Pessoa has been refined in Section 3.1.2 [47].
The structure of the network model, particularly the boundary conditions have been
explicitly defined. New methods have been put forward for determining the network
resistor values. For laminar flow, the resistance values can be computed from one
run of the lubrication theory model, speeding up the calibration process. For tur-
bulent flow, resistor values can be determined without any calibration, as opposed
to requiring two manual and empirical calibration steps, one for laminar flow, and

145
one for turbulent flow. This improvement has been made possible by the insight that
the high-Re losses are primarily driven by mixing dissipation of kinetic energy at the
seal's outlet. The laminar and turbulent theories can now be combined using a new
metric for predicting the transition between the two flow regimes. The long orifice
model network model was then used to design a water-moving PCP, in Section 4.2,
which matched predicted performance to within 20% over speeds ranging from 600
to 2,300 rpm and pressures ranging from 20 to 150 kPa.

In Chapter 2, the lubrication theory model has been extended to account for
imperfections in the surface of the rotor and imperfections in the trajectory of the
rotor. This extension has two implications. First, it verifies the accuracy of lubri-
cation theory for low Re PCP operation. The difference between Andrade's model
and Gamboa's experiments can be described in the uncertainty around rotor motion
alone [3, 19]. I then re-created Gamboa's experiment with a constrained rotor trajec-
tory and precisely measured the surfaces of the rotor and stator in Section 4.3, with
agreement to within geometrical and flow measurement error. This result validates
the lubrication theory model. The method for directly computing the torque on the
rotor using the lubrication theory model is described in detail in Section 2.3. When
combined with mechanical contact, and the performance of other pump elements, the
computations enable designers to estimate the mechanical efficiency of a PCP design.
The torque model is also validated against data from the rail bearing experiment
with qualitative agreement on the underlying mechanisms, although this work does
not fully separate out the non-hydraulic losses. Steps toward a turbulent version of
the lubrication theory model have been put forward in Section 3.2.1.

Insights from both experiments, the lubrication theory model structure, and the
turbulent network model structure, has led to a proposed set of PCP scaling laws for
understanding the trade-off between flow, pressure, speed, and total efficiency in both
turbulent and laminar operation. The scaling laws are verified by experimental data
currently available to the author. Although these scaling laws need more extensive
validation, they provide a new toolset for talking about rigid PCP design. PCP design
engineers can now better understand the characteristics of their built pumps. These

146
scaling laws provide a new tool for understanding how to operate a PCP. They also
provide a structured framework for defining technical goals, i.e. reduce the parasitic
torque, or increase the slip sealing performance.
To enable long operating life with inexpensive plastic components, and/or high-
speed operation, a novel hypocycloidal crank-shaft bearing is presented in Section 4.1.
Three embodiments of the hypocycloidal crankshaft bearing (roller, rail, and flexural)
have been prototyped. Both the roller and rail embodiments have been implemented
in working PCP prototypes.

5.2 Future Work

The goal of this work was to advance the understanding of PCPs such that a design
engineer could create a cost-effective water-moving PCP. While this thesis represents a
significant advance towards that goal, more work is necessary to make the technology
approachable for pump designers. This challenge is a rich opportunity for research in
modeling and mechanism design.
Given the success of compliant PCP stators in artificial lift for oil, analytical tools
for high Re flows are very interesting for water PCP design. The lubrication theory
model needs to be coupled with a compliant stator model to couple the fluid flow
with structural deformations. Because the lubrication theory model is only directly
applicable to low Re flows right now, the compliant PCP model will need extension for
high Re flows. The simplest approach would be to use the low Re simulation to predict
unique K values for the sealing line using a methodology based on Section 5.2. Those
K values can be used in a long-orifice network model, Section 3.1.4, to estimate the
sealing performance of the compliant PCP. The process for computing K will need to
be modified to account for the possibility that each sealing line in the PCP is unique.
The long orifice network model appears to work quite well for rigid stator PCPs, but
it may be necessary to implement a general turbulent lubrication model, as described
in Section 3.2.1, to get sufficient predictive capability. As the general turbulent PCP
lubrication theory model is developed, the cell-averaged inertial terms need to be

