Fulltext

Pump as Turbine
Symmetry Prediction Method for Pump as

Turbine Characteristics
Sondre Skjoldli
Master of Energy and Environmental Engineering

Submission date: June 2018
Supervisor: Torbjørn Kristian Nielsen, EPT
Co-supervisor: Mdee Ombeni, UDSM
Norwegian University of Science and Technology

Department of Energy and Process Engineering
i
ii
iii
iv
Preface and acknowledgements
The following Master’s thesis has been carried out in the spring of 2018, and the work has been
done for the Waterpower Laboratory at the Norwegian University of Science and Technology
(NTNU). The project is also a collaboration with a Norad supported program for Pumps as Tur-
bines.
The authors would like to thank supervisor Torbjørn K. Nielsen for his valuable help and
guidance throughout the project, as well as the laboratory technicians for their countless ef-
forts. Additionally, Ph.D. candidate Carl W. Bergan has provided us with significant aid dur-
ing the experiments and testing procedure. Furthermore, we would like to thank co-supervisor
Mdee Ombeni for his contribution, as well as the technicians at the University of Dar es Salaam
(UDSM), Tanzania.
Trondheim, 2018-06-11
Øyvind Albert and Sondre Skjoldli
v
Abstract
As a cheap and available source of renewable energy, regular centrifugal pumps may be run in
reverse to act as hydro turbines. Pumps as turbines (PAT) are especially relevant for isolated
rural areas, or in developing countries, where efficiency is not necessarily the highest priority.
The main challenge, however, is to be able to pick a suitable pump for a given site, which coin-
cides with the available head and discharge. In order to make this decision, one has to be able
to predict how a certain pump will perform as a turbine. A lot of work has been done in the
past to establish such prediction methods, based on tests and empirical correlations. Many of
these methods only predict one optimal point of operation, but as pumps do not have any guide
vanes, it will be difficult to constantly operate a PAT at this specific point. Therefore, this may
be insufficient. With this in mind, and because of the inaccuracies of these methods, a new ap-
proach has been suggested - an approach in which full head-flow characteristics are predicted
based solely on the pump’s impeller geometry.
By assuming symmetry in the velocity diagrams of pump and turbine operation, a method
for establishing the PAT characteristic was made. For validation purposes, this method was ap-
plied to one specific pump. Experimental results revealed a slight asymmetry, and shortcomings
in the initial assumptions. Therefore, the model was modified thereafter, with the introduction
of an empirical constant. The predicted characteristic lacks some accuracy in comparison to the
real PAT characteristic found through testing, but its optimal point of operation is in line with
previous established prediction methods. Still, the modifications, and the empirical constant,
may be enhanced further through comprehensive testing of numerous pumps, of various type.
vi
Sammendrag
Som en billig og tilgjengelig fornybar energikilde, kan vanlige sentrifugalpumper brukes som
vannkraftturbiner. Å bruke en pumpe som turbin (PAT) er spesielt relevant for avsidesliggende
områder, og utviklingsland, hvor turbinens effektivitet ikke nødvendigvis er av høyeste prioritet.
Hovedutfordringen ligger i å velge riktig pumpe for et gitt anlegg, som sammenfaller med om-
rådets vannføring og trykk. For å kunne ta denne avgjørelsen, må man kunne forutsi hvordan
en spesifikk pumpe vil prestere som en turbin. Det er tidligere gjort mye arbeid for å etablere
slike predikasjonsmetoder, basert på tester og empiri. Mange av disse metodene predikerer kun
et optimalt driftspunkt, men ettersom pumper ikke har justerbare ledeskovler, er det vanske-
lig å operere en PAT i dette ene punktet. De kan derfor vise seg å være utilstrekkelige. På
grunn av dette, samt unøyaktighetene knyttet til disse metodene, har det blitt foreslått en ny
fremgangsmåte - en metode som predikerer en full turbinkarakteristikk basert på geometrien til
pumpens løpehjul.
Ved å anta symmetri mellom hastighetsdiagrammene i pumpe- og turbindrift, har en modell
blitt etablert for predikere PAT-karakteristikken. For å validere metoden, har den blitt anvendt
på en spesifikk pumpe. Eksperimentelle resultater avslørte en viss asymmetri, og mangler i
de initielle antagelsene. Modellen ble modifisert deretter, og en empirisk konstant ble intro-
dusert. Sammenlignet med den reelle test-karakteristikken, er den predikerte karakteristikken
noe unøyaktig. Likevel er det estimerte optimale driftspunktet i tråd med tidligere etablerte
predikasjonsmetoder. Modifikasjonene, og den empiriske konstanten, kan imidlertid blir videre
forbedret gjennom testing av flere ulike pumper.
List of Tables
2.1 Different prediction methods investigated by Williams (1994)[1]. . . . . . . . . . . . 7
3.1 Test-Pump 2 impeller geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 Pump impeller geometry and rotational speed of Pedrollo FG 32/160B. . . . . . . . 22
3.3 Total uncertainty for the experiment parameters. . . . . . . . . . . . . . . . . . . . . 27
4.1 Comparison of test and estimated values in pump operation. . . . . . . . . . . . . . 31

4.2 Comparison of PAT parameters - tests and predictions. . . . . . . . . . . . . . . . . . 36
B.1 Component errors in the calibration of an instrument. . . . . . . . . . . . . . . . . . 63

B.2 Calibration uncertainties for pressure transducer upstream of the PAT. . . . . . . . 64
B.3 Calibration uncertainties for pressure transducer downstream of the PAT. . . . . . 64
B.4 Calibration uncertainties of the flow sensor. . . . . . . . . . . . . . . . . . . . . . . . 65
B.5 Calibration uncertainties for the torque sensor in both pump and turbine operation. 66
viii
List of Figures
2.1 A centrifugal pump operating in a) pump mode and b) turbine mode [2]. . . . . . . 4
2.2 Typical performance curves of pumps and turbines, adapted from [3]. . . . . . . . 9
2.3 Illustration of the phenomenon presented by Jain and Patel (2013)[2]. . . . . . . . . 10
2.4 Obtaining the PAT velocity diagrams by symmetry (not to scale). . . . . . . . . . . . 11
3.1 PAT test set-up UDSM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2 PAT test rig at UDSM: Water enters the PAT from the left in the picture, and exits
in the foreground. To the right the shaft goes from the PAT to the generator, with a
torque transducer in between. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 a) Insertion of permanent magnets into the rotor - b) stator. . . . . . . . . . . . . . . 20
3.4 The pump impeller of Pedrollo FG 32/160B. . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Illustration of test rig constructed at NTNU, designed in Creo. . . . . . . . . . . . . 24
4.1 H −Q characteristic and efficiency curve from pump testing. . . . . . . . . . . . . . 30

4.2 H −Q characteristics, both predicted and acquired through testing. . . . . . . . . . 32
4.3 Pump as turbine efficiency curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4 Velocity diagrams from test results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 H − Q characteristics, improved predicted curves and the one acquired through
testing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
ix
Nomenclature
Acronyms
BEP Best Efficiency Point
CFD Computational Fluid Dynamics
PAT Pump As Turbine
Greek Symbols
α Angle between absolute velocity and peripheral velocity °
β Impeller blade angle °
η Efficiency
κ Guide vane opening degree
ω Angular velocity rad/s
ψ Machine coefficient
ρ Density kg/m3
σ Dimensionless throttling dependency of angular speed of rotation
τ Torque Nm
ω̃ Dimensionless angular velocity
Roman Symbols
h̃ Head correction factor
q̃ Discharge correction factor
A Cross-sectional area m2
B Impeller height m
c Absolute velocity m/s
x
Nomenclature xi
D Diameter m
g Gravitational acceleration m/s2
H Head m
h Dimensionless head
k Empirical constant
n Rotational speed rpm
Ns Specific speed
n ed Speed factor
P Power W
p Pressure Pa
Q Discharge m3 /s
q Dimensionless discharge
Q ed Discharge factor
t Time s
Tw t Time constant representing hydraulic inertia s
u Peripheral velocity m/s
v Relative velocity m/s
Z Height m
Subscripts
1 Runner inlet
2 Runner outlet
bep Best efficiency point value
h Hydraulic
in Refers to the inlet
m Meridional component
out Refers to the outlet
p Refers to pump operation
R Rated value
t Refers to turbine operation
u Tangential component
Contents
Preface and acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Sammendrag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1 Introduction 1
2 Theoretical background 3
2.1 The concept of a Pump as Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Advantages and disadvantages of a PAT . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Earlier work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Pump and turbine performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Symmetry method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Design equations and assumptions . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Producing the PAT’s H −Q curve . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Experimental preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
xii
CONTENTS xiii
3 Methodology 17
3.1 Experimental setup UDSM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Pedrollo FG 32/160B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Experimental setup NTNU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Data acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4.1 Pump operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4.2 Pump as Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Calibration and uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Results 29
4.1 Pump testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 PAT testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.1 Velocity diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3 Potential improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5 Discussion 37
5.1 Pump testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 PAT testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2.1 First modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2.2 Second modification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2.3 Assumptions and inaccuracies . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2.4 Additional challenges and further work . . . . . . . . . . . . . . . . . . . . . . 42
6 Conclusion 43
Bibliography 45
A Paper written for CRHTVIII’18 47
B Calibration and Uncertainty Analysis 63

B.1 Pressure sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
B.2 Flow sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
B.3 Torque sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
CONTENTS
B.4 RPM sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

B.5 Temperature sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
B.6 Uncertainty of derived values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
B.6.1 Pressure head . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
B.6.2 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
B.7 Uncertainty from measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B.8 Calibration certificates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
B.8.1 Calibration certificate Druck DPI601 . . . . . . . . . . . . . . . . . . . . . . . 70
B.8.2 Calibration certificate OPTIFLUX 2000 C . . . . . . . . . . . . . . . . . . . . . 71
B.8.3 Calibration certificate Druck PTX 1400 . . . . . . . . . . . . . . . . . . . . . . 72
B.8.4 Calibration certificate GE UNIK 5000 . . . . . . . . . . . . . . . . . . . . . . . 73
B.8.5 Calibration certificate HBM T22 for pump operation . . . . . . . . . . . . . . 74
B.8.6 Calibration certificate HBM T22 for turbine operation . . . . . . . . . . . . . 75
C LabVIEW 77
C.1 Front panel and block diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
D Matlab codes 81
D.1 PAT prediction and experimental results compared . . . . . . . . . . . . . . . . . . . 81
D.2 Pump experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
D.3 Test velocity diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
D.4 Error function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
E Risk assessment 93
CHAPTER 1
Introduction
Micro-hydropower has a great potential, and can be very beneficial in order to help electrify
isolated communities, with no connection to the power grid. This can especially be relevant
in developing countries, small villages, and hilly rural areas. Even though running costs of such
plants are low and affordable, the biggest hurdle to some of these communities is the high initial
investment cost [4]. One appropriate solution is to use a pump as a turbine (PAT), where a
regular centrifugal pump will be able to generate energy by being run in reverse.
Pumps are significantly cheaper than regular hydro turbines, they are simple and robust
machines, and easily available throughout the world. Also, because they are more widespread,
spare parts and qualified personnel to conduct repairs are also more available [5]. However, the
main challenge of installing a PAT is the difficulty of predicting how a certain pump will perform
as a turbine. Determining this performance is crucial, because unlike a turbine, a centrifugal
pump does not have any adjustable guide vanes. Therefore, if the running speed is fixed, a PAT
is only able to perform efficiently for one set of head and flow values [1]. In consequence, it is
critical to be able to pick a pump that correlates with the available head and discharge at the
planned PAT site.
Some researchers have developed prediction methods that relate the best efficiency point
1
CHAPTER 1. INTRODUCTION 2
(BEP) in pump operation to the one in PAT operation. However, it has been shown that these
empirical methods deviate by ±20% from experimental data [5]. In recent years, attempts of
predicting the pump as turbine performance have been made through computational fluid dy-
namics (CFD). Still, these results are not reliable unless they are verified against experimental
data. This has motivated a new approach, an approach in which the PAT performance is esti-
mated based on the pump’s impeller geometry.
This proposed prediction method is based on assuming symmetry in the velocity diagrams
of pump and turbine operation. Even though a slight asymmetry might be expected, its impact
is presumed to be low. However, this may be accounted for postliminary. Additionally, it differs
from many of the old empirical correlations, in the way that it predicts a full head-flow (H − Q)
characteristic, instead of only one best efficiency point. This is especially important, as it is hard
to constantly operate a PAT at its BEP. In order to optimize PAT usage an accurate prediction
model is necessary. The goal is therefore to verify the symmetry prediction method presented,
and to investigate the method’s validity. This will be done through experimental PAT tests in the
laboratory.
CHAPTER 2
Theoretical background
Some parts of the following section was developed for the paper Symmetry prediction method
for pump as turbine characteristics and presented at the 8th edition of the annual Symposium
on Current Research in Hydraulic Turbines (CRHTVIII’18) at Kathmandu University in March of
2018. However, the authors feel this is material which is necessary to present again, in order to
fully comprehend the concept of a PAT, as well as the developed PAT prediction method. The
paper may be read in full in appendix A.
2.1 The concept of a Pump as Turbine
In a time where renewable power production is becoming increasingly important, the motiva-
tion to reduce the costs as far as possible also gets a lot of attention. One appropriate technology
is to use a pump as a turbine. This will include using a regular centrifugal pump, and running it
in reverse, as illustrated in figure 2.1. In that way, one will be able to extract energy from a fluid
instead of putting energy into it. The idea is not at all new, but could prove to be increasingly
relevant and important for developing countries, and isolated rural areas, where the electricity
demand is ever growing [2].
3
CHAPTER 2. THEORETICAL BACKGROUND 4
Even though operational and running costs of a hydropower plant are low and affordable, the
high initial capital cost in the development of such a plant can truly be a barrier [4]. Additionally,
as large-scale generation is not always feasible, there is an increasing interest in small-scale
hydropower plants. In these small-scale plants, it can often be hard to justify the construction
costs in comparison to the total power generation possible [6]. It is in these situations a PAT can
show its full potential, and can prove to be a very suitable technology, because of its low price.
The price per kW produced by small-scale hydropower plants, are usually higher than that of
large hydropower plants [2]. Therefore, installment of a PAT could be essential in reducing these
costs.
Unlike hydraulic turbines, centrifugal pumps are mass produced, which in turn make them
a lot more economically viable. However, the solution is mostly relevant for power plants where
the efficiency is not of the highest priority, as a pump run in reverse never will achieve efficien-
cies of the same order as a turbine designed for specific conditions [2].
a b
Discharge Discharge Inlet Inlet
Impeller Impeller
Suction Outlet
Casing Casing
Figure 2.1: A centrifugal pump operating in a) pump mode and b) turbine mode [2].
2.1.1 Advantages and disadvantages of a PAT
If a centrifugal pump is to be used as a PAT, the advantages and disadvantages have to be care-
fully weighed up against each other. From an economic investment point of view, PATs have a
clear advantage, as centrifugal pumps are mass produced all over the world, and are manufac-
tured for a wide range of heads and flows. By being able to pick a centrifugal pump right off
the shelf, a PAT will be a significantly cheaper option than a turbine designed for specific con-
ditions. In other words, they are more accessible and investments costs are lower. What’s more
is that they are easy to install, easy to maintain, and that spare parts are easily available [4]. As
pumps are a more widespread technology than turbines, qualified people to conduct repairs
and maintenance are also more available.
It has been shown that in low capacity power plants, up to 500 kW , installment of a PAT may
reduce the capital costs by the order of 10-1, compared to a conventional hydro turbine [6]. As
investment costs of hydro turbines in this range often can be quite high, the payback period can
be reduced from as much as 15 years to 2 years, if a PAT is installed instead [2].
Additionally, the pump with its integrated motor may be used as a turbine and generator
set. Thus, instead of buying an expensive synchronous generator, one may apply the provided
induction motor as an asynchronous generator [4].
There are several disadvantages however. Even though Fernandez et al. (2004)[6] state that
the efficiency of a pump running in reverse has almost the same efficiency as in pump mode, it
will most certainly not perform as well as a custom made turbine. It may seem as if the pump
industry is not as concerned as the turbine industry, with gaining the highest efficiency possible.
Whereas a hydro turbine has a very smooth surface to minimize losses, a mass produced pump
will often have a high level of roughness on the impeller. Even though it may be cumbersome,
manual grinding and smoothing of the PAT’s impeller may prove advantageous and worthwhile.
Moreover, pumps experience the highest pressure at their discharge, where turbines experi-
ence the highest pressure at the intake. The seals in both pumps and turbines are designed for
a specific expected pressure distribution [7]. However, the pressure build-up the pump is de-
signed for does not necessarily equal the pressure drop when using it as a turbine. It is possible
that at the PAT outlet, some points will experience higher pressures than those for which the
seals were designed for. If so, the seal must be redesigned or reinforced at these points.
Another key thing is that a centrifugal pump, compared to a regular hydro turbine, does not
have any guide vanes. Guide vanes are a number of blades that can be adjusted in order to
increase or decrease the flow rate through the turbine [8]. The vanes are placed between two
parallel covers normal to the turbine shaft. Pumps do not have such guide vanes, meaning they
cannot control the flow in the same way as turbines. Whereas a regular hydro turbine can adjust
incoming flow, and is able to perform efficiently for a range of flow rates, a PAT does not have
this option [9]. This will enable regular hydro turbines to produce power at BEP, even through
varying conditions. By implication, the range of suitable operating flow rates is much more lim-
ited for a PAT. One solution to this problem is to have multiple PATs coupled in parallel. With
such a rig, one will be able to activate the number of pumps required to handle the incoming
flow rate most efficiently. Having said that, a single PAT may perform close to maximum effi-
ciency if a suitable pump is chosen for the given site conditions. Especially if the site has a close
to fixed water supply throughout the year.
The primary disadvantage of a PAT however, is the difficulty of predicting how a certain
pump will behave as a turbine [4]. Thus, if a centrifugal pump is to be picked right off the shelf,
one has to be able to accurately predict its performance. The main challenge therefore lies in
being able to pick a suitable pump for a given site, which has a turbine characteristic that coin-
cides with the available head and discharge. A lot of work has been done in the past to establish
such prediction methods, based on tests and empirical correlations.
2.2 Earlier work
As the main disadvantage of a PAT is the difficulty of predicting the turbine characteristics that
are needed for a given site, a lot of work has been done throughout the years to establish predic-
tion methods which calculates the performance [1]. The actual turbine performance must be
found through testing, but this is a costly and time consuming process, as well as it requires that
the pump has already been purchased. Therefore, many different empirical correlations have
been produced, which in turn produce a wide range of results. A common factor, however, is
that the optimal operating point of a PAT is higher in both head and flow, than that of the pump
it originates from [2]. This is to account for the head reduction caused by the different losses. In
recent years, turbine performance have been predicted through CFD, but these results are not
reliable unless they are verified against experimental data [5].
The two main empirical approaches that have been taken to predict turbine performance are
either by relating the head and flow ratios to the pump’s efficiency, or by using its specific speed
[1]. Williams (1994)[1] investigated in total eight different prediction techniques, and compared
the accuracy of the models. Table 2.1 gives an outline of the different methods, as well as the
Ht
appropriate relationship of the head correction factor h̃ = Hp
and the discharge correction factor
Q
q̃ = Q pt .
Table 2.1: Different prediction methods investigated by Williams (1994)[1].
Name of method/investigator Based on Head correction factor h̃ Discharge correction factor q̃
1 1
Childs BEP
ηp ηp
1 1
Hancock BEP
ηt ηt
1 1
Stepanoff BEP p
ηp ηp
1 1
Sharma BEP
η1.2
p η0.8
p
1 0.85η5p +0.385
Alatorre-Frenk BEP
0.85η5p +0.385 2η9.5
p +0.205
2.5 2.4
Schmiedl BEP −1.4 + −1.5 + 2
η hp η hp
Grover Specific speed 2.693 − 0.0229N st 2.379 − 0.0264N st

6 1.6
Hergt Specific speed 1.3 − 1.3 −
N st − 3 N st − 5
According to Williams’ study, the method of Sharma proved to be the most accurate one. The
study involved comparing the turbine prediction methods on 35 different pumps, with available
test data. As Sharma’s method proved to be the most accurate, it is the only one that will be
looked at here. He relates the discharge and head correction factors to the pump efficiency as
shown in equations 2.1 and 2.2, where H and Q is the best efficiency point values of head and
discharge, and subscripts p and t refer to pump and turbine mode, respectively.
Qt 1
q̃ = = 0.8 (2.1)
Qp ηp
Ht 1
h̃ = = 1.2 (2.2)
Hp ηp
Even though a few of the other methods proved competitive, Sharma’s method was found
to be the most accurate of the eight approaches. Still, 20 percent of the tested pumps fell out-
side what was said to be the “acceptable” prediction limits. Therefore, it will always be wise to
conduct thorough tests before installing a certain PAT [1].
Although the accuracy of the different correlations can be questioned, they may serve as a
rough guide when designing a PAT-site [10]. However, the questionable precision, along with the
large number of different pumps that need to be tested to create a trustworthy model, motivates
a new approach. An approach in which the turbine performance is determined solely based on
the pump’s geometry. Additionally, common for the eight different prediction models in table
2.1 is that they only predict one set of BEP values. They do not provide any answer of how head
and flow are related on either side of this point. As stated previously, because of the lack of guide
vanes, it will be more challenging to operate a PAT at its best efficiency point, than for a turbine.
Therefore, developing a new method that predicts a full turbine characteristic may truly prove
advantageous.
2.3 Pump and turbine performance
Figure 2.2 shows typical performance characteristics of pumps and turbines at constant rota-
tional speed. It describes the variation of flow with head, power and efficiency. As seen from the
graph to the left in the figure, the pump height decreases with increasing volume flow. In turbine
operation this relationship is reversed, where both variables are strictly rising. The highest value
of η indicates where the location of the best efficiency points are found in both modes. This is
the desired point of operation, and is described by Hbep and Q bep .
The P -curve shows typical power distributions for different flow rates. In pump operation
this line represents consumed power, while in turbine mode power is generated. The power
curve of the turbine naturally increases for increased head and flow, but the maximum efficiency
is not necessary where the power is at its maximum.
When operating a pump as a turbine it can be expected that the H −Q curve in turbine mode
is similar to that of a regular turbine [9]. Therefore, this curve is relevant and comparable when
considering the validity of the following PAT experiments.
n = constant n = constant
H H
P H P
η η
H
Hbep P
Hbep
P
η
η
Qbep Q Qbep Q
Figure 2.2: Typical performance curves of pumps and turbines, adapted from [3].
Another interesting aspect to note when investigating a pump running as a turbine, is what
was presented by Jain and Patel (2013)[2]. Figure 2.3 shows the characteristics of a PAT in both
pump and turbine mode, labelled with positive and negative flow rates, respectively. It also
includes the different losses a pump and turbine are subjected to. It was found that at zero flow,
and at constant but opposite rotational speed, there is a difference in pressure head between
the two operations. This gap is represented by the red line in figure 2.3. Intuitively, one would
expect these two characteristics to intersect at the ordinate axis, at the same head. However, the
magnitude of the various losses in turbine and pump mode are not necessarily equal, which
could be a reason for the head difference. This phenomenon could also have a meaningful
impact when trying to predict the transition from pump to turbine operation.
Figure 2.3: Illustration of the phenomenon presented by Jain and Patel (2013)[2].
2.4 Symmetry method
As described previously, the main challenge of a PAT is to predict the performance of a specific
pump in turbine mode. The prediction method presented here is based on assuming symme-
try in the velocity diagrams from pump and turbine operation. By this assumption, the inlet
diagram of a turbine can be directly mirrored from the outlet diagram of a pump. The same
procedure can then also be applied to find the turbine outlet diagram, from the pump inlet dia-
gram. In reality, the optimal operating point in PAT operation is anticipated to lie higher than in
pump operation, as already mentioned. Thus, a slight asymmetrical relationship might be ex-
pected. Still, symmetry is the initial assumption, as the asymmetry is expected to have a limited
impact on the prediction. However, if experiments prove otherwise, this can be accounted for
accordingly.
(a)
(b)
Figure 2.4: Obtaining the PAT velocity diagrams by symmetry (not to scale).
The two respective diagrams and their mirrored opposites are presented in figure 2.4a and
2.4b. It is important to note that with this approach the slip and losses, such as friction and shock
losses, are not directly accounted for. Slip would have altered the blade angle to the relative flow
angle, and consequently changed the velocity diagram in turbine mode. The omitting of the slip
however, may be justified as the effect of slip in turbine operation, compared to pump operation,
will be modest [3]. This, and further losses, will be a topic of discussion later.
Different from the methods presented previously, the symmetry method is not based on
either pump efficiency or specific speed. As it is solely dependent on the pump geometry, it re-
quires physical measurement of the impeller to acquire the necessary input. This is because the
impeller geometry data is seldom published by pump manufacturers. The necessary parame-
ters consist of inlet and outlet diameter, inlet and outlet height, and outlet blade angle. Also,
the rotational speed is needed, as well as an assumption of the inlet meridional velocity - an as-
sumption which also will be subjected to further discussion later on. From this, the BEP values
of head and flow for a pump running as a turbine can be calculated through trigonometry and
standard hydropower design equations.
2.4.1 Design equations and assumptions
As the prediction method is based on a symmetrical transition between parameters in pump

and turbine mode, it requires determination of the different velocity components, as well as
head and flow at the best efficiency point. These are calculated through trigonometry and stan-
dard turbine design equations, as presented by Brekke (2003)[11]. Firstly, the peripheral ve-
locity is found as u = ωr , where ω is the angular velocity, and r the radius. Then, the pump’s
inlet meridional velocity component at BEP, c m1,p , is assumed based on numbers given by Bye
(1967)[12]. This is an assumption which will be discussed in great detail later on. From this, and
because the β-angles in the geometry is already known, the other parameters in the velocity
triangle are calculated.
Further, a 10% acceleration is assumed through the runner in pump operation. Hence c m2,p =
1.1c m1,p . Then, when symmetry is applied to the velocity components in pump mode, the rela-
tionship is reversed, and turns to c m1,t = 1.1c m2,t .
To be able to predict the rated flow at a certain PAT’s best efficiency point, the meridional ve-
locity component, along with the cross-sectional area, gives an estimate through conservation
of mass. To predict the rated head, Euler’s equations may be applied. Euler’s pump and turbine
equations are defined as follows:
g Hp
η h,p = (2.3)
u 2,p c u2,p − u 1,p c u1,p
u 1,t c u1,t − u 2,t c u2,t

η h,t = (2.4)
g Ht
Where η h is the hydraulic efficiency, g is the acceleration due to gravity, and c u is the tangential
component of the absolute velocity.
Another assumption worth noting is the assumption of no rotation at the inlet of the pump
and the outlet of the PAT. With this swirl free assumption, equations 2.3 and 2.4 simplifies as
c u1,p = 0 and c u2,t = 0 [11]. Also, the symmetry method assumes a rather bold hydraulic pump
efficiency of η h,p = 1 in determination of the velocity components. This is also an assumption
which will be subjected to discussion in due time.
2.4.2 Producing the PAT’s H −Q curve
Up until now it is only the performance at the best efficiency point that has been addressed.
However, it is very much of interest to be able to predict how the PAT will perform at either side
of this point, as BEP operation is hard to maintain for a PAT with no guide vanes. This is possible
by manipulating the dimensionless momentum equation presented by Nielsen (2015)[13]:
dq q|q|
Tw t = h − 2 − σ(ω̃2 − 1) = 0 (2.5)
dt κ
where T w t is a time constant representing the hydraulic inertia. This value is not needed how-
ever, as the left hand side of the equation can be set to zero to find the stationary characteristic.
Furthermore, the dimensionless properties for flow, head and angular speed of rotation in equa-
tion 2.5 are defined as q = Q t /Q R,t , h = H t /HR,t and ω̃ = ω/ωR . Now, it can be assumed that the
PAT has its BEP at q = 1, h = 1, ω̃ = 1, and that the derivative of the hydraulic efficiency in Euler’s
turbine equation (2.4) with regards to ω is zero, ∂η/∂ω = 0. In that way, the PAT’s dimensionless
throttling dependency of angular speed of rotation, σ, is defined as follows:
η hR − ψ
σ= (2.6)
η hR + ψ
By setting the hydraulic efficiency, η hR = 1 for the BEP in equation 2.6, and defining the machine
coefficient as
u 2,t 2
ψ= , (2.7)
g HR,t
σ for the PAT can be found. Further, the opening degree of the turbine, κ, equals 1 at the best ef-
ficiency point, and is defined as seen in equation 2.8. Q R is the rated volume flow in the turbine’s
best efficiency point.
pQ t
2g H t
κ= Q R,t
(2.8)
p
2g HR,t
By varying the rotational speed n, and solving the stationary version of equation 2.5 for q, the
discharge Q t for different rotational speeds can be determined. IEC 62097 [14] specifies the two
dimensionless factors for speed and discharge, shown in equation 2.9 and 2.10 respectively.
nD 2,t
n ed = p (2.9)
g HR,t
Qt
Q ed = 2
p (2.10)
D 2,t g HR,t
From these two equations the turbine’s Q ed − n ed relationship can be established. The H − Q
curve can then be produced by modifying equations 2.9 and 2.10. H t is found by keeping the
rotational speed constant, and only varying the speed factor. Q t , on the other hand, is found for
changing values of Q ed and corresponding values of H t . The two final equations, which enables
graphing of the turbine’s H −Q relationship, are presented in equations 2.11 and 2.12.
nD 2,t 2
³ ´
n ed
Ht = (2.11)
g
2
p
Q t = Q ed D 2,t g Ht (2.12)
2.5 Experimental preparation
The only way to properly validate the aforementioned prediction method, is to conduct thor-
ough tests in the laboratory, where the performance of the PAT is tested for a range of heads
and flows. By producing the actual H − Q curve for a pump working in turbine mode, it will be
possible to compare the accuracy of the theoretical model to what happens in reality.
There are three essential test variables that are necessary to construct the PAT’s H − Q rela-
tionship. Head and flow are self explanatory, but the rotational speed is also important to mon-
itor, as it will need to be held constant for one specific H − Q curve. Furthermore, in addition
to validating the prediction model, it will be of great interest to measure the power output, and
investigate the actual PAT efficiency. The maximum efficiency is also important to investigate,
in order find the head and flow values at the best efficiency point.
In order to calculate the pressure head, Bernoulli’s equation (2.13) may be used. In equation
2.13 ∆p is the pressure difference, p i n − p out , ∆Z is the height difference, while A i n and A out is
the cross-sectional area of these two locations. Further, V is the velocity of the water flow, and ρ
is the water density.
µ ¶
2 1
Q
∆p ∆V 2 ∆p A 2i n −A 2out
H= + ∆Z + = + ∆Z + (2.13)
ρg 2g ρg 2g
This implies that in order to calculate the head, the pressure at two locations, the inlet and
outlet of the PAT, as well as the volume flow Q, is required. Thus, it is essential that the laboratory
test rig is equipped with two pressure sensors and a flow meter.
Furthermore, as it is of interest to measure the actual PAT efficiency, another relationship is
necessary. The power of the rotating shaft may be expressed as in equation 2.14, where τ is the
torque and ω is the angular velocity [11].
P = τω (2.14)
Additionally, the power may also be expressed by the use of equation 2.15 for a turbine, and
equation 2.16 for a pump. With this, the efficiency, η, of the system can be calculated for both
turbine and pump mode as long as the torque, τ, is known. Therefore, it is necessary to mount
a torque transducer on the shaft between the pump and electric motor, to register and log the
moment of force.
P = η t ρgQ H (2.15)
ρgQ H
P= (2.16)
ηp
16
CHAPTER 3
Methodology
With the purpose of validating the symmetry prediction method, laboratory tests were sched-
uled to be conducted at the University of Dar es Salaam (UDSM), Tanzania, in January 2018. At
UDSM we would aid co-supervisor Mdee Ombeni in preparing a test rig, equip it with the nec-
essary measuring equipment, and obtaining the required data. As previously stated, the PAT’s
H −Q relationship is of utmost importance for validation purposes, but investigation of the PAT
efficiency will also be of interest.
3.1 Experimental setup UDSM
The pump available for PAT testing at UDSM, hereby referred to as Test-Pump 2, was an old
pump with no known pump characteristics. The first step was to remove the spiral casing, and
manually measure the relevant impeller lengths and angles. The important pump dimensions,
as well as the operating conditions, are presented in table 3.1.
After that, the plan was to test the pump in PAT operation in a test rig erected by co-supervisor
Mdee Ombeni. The test rig was equipped with the necessary instrumentation, and a sketch of
the set-up can be seen in figure 3.1. A picture of a portion of the rig can be seen in figure 3.2.
17
CHAPTER 3. METHODOLOGY 18
Table 3.1: Test-Pump 2 impeller geometry.
Parameter Value Unit