147
considered as well. If the losses are mixing driven, inertial terms may capture high
Re behavior without a need to include a computationally expensive RANS closure.
This work and other investigations assert both that the rotor motion deviates from
a perfectly centered orbit and that those error motions do not have negligible effects
on PCP behavior. While some full-CFD simulations for rigid stator PCPs account
for this rotor motion [5], the rotor motion should be deterministically coupled with
the rotor error motion lubrication theory model presented in Section 2.2.3.
The pressure loss coefficient, Cp, for the long orifice network model is currently
fixed at 0.8. While this model has produced satisfactory results in the three water-
moving PCPs studied in this thesis, a more in-depth model for C, would improve
confidence in the model and improve the scientific understanding of how high Re
PCPs behave.
Similarly, the PCP affinity laws proposed in Section 4.3.5 do not have enough
experimental support to be properly recognized as affinity laws by the field. Although
they are inspired by rigorous analytical models, much more experimental verification
is necessary. Such experiments might show that the model is inaccurate, or cast light
on other phenomena that need to be considered. Furthermore, this experimental work
informs the scaling of the affinity law coefficients, which would be very informative
to PCP designers.
While this thesis proposes a new hypocycloidal bearing for constraining rotor
motion and preventing contact with the stator, a bearing with sufficient mechanical
efficiency to enable a >60% efficient water PCP was not developed. A focused effort
on modeling the friction, drag, and parasitic torques associated with hypocycloidal
bearings would allow for PCP designers to balance cost, pump life, and efficiency.
Furthermore, modeling the mechanical efficiency losses associated with rotor-stator
contact will dictate when is it even appropriate to use a hypocycloidal PCP bearing.
By eliminating contact between the rotor and stator, new rotor designs such as pres-
sure compensation, surface treatment, or even hair could increase sealing performance
with little additional cost.
These models, experiments, and design tools are critical for efficiently iterating

148
on pump designs. Without this understanding, the pump community will be missing
a promising solution for fluid systems across the world.

149
Calibration and Spec Sheets

150
 macnaught

Flow Meter Test Report

zione della Metroligio di Calibrazione Rapport Putrologique d'Otolonnoge Bericht der Kolibrierungs-Messkunde
PTnforme de Metroligio de Calibrocidn Relatdrio de Metroligio de Calibragem

Report No. RelaIOne No, Rapport No Informe No Bericht Nr. RelatrioN' 513447

Model Modello Modile Modelo Modell Modelo MX2SF-2SA

Serial No. N. de Serie No. de Sirie No. de Serie Serien Nr. N9 de Sdrie D513447
Test Media Mezzo di Colldo Methode de Preuve Media poro ta Pruebo Testmedien Mitodo do Provo

Calibration Fluid 6 cPs
meets Exxsol D130 requirements
Calibrazione fiqvido 6cPs conforme ot requisiti Exosol D130
Fluide d'talonnoge 6cPs conforme oux exigences Exysol D130
Kolibrierungsfiussigkeitentspricht den 6 cPs Exxsol D130 onforderungen
Fluido de calibroci6n6 cPs reoune los requisitaos del Eaxsol 0130
Fluiwda de Coalbrogem 6 cps compre com as easgincias do Exosol 0130

Risujltot, Colhbrozione Rdsultato d'drolannoge Kalibrnerwngseesultate

Test Results Resullados Coiro
ReoultaO de Calibraci6n Resultodosdo
n Resultads do Colibrogem
US Gallons per Minute
Test No. 1 2 3 4 5 Colloudo Pruebo Proao

15.9 - - velocitd del Flusso Toux de Debit Velocidad de Flujo

Flow Rate -
Durchjlulmenge Ritmo do Fluxo
K-Factor 136.81 - - - - FttoreK coeffirent Faktor K
(pulses per Litre or Gallon)
29.77 - - - Volume di pravo Volume d'essai
Test Volume
Gemessene Menge

Accuracy Of Reading (%) +/- 0.50 - - - Prectsione erecision Exactftuo

Genouigkeit de Exactiddo

Pressure Drop (kPa) 22 - - - - AP (Delta P)

Toraturo Iquido Fluide d'etolonnage FluidO de coalbracion

Fluid Temperature (*C) 24.0 - - - -
Kolibsierungsfluftd keat Fluido do Colbragem

Tested By: Arvind.M Signatory: A. Ziaei

Date: 24/11/17 Date: 12/12/2017

The tests, measurements or calibrations covered by this document have been performed by volumetric comparison with
master meters. The master instruments used for testing meet 1S017025
and are traceable to international standards.