Inlet diameter D 1,p 164.5 mm
Outlet diameter D 2,p 200 mm
Inlet height B 1,p 74 mm
Outlet height B 2,p 39 mm
Outlet blade angle β2,p 37.9 °(degrees)
Rotational speed n 1450 r pm
Figure 3.1: PAT test set-up UDSM.
To feed the PAT, a pump designed for flow Q = 350m 3 /h and head H = 10m was used. This
is what is labeled as F-Pump in figure 3.1. It has been estimated that the feed pump will need
approximately four times the power rating of the PAT, to be able to produce heads and flows high
enough above the BEP [9].
Moreover, pressure sensors were mounted on both the high and low pressure side of the
PAT, labelled P1 and P2 respectively. However, the possibility of calibrating the equipment was
somehow inadequate. For the pressure transducers, UDSM had access to a pressure calibration
unit, but at the time of our visit this piece of equipment was faulty. Hence, the manufacturer’s
calibration had to be trusted initially. The flow meter was located in the horizontal pipe section
between the feed pump and the PAT. This sensor was not re-calibrated either, and will not be
until UDSM receives the necessary calibration apparatus.
Figure 3.2: PAT test rig at UDSM: Water enters the PAT from the left in the picture, and exits in the
foreground. To the right the shaft goes from the PAT to the generator, with a torque transducer
in between.
Further, the torque transducer was mounted on the shaft between the PAT and the genera-
tor. In the absence of essential equipment to calibrate this, one would again have to rely on the
manufacturer’s calibration, for the time being. Also mounted on the shaft was a small piece of
reflector tape, in order to keep track of the rotational speed. As it is of interest to keep the rota-
tional speed constant when running the tests, and acquiring the PAT characteristics, an optical
sensor, which counted every passing of this small reflector tape, was introduced. By coupling
this to a clock function, this would serve as an adequate rpm-indicator.
In order to obtain and collect the data an Arduino Nano unit was used. An Arduino is an
open-source electronics platform, which made it possible to transform the analog measure-
ments into digital output [15]. This is a very cheap and available technology, and such a unit
was used to log all of the data from the sensors mentioned above.
As for the generator, the initial idea was to use the pump’s induction motor as an asyn-
chronous generator. This solution is far cheaper than acquiring an expensive synchronous gen-
erator, especially for sizes up to 30kW [4]. In addition to being the most cost efficient option,
(a) (b)
Figure 3.3: a) Insertion of permanent magnets into the rotor - b) stator.
it will also result in an easier installation, as the PAT and generator come as one unit. However,
in such a set up there may be problems in achieving excitation. A solution to this is to use an
arrangement of capacitors to introduce the required reactive power [4]. Another possibility, in-
troduced by co-supervisor Mdee Ombeni, was an approach in which the generator rotor was
equipped with four permanent magnets. Thus modifying the asynchronous induction machine
into a synchronous generator. This can be seen in figures 3.3a and 3.3b, which displays the rotor
with the permanent magnets, and stator, respectively. At this stage however, the output voltage
only reached values of between 5V and 7V , due to various complications. Therefore, in order
to fix this, three 100µF capacitors were used to achieve the expected voltage output of around
240V .
In the end, due to the problems encountered, too many uncertainties, and limited time, no
useful data was attained during the time at UDSM. The last straw, however, was the limitations
of the available feed pump and motor. Run at maximum capacity it was only able to deliver
about 120m 3 /h. In comparison, it was estimated that the flow had to be around 240m 3 /h in
order to reach the predicted best point of this specific PAT. Thus, no proper conclusions could
be drawn of the symmetry method’s accuracy. This was a setback for the project, which had to
take a new approach. The new plan included ordering a new pump, and to prepare a new test
rig for testing and validation at NTNU.
3.2 Pedrollo FG 32/160B
In order to conduct the tests at the waterpower laboratory at NTNU, a new pump had to be
acquired. It was desired that the pump had a size which allowed it to be easily transported to
UDSM after the tests, thus the decision fell on the modest sized Pedrollo FG 32/160B. The first
thing that had to be done was to remove the spiral casing, and manually measure the impeller
geometry. The geometry specifications relevant to the symmetry method can be found in table
3.2.
This pump is designed for two rotational speeds, 1450 and 2900 r pm. As mentioned pre-
viously, it is far more economical to use a pump’s induction motor as a generator when setting
up a PAT driven power plant. A criteria for doing so is to operate the PAT at the same rotational
speed as the induction motor is designed for. When investigating the symmetry method’s valid-
ity, running the pump at 2900 r pm required sensors and equipment of a higher range than what
was available during the testing period. Therefore, the choice landed on running the pump and
PAT at a speed of 1450 r pm. However, the characteristic curves and performance data supplied
by the manufacturer were only attainable at 2900 r pm, where the pump is said to have an ef-
ficiency of 58%. As the efficiency can expected to be of approximately the same order for both
rotational speeds, an assumption was made to assign the same efficiency, 58%, to the Pedrollo
FG 32/160B at 1450r pm as well. At this best efficiency point, the pump’s data sheet indicated a
discharge of Q p = 9m 3 /h and a head of H p = 5.75m.
After measuring the pump impeller geometry, the values of table 3.2 were inserted into
the symmetry method, producing the predicted PAT characteristic curve. Sharma’s prediction
method, as seen in equations 2.1 and 2.2, was also applied, with the input η p = 0.58.
Table 3.2: Pump impeller geometry and rotational speed of Pedrollo FG 32/160B.

Inlet diameter D 1,p 51 mm
Outlet diameter D 2,p 153 mm
Inlet height B 1,p 30.4 mm
Outlet height B 2,p 5 mm
Outlet blade angle β2,p 29 °(degrees)
Rotational speed n 1450 r pm
Figure 3.4: The pump impeller of Pedrollo FG 32/160B.
3.3 Experimental setup NTNU
After acquiring the new pump, the new test rig had to be built. The test rig was modelled in
Creo Paramterics 3D Modelling Software, as seen in figure 3.5. It was erected as the illustration
shows, and equipped with the necessary instrumentation. During testing the pump was run
both as a pump and in turbine mode as a PAT, hence, the flow direction differed. In turbine
operation the water came down from the top of the figure into a 600mm diameter pipe. From
here the cross-section of the piping decreases to a diameter of 200mm where the valve was
located. After the valve, the diameter was further coned down to 50mm, where the flow faced
a bend. Further, the water was transported rightwards in the figure, passing through the flow
sensor, before the cross-section once again was reduced to a diameter of 32mm, which matched
the inlet of the PAT. Then, the water passed through the impeller, before leaving the PAT through
a 50mm diameter pipe into the reservoir.
In pump operation, the flow direction, and the path of the water, was opposite. The water
was pumped up from the bottom reservoir and out through the top of the illustrated piping.
From here the water was transported along a piping system not showed in the figure, before it
was returned down into the reservoir.
The laboratory, where the tests were conducted, was equipped with a pressure tank, which
supplied water for PAT testing. For pump operation testing, the bottom reservoir was used. In
both modes the flow rate was controlled manually by using the valve. The flow rate was then
measured with the Optiflux 2000 C flow sensor. In order to establish corresponding pressure
head values to each flow measurement, the inlet and outlet pressures had to be logged at the
positions labelled as P 1 and P 2 in figure 3.5. The two pressure sensors consisted of one GE UNIK
5000 with a range of 0 − 5bar gage, and one GE Druck PTX 1400 designed for a gage pressure
0 − 4bar .
To be able to calculate the efficiency of the PAT and pump, the torque was measured with a
HBM T22/200Nm sensor, which was mounted with flexible couplings on the shaft between the
pump and the motor. Also, an rpm-sensor was mounted at the same location. This sensor was a
similar solution to what was used at UDSM, in which an optical sensor counted every passing of
a small reflector tape on the shaft. All the sensor data were monitored and logged in a specially
designed LabVIEW program, attached in appendix C.
To the right of the pump, a 3kW Lönne electric motor was positioned. This was further
coupled to a frequency converter, which made it possible to keep the rotational speed close to
constant throughout the testing.
Ø600
Valve
Ø200 2.0m
0.5m 1.5m
Ø50 Turbine operation

Ø32 P1
Flow sensor Pump operation Torque and

PAT rpm sensor Motor
P2
2.0m
Ø50
Figure 3.5: Illustration of test rig constructed at NTNU, designed in Creo.
3.4 Data acquisition
3.4.1 Pump operation
When testing the pump, and obtaining its H − Q characteristics, the amount of water pumped
up was controlled by turning the valve, located to the left in figure 3.5. By controlling the volume
flow, the corresponding head values could be calculated for different opening degrees.
As previously stated, the symmetry method’s input is the pump geometry, rotational speed
and an assumption of the inlet meridional velocity. In the development of the prediction method,
and in the paper prepared for the conference at Kathmandu University (appendix A), the as-
sumed inlet meridional velocity was based on numbers given by Bye (1967)[12]. For a modest
sized pump like the Pedrollo FG 32/160B, c m1,p was set a value of 2.5m/s. During testing how-
ever, it became clear that this assumption was way to large for a pump with flow rates of this
magnitude. Therefore, the assumption of c m1,p needed to be modified. The alternative solution
was to use mass conservation of the flow to estimate a more appropriate velocity component,
based on the flow at the pump’s BEP. As the inlet and outlet geometry was known, c m1,p could
be directly calculated as:
Qp
c m1,p = (3.1)
πB 1 D 1
By using the equation above, and inserting the geometry parameters from table 3.2 as well as
the rated flow given by the manufacturer, the meridional velocity component came out to be
c m1,p = 0.513m/s.
Another problem that arose during testing was complications with the torque sensor. The
output signal from the sensor was extremely unstable, and not anywhere close to expected
torque values. Based on previous experience in the laboratory, one possible reason for this was
said to be the frequency converter, which has a tendency to disturb the sensor signals. By not
being able to correctly measure the torque, and consequently the power, the efficiency and BEP
would be hard to find. Due to time limitations, a proper solution of making the torque sensor
work was not found. However, the frequency converter had an option of measuring the power,
which in turn was used to calculate the efficiency. It should be stressed that this efficiency is
not an entirely correct representation, as it contains the losses in both the motor/generator and
the frequency converter itself. Still, as the main intention was to find head and flow values at
BEP, and not necessarily accurate efficiency values, this solution would serve as a respectable
indication.
3.4.2 Pump as Turbine
During PAT testing, and similar to the testing in pump mode, the water flow was controlled by
adjusting the valve. The water was supplied from a pressure tank in the laboratory, which have
an available head of approximately 20m. It was important to achieve a pressure head of this
magnitude, as the PAT was not operative at 1450r pm for lower pressures. Similar to testing
in pump operation, the efficiency values were calculated based on the power measured by the
frequency converter.
3.5 Calibration and uncertainty
Before the tests could be performed, the appropriate sensors had to be calibrated. This calibra-
tion comes with an uncertainty. All measurements done during the tests also have uncertainties
due to inaccuracies in the instruments and random variations of the measured property. Be-
cause of this, it is always necessary to conduct an uncertainty analysis. A thorough description
of the calibration, and the uncertainty analysis, can be found in appendix B, along with the cal-
ibration certificates. The results from this analysis is summarized in table 3.3 below.
As already stated, due to complications with the torque sensor, the efficiency was not found
the conventional way. It was instead found through the power calculated by the frequency con-
verter, which will contain losses from both the motor/generator and the converter itself. Thus,
there are major uncertainties associated with the efficiency value η. However, as the value of
the efficiency is not the main concern, but rather the location of the best efficiency point, this
value will still be useful. To account for the uncertainty in the location of the BEP, an error was
assumed in the manual reading from the display on the frequency converter. The power mag-
nitude was varying considerably, hence a rather large uncertainty of ±5% was assumed.
Table 3.3: Total uncertainty for the experiment parameters.
Quantity Calibration Uncertainty Mean Measured Uncertainty Total Mean Uncertainty

P1 ±0.495% ±3.855% ±3.886%
P2 ±1.154% ±3.182% ±3.384%
Q ±0.231% ±0.003% ±0.231%
H − ±0.462% ±0.462%
η − − ±5%
As table 3.3 reveals, the uncertainty was also high in the two pressure sensors. It was ob-
served that the signals were quite unstable during testing, and the uncertainty depicts the ran-
dom errors from the high degree of variation in the system.
28
CHAPTER 4
Results
As the initial tests at UDSM in Tanzania did not produce any valuable outcome, the only ap-
propriate results are from the tests done at NTNU. Thus, the following results, which will be
discussed and used to investigate the symmetry method’s validity, is solely from the tests con-
ducted there.
4.1 Pump testing
By testing the Pedrollo FG 32/160B in pump operation, its H − Q characteristic was produced.
This can be seen in figure 4.1. Also seen in figure 4.1, is the pump’s efficiency curve with respect
to flow. It can be observed that both curves are similar in shape to what was expected from fig-
ure 2.2. By studying the power computed by the frequency converter, a rough estimate of the
maximum pump efficiency could be calculated. The estimate came out to be η p = 0.45. This
in turn supplied the head and flow values at the predicted BEP, H = 5.91m and Q = 8.95m 3 /s
respectively. Compared to the pump efficiency of η p = 0.58 given by the manufacturer, the effi-
ciency calculated from the frequency converter is far off. However, the BEP values of head and
flow are strikingly similar, as seen in table 4.1. The estimated value of head deviates 2.7%, while
29
CHAPTER 4. RESULTS 30
flow deviates by 0.6%, compared to the experimental values. This shows that even though the
efficiency is lower due to the various losses, it still serves as a good indication of where the BEP
lies.
Using the rated volume flow found through testing, and applying equation 3.1, the actual
c m1,p at BEP was calculated. This can be seen in comparison with the previously estimated
c m1,p -value, based on the rated values in the pump’s data sheet. The latter of which was used
as input to the symmetry prediction method. Both meridional velocities are also presented in
table 4.1, where the estimated value deviates 0.4% from the test value.
Figure 4.1: H −Q characteristic and efficiency curve from pump testing.

Table 4.1: Comparison of test and estimated values in pump operation.
Parameter Test value Estimated (given) value Unit

Rated head H p 5.91 5.75 m
Rated flow Q p 8.95 9 m 3 /h
Inlet meridional velocity c m1,p 0.511 0.513 m/s
Pump maximum efficiency η p 0.45 0.58 −
4.2 PAT testing
Similar to the pump testing procedure, the PAT’s H − Q characteristic was established through
adjusting the valve and incoming flow. This characteristic can be seen in figure 4.2, which
also follows the typical performance curve from 2.2. Moreover, in the same plot, the symme-
try method’s prediction is presented along with its predicted best efficiency point, marked with
a red asterisk. Additionally, the pump efficiency found in the data sheet of Pedrollo FG 32/160B
was inserted into Sharma’s prediction method equations (2.1 and 2.2), which produce a BEP as
marked by the pink asterisk. It is clear that the characteristic predicted by the symmetry method
is off-target, compared to the actual characteristic attained from the tests. Thus, investigation
of how to improve the method’s precision is required.
Figure 4.2: H −Q characteristics, both predicted and acquired through testing.
Figure 4.3 presents the efficiency calculated in PAT operation. The maximum efficiency is
estimated to be at η t = 0.395, which provides the BEP values of head and flow in figure 4.2.
It is important to note the uncertainty of the efficiency in figure 4.3, and that the location of
the best efficiency point in figure 4.2 could change in line with this uncertainty. It can also
be observed that the maximum efficiency of the PAT is below that of the pump. However, the
value must not be taken too literally, as the method of calculating the efficiency is subjected
to considerable losses. This was expressed in the previous section, where the pump maximum
efficiency estimated with this method, being 45%, and the efficiency given by the manufacturer,
58%, gave practically the same head and flow values.
Figure 4.3: Pump as turbine efficiency curve.
4.2.1 Velocity diagrams
Based on the measured best efficiency flow in both pump and turbine operation, impeller geom-
etry and operating conditions, the velocity diagrams could be constructed. The upper portion
of figure 4.4 shows the velocity diagram of the pump outlet, and the PAT inlet. Similarly, the
lower portion displays the pump inlet and PAT outlet. It can be observed that the diagrams are
not strictly symmetrical, as assumed in the symmetry method, and do not look exactly like the
diagrams in figure 2.4. The turbine meridional velocities are significantly larger than those in
pump mode. As c m1,t is larger than c m2,p , angle α1,t is also larger than α2,p , which makes the
mirroring of α not entirely acceptable. This is something that has to be accounted for in order
to improve the accuracy of the symmetry method.
Likewise, as the flow and meridional component is larger in turbine mode, c m2,t is also
greater than its pump counterpart. This is illustrated in the lower part of figure 4.4. Moreover,
the common assumption of no swirl at the pump inlet and turbine outlet is still enforced, which
eliminates any c u -components in this diagram. In reality, one might expect some rotation at the
PAT outlet, which in turn would give the flow a relative flow angle β02,t , and introduce a tangen-
tial component of the absolute velocity. An estimate of c u2,t can be made by the use of Euler’s
equation 2.4, if a good measure of the hydraulic efficiency is known.
Test velocity diagrams: Pump outlet, PAT inlet
c2,P
v2,P cm2,P
β2,P α2,P cu1,T u1,T
u2,P cu2,P α1,T β1,T
cm1,T
c1,T
v1,T
Test velocity diagrams: Pump inlet, PAT outlet
v1,P
cm1,P
β1,P u2,T
u1,P β2,T
cm2,T
v2,T
Figure 4.4: Velocity diagrams from test results.