Macnaught Pty Ltd, 41.49 Henderson Street, Turrella, NSW 2205 Australia.

151
152
153
154
155
 Definition of Terms N

Output Voltage Range of output voltage to which the sensor is calibrated. Any output voltage that exceeds this
range may give an inaccurate measurement.
Output Current Range of output current to which the sensor is calibrated. Any output current that exceeds this
range may give an inaccurate measurement.
Range Total displacement range over which the sensor is calibrated.
Near Gap Absolute gap from the probe surface to the target at the point of the calibrated range which is
nearest the target
Bandwidth When measuring a vibrating target, the vibrating frequency at which the sensor will indicate ap-
proximately 70% (-3dB) of the actual displacement/vibration. For example, if the bandwidth is
1kHz, and the target is vibrating 1.0mm at 1 kHz. the measurement will indicate about 0.7mm.
As frequency increases beyond the bandwidth, the output will continue to decrease.
Sensitivity Change in output voltage caused by a given change in displacement, measured between the
endpoints of the calibrated range.
For example. if the Output Voltage is 0-10 Volts and the Range is 1mm, then the Sensitivity is
1OVolts per 1.0mm
Resolution Resolution indicates the smallest reliable measurement possible. Resolution is a function of the
inherent electrical noise in electronic amplifiers. For this reason, resolution is simply a measure-
ment of the noise voltage of the probe driver. Because electrical noise is affected by bandwidth,
resolution is also bandwidth dependent. Resolution measurements on this calibration sheet are
taken at bandwidth indicated in the calibration information.
Peak-to-Peak Peak-to-peak resolution (output noise) is measured as the maximum peak-to-peak output voltage
Resolution occurring over a period of time. long enough to include low frequency components. with the
probe fixed on a stationary target.
RMS Resolution RMS resolution (output noise) is the standard deviation of output voltages sampled over a period
of time. long enough to include low frequency components. with the probe fixed on a stationary
target.
Linearity Error Worst-case deviation of the calibration data points from a straight line (least squares fit of the
data points). If all the points of the calibration lie on a straight line. the linearity error is zero.
Linearity error is reported as a percent of full scale. For example, if the worst-case deviation from
the best fit line is 0.001mm and the measurement range is 1mm, the linearity error is: 0.001mm/
(1mm) x 100%= 0.1% of full scale.
Linearity Error is not affected by sensitivity errors or DC offset errors and is therefore only one
component in the determination of "accuracy."
Error Band Worst-case deviation of the actual calibration points from the expected (ideal) calibration points.
For example, if the worst-case error in a 1mm range calibration is 0.005mm, the Error Band
is: (0.005mm) i (1mm) x 100% = 0.5% of full scale. The result is reported as a bipolar range:
0.5% of full scale in this example. Error band is a more complete prediction of gauge accuracy
than linearity.
Combined Accuracy of the calibration device considering the potential error sources of the device itself and
Uncertainty of the NIST traceable laser interferometer that is used to calibrate it.
Calibration
Environmental Temperature and humidity are monitored and recorded on the calibration certificate.
conditions
Equipment IDs Identification numbers for all the equipment used for calibration are recorded for NIST traceability
reasons.
Calibration There are different methods for performing calibrations depending on the specific equipment and
Procedure ID details of the calibration. This lists the specific procedure used to perform the calibration.

156
 34Y214S-LW8 w/M LA1 0641 ,Div by 5,ParaIleI
1000- 500
900 - 450
800 -- 400
700 - - 350
- TORQUE

-
600 - - POWER - 300
-500 - - 250%

1 400 - 200
----
o02 -
-150
200 - - 100
100- -50

0 0
0 5 10 15 20 25 30 35 40 45 50
SPEED (RPS)

Figure -1: The torque speed curve for the 34Y214 motor wired in parallel.

157
158
Bibliography

[1] All India Report on Input Survey 2011-12, 2013.

[2] P E Allaire. Design Of Journal Bearings For Rotating Machinery. page 22.

[3] Selma F. Andrade, Juliana V. Valrio, and Mrcio S. Carvalho. Mecnica Com-
putacional, Volume XXIX. Number 87. Application of Computational Methods
in Petroleum Engineering (A). 2010.

[4] I.R. Belcher. An Investigation Into The Operating Characteristics of the Pro-
gressive Cavity Pump. PhD thesis, Cranfield Institute of Technology, May 1991.

[5] Mikael Berton, Olivier Allain, Caroline Goulay, Pierre Lemetayer, and others.
Complex fluid flow and mechanical modeling of metal progressing cavity pumps
PCP. In SPE Heavy Oil Conference and Exhibition. Society of Petroleum Engi-
neers, 2011.

[6] Gregory Bixler, Stacey Stathulis, Thomas Haubert, Daniel Lakovic, and Ga-
jan Sivandran. The LifePump Innovation for Developing Countries. Journal

-
American Water Works Association, 107(5):48-54, May 2015.

[7] Sunil Chandel, S. N. Singh, and V. Seshadri. A Comparative Study on the

Performance Characteristics of Centrifugal and Progressive Cavity Slurry Pumps
with High Concentration Fly Ash Slurries. ParticulateScience and Technology,
29(4):378-396, July 2011.

[8] A. Chasalevris and D. Sfyris. Evaluation of the finite journal bearing charac-
teristics, using the exact analytical solution of the Reynolds equation. Tribology
International, 57:216-234, January 2013.

[9] Jie Chen, He Liu, Fengshan Wang, Guocheng Shi, Gang Cao, and Hengan Wu.
Numerical prediction on volumetric efficiency of progressive cavity pump with
fluidsolid interaction model. Journal of Petroleum Science and Engineering,
109:12-17, September 2013.

[10] Young-Do Choi, Junichi Kurokawa, and Jun Matsui. Performance and Internal
Flow Characteristics of a Very Low Specific Speed Centrifugal Pump. Journal
of Fluids Engineering, 128(2):341, 2006.

159
[11] Victor WF de Azevedo, Joo A. de Lima, Emilio E. Paladino, and others. A 3d
Transient Model for the Multiphase Flow in a Progressing-Cavity Pump. SPE
Journal, 2016.

[12] Angus Deaton and Jean Drze. Food and Nutrition in India: Facts and Interpre-
tations. (7):24, 2009.

[13] Dennis Denney and others. Pressure Distribution in Progressing-Cavity Pumps:

Performance and Run Life. Journalof Petroleum Technology, 64(07):85-87, 2012.

[14] M. F. Edwards, M. S. M. Jadallah, and R. Smith. Head Losses in Pipe Fittings at

Low Reynolds Numbers. Chemical Engineering Research and Design, 63(1):43-
50, January 1985.

[15] V.G. Fester, D.M. Kazadi, B.M. Mbiya, and P.T. Slatter. Loss Coefficients
for Flow of Newtonian and Non-Newtonian Fluids Through Diaphragm Valves.
Chemical Engineering Research and Design, 85(9):1314-1324, January 2007.

[16] P. Flores, J. Ambrsio, J. C. P. Claro, H. M. Lankarani, and C. S. Koshy. Lubri-

cated revolute joints in rigid multibody systems. NonlinearDynamics, 56(3):277-
295, May 2009.

[17] UN High Commissioner for Refugees (UNHCR). The Sustainable Development

Goals and Addressing Statelessness, March 2017.

[18] Jose Gamboa, Aurelio Olivet, and Sorelys Espin. New Approach for Modeling
Progressive Cavity Pumps Performance. SPE International, October 2003.

[19] Jose Gamboa, Aurelio Olivet, Juan Iglesias, and Pedro Gonzalez. Understand-
ing the performance of a progressive cavity pump with a metallic stator. In
Proceedings of the 20th InternationalPump Users Symposium, 2002.

[20] Jose Gamboa, Aurelio Olivet, Juan Iglesias, and Pedro Gonzalez. Understanding
the Performance of a Progressive Cavity Pump with a Metallic Stator. 2003.

[21] Fred Glover, OptTek Systems, Leon Lasdon, John Plummer, Abraham Duarte,
Universidad Rey Juan Carlos, Rafael Marti, Universidad de Valencia, Manuel La-
guna, and Cesar Rego. Pseudo-Cut Strategies for Global Optimization. page 13,
2011.

[22] E. M. Greitzer, C. S. Tan, and M. B. Graf. Internal Flow: Concepts and Appli-
cations. Cambridge University Press, Cambridge, 2004.