4.3 Potential improvements
By looking at figure 4.4, and paying attention to the axes, it becomes clear the meridional veloci-
ties do not follow the initial assumption of c m2,p = 1.1c m1,p . One way to modify this assumption
is to utilize the fact that the flow rate is unchanged through the impeller. In that way the rela-
tionship between c m1,p and c m2,p is only dependent on the inlet and outlet geometry.
Qp
c m2,p A 2,p A 1,p πB 1,p D 1,p B 1,p D 1,p
= = = = (4.1)
c m1,p Qp A 2,p πB 2,p D 2,p B 2,p D 2,p
A 1,p
For this specific pump, the relationship turned out to be as in equation 4.2. By implication,
and through the assumptions of the symmetry method, the meridional components in turbine
mode has the inverse relationship.
c m2,p c m1,t
= = 2.03 (4.2)
c m1,p c m2,t
If the new relationship in equation 4.1 is applied to the symmetry method, the predicted
characteristic moves closer towards the H − Q curve from the experiments, as represented by
the dashed line in figure 4.5.
After this alteration it was also observed that the c m,t -components from the tests were greater
than the c m,p -components from the pump tests. To account for this, the prediction of c m1,t is
modified accordingly:
c m1,t = kc m2,p (4.3)
The constant k is an empirical value which accounts for the asymmetry from pump to turbine
operation. From this single PAT test, the value of the constant was found to be k = 2.005. The
predicted PAT characteristic with this additional improvement is shown in figure 4.5, repre-
sented by the solid blue line. It is seen that the flow predicted is a lot closer to the actual best
efficiency point than what was found earlier. On the other hand, the pressure head is still a ways
from the test BEP. This will be subjected to discussion later on.
A summary of the significant parameters are presented in table 4.2. It contains values from
the symmetry method, the real test values, as well as the new and improved values after the
second modification.
Figure 4.5: H −Q characteristics, improved predicted curves and the one acquired through test-
ing.
Table 4.2: Comparison of PAT parameters - tests and predictions.
Parameter Symmetry method prediction Improved symmetry method Test value

c m1,t 0.565m/s 2.089m/s 2.075m/s
c m2,t 0.513m/s 0.513m/s 1.024m/s
α2,p 3.05° 6.11° 6.06°
α1,t 3.05° 12.11° 14.76°
HR,t 12.53m 11.52m 18.62m
Q R,t 4.88m 3 /h 18.07m 3 /h 17.95m 3 /h
CHAPTER 5
Discussion
As already explained, the initial tests at UDSM did not produce any valuable results. However,
the tests granted us a beneficial understanding on how to plan the experiments that were con-
ducted at NTNU later on. Also, the visit proved to be a useful experience in order to understand
the conditions and limitations of doing scientific research in developing countries. Additionally,
the stay confirmed what has been suggested previously, that PATs can be a favorable technology
in areas where proper equipment and competence is scarce, and simpler and more robust solu-
tions are a necessity.
5.1 Pump testing
Even though the main goal of the testing was to verify how well the symmetry method predicted
the actual PAT characteristic, the Pedrollo FG 32/160B was first tested as a pump. This was done
in order to investigate if the real best efficiency point of the pump was the same as the one given
by the manufacturer, and to evaluate the c m1,p -assumption. What was found was that the two
best efficiency points were very much in accordance with each other. However, the location of
the BEP may of course vary in accordance with the uncertainty, but it is still within an acceptable
37
CHAPTER 5. DISCUSSION 38
range.
Initially, as an input to the symmetry method, c m1,p was assumed based on suggested values
given by Bye (1967)[12]. However, during the pump tests it was observed that the assumed value
was way to large, and had to be modified. As already mentioned, the new approach was then to
use equation 3.1 to calculate c m1,p based on the pump’s rated flow. As the test results were so
similar to what was stated as the pump’s BEP in the data sheet, the real c m1,p that was calculated
only differed about 0.4% from the one calculated in advance, and applied in the prediction. This
can be seen in table 4.1. Also seen in the table is that this modification influenced the α-angles,
and pushed the predicted values closer to the real test values. Therefore, in this case, it can be
justified to use the information from the pump manufacturer as input to the model.
Despite acquiring very similar BEP values, the calculated efficiency from the tests did not
match the pump’s rated efficiency. This was most likely because the efficiency was found through
the frequency converter, and not by the conventional approach of measuring the shaft torque.
As stated previously, measuring the power through the frequency converter contain the losses
from both the motor/generator and the converter itself, and is therefore not an entirely legiti-
mate representation. Still, as the BEP values turned out to be as similar as they were, it indicates
that the location of the maximum efficiency found was comparable.
5.2 PAT testing
After the pump tests, where the approach for finding c m1,p was revised, the characteristic in fig-
ure 4.2 was predicted. Also in the same plot, the characteristics from the experimental data is
presented, along with Sharma’s best efficiency prediction. It is clear that the symmetry method
predicts a curve, and a BEP, which is far below what was found during testing, and what Sharma
predicts. This is especially seen in the amount of volume flow. The point of highest efficiency
was determined by the relationship in figure 4.3, and came out to be 39.5% at a flow of 17.95m 3 /s.
In the region around this point, the efficiency varies little with respect to flow. Therefore, a tiny
shift in efficiency will result in a large shift in the flow, and consequently the location of the PAT’s
best efficiency point. This is especially important to note, considering the questionable way the
power was logged, and found, through the frequency converter.
It may also be observed that the best efficiency point predicted by the symmetry method
has a lower flow than the pump’s BEP, presented in figure 4.1. This is not in agreement with what
has been found by previous studies on pumps as turbines, where the PAT BEP is expected to be
higher in both head and flow, than the BEP in pump operation [6].
As the experimental BEP had a flow of more than three times the one predicted by the sym-
metry method, it was evident that some modifications had to be done. By investigating the
velocity diagrams in figure 4.4, it is obvious why the predicted flow is of the magnitude that it
is. Whereas the predicted curve is based on absolute symmetry between pump and turbine op-
eration, the velocity diagrams from the tests indicate a clear asymmetrical relationship. One
possible explanation for this relationship, could be because of the gap in head between pump
and turbine operation, seen in figure 2.3. The gap indicates that the transition between pump
and turbine mode is not necessarily straight forward, and that the symmetry assumption might
be a little weak. Thus, the goal became to find possible improvements which could strengthen
the model.
5.2.1 First modification
One of the assumptions done in the early stages of the development of the model, was that the
pump would accelerate the meridional velocity component by 10%. The velocity diagrams in
figure 4.4 reveal that the acceleration is significantly higher. Thus, the modification presented
in equation 4.1 was applied. This new way of determining the c m,p -ratio is in line with the sym-
metry method’s approach, namely to use impeller geometry data as the only input. The dashed
line in figure 4.5 represents the new predicted characteristic, with this modification applied. It
can be observed that the BEP flow is almost double to what it was before, but still only half of
what the tests indicate. At this point, the predicted BEP in turbine operation increases past the
BEP flow in pump operation, making the symmetry prediction model more analogous to what
has been found in previous studies. Still, this modification does not address the asymmetry in
the velocity diagrams. Hence, an additional modification was needed in order to account for
this.
5.2.2 Second modification
When studying figure 4.4, one can see that the reason for the asymmetry in the velocity diagrams
is the results of a much larger meridional velocity component in turbine BEP operation, com-
pared to pump operation. The reason could be because of the previous mentioned gap in figure
2.3. Consequently, it is not entirely correct to directly mirror c m2,p = c m1,t and c m1,p = c m2,t . In
fact, the c m,t -components turned out to be about two times larger than the c m,p -components.
To deal with this, the empirical relationship in equation 4.3 was enforced.
From this single PAT test, the value of the constant was found to be k = 2.005. However,
establishing a general empirical factor based on one test alone, is dubious at best. Consequently,
more PATs must be tested in order to conclude on a more universal constant k.
Introducing an empirical relationship like this defines a turning point the in development of
the symmetry model. By moving away from the symmetry assumption, the model’s name loses
credibility, and it takes a step towards the many already existing prediction methods presented
earlier. Still, the only input to the model remains the impeller geometry, but the value of the
constant k will need to be refined through further testing.
The most recent predicted characteristic, with the first and second modifications combined,
result in the solid blue line in figure 4.5. One can see that the latest modification accounts for
the difference in flow, and the newly predicted BEP is very similar in volume flow to the test BEP.
On the other hand, the predicted pressure head has remained the same value throughout, and
is about seven meters below the test’s best efficiency point.
At this stage it is of interest to draw a comparison between the symmetry model, and Sharma’s
prediction model. Sharma’s model, which Williams (1994)[1] considered to be the most accurate
existing model at the time, also predicts a head of the same range - far below what was found
during testing. Sharma’s predicted BEP is expressed by the pink asterisk in figure 4.5.
5.2.3 Assumptions and inaccuracies
One possible explanation of why Sharma’s and the symmetry prediction models fall short in
evaluating the BEP head, might be because of uncertainty in the location of the BEP itself. As
already mentioned, the highest efficiency was found in a region with little variation with respect
to flow. Because of this, a small variation in efficiency, will result in a significantly larger change
in flow. This change in flow will in turn result in an even larger shift in the BEP head. The reason
for this is the steepness of the test characteristic curve in figure 4.5. In short, an inaccurate mea-
sure of the efficiency will have a great impact on the rated head. Because of the large assumed
uncertainty in η, seen in figure 4.3, the flow could range from about 16.5 − 21m 3 /h. If the lower
limit is the case, the location of the experimental best efficiency point in figure 4.5 could move
all the way down to a head of approximately 15m. Thus, the experimentally measured head at
BEP moves closer towards what has been predicted by both Sharma, and the symmetry method.
Another factor that could impact the head prediction is one of the assumptions in the sym-
metry method. As described in section 2.4.1, a swirl free flow is assumed at the inlet of the
pump, and outlet of a turbine. If rotation at the turbine outlet is included, equation 2.4 gets
a c u2,t -component, which could alter the prediction of HR . As mentioned previously, an esti-
mate of c u2,t can be made by the use of Euler’s equation 2.4, if a good measure of the hydraulic
efficiency is known.
A separate aspect that hasn’t been addressed is the effect of friction and shock losses. In
turbine mode, these losses will contribute to a reduction of the net head. Therefore, it might
be expected that the actual best efficiency point in PAT mode lies even higher than what has
been predicted, to account for these losses. Further, when applying symmetry, a decision was
made not to include the slip from pump operation. Slip would have altered the blade angle to
the relative flow angle, and consequently changed the velocity diagram in turbine mode. This
can be justified as the effect of slip in turbine operation, compared to pump operation, will be
modest, and therefore may be neglected [3].
Finally, the hydraulic efficiency was for simplicity set to 1, even though the pump was mea-
sured to have a maximum efficiency of η p = 0.45. By including a more trustworthy hydraulic
efficiency into Euler’s pump and turbine equations (2.3 and 2.4), the predicted head in turbine
mode would increase, and the accuracy of the prediction model could be enhanced.
5.2.4 Additional challenges and further work
As seen from the results, the PAT’s maximum efficiency was measured to be of approximately the
same order as the pump’s efficiency. Even though the actual value of the efficiency measurement
may be questioned, it is fair to say that the PAT’s efficiency will not exceed the efficiency supplied
by the pump manufacturer. Therefore, as a pump’s efficiency is significantly lower than a custom
made turbine, one will have to consider if it is profitable to invest in a PAT-driven power plant.
However, as stated earlier, this is mainly a solution appropriate for developing countries and
rural areas, where low investment costs are far more important than efficiency.
On a separate note, a drawback to the symmetry prediction method presented, is that the
impeller geometry has to be manually measured. By having to remove the spiral casing of the
pump, the usage of the model is not exactly straight forward. This will usually require actually
purchasing the pump, as pump manufacturers seldom publish this information. Hence, the
goal of being able to pick a suitable pump right off the shelf, which correlates to a site’s given
conditions, is not satisfied. For that reason, the selection process is optimal only if pump man-
ufacturers make such data available. Another idea is for pump manufacturers to apply such a
prediction method to their own pumps, and publishing predicted PAT characteristics to poten-
tial buyers, instead of having to test the pumps themselves.
Potential further development of the symmetry method may include obtaining a more gen-
eral constant k, found in equation 4.3. As for now, the constant’s value is based on one single
PAT test, but by testing several pumps as turbines, the credibility of which will be strengthened.
In that way, different k values for different sizes and types of pumps may be established. Further,
if a way is found to quantify the different losses in PAT operation, a better prediction should be
achievable. Lastly, if future tests are carried out with an operational torque transducer, a better
estimation of the uncertainty of η could be made. Also, the location of the best efficiency point
could become more accurate.
CHAPTER 6
Conclusion
The objective of this thesis was to investigate the validity of the already developed symmetry
method, which is based solely on the pump’s impeller geometry. For this reason, experimental
tests were conducted in the laboratory to study the PAT’s H − Q characteristic. The tests re-
vealed that some of the assumptions in the model were inadequate, and had to be adjusted and
improved accordingly.
In order to analyze the model, and its assumptions, the pump was tested in both pump and
turbine operation. The results from these tests made it clear that the assumption of total sym-
metry in the velocity diagrams between pump and turbine operation, was not completely valid.
Therefore, to account for this, two potential improvements were proposed. These improve-
ments enhanced the prediction model in the way that the predicted PAT characteristic moved
closer towards to the experimentally produced characteristic. However, in order to strengthen
the new improvements, and to make the prediction model more general, additional tests must
be carried out. This is especially important with respect to finding a more general empirical
constant k, as the result from one single PAT test is uncertain at best.
It is also uplifting that the predicted characteristic is in line with already established predic-
tion techniques, and produce a quite similar best efficiency point. However, because of the lack
43
CHAPTER 6. CONCLUSION 44
of guide vanes, it is difficult to operate a pump as turbine at its BEP. Therefore, it can truly be
advantageous that the symmetry method produces a full characteristic, to be able to foresee the
PAT’s performance at either side of this point.
One drawback with the method is the fact that the impeller geometry has to be measured
manually. This will often include purchasing and dismantling the pump. Hence, the model will
be significantly easier to use if pump manufacturers begin to make such geometry information
public. In that case, and if further refinement of the constant k is done, the proposed prediction
method could become both accurate and simple to use. Also, even though the efficiency of a
PAT is low compared to specifically designed hydro turbines, it is still a highly relevant option.
This is especially profitable in developing countries and rural areas, as these areas will benefit
greatly from an accurate prediction model, in order to maximize the power output available.
Bibliography
[1] A. Williams. The turbine performance of centrifugal pumps: a comparison of prediction

methods. Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power
and Energy, 208(1):59–66, 1994.
[2] S. V. Jain R. N. Patel. Investigations on pump running in turbine mode: A review of the
state-of-the-art. Renewable and Sustainable Energy Reviews, 30:841–868, 2014.
[3] J. M. Chapallaz P. Eichenberger G. Fischer. Manual on pumps used as turbines. Vieweg,

1992.
[4] A. Williams. Pumps as turbines for low cost micro hydro power. Renewable Energy, 9(1-4):
1227–1234, 1996.
[5] S. Derakhshan A. Nourbakhsh. Experimental study of characteristic curves of centrifugal

pumps working as turbines in different specific speeds. Experimental thermal and fluid
science, 32(3):800–807, 2008.
[6] J. Fernandez E. Blanco J. Parrondo M. T. Stickland T. J. Scanlon. Performance of a centrifu-

gal pump running in inverse mode. Proceedings of the Institution of Mechanical Engineers,
Part A: Journal of Power and Energy, 218(4):265–271, 2004.
[7] P. N. Garey. Using pumps as hydro-turbines. Hydro Review, pages 52–61, 1990.
45
BIBLIOGRAPHY 46
[8] A. Kjølle. Mechanical equipment. Waterpower laboratory, NTNU, 2001.
[9] A. Williams. Pumps as turbines: a user’s guide. Intermediate Technology, 1995.
[10] B. H. Teuteberg. Design of a pump-as-turbine microhydro system for an abalone farm. PhD
thesis, Stellenbosch: University of Stellenbosch, 2010.
[11] H. Brekke. Pumper & turbiner. Vannkraftlaboratoriet NTNU, 2003.
[12] G. Bye. Pumper. Universitetsforlaget, 1967.
[13] T. K. Nielsen. Simulation model for francis and reversible pump turbines. International
Journal of Fluid Machinery and Systems, 8(3):169–182, 2015.
[14] International Electrotechnical Commission. IEC 62097: Hydraulic machines, radial and
axial - Performance conversion method from model to prototype. 2009.
[15] M. Margolis. Arduino Cookbook: Recipes to Begin, Expand, and Enhance Your Projects. "
O’Reilly Media, Inc.", 2011.
[16] International Electrotechnical Commission. IEC 60193: Hydraulic turbines, storage pumps
and pump-turbines - model acceptance tests. 1999.
[17] B. W. Solemslie. Compendium in instrumentation, calibration uncertainty analysis. 2010.
[18] P. T. Storli. Modelltest av francis turbin, 2006.

APPENDIX A
Paper written for CRHTVIII’18
The following section contains a research paper written for the 8th edition of the annual Sym-
posium on Current Research in Hydraulic Turbines (CRHTVIII’18). The results presented were
obtained during the previous project work, where the symmetry prediction method was estab-
lished. As the derived method, and the theoretical concepts, are quintessential for this thesis, it
was considered necessary to present them once more. Therefore, some sections of the following
paper may be very similar the thesis you have just read.
47
APPENDIX A. PAPER WRITTEN FOR CRHTVIII’18 48
Symmetry prediction method for pump as turbine

characteristics
Øyvind Albert*, Sondre Skjoldli*, Torbjørn K. Nielsen
Department of Energy and Process Engineering, Norwegian University of Science and
Technology, Trondheim, Norway
*Corresponding authors (oyvinalb@stud.ntnu.no, sondsk@stud.ntnu.no)
Abstract. As a cheap and available source of renewable energy, regular centrifugal pumps may
be run in reverse to act as hydro turbines. Pumps as turbines (PAT) are especially relevant for
isolated rural areas, or in developing countries, where efficiency is not necessarily the highest
priority. The main challenge, however, is to be able to pick a suitable pump for a given site,
which coincides with the available head and discharge. In order to make this decision, one has
to be able to predict how a certain pump will perform in turbine mode. A lot of work has been
done in the past to establish such prediction methods, based on tests and empirical correlations.
However, the inaccuracies of these methods motivate a new approach, where the performance
in turbine mode is based solely on the pump’s geometry.
By assuming symmetry at the outlet of a pump to the inlet of a turbine, a method for
establishing a PAT’ s characteristics, such as Qed − ned and H − Q curves, was made. To
generalize the method, it was applied to both high and low head pumps. These characteristics
were then compared to fictional Francis runners, designed for the same best efficiency point
(BEP) values of head and flow. Further, the method was applied to one specific pump, which
is scheduled to be subjected to physical tests. Analogous to literature, it was predicted that
the BEP operating values of this PAT lies above its original BEP values in pump mode. From
these rated BEP values, corresponding PAT characteristics were predicted. Finally, to be able
to verify and validate the prediction model, thorough tests are scheduled to be carried out.
1. Introduction
In a time where renewable power production is becoming increasingly important, the motivation
to reduce the costs as far as possible also gets a lot of attention. One appropriate technology
is to use a pump as a turbine (PAT). By using a regular centrifugal pump, and running it in
reverse, one will be able to extract energy from a fluid instead of putting energy into it. The
idea is not at all new, but could prove to be increasingly relevant and important for developing
countries, and isolated rural areas, where the electricity demand is ever growing [1].
Even though operational and running costs of a hydropower plant are low and affordable, the
high initial capital cost in the development of such a plant can truly be a barrier [2]. Additionally,
as large-scale generation is not always feasible, there is an increasing interest in small-scale
hydropower plants. In these small-scale plants, it can often be hard to justify the construction
costs in comparison to the total power generation possible [3]. It is in these situations a PAT can
show its full potential, and can prove to be a very suitable technology, because of its low price.
The price per kW produced by small-scale hydro power plants, are usually higher than that of
large hydro power plants [1]. Therefore, installment of a PAT could be essential in reducing
these costs.
The main obstacle when planning a PAT driven power plant is the difficulty of predicting the
optimal turbine characteristics [2]. The primary challenge therefore lies in being able to pick a
suitable pump for a given site, which coincides with the available head and discharge. In order
to make this decision, one has to be able to predict how a certain pump will perform in turbine
mode.
2. Theoretical background
If a centrifugal pump is to be used as a PAT, the advantages and disadvantages has to be
carefully weighed up against each other. Firstly, when looking at it from an economic point of
view, PATs have a clear advantage. Centrifugal pumps are mass produced all over the world,
and are manufactured for a wide range of heads and flows. By being able to pick a centrifugal
pump right off the shelf, a PAT will be a significantly cheaper option than a turbine designed
for specific conditions. In other words, they are more cost efficient as well as more accessible.
What’s more is that they are easy to install, easy to maintain, and that spare parts are easily
available [2]. As pumps are a more widespread technology than turbines, qualified personnel to
conduct repairs and maintenance is also more available.
There are several disadvantages however, besides the aforementioned challenge of predicting
the PAT performance. Even though Fernandez et al. (2004)[3] state that the efficiency of a
pump running in reverse has almost the same efficiency as in pump mode, it will most certainly
not perform as well as a custom made turbine. It may seem as if the pump industry is not as
concerned as the turbine industry with gaining the highest efficiency possible. Whereas a hydro
turbine has a very smooth surface to minimize losses, a mass produced pump will often have a
high level of roughness on the impeller.
Another key thing is that a centrifugal pump, compared to a regular hydro turbine, does
not have any guide vanes. Whereas a regular hydro turbine can adjust incoming flow, and is
able to perform efficiently for a range of flow rates, a PAT does not have this option [4]. This
will enable regular hydro turbines to produce power at BEP, even through varying conditions.
By implication, the range of suitable operating flow rates is much more limited for a PAT. One
solution to this problem is to have multiple PATs coupled in series. With such a rig, one will be
able to activate the number of pumps required to handle the incoming flow rate most efficiently.
Having said that, a single PAT may perform close to maximum efficiency if a suitable pump
is chosen for the given site conditions. Especially if the site has a close to fixed water supply
throughout the year.
2.1. Performance of a Francis turbine

Using a pump as a turbine requires understanding of how the performance is found for a regular
turbine. It will be of importance to apply this to the turbine mode of a PAT, in which the
performance can be predicted. The turbine’s runner will transform the available hydraulic energy
to mechanical energy, and the hydraulic efficiency is a measure of how good this transformation
is. The hydraulic efficiency is defined as:
u1,t cu1,t − u2,t cu2,t
ηh,t = (1)
gHn,t
In equation 1, u is the peripheral velocity, while cu is the peripheral component of the
absolute velocity. Subscripts 1, t and 2, t refer to the inlet and outlet of the turbine, respectively.
Furthermore, Hn is the available net head, while g is the acceleration due to gravity.
Assuming zero losses in the turbine, the hydraulic efficiency is ηh,t = 1, where the runner
blades will have angles perfect for transforming hydraulic energy to mechanical [5]. As the
turbine transforms hydraulic energy to mechanical energy, it works as a throttle in the system.
Figure 1: Showing the tendency of how the flow, Q, varies with the rotational speed, n, for low
head and high head Francis turbines (Adapted from [5]).
Nielsen (2015)[5] shows that this throttling is a function of the rotational speed and depends on
the geometry of the turbine, where the relationship between flow and rotational speed (Q − n)
is dependent on the D1,t /D2,t -ratio. For high head Francis turbines, the flow has a tendency to
decrease as a function of the rotational speed, seen in figure 1. Likewise, for low head turbines,
where the D1,t /D2,t -ratio is low, the flow increases with n.
2.2. Earlier work

As the main disadvantage of a PAT is the difficulty of predicting the turbine characteristics
that are needed for a given site, a lot of work has been done throughout the years to establish
prediction methods which calculates the performance [6]. The actual turbine performance must
be found through testing, but as this is a costly and time consuming process, as well as it
requires that the pump has already been purchased, many different empirical correlations have
been produced. These in turn produce a wide range of results. A common factor, however, is
that the optimal operating point of a PAT is higher in both head and flow, than that of the
pump it originates from [1]. This is to account for the head reduction caused by the various
losses in turbine operation.
The two main approaches that has been taken to predict turbine performance are either using
the pump efficiency, or by relating the head and flow ratios to the specific speed [6]. Williams
(1994)[6] investigated in total eight different prediction techniques, and compared the accuracy
of the models. Table 1 gives an outline of the different methods, as well as the appropriate
Ht Qt
relationship of the head correction factor h̃ = H p
and the discharge correction factor q̃ = Qp
.
According to Williams’ (1994)[6] study, the method of Sharma proved to be the most accurate.
The test involved comparing the turbine prediction methods on 35 different pumps, with
available test data. As Sharma’s method proved to be the most accurate, it is the only one
that will be looked at here. He relates the discharge and head correction factors to the pump
efficiency as shown in equations 2 and 3.
Qt 1
q̃ = = 0.8 (2)
Qp ηp
Ht 1
h̃ = = 1.2 (3)
Hp ηp
Even though a few of the other methods proved competitive, Sharma’s method was found
to be the most accurate of the eight approaches. Still, 20 percent of the tested pumps fell
outside what was said to be the acceptable prediction limits. Therefore, it will always be wise
to conduct thorough tests before installing a certain PAT [6]. Also, what is important to note,
is that all these methods do only produce one best efficiency operating point, and not a full
turbine characteristic.
Although the accuracy of the different correlations can be questioned, they may serve as a
rough guide when designing a PAT-site [7]. However, the questionable precision, along with the
large number of different pumps that need to be tested to create a trustworthy model, motivates
a new approach. An approach in which the turbine performance is determined solely based
on the pump geometry, as well as producing turbine characteristics on either side of the best
efficiency point.
Table 1: Different prediction methods investigated by Williams (1994) [6].
Name of method/investigator Based on Head correction factor h̃ Discharge correction factor q̃

1 1
Childs BEP ηp ηp
1 1
Hancock BEP ηt ηt
1 √1
Stepanoff BEP ηp ηp
1 1
Sharma BEP ηp1.2 ηp0.8
1 0.85ηp5 +0.385
Alatorre-Frenk BEP 0.85ηp5 +0.385 2ηp9.5 +0.205
Schmiedl BEP −1.4 + η2.5 hp
−1.5 + η2.4 2
hp
Grover Specific speed 2.693 − 0.0229Nst 2.379 − 0.0264Nst
Hergt Specific speed 1.3 − Nst6−3 1.6
1.3 − Nst −5
3. Methodology
The method presented in this study, takes a different approach than previous prediction
techniques. It is based on assuming symmetry in the velocity diagram of a pump and a turbine.
By this assumption, the inlet diagram of a turbine could be directly mirrored from the outlet
diagram of a pump. The same procedure was then also applied to find the turbine outlet diagram,
from the pump inlet diagram. The two respective diagrams and their mirrored opposites are
presented in figure 2a and 2b. It is important to note that with this approach the slip and
losses such as friction and shock losses, are not directly accounted for. This will be a topic of
discussion later.
The BEP values of head and flow for a PAT can thus be calculated directly from any arbitrary
centrifugal pump, provided that the pump geometry is known. This is information that can be
troublesome to find, as it is not usually supplied by pump manufacturers. For this reason,
professor Torbjørn K. Nielsen supplied a spreadsheet which calculates the main dimensions, and
designs a centrifugal pump. From this, the main dimensions of the PAT such as diameters,
heights, and angles can be taken directly from the pump dimensions. While doing so, extra care
has to be taken when defining the inlet and outlet regions, as the outlet dimensions of the pump
becomes the inlet dimensions of the turbine, and vice versa. Further, by using standard Francis
(a) (b)
Figure 2: Obtaining the PAT velocity diagram by symmetry (not to scale).
turbine design equations, and assuming no rotation at the outlet, the rated values HR and QR
for the PAT were found. These values are factors of the rotational speed, which are specified to
be the same in turbine mode as in pump mode. This is because it is of interest to use the pump
motor as a generator for the PAT set up, to limit the capital costs even more.
3.1. Designing a Francis turbine

The geometry of a centrifugal pump is quite similar to that of a Francis runner, only with the
fluid flowing in the opposite direction. Therefore, after applying the symmetry method, it was
of interest to compare the PAT to a Francis turbine specifically designed for the same head
and flow. Naturally, such a Francis turbine will have a different geometry than the PAT. When
designing a Francis turbine at its best efficiency, one assumes that there is no rotation at the
outlet, and cu2 was set to zero. Further, the hydraulic efficiency was assumed to be 0.96, and
the reaction ratio R = 0.5 as it is common to assume that approximately 50% of the total net
specific energy is converted to mechanical energy in the runner [8]. Knowing this, the following
equation for the reaction ratio, taken from [8], could be applied:
R = 2u1,t cu1,t − c2u1,t (4)

By recognizing the first term on the right hand side of equation 4 as the theoretical hydraulic
efficiency, the reduced absolute tangential and peripheral velocities could be found. Then,
together with the same values for rotational speed, head and flow as for the PAT’s BEP values,
the main dimensions of the Francis turbine were calculated. However, a few of the dimensions
had to be assumed first. According to Brekke (2003)[8], it is common to choose a β2,t angle
between 13◦ and 19◦ , and a u2,t component between 35 and 43 m/s. These are values that have
proven adequate through years of experience. Therefore, when designing this fictional Francis
runner, the chosen parameters were set to β2,t = 16◦ and u2,t = 40m/s. It is also the norm to
demand a slight acceleration of the meridional velocity component cm through the runner, as
an accelerating flow reduces the chance of a retardation. Thus, as cm2,t could be found through
the angle and peripheral velocity stated above, cm1,t could be determined when assuming a 10%
velocity increase from inlet to outlet.
3.2. Producing the PAT’s H − Q curve

Up until now it is only the performance at BEP that has been addressed. However, it is very
much of interest to be able to predict how the PAT will perform at either side of this point.
This is possible by manipulating the dimensionless momentum equation presented by Nielsen
(2015)[5]:
dq q|q|
Twt = h − 2 − σ(ω̃ 2 − 1) = 0 (5)
dt κ
In equation 5, Twt is a time constant representing the hydraulic inertia, while q, h and ω̃
are dimensionless properties for flow, head and angular speed respectively. Further, σ is the
dimensionless throttling dependency of angular speed of rotation, and κ the opening degree of
the turbine. A more thorough explanation of the method, as well as variable definitions, may
be found in Appendix A.
Now, by varying the rotational speed n, and solving the stationary version of equation 5 for
q, the discharge Qt for different rotational speeds was found. IEC 62097 [9] specifies the two
dimensionless factors for speed and discharge, shown in equation 6 and 7 respectively.
nD2,t
ned = p (6)
gHR,t
Qt
Qed = 2
p (7)
D2,t gHR,t
From these two equations the turbine’s Qed − ned relationship was established, and from this
relationship the PAT’s H − Q curve was predicted.
For validation purposes, this approach was initially carried out for two different pumps. As
the access to available pump geometries was limited, the geometries for the two different pumps
were fictional, and calculated through the earlier mentioned design spreadsheet supplied by
professor Torbjørn K. Nielsen. The method was tested for a radial high head pump, as well as
a more axial pump with lower head. This was done in order to compare the resulting Qed − ned
and H − Q curves to literature, and expected results.
In order to follow the outlined approach, and to assume symmetry in the pump outlet to the
turbine inlet, only a few pump parameters are needed. The hydraulic efficiency was assumed to
be 1. Additionally, the pump meridional velocity component, cm1,p , is also an input parameter
in the model. This value needed to be assumed. For the two fictional pumps the meridional
velocity was prescribed a value of cm1,p = 9m/s based on Nielsen’s assumptions. All other
required input parameters are presented in table 2.
Table 2: Fictional pump geometry parameters and operating conditions.
Parameter Pump 1 Pump 2 Unit

Inlet diameter D1,p 1.457 1.5 m
Outlet diameter D2,p 1.737 3 m
Outlet height B2,p 0.284 0.3 m
Outlet blade angle β2,p 12 12 ◦ (degrees)
Inlet meridional velocity cm1,p 9 9 m/s

Rotational speed n 600 600 rpm
3.3. Test-Pump 1
To be able to properly validate and verify the symmetry prediction method presented,
experiments has to be carried out. These experiments are scheduled to be conducted in the
laboratory at the University of Dar es Salaam, Tanzania and at NTNU. The PAT test rig in
Dar es Salaam is equipped with two pumps, where one of them will work as a PAT. This is
an old pump, where both the name and pump characteristic is unknown. However, it has been
found that at its best efficiency in pump mode, it has a rated head of HR,p = 10m and flow
QR,p = 0.05m3 /s. The symmetry method was therefore applied to this pump geometry, hereby
referred to as Test-Pump 1. As this pump had a modest discharge, Qp , compared to the two
fictional pumps, the assumption of cm1,p needed to be adjusted in order to predict the turbine
performance accurately. According to Bye (1967)[10], the absolute inlet velocity, c1,p , can take
the value of 2.3 − 5m/s for pumps at low discharge. As c1,p = cm1,p at the pump inlet, the
meridional velocity, cm1,p was set to 2.5m/s for this pump.
Test-Pump 2, which is the pump scheduled to be tested at NTNU, has been ordered from
the manufacturer. However, the geometry data has not yet been measured. Therefore, further
investigation in this paper will be on Test-Pump 1, where the most important pump parameters,
the pump dimensions, as well as the operating conditions, are presented in table 3. It is important
to remember that the input parameters for the pump become the outlet parameters for the PAT.
Thus, one has to be careful not to adopt incorrect values.
Table 3: Geometry parameters and operating conditions of Test-Pump 1.

Inlet diameter D1,p 148.1 mm
Outlet diameter D2,p 197.3 mm
Outlet height B2,p 34.3 mm
Outlet blade angle β2,p 20.17 ◦ (degrees)
Inlet meridional velocity cm1,p 2.5 m/s

Rotational speed n 1450 rpm
Pump efficiency ηp 0.804 -
4. Results
4.1. Comparable results; PAT versus Francis turbine
By assuming symmetry in the velocity diagrams from the outlet of the pump to the inlet of the
turbine, and using the pump’s outlet dimensions as the inlet dimensions in turbine mode, the
PAT characteristics at BEP could be determined. By obtaining the values for the rated flow
and head, appropriate Qed − ned and H − Q curves were calculated. This was done in order to
predict the PAT’s performance outside of the best efficiency point. Further, it was of interest
to design a Francis runner for the same head and flow as was found for the PAT. By doing
so, conclusions could be drawn between the performance of a pump used as a turbine, and a
Francis turbine specifically designed for certain conditions. This would also reveal some of the
limitations of the PAT.
As mentioned previously, two different fictional pumps were used to produce the initial results.
These consisted of both a high head pump, and one of lower head. The resulting BEP values
for head and discharge in turbine mode were found to be HR,t = 44.5m and QR,t = 15.3m3 /s
(a) (b)
Figure 3: Qed − ned and H − Q curves for HR,t = 44.5m and QR,t = 15.3m3 /s.
(a) (b)
Figure 4: Qed − ned and H − Q curves for HR,t = 458m and QR,t = 28m3 /s.
for the lower head case. For the high head case the equivalent BEP values were HR,t = 458m
and QR,t = 28m3 /s. These results are summarized in table 4.
Table 4: Results of the two fictional PATs.
Parameter PAT 1 PAT 2 Unit

Rated head HR,t 44.5 458 m
Rated discharge QR,t 15.3 28 m3 /s
Speed number ∗ Ω 1.53 0.36 -
Figure 3a displays the Qed − ned relationship for the lower head case, both for the PAT and
the designed Francis turbine. From this result, the H − Q curves presented in figure 3b was
calculated. Similarly, figure 4a and 4b follow the same approach, but for the high head case.
4.2. Test-Pump 1
After seeing that the method behaved in a desired manner, further calculations were done on one
specific pump, the one referred to as Test-Pump 1. With this geometry, as specified in table 3,
the BEP values for head and discharge came out to be HR,t = 11.44m and QR,t = 0.0585m3 /s.
Additionally, the speed number of this PAT was calculated to ∗ Ω = 0.63.
Table 5: Results of Test-Pump 1 working as a PAT.
Parameter Test-Pump 1 Unit

Rated head HR,t 11.44 m
Rated discharge QR,t 0.0585 m3 /s
Speed number ∗ Ω 0.63 −
In the same manner as for the two fictional PATs, the Qed − ned and H − Q curves were
calculated based on the BEP values for Test-Pump 1. These are presented in figures 5a and
5b, respectively. These are the PAT characteristics that future test results will be compared up
against.
(a) (b)
Figure 5: Qed − ned and H − Q curves for Test-Pump 1.
4.3. Applying Sharma’s prediction method to Test-Pump 1

To be able to conclude if this prediction method has any future, it was of great significance to
compare it to the old empirical correlations. As Sharma’s performance prediction method was
found to be most accurate of the eight methods Williams (1994)[6] looked at, it was the only
one considered here.
By equations 2 and 3, as well as the rated values for head and discharge for the pump at
BEP, the PAT’s performance at BEP could be calculated as follows:
Hp 10
Ht = = = 12.99m (8)
ηp1.2 0.8041.2
Qp 0.05
Qt = = = 0.0595m3 /s (9)
ηp0.8 0.8040.8
5. Discussion
5.1. PAT versus Francis turbine
By looking at the results presented for the two fictional PATs, it can be seen that the symmetry
prediction model is in agreement with literature. The Qed − ned diagrams in figures 3a and 4a
follow the same speed-flow relationship as can be seen in figure 1. This can be observed for both
the high head and lower head case. The H − Q curves in figures 3b and 4b are also analogous
in shape to what an authentic turbine may produce [8].
Taking a closer look at the lower head case in figure 3a, and comparing the PAT to a Francis
turbine designed for the same BEP values of head and flow, it becomes clear that the PAT is
inferior. The pump’s angles and geometry makes the PAT less suitable in turbine mode than
a Francis turbine, which in turn will result in a lower power production. For increasing speed
values, the Francis turbine will be able to handle a higher discharge. Therefore, for the same
rated head, the Francis runner will be able to produce more power. Also, as the Francis runner
is expected to have a higher efficiency than the PAT, this tendency will be amplified.
For the high head case in figure 4a, the relationship is different. The PAT curve is significantly
steeper than the curve for the Francis turbine. As the rotational speed increases to the point
where Qed becomes negative, the flow changes direction. Past this point, the machines are
working as pumps. This ”pumping effect” is when the rotational speed is increased enough so
that the fluid is throttled through the runner [11]. Naturally, the PAT will work better in this
region, as it is a pump originally, and will handle larger flow values for a given rotational speed.
Additionally, for low speeds the PAT lies above the Francis turbine curve, which at first
glance may be considered strange. This could be because of the geometry, where the relationship
between inlet and outlet diameters make the PAT more radial than the Francis runner, and thus
makes it better designed for high head values. However, this is only in the low speed region.
At this point one may also draw a connection between a PAT and a reversible pump turbine
(RPT), as they will have certain geometrical similarities. A RPT will have a similar Qed − ned
characteristic, as it too will be steeper than the Francis turbine [11].
The corresponding H − Q curves are presented in figures 3b and 4b. They show a similar
relationship for both cases, but with the high head case graphed for a larger discharge range.
In both cases, the geometry, and D1,t /D2,t -ratio, make the PAT more radial than the Francis
runner. This in turn makes the H − Q curve steeper for the PAT. For the high head case, where
the PAT is more radial than the low head, this effect is stronger, and the curve is even steeper.
An implication of this is that the PAT is harder to operate. As the head-flow relationship
is steep, a slight change in any of the two variables will take the operation well away from its
BEP. For a Francis turbine, this could be solved by modifying the guide vanes. But, for a PAT
with no guide vanes, it is clear that this will have a considerable negative effect on the power
production during varying conditions.
5.2. PAT characteristics of Test-Pump 1

By using the input parameters for the pump that is scheduled to be tested in Dar es Salaam,
the applied method predicts characteristics as shown in figure 5. The best efficiency point in
turbine mode was found to be at HR,t = 11.44m and QR,t = 0.0585m3 /s. In comparison, the
rated head and flow in regular pump mode is HR,p = 10m and QR,p = 0.05m3 /s. Thus, the
operation point at BEP lies higher, both for head and discharge, in PAT mode compared to
pump mode. This is analogous to what has been stated previously.
To verify if the presented prediction method is within reasonable limits, the actual H − Q
performance of the PAT must be established in the laboratory. This will reveal the method’s
credibility, and if the assumptions have been acceptable.
5.3. Sharma’s prediction method

The prediction method of Sharma estimated the PAT’s best efficiency point to be at HR,t =
12.99m and QR,t = 0.0595m3 /s. As expected, this method also predicts BEP values for the PAT
above pump BEP values. When comparing the symmetry prediction method outlined in this
report, to that of Sharma’s, it deviates 11.9% in head and 1.7% in discharge when looking at
Test-Pump 1. At this stage, however, comparing the two prediction methods give limited closure.
Therefore, it will be interesting to conduct proper test, and to compare the two methods to the
real characteristic, for this specific pump.
Table 6: Prediction method comparison of Test-Pump 1 : Sharma vs symmetry.
Parameter Sharma Symmetry Deviation

Rated head HR,t 12.99m 11.44m 11.9%
Rated discharge QR,t 0.0595m3 /s 0.0585m3 /s 1.7%
5.4. Assumptions, losses, and inaccuracies

A pitfall to the presented symmetry method is that the meridional velocity cm1,p has to be
assumed. For Test-Pump 1 this velocity was assumed to be cm1,p = 2.5m/s, because of
the modest amount of discharge. This assumption was based on numbers presented by Bye
(1967)[10]. There are uncertainties related to this assumption, and the integrity of which may
be questioned. Hence, if a range of different pumps are to be examined with this method, new
meridional velocities have to be assumed. Therefore, it will be of interests to establish a more
general criterion, and a better way to estimate this parameter. One possible solution may be
to calculate the meridional velocity based on the flow and the geometry of the impeller inlet.
However, the impeller inlet height was not available during the time of this study.
Another aspect that hasn’t been addressed is the effect of friction and shock losses. In
turbine mode, these losses will contribute to a reduction of the net head. Therefore, it might
be expected that the actual BEP of the test-pump in PAT mode lies even higher than what has
been predicted, to account for these losses. If these losses are quantified, the symmetry method
may be used to create a span of BEP values. This span would range from zero losses, to a
maximum amount of expected losses. Then, one could be more confident that the actual BEP
lies within this interval.
Further, the assumption of total symmetry between pump and turbine mode may also be
questioned. When applying symmetry, a decision was made not to include the slip from pump
operation. Slip would have altered the blade angle to the relative flow angle, and consequently
changed the velocity diagram in turbine mode. This can be justified as the effect of slip in turbine
operation, compared to pump operation, will be modest, and therefore may be neglected [12].
Finally, the hydraulic efficiency was for simplicity set to 1. Even though it has been proven
that the efficiency of the PAT will be almost the same as in pump operation, the actual efficiency
of a PAT will be significantly lower than 100%. The accuracy of the prediction method could
therefore be enhanced by including a more realistic pump efficiency.
6. Conclusion
The objective was to investigate the possibility of predicting a PAT’s performance, only given the
pump’s geometry. For this, a general method based on symmetry has been suggested. However,
in addition to the measurable lengths and angles of the pump, a few assumptions had to be
made. These may weaken the model, and contribute a more inaccurate prediction.
In order to be able to validate and verify the prediction model, tests are scheduled to be
conducted. As these tests will be run on the pump referred to as Test-Pump 1 at the University
of Dar es Salaam, proposed characteristics of this pump were produced.
Several losses and uncertainties have been discussed as well, where the ones left out may
very well have a substantial impact on the predicted characteristics. However, supported by
literature, the predicted BEP points in turbine mode are higher than for pump mode. Also,
by applying one of the established empirical prediction methods, it can be observed that the
symmetry method follows the same trend.
Lastly, what remains is therefore to conduct the necessary tests for validation.
Appendix A. Producing the PATs H-Q curve

Start by manipulating the dimensionless momentum equation presented by Nielsen (2015)[5]:
dq q|q|
Twt = h − 2 − σ(ω̃ 2 − 1) = 0 (A.1)
dt κ
Where Twt is a time constant representing the hydraulic inertia. This value was not needed
however, as the left hand side of the equation could be set to zero to find the stationary
characteristic. Furthermore, the dimensionless properties for flow, head and angular speed
of rotation in equation A.1 are defined as q = Qt /QR,t , h = Ht /HR,t and ω̃ = ω/ωR . Now,
by assuming that the PAT has BEP at q = 1, h = 1, ω̃ = 1, and that the derivative of the
hydraulic efficiency in the Euler turbine equation (1) with regards to ω is zero, ∂η/∂ω = 0, the
PAT’s dimensionless throttling dependency of angular speed of rotation, σ, is defined as seen in
equation A.2.
ηhR − ψ
σ= (A.2)
ηhR + ψ
By setting the hydraulic efficiency, ηhR = 1 for BEP in equation A.2, and defining the machine
coefficient as
u2,t 2
ψ= , (A.3)
gHR,t
σ for the PAT could be found. Further, the opening degree of the turbine, κ, equals 1 for the
BEP, and is defined by:
√ Qt
2gHt
κ= (A.4)
√QR,t
2gHR,t
By varying the rotational speed n, and solving the stationary version of equation A.1 for q, the
discharge Qt for different rotational speeds were readily found. IEC 62097 [9] specifies the two
dimensionless factors for speed and discharge, shown in equation A.5 and A.6 respectively.
nD2,t
ned = p (A.5)
gHR,t
Qt
Qed = 2
p (A.6)
D2,t gHR,t
From these two equations the turbine’s Qed − ned relationship was established. The H − Q curve
could then be produced by first solving equation A.5 for Ht :
nD2,t 2
( ned )
Ht = (A.7)
g
By keeping the rotational speed constant, and only varying the speed factor, Ht was found.
Finally, the definition of the discharge factor was modified to equation A.8 to give a result for
Qt for changing values of Qed and corresponding Ht . This enabled graphing of the turbine’s
H − Q relationship. p
2
Qt = Qed D2,t gHt (A.8)
References
[1] Jain S V and Patel R N 2014 Renewable and Sustainable Energy Reviews 30 841–868
[2] Williams A 1996 Renewable Energy 9 1227–1234
[3] Fernandez J, Blanco E, Parrondo J, Stickland M and Scanlon T 2004 Proceedings of the Institution of
Mechanical Engineers, Part A: Journal of Power and Energy 218 265–271
[4] Williams A et al. 1995 Pumps as turbines: a user’s guide (Intermediate Technology)
[5] Nielsen T K 2015 International Journal of Fluid Machinery and Systems 8 169–182
[6] Williams A 1994 Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power and
Energy 208 59–66
[7] Teuteberg B H 2010 Design of a pump-as-turbine microhydro system for an abalone farm Ph.D. thesis
Stellenbosch: University of Stellenbosch
[8] Brekke H 2003 Vannkraftlaboratoriet NTNU
[9] IEC62097 2009 Hydraulic machines, radial and axial - Performance conversion method from model to
prototype
[10] Bye G 1967 Pumper:(Omsl.: Elmer Rodin) (Universitetsforlaget)
[11] Nielsen T K and Olimstad G
[12] Chapallaz J M, Eichenberger P and Fischer G 1992 Manual on pumps used as turbines (Vieweg)
62
APPENDIX B
Calibration and Uncertainty Analysis
When calibrating an instrument, different sources of error leads to the total uncertainty. Table
B.1 describes the possible errors defined by IEC 60193 [16]. To find the total uncertainty, all
the different uncertainties are combined in the Root-sum-square (RSS) method, presented in
equation B.6.
Table B.1: Component errors in the calibration of an instrument.
Uncertainty Description
± f Xa Systematic error in primary calibration method
± f Xb Random error in primary calibration method
± f Xc Systematic error (repeatability) of the secondary instrument
± f Xd Random error of the secondary instrument
± f Xe Physical phenomena and external influences
±fXf Error in physical properties
63
APPENDIX B. CALIBRATION AND UNCERTAINTY ANALYSIS 64
B.1 Pressure sensors
Both the upstream and downstream pressure transducers were calibrated using a calibration
hand pump, a Druck DPI 601. The primary calibrator uncertainty, thus the hand pump un-
certainty, was found in calibration report B.8.1. For calibration of the transducers, a LabView
program was utilized, which calculated the sensor regression line, as well as the systematic and
random error in the calibrations.
The pressure sensor used upstream of the PAT was a GE UNIK5000 transducer, ranging from
0 − 5 bar g ag e. Its calibration report can be found in B.8.4. The uncertainties for the upstream
pressure transducer are shown in table B.2.
Table B.2: Calibration uncertainties for pressure transducer upstream of the PAT.
Uncertainty Description Magnitude

f P 1,a Systematic error in pressure calibrator ±0.050%
f P 1,r eg r essi on Systematic and random error in instrument max(±0.492%)
From this, the total calibration uncertainty for the upstream pressure transducer is found
through the RSS method:
p
max( f P 1,c al ) = ± (0.050)2 + (0.492)2 = ±0.495% (B.1)
The pressure sensor used downstream of the PAT was a Druck PTX1400 transducer, ranging
from 0 − 4 bar g ag e. The calibration report for this sensor is found in B.8.3. The uncertainties
of this pressure transducer are shown in table B.3.
Table B.3: Calibration uncertainties for pressure transducer downstream of the PAT.

f P 2,a Systematic error in pressure calibrator ±0.050%
f P 2,r eg r essi on Systematic and random error in instrument max(±1.153%)
The table above gives the total calibration uncertainty of the downstream pressure trans-
ducer as:
p
max( f P 2,c al ) = ± (0.050)2 + (1.153)2 = ±1.154% (B.2)
B.2 Flow sensor
For the PAT-experiment, an OPTIFLUX 2000C 50mm flow sensor was utilized. The calibration
of this flow sensor was done at a previous point in time, and the report can be found in B.8.2.
In the report, both systematic and random error in the primary instrument, a piston prover, as
well as the systematic and random error in the secondary instrument, the flow sensor, can be
found. The specific uncertainties are tabulated in table B.4.
Table B.4: Calibration uncertainties of the flow sensor.

f Q ab Systematic and random error in piston prover ±0.020%
f Q r eg r essi on Systematic and random error in instrument max(±0.23%)
Thus, the total calibration uncertainty becomes:
p
max( f Q cal ) = ± (0.020)2 + (0.23)2 = ±0.231% (B.3)
B.3 Torque sensor
The sensor used for measuring the torque at the shaft, was an HBM T22/200NM transducer. In
table B.5, the uncertainties have been estimated for the calibration in both pump and turbine
mode. It was calibrated by applying torque to the shaft, where the direction differed from pump
and turbine operation. This was done by connecting a horizontal metal bar to the shaft. A
weight bed was connected to the bar via a wire so that the torque could be measured by adding
weights to the weight bed. In reports B.8.5 and B.8.6, the calibration certificates can be found.
The systematic error in weights and weight bed, as well as the systematic error in the length of
the arm, was assumed as ±0.01% and ±0.013%, according to [17], respectively.
Table B.5: Calibration uncertainties for the torque sensor in both pump and turbine operation.

f τW Systematic error in weights and weight bed ±0.01%
f τar m Systematic error in the length of the arm max(±0.013%)
f τr eg r essi on,pump Systematic and random error in the instrument max(±1.776839%)
f τr eg r essi on,P AT Systematic and random error in the instrument max(±0.563075%)
The maximum calibration uncertainty is found by combining these errors, and equation B.4
and B.5 present the torque’s uncertainty in pump and PAT operation, respectively:
p
max( f τc al ,pump ) = ± (0.01)2 + (0.013)2 + (1.776839)2 = ±1.7769% (B.4)
p
max( f τc al ,P AT ) = ± (0.01)2 + (0.013)2 + (0.563075)2 = ±0.5633% (B.5)
B.4 RPM sensor
The rotational speed was measured with an optical sensor, and a piece of reflective tape, taped
to the shaft. The uncertainty of the measurements are assumed to be ±0.025%, as recommended
by IEC 60193 [16].
B.5 Temperature sensor
A thermometer was used to measure the water temperature in order to find a tabulated density
values. However, the systematic error of estimating the density is in order of ±0.01%, which is
not significant in the total uncertainty of the derived parameters, and is therefore neglected [18].
B.6 Uncertainty of derived values
By using Gauss’ law of error propagation, the uncertainty of derived parameters such as head
and efficiency can be found. The error ∆Y caused by the individual errors ∆x i is calculated as
seen below:
s
¶2 µ ¶2 ¶2
δy 1 δy 2 δy n
µ µ
∆Y = ± ∆x 1 + ∆x 2 + ... + ∆x n (B.6)
δx 1 δx 2 δx n
where variable Y = Y + ∆Y is a function of x 1 , x 2 , ...x n , such that
Y = y(x 1 ± ∆x 1 , x 2 ± ∆x 2 , ..., x n ± ∆x n ) (B.7)
∆Y
The uncertainty is here given by f Y = Y . The uncertainty of the density of water is assumed
negligible for the following calculations, because, as stated above, it is not significant in the total
uncertainty of the derived parameters [18].
B.6.1 Pressure head
The pressure head is calculated by the use of Bernoulli’s equation:
µ ¶
2 1
Q
∆p A 2i n −A 2out ∆p Q2
H= + + ∆z = + + ∆z (B.8)
ρg 2g ρg 2ag
(A out A i n )2
where, in turbine operation, ∆p = p i n − p out , ∆z = z i n − z out , and a = A 2out −A 2i n
. From this point
on, the notation will change from ∆p = P and ∆z = Z . The error in the height difference Z is
found by assuming an error of ±0.1mm when measured by a ruler. In this experiment, ∆z was
0.1mm
found to be 0.37m, and therefore f z = 370mm = ±0.027%. The total error in head is then given as:
sµ ¶2 ¶2 ¶2 sµ ¶2 ¶2
δH δH δH Q
µ µ µ
1
∆H = ± ∆P + ∆Q + ∆Z =± ∆P + ∆Q + ∆Z 2 (B.9)
δP δQ δZ ρg ag
The uncertainty of the pressure head is then calculated as follows.
v v
u 1 ∆P 2 + Q ∆Q 2 + ∆Z 2
u³ ´ ³ ´ u ³ ´2 ³ ´2
1 Q
∆P + ∆Q + ∆Z 2
u
∆H u ρg ag u ρg ag
fH = = ±t = ±t 2
u u
´2 PQ 2 Q4 Q2 Z
H Q2 P
+ ρag 2 + 2P Z
³
P
+ + Z ρ2 g 2 ρg
+ 4a 2 g 2 + ag + Z 2
ρg 2ag
v
¡ ∆P ¢2 ∆Q 2 ∆Z 2
u ³ ´ ³ ´¡ ¢
1
u
u ρ 2 g 2Q 4 Z 2 P
+ P 2 a 21g 2 Z 2 Q
+ 1
P 2Q 4 Z
= ±t 1 1
(B.10)
Q 4 ρ2 Z 2 g 2
+ PQ 2 Z 2 ρag 2
+ PQ 42Z ρg + 4P 2 Z12 a 2 g 2 + P 2Q 21Z ag + Q 41P 2
∆Q
By inserting f P = ∆P
P
, fQ = Q
and f P = ∆Z
Z
, the expression turns to:
v
P 2 a 2 f p 2 +Q 4 ρ 2 f Q 2 + Z 2 ρ 2 g 2 a 2 f Z 2
u
u
f H = ±t (B.11)
P 2 a 2 + PQ 2 ρa + 2P Z ρg a 2 + 14 Q 4 ρ 2 +Q 2 Z ρ 2 ag + Z 2 ρ 2 g 2 a 2
p
Note: f P = ( f P 1 )2 + ( f P 2 )2 .
By inserting the values of f P , f Q and f Z , as well as measured values for P , Q and Z , equation
B.11, can be calculated to find the total uncertainty of the pressure head.
B.6.2 Efficiency
A similar uncertainty derivation as above is also needed for efficiency calculations. However,
as the initial approach of measuring the power from the shaft torque was not feasible, this has
been left out. Instead, an attempt was made in order to quantify the uncertainty in the manual
reading of the power from the frequency converter display. Based on the variation in the power
magnitude, an uncertainty of ±5% in the efficiency calculation was assumed.
B.7 Uncertainty from measurements
The error of a measurement is defined as the difference between the measured value and the
actual value of the physical property [17]. In this analysis three types of errors have been con-
sidered - spurious errors, systematic errors and random errors.
Spurious errors are errors caused by human failures in an experiment, or by a failure in the
measuring equipment itself. If some points in the measurement series is out of line with the
rest of the measurements, they may be rejected [17]. Such points are called outliers. Systematic
errors exists when using poorly calibrated instruments, and some of the sources of uncertainty
are hysteresis and linearity issues in the instrument. It is therefore important to have enough
measurement points in the calibration, to limit these errors.
Random errors come from repeatability, or test-retest reliability, and is the phenomenon
when an instrument’s output varies for the same measurement conditions, namely the same in-
put [17]. When the number of measurements is large, the error may be found by using a Student-
t distribution in order to describe the distribution around the mean. The Student-t distribution
is quite similar in shape to a normal distribution, given that the number of measurement points
is high, but the ends are spread more when the number of samples is low.
The expression for the confidence interval of the mean of the measurement with a 1 − α
confidence is shown in equation B.12 [17].
SX SX
µ ¶
P X̄ − t α p ≤ µ ≤ X̄ + t α p = 1−α (B.12)
2 2
N N
With S X defined as v
N
u
u 1 X
SX = t (x i − X̄ )2 (B.13)
N − 1 i =1
where N is the number of samples and t α/2 is the t-value for the Student-t distribution with the
confidence level of 1 − α. In this study α was set to 5%.
B.8 Calibration certificates
B.8.1 Calibration certificate Druck DPI601

B.8.2 Calibration certificate OPTIFLUX 2000 C

B.8.3 Calibration certificate Druck PTX 1400
DruckPTX1400kalibrering 24.04.18
CALIBRATION REPORT
CALIBRATION PROPERTIES
Calibrated by: Sondre Skjoldli, Øyvind Albert
Type/Producer: Druck PTX 1400
SN: Z00227/07
Range: 0-2,5 bar g
Unit: Bar
CALIBRATION SOURCE PROPERTIES

Type/Producer: Druck DPI601
SN: 14206/96-1
Uncertainty [%]: 0,05
POLY FIT EQUATION:

Y= -106,73877956E+0X^0 + 50,04254576E+0X^1
CALIBRATION SUMARY:
Max Uncertainty : 1,153234 [%]
Max Uncertainty : 0,442137 [Bar]
RSQ : 0,999966
Calibration points : 20
Figure 1 : Calibration chart (The uncertainty band is multiplied by 10 )
COMMENTS:
Atmospheric Pressure when calibrated: 98,538 kPa
The uncertainty is calculated with 95% confidence. The uncertainty includes the randomness in the calibrated instrument during the calibration, systematic
uncertainty in the instrument or property which the instrument under calibration is compared with (dead weight manometer, calibrated weights etc.), and due to
regression analysis to fit the calibration points to a linear calibration equation. The calculated uncertainty can be used as the total systematic uncertainty of the
calibrated instrument with the given calibration equation.
_______________________________________
Sondre Skjoldli, Øyvind Albert
B.8.4 Calibration certificate GE UNIK 5000
GEUNIK5000kalibrering 24.04.18
CALIBRATION REPORT
Calibrated by: Sondre Skjoldli, Øyvind Albert
Type/Producer: GE UNIK 5000
SN: 4321073
Range: 0-2,5 bar g
Unit: Bar

Type/Producer: Druck DPI601
SN: 14206/96-1
Uncertainty [%]: 0,05
POLY FIT EQUATION:

Y= -125,51513139E+0X^0 + 62,47049663E+0X^1
CALIBRATION SUMARY:
Max Uncertainty : 0,127840 [Bar]
RSQ : 1,000000
COMMENTS
Atmospheric Pressure when calibrated: 98,538 kPa
_______________________________________
Sondre Skjoldli, Øyvind Albert
B.8.5 Calibration certificate HBM T22 for pump operation
Torque pump operation
CALIBRATION REPORT
Calibrated by: Sondre Skjoldli and Øyvind Albert
Type/Producer: HBM T22/200NM
SN: 01709720
Range: 4-37,5
Unit: Nm

Type/Producer: Calibrated Weights
SN: -
Uncertainty [%]: -
POLY FIT EQUATION:

Y= + 1,32395301E+0X^0 + 40,22771287E+0X^1
CALIBRATION SUMARY:
Max Uncertainty : 0,078748 [Nm]
RSQ : 0,999871
_______________________________________
Sondre Skjoldli and Øyvind Albert
B.8.6 Calibration certificate HBM T22 for turbine operation
Torque turbine operation
CALIBRATION REPORT
Calibrated by: Sondre Skjoldli and Øyvind Albert
Type/Producer: HBM T22/200NM
SN: 01709720
Range: 4-37,5
Unit: Nm

Type/Producer: Calibrated Weights
SN: -
Uncertainty [%]: -
POLY FIT EQUATION:

Y= -1,33051734E+0X^0 - 40,31310084E+0X^1
CALIBRATION SUMARY:
Max Uncertainty : 0,024987 [Nm]
RSQ : 0,999988
_______________________________________
Sondre Skjoldli and Øyvind Albert
76
APPENDIX C
LabVIEW
77
APPENDIX C. LABVIEW 78
C.1 Front panel and block diagram

APPENDIX C. LABVIEW 79
80
APPENDIX D
Matlab codes
D.1 PAT prediction and experimental results compared
1 clear a l l
2 clc
3 close a l l
4
5 g =9.82146516;
6 rho = 998.5986;
7 Ain = pi * ( 0 . 0 3 2 ^ 2 ) / 4 ;
8 Aout = pi * ( 0 . 0 5 ^ 2 ) / 4 ;
9 a = ( Aout * Ain ) ^2/( Aout^2−Ain ^2) ; %In turbine mode
10 Z = −0.37; %In turbine mode
11
12 %Pump input parameters ( e f f i c i e n c y assumed eta =1)

13 n_R=1450; %Rated r o t a t i o n a l speed
14 D1P=0.051; %I n l e t diameter
15 D2P=0.153; %Outlet diameter
16 B1P=0.0304; %For regne ut cm1p
17 beta2P =29; %mean( [ 2 9 . 5 , 29.25 , 28 , 29.75 , 29 , 2 8 . 7 5 ] ) %Blade angle
18 B2P=0.005; %Outlet height
19
20 %Empirical c o r r e l a t i o n s
21 eta_pump = 0 . 5 8 ; %XX
81
APPENDIX D. MATLAB CODES 82
22 eta_h_pump = 0 ;
23 H_pump = 5 . 7 5 ; %XX
24 Q_pump = 0 . 0 0 2 5 ; %XX
25
26 %Calculated pump parameters

27 cm1P=Q_pump/ ( pi * D1P * B1P) ;
28 omega_R=(2 * pi * n_R ) / 6 0 ;
29 u2P=(omega_R * D2P) / 2 ;
30 %cm2P=1.1 * cm1P ;
31 cm2P=2.03 * cm1P ;
32 cu2P=u2P−(cm2P/ tan ( beta2P * pi /180) ) ;
33 alpha2P=atan (cm2P/cu2P ) * (180/ pi ) ;
34
35 %PAT calculated values

36 D1=D2P ;
37 D2=D1P ;
38 B1=B2P ;
39 beta1=beta2P ;
40 u1=omega_R * (D1/2) ;
41 cm1=cm2P ;
42 cu1 = cu2P ;
43 alpha1=atan (cm1/cu1 ) * (180/ pi ) ;
44 u2=omega_R * (D2/2) ;
45 H_R=u1 * cu1/g ;
46 Q_R=cm1* pi *D1* B1 ;
47 phi =(u2^2) / ( g *H_R) ;
48 sigma=(1−phi ) /(1+ phi ) ;
49
50 n= [ 9 5 0 : 1 : 2 1 5 0 ] ;
51 %PAT
52 f o r i =1: length (n)
53 i f (1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) <0
54 q ( i )=−s q r t ( abs(1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) ) ;
55 Q( i ) =q ( i ) * Q_R ;
56 n_ed ( i ) =n( i ) * (D2/ s q r t ( g *H_R) ) ;
57 Q_ed( i ) =Q( i ) / (D2^2 * s q r t ( g *H_R) ) ;
58 H( i ) = ( ( ( n_R *D2) /n_ed ( i ) ) ^2) /g ;
59 Q( i ) =Q_ed( i ) *D2^2 * s q r t ( g *H( i ) ) ;
60 else
61 q ( i ) = s q r t (1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) ;
62 Q( i ) =q ( i ) * Q_R ;
63 n_ed ( i ) =n( i ) * (D2/ s q r t ( g *H_R) ) ;
64 Q_ed( i ) =Q( i ) / (D2^2 * s q r t ( g *H_R) ) ;
65 H( i ) = ( ( ( n_R *D2) /n_ed ( i ) ) ^2) /g ;
66 Q( i ) =Q_ed( i ) *D2^2 * s q r t ( g *H( i ) ) ;
67 end
68 end
69
70 newcm1=2.005 * cm2P ;
71 newalpha1=atan (newcm1/cu1 ) * (180/ pi ) ;
72 newH_R=u1 * cu1/g ;
73 newQ_R=newcm1* pi *D1* B1 ;
74
75 f o r i =1: length (n)

76 i f (1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) <0
77 q ( i )=−s q r t ( abs(1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) ) ;
78 newQ( i ) =q ( i ) *newQ_R ;
79 n_ed ( i ) =n( i ) * (D2/ s q r t ( g *H_R) ) ;
80 Q_ed( i ) =newQ( i ) / (D2^2 * s q r t ( g *H_R) ) ;
81 newH( i ) = ( ( ( n_R *D2) /n_ed ( i ) ) ^2) /g ;
82 newQ( i ) =Q_ed( i ) *D2^2 * s q r t ( g *newH( i ) ) ;
83 else
84 q ( i ) = s q r t (1 −(sigma * ( ( n( i ) /n_R ) ^2−1) ) ) ;
85 newQ( i ) =q ( i ) *newQ_R ;
86 n_ed ( i ) =n( i ) * (D2/ s q r t ( g *H_R) ) ;
87 Q_ed( i ) =newQ( i ) / (D2^2 * s q r t ( g *H_R) ) ;
88 newH( i ) = ( ( ( n_R *D2) /n_ed ( i ) ) ^2) /g ;
89 newQ( i ) =Q_ed( i ) *D2^2 * s q r t ( g *newH( i ) ) ;
90 end
91 end
92
93
94 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
95 %E x t r a c t i n g data from l o g f i l e s
96
97 t x t f i l e s = dir ( ’ * . txt ’ ) ;
98 x =1;
99 for f i l e = t x t f i l e s ’
100 NumHeaders = 8 ;
101 NumDataLines = 1000;
102 fmt = [ ’%f %f %f %f %f %f %f ’ ] ;
103 f i d = fopen ( f i l e .name) ;
104 data = textscan ( f i d , fmt , NumDataLines , ’ HeaderLines ’ , NumHeaders) ;
105 fclose ( fid ) ;
106
107 %P _ i n l e t P_outlet Flow Torque RPM Head Efficiency

108 f o r y = 1:7
109 datamean ( y ) = mean( data { y } ) ;
110 datamatrix ( x , y ) = datamean ( y ) ;
111 [ f ( x , y ) ] = xErr ( data { y } , 0 . 0 5 ) ;
112 end
113 x=x +1;
114 end
115
116 P1 = datamatrix ( : , 1 ) . * 1 0 0 0 ;
117 P2 = datamatrix ( : , 2 ) . * 1 0 0 0 ;
118 Qtest = datamatrix ( : , 3 ) . * (3600/1000) ;
119 T = datamatrix ( : , 4 ) ;
120 rpm = datamatrix ( : , 5 ) ;
121 Htest = datamatrix ( : , 6 ) ;
122 eta = datamatrix ( : , 7 ) ;
123
124 %Finding BEP : max e f f i c i e n c y value and location

125 PowT = [ 0 . 6 0.51 0.42 0.36 0.27 0.22 0.19 0.15 0.11 0.065 0.025 0.005 −0.02 −0.06 −0.08 −0.1 −0.11 −0.12];
126 f o r i = 1 : length ( Qtest )
127 etaT ( i ) = (PowT( i ) * 1000) / ( rho * g * ( Qtest ( i ) /3600) * Htest ( i ) ) ;
128 end
129 [M, I ]=max( etaT ) ;
130 Q_Rtest=datamatrix ( I , 3 ) * (3600/1000) ;
131 H_Rtest=datamatrix ( I , 6 ) ;
132
133 %c a l i b r a t i o n u n c e r t a i n t i e s
134 fP1cal = 0.00495;
135 fP2cal = 0.01154;
136 fQcal = 0.00231;
137 f T c a l = 0.005633;
138 fZ = 0.00027;
139
140 %t e s t u n c e r t a i n t i e s
141 f (: ,1)=f (: ,1) ./10;
142 %Total uncertainty
143 fQtot = s q r t ( fQcal^2+ f ( : , 3 ) . ^ 2 ) ;
144 f P 1 t o t = s q r t ( fP1cal ^2+ f ( : , 1 ) . ^ 2 ) ;
146 f P t o t = s q r t ( f P 1 t o t .^2+ f P 2 t o t . ^ 2 ) ;
147 fetaT = 0 . 0 5 ;
148 %mean
149 meanfP1 = mean( f ( : , 1 ) ) ;
150 meanfP2 = mean( f ( : , 2 ) ) ;
151 meanfQ = mean( f ( : , 3 ) ) ;
152 %t o t a l mean
153 maxfQtot = s q r t ( fQcal^2+meanfQ^2) ;
154 maxfP1tot = s q r t ( fP1cal ^2+meanfP1^2) ;
156
157 %derived u n c e r t a i n t i e s
158 deltaP = P1−P2 ;
159 f o r i =1: length ( Qtest )
160 fHder ( i ) = s q r t ( ( a^2 * deltaP ( i ) ^2 * f P t o t ( i ) ^2 + Qtest ( i ) ^4 * rho^2 * fQtot ( i ) ^2 + Z^2 * rho^2 * g^2 * a^2 * fZ ^2) / ( deltaP
( i ) ^2 * a^2 + deltaP ( i ) * Qtest ( i ) ^2 * rho * a + 2 * deltaP ( i ) * Z * rho * g * a^2 + ( Qtest ( i ) ^4 * rho ^2/4) + ( Qtest ( i ) ^2 * Z
* rho^2 * a * g ) + Z^2 * rho^2 * g^2 * a^2) ) ;

161 end
162 maxfHder = max( fHder ) ;
163
164 %Empirical c o r r e l a t i o n s
165 Hsharma = H_pump/eta_pump ^ ( 1 . 2 ) ;
166 Qsharma = Q_pump/eta_pump ^ ( 0 . 8 ) ;
167
168 %P l o t t i n g predicted Q_ed−n_ed

169 figure ;
170 plot ( n_ed , Q_ed)
171 t i t l e ( ’Q_{ ed } − n_ { ed } diagram ’ )
172 y l a b e l ( ’Q_{ ed } ’ )
173 x l a b e l ( ’ n_ { ed } ’ )
174 legend ( ’ Predicted PAT c h a r a c t e r i s t i c ’ , ’ Location ’ , ’ northeast ’ )
175
176 %P l o t t i n g predicted H−Q

177 ind2 = H>0 & H<30;
178 figure ;
179 plot (Q( ind2 ) . * 3 6 0 0 ,H( ind2 ) , ’−−b ’ )
180 hold on
181 plot (Q_R. * 3 6 0 0 ,H_R, ’ r * ’ , ’ markers ’ , 10)
182 t x t 1 = [ ’ \uparrow Symmetry BEP : H_R= ’ num2str ( round (H_R, 2 ) ) , ’m , Q_R= ’ num2str ( round (Q_R. * 3 6 0 0 , 2 ) ) , ’m^3/h ’ ] ;
183 t x t 1 = [ ’Symmetry BEP : ’ ] ;
184 t x t 4 = [ ’H_R= ’ num2str ( round (H_R, 2 ) ) , ’m’ ] ;
185 t x t 5 = [ ’Q_R= ’ num2str ( round (Q_R. * 3 6 0 0 , 2 ) ) , ’m^3/h \ rightarrow ’ ] ;
186 t e x t ( 3 , 1 3 , { txt1 , txt4 , t x t 5 } , ’ FontSize ’ , 1 1 )
187
188 %P l o t t i n g Sharma
189 plot (Qsharma. * 3 6 0 0 ,Hsharma , ’m* ’ , ’ markers ’ , 10)
190 %t x t 2 = [ ’ \ l e f t a r r o w Sharma BEP : H_R= ’ num2str ( round (Hsharma , 2 ) ) , ’m, Q_R= ’ num2str ( round (Qsharma. * 3 6 0 0 , 2 ) ) , ’m
^3/h ’ ] ;
191 %t e x t (Qsharma. * 3 6 0 0 ,Hsharma , txt2 , ’ FontSize ’ , 11)
192 %t x t 2 = [ ’ Sharma BEP : ’];
193 %t x t 4 = [ ’H_R= ’ num2str ( round (Hsharma , 2 ) ) , ’m’ ] ;
194 %t x t 5 = [ ’Q_R= ’ num2str ( round (Qsharma. * 3 6 0 0 , 2 ) ) , ’m^3/h \ rightarrow ’ ] ;
195 %t e x t ( 6 , 1 3 , { txt2 , txt4 , t x t 5 } , ’ FontSize ’ , 1 1 )
196
197 %P l o t t i n g t e s t d a t a
198 errorbar ( Qtest , Htest , fHder ’ . * Htest , fHder ’ . * Htest , fQtot . * Qtest , fQtot . * Qtest , ’ k ’ ) ;
199 plot ( Qtest , Htest , ’ r ’ )

200 plot ( Q_Rtest , H_Rtest , ’b * ’ , ’ markers ’ , 10)
201 t x t 3 = [ ’ \ l e f t a r r o w BEP : H= ’ num2str ( round ( H_Rtest , 2 ) ) , ’m , Q= ’ num2str ( round ( Q_Rtest , 2 ) ) , ’m^3/h ’ ] ;
202 t e x t ( Q_Rtest , H_Rtest , txt3 , ’ FontSize ’ , 11)
203
204 %P l o t t i n g NEW predicted H−Q

205 ind2 = H>0 & H<30;
206 plot (newQ( ind2 ) . * 3 6 0 0 ,newH( ind2 ) , ’b ’ )
207 plot (newQ_R. * 3 6 0 0 ,newH_R, ’ r * ’ , ’ markers ’ , 10)
208 t x t 6 = [ ’ \uparrow Symmetry BEP : H_R= ’ num2str ( round (newH_R, 2 ) ) , ’m , Q_R= ’ num2str ( round (newQ_R. * 3 6 0 0 , 2 ) ) , ’m
^3/h ’ ] ;
209 %t x t 1 = [ ’ Symmetry BEP : ’];
210 %t x t 4 = [ ’H_R= ’ num2str ( round (H_R, 2 ) ) , ’m’ ] ;
211 %t x t 5 = [ ’Q_R= ’ num2str ( round (Q_R. * 3 6 0 0 , 2 ) ) , ’m^3/h \ rightarrow ’ ] ;
212 %t e x t (Q_R. * 3 6 0 0 ,H_R, t x t 1 )
213 %t e x t ( 1 , 1 4 , txt1 , ’ FontSize ’ , 11)
214 %t e x t ( 5 , 1 3 , { txt1 , txt4 , t x t 5 } , ’ FontSize ’ , 1 1 )
215 t e x t ( 1 8 , 1 0 , { t x t 6 } , ’ FontSize ’ , 1 1 )
216
217
218 t i t l e ( ’H−Q diagram f o r n_ {R}=1450 rpm ’ )

219 y l a b e l ( ’H [m] ’ )
220 x l a b e l ( ’Q [m^3/h ] ’ )
221 legend ( ’ Predicted c h a r a c t e r i s t i c 1 s t improvement ’ , ’ Predicted BEP ’ , ’Sharma BEP ’ , ’ Error in H and Q ’ , ’ Test
c h a r a c t e r i s t i c ’ , ’ Test BEP ’ , ’ Predicted c h a r a c t e r i s t i c 2nd improvement ’ , ’ Location ’ , ’ northwest ’ )
222
223 %P l o t t i n g e f f i c i e n c y
224 figure ;
225 ind3 = etaT >−0.15 & etaT <0.5 ;
226 errorbar ( Qtest ( ind3 ) , etaT ( ind3 ) , fetaT . * etaT ( ind3 ) , fetaT . * etaT ( ind3 ) , fQtot ( ind3 ) . * Qtest ( ind3 ) , fQtot ( ind3 ) . * Qtest
( ind3 ) , ’ k ’ ) ;
227 hold on
228 plot ( Qtest ( ind3 ) , etaT ( ind3 ) , ’b ’ , ’ linewidth ’ , 1 )
229 grid on
230 plot ( Q_Rtest ,M, ’ r * ’ , ’ markers ’ , 10)
231 t x t 2 = [ ’ \uparrow \ eta = ’ num2str ( round (M, 3 ) ) ] ;
232 t e x t ( 1 7 . 9 , 0 . 3 6 , txt2 , ’ FontSize ’ , 12)
233 t i t l e ( ’ \ eta − Q diagram f o r n_ {R}=1450 rpm ’ )
234 y l a b e l ( ’ \ eta ’ , ’ FontSize ’ , 18)
235 x l a b e l ( ’Q [m^3/h ] ’ )
236 legend ( ’ Error in \ eta and Q ’ , ’Pump e f f i c i e n c y curve ’ , ’ Location ’ , ’ southeast ’ )
D.2 Pump experimental results
1 clear a l l
2 close a l l
3 clc
4
5 rho = 998.5986;
6 g = 9.82146516;
7 Aout = pi * ( 0 . 0 3 2 ^ 2 ) / 4 ;
8 Ain = pi * ( 0 . 0 5 ^ 2 ) / 4 ;
9 a = ( Aout * Ain ) ^2/( Ain^2−Aout^2) ;
10 Z = 0.37;
11
12 t x t f i l e s = dir ( ’ * . txt ’ ) ;
13 x =1;
14 for f i l e = t x t f i l e s ’
15 NumHeaders = 8 ;
16 NumDataLines = 1000;
17 fmt = [ ’%f %f %f %f %f %f %f ’ ] ;
18 f i d = fopen ( f i l e .name) ;
19 data = textscan ( f i d , fmt , NumDataLines , ’ HeaderLines ’ , NumHeaders) ;
20 fclose ( fid ) ;
21
22 %P _ i n l e t P_outlet Flow Torque RPM Head Efficiency

23 f o r y = 1:7
24 datamean ( y ) = mean( data { y } ) ;
25 datamatrix ( x , y ) = datamean ( y ) ;
26 [ f ( x , y ) ] = xErr ( data { y } , 0 . 0 5 ) ;
27 end
28 x=x +1;
29 end
30
31
32 P1 = datamatrix ( : , 1 ) * 1000;
33 P2 = datamatrix ( : , 2 ) * 1000;
34 Q = datamatrix ( : , 3 ) ;
35 T = datamatrix ( : , 4 ) ;
36 rpm = datamatrix ( : , 5 ) ;
37 H = datamatrix ( : , 6 ) ;
38 eta = datamatrix ( : , 7 ) ;
39
40 %Finding BEP : max e f f i c i e n c y value and location

41 PowP = [ 0 . 3 5 0.345 0.34 0.335 0.335 0.33 0.33 0.32 0.32 0.32 0.305 0.305 0.30 0.30 0.295 0.285 0.28 0.265 0.26
0.255 0.2375 0.2375 0.225 0 . 2 1 5 ] ;
42 f o r i = 1 : length (Q)
43 etaP ( i ) = ( rho * g * (Q( i ) /1000) *H( i ) ) / (PowP( i ) * 1000) ;

44
45 end
46 [M, I ]=max( etaP ) ;
47 Q_R=datamatrix ( I , 3 ) ;
48 H_R=datamatrix ( I , 6 ) ;
49 etaP ( 1 0 ) = ( etaP ( 9 ) +etaP ( 1 1 ) ) / 2 ; %%%
50
51 %c a l i b r a t i o n u n c e r t a i n t i e s
52 fP1cal = 0.00495;
53 fP2cal = 0.01154;
54 fQcal = 0.00231;
55 f T c a l = 0.005633;
56 fZ = 0.00027;
57
58 %Total uncertainty
59 fQtot = s q r t ( fQcal^2+ f ( : , 3 ) . ^ 2 ) ;
62 f P t o t = s q r t ( f P 1 t o t .^2+ f P 2 t o t . ^ 2 ) ;
63
64 %mean
65 meanfP1 = mean( f ( : , 1 ) ) ;
66 meanfP2 = mean( f ( : , 2 ) ) ;
67 meanfQ = mean( f ( : , 3 ) ) ;
68 %t o t a l mean
69 maxfQtot = s q r t ( fQcal^2+meanfQ^2) ;
72
73 %derived u n c e r t a i n t i e s
74 deltaP = P2−P1 ;
75 f o r i =1: length (Q)
76
77 fHder ( i ) = s q r t ( ( a^2 * deltaP ( i ) ^2 * f P t o t ( i ) ^2 + Q( i ) ^4 * rho^2 * fQtot ( i ) ^2 + Z^2 * rho^2 * g^2 * a^2 * fZ ^2) / ( deltaP ( i ) ^2 * a
^2 + deltaP ( i ) *Q( i ) ^2 * rho * a + 2 * deltaP ( i ) * Z * rho * g * a^2 + (Q( i ) ^4 * rho ^2/4) + (Q( i ) ^2 * Z * rho^2 * a * g ) + Z^2 * rho
^2 * g^2 * a^2) ) ;
78
79 end
80 maxfHder = max( fHder ) ;
81
82 figure ;
83 subplot ( 2 , 1 , 1 )
84 hold on
85 grid on
86 errorbar (Q* (3600/1000) ,H, fHder ’ . * H* (3600/1000) , fHder ’ . * H* (3600/1000) , fQtot . * Q* (3600/1000) , fQtot . * Q* (3600/1000) ,
’k ’ ) ;
87 plot (Q* (3600/1000) ,H, ’b ’ )
88 plot (Q_R * (3600/1000) , H_R, ’ r * ’ , ’ markers ’ , 10)
89 t x t 1 = [ ’ \downarrow BEP : H= ’ num2str ( round (H_R, 2 ) ) , ’m , Q= ’ num2str ( round (Q_R * (3600/1000) , 2 ) ) , ’m^3/h ’ ] ;
90 t e x t ( 8 . 8 6 , 6 . 4 , txt1 , ’ FontSize ’ , 11)
91 t i t l e ( ’H−Q diagram f o r n_ {R}=1450 rpm ’ )
92 y l a b e l ( ’H [m] ’ )
93 x l a b e l ( ’Q [m^3/h ] ’ )
94 legend ( ’ Error in H and Q ’ , ’Pump t e s t c h a r a c t e r i s t i c ’ , ’ Location ’ , ’ southwest ’ )
95
96 fetaP = 0 . 0 5 ;
97 subplot ( 2 , 1 , 2 )
98 hold on
99 errorbar (Q* (3600/1000) , etaP , fetaP . * etaP , fetaP . * etaP , fQtot