[23] Jennifer Holthaus, Bikash Pandey, Robert Foster, Bernard Ngetich, James
Mbwika, Evgenia Sokolova, and Philip Siminyu. Accelerating Solar Water
Pump Sales in Kenya: Return on Investment Case Studies. In Proceedings of
SWC2017/SHC2017, pages 1-10, Abu Dhabi, 2017. International Solar Energy
Society.

160
[24] Jonathan Hopkins. Design of Parallel Flexure Systems via Freedom and Con-
straint Topologies (FACT). Master's thesis, Massachusetts Institute of Technol-
ogy, June 2007.

[25] Yukio Hori. Hydrodynamic Lubrication. Springer, Tokyo ; New York, 2006.
OCLC: ocm63518748.

[26] Meng-Hui Hsu, Hong-Sen Yan, Jen-Yu Liu, and Long-Chang Hsieh. Epicycloid
(Hypocycloid) Mechanisms Design. Hong Kong, page 6, 2008.

[27] I. E. Idelchik and Erwin Fried. Handbook of hydraulic resistance. Washington

Hemisphere Pub. Corp., c1986., 1986.

[28] Igor J. Karassik, editor. Pump handbook. McGraw-Hill, New York, 3rd ed edition,
2001.

[29] Steven T. Klein. Development of Composite Progressing Cavity Pumps. SPE

Eastern Regional Meeting, October 2002.

[30] Emily Kunen, B. Pandey, Robert Foster, Jennifer Holthaus, Binod Shrestha,
and Bernard Ngetich. Solar Water Pumping: Kenya and Nepal Market Accel-
eration. In Proceedings of the ISES Solar World Congress 2015, pages 1-12,
Daegu, Korea, 2015. International Solar Energy Society.

[31] Smt G S Lakshmi. Energy Statistics. page 121, 2017.

[32] Eric Lengyel. Mathematics for 3D game programming and computer graphics.
Course Technology, Cengage Learning, Boston, MA, 3rd ed edition, 2012.

[33] Pierre F. J. Lermusiaux. Numerical fluid mechanics. MIT OpenCourseWare,

May 2015.

[34] ADRIAN J. LEW, Gustavo C. Buscaglia, and Pablo M. Carrica. A Note on the
Numerical Treatment of the k-epsilon Turbulence Model*. InternationalJournal
of ComputationalFluid Dynamics, 14(3):201-209, 2001.

[35] Y. Li and F. Hsieh. Modeling of flow in a single screw extruder. Journal of Food
Engineering, 27(4):353-375, January 1996.

[36] R. Moineau. A New Capsulism. PhD thesis, The University of Paris, 1930.

[37] K.R. Mrinal, Md. Hamid Siddique, and Abdus Samad. A Transient 3d CFD
Model of a Progressive Cavity Pump. Proceedings of ASME Turbo Expo 2016:
Turbomachinery Technical Conference and Exposition, June 2016.

[38] Brennan Jim Nelik, Lev. Progressing Cavity Pumps, Downhole Pumps and Mud-
motors. Gulf Publishing Company, 2005.

161
[39] T. Nguyen, H. Tu, E. Al-Safran, and A. Saasen. Simulation of single-phase liquid
flow in Progressing Cavity Pump. Journalof Petroleum Science and Engineering,
147:617-623, November 2016.

[40] Tan Nguyen, Eissa Al-Safran, Arild Saasen, and Olav-Magnar Nes. Modeling the
design and performance of progressing cavity pump using 3-D vector approach.
Journal of Petroleum Science and Engineering, 122:180-186, October 2014.

[41] John C Nicholas. Hydrodynamic Journal Bearings - Types, Characteristics and

Applications. page 22.

[42] Emilio Paladino, Joao A. Lima, Paulo AS Pessoa, and Rairam Almeida. A Com-
putational Model for the Flow Within Rigid Stator Progressing Cavity Pumps.
78(1):178-192, 2011.

[43] Emilio E. Paladino, Rairam Francelino Cunha de Almeida, and Felipe Pin-
heiro Mota Assmann. Mesh Generation For Numerical Simulation Of Fluid-
Structure Interaction Within Progresing Cavity Pumps. 2009.

[44] Emilio E. Paladino, Joo A. de Lima, Paulo AS Pessoa, and Rairam FC Almeida.
Computational Three Dimensional Simulation of the Flow Within Progressing
Cavity Pumps. In 20th InternationalCongress of Mechanical Engineering, Gra-
mado, RS, Brazil, 2009.

[45] Robert M. Panas and Jonathan B. Hopkins. Eliminating Underconstraint in

Double Parallelogram Flexure Mechanisms. Journal of Mechanical Design,
137(9):092301, September 2015.

[46] William Perry and Jane Lubchenco. NAE Grand Challenges For Engineering
Committee. page 56.
[47] Paulo AS Pessoa, Emilio E. Paladino, and J. A. Lima. A simplified model
for the flow in a Progressive Cavity Pump. In 20th International Congress of
Mechancical EngineeringCOBEM, Gramado/RS, Brazil, pages 15-20, 2009.

[48] G. Samuel and Ken J. Saveth. Optimal Design of Progressing Cavity

Pumps(PCP). Journal of Energy Resources Technology, 128(4):275, 2006.

[49] Kevin Simon. Applications of Design For Value to Distributed Solar Generation
in Indian Food Processing and Irrigation. PhD thesis, Massachusetts Institute
of Technology, Cambridge, MA, June 2015.

[50] Yujie Song and Houzhen Wen. Research on the Line Style of Shaft Housing Pairs
of Progressing Cavity Pump. In ASME 2010 29th International Conference on
Ocean, Offshore and Arctic Engineering, pages 669-672. American Society of
Mechanical Engineers, 2010.

[51] P. Spalart and S. Allmaras. A one-equation turbulence model for aerodynamic

flows. American Institute of Aeronautics and Astronautics, January 1992.

162
[52] Tat Joo Teo and Alexander H. Slocum. Principle of elastic averaging for rapid
precision design. Precision Engineering, 49:146-159, July 2017.

[53] Virginia Torczon. On the Convergence of Pattern Search Algorithms. SIAM

Journal on Optimization, 7(1):1-25, February 1997.

[54] G Vetter and W. Wirth. Suitability of Eccentric Helical Pumps For Turbid Water
Deep Well Pumping in Photovoltaic Systems. Solar Energy, 1993.

[55] Gerhard Vetter, Ralph Kiebling, and Wolfgang Wirth. Abrasive Wear in Pumps
- A Tribometric Approach to Improve Pump Life. Proceedings of the Thirteenth
InternationalPump Users Symposium, pages 143-158.

[56] Gerhard Vetter and D. Paluchowski. Modelling of NPSHR For Progressing Cav-
ity Pumps. ASME Fluids Engineering Division Summer Meeting, June 1997.

[57] Gerhard Vetter and Wolfgang Wirth. Understanding Progressing Cavity Pumps
Characteristics And Avoid Abrasive Wear. Proceedings of the Twelfth Interna-
tional Pump Users Symposium, pages 47-59.

[58] Pierre-Louis Viollet. Water Engineering in Ancient Civilizations: 5,000 Years

of History. CRC Press, July 2007. Google-Books-ID: dEjLBQAAQBAJ.
[59] Peter Ward and William Dunford. Solar Powered Groundwater Pumping for
Medium Heads. Challenges in African Hydrology and Water Resources, 144,
1984.

[60] Peter Ward, William Dunford, and David Pulfrey. Performance of Small Pro-
gressive Cavity Pumps with Solar Power. CanadianJournalof Civil Engineering,
pages 284-287, 1987.

[61] Timothy J. Wray and Ramesh K. Agarwal. Low-Reynolds-Number One-Equation

Turbulence Model Based on k- Closure. AIAA Journal, 53(8):2216-2227, April
2015.

[62] Stavros I. Yannopoulos, Gerasimos Lyberatos, Nicolaos Theodossiou, Wang Li,

Mohammad Valipour, Aldo Tamburrino, and Andreas N. Angelakis. Evolution of
Water Lifting Devices (Pumps) over the Centuries Worldwide. Water, 7(9):5031-
5060, September 2015.

[63] Lei Zheng, Xiaodong Wu, Guoqing Han, Huachang Li, Yi Zuo, and Dake Zhou.
Analytical Model for the Flow in Progressing Cavity Pump with the Metal-
lic Stator and Rotor in Clearance Fit. Mathematical Problems in Engineering,
2018:1-14, July 2018.

[64] X.Z. Zhou, G.C. Shi, G. Cao, C.L. Sun, Y. He, H. Liu, and H.A. Wu. Three
dimensional dynamics simulation of progressive cavity pump with stator of even
thickness. Journal of Petroleum Science and Engineering, 106:71-76, June 2013.

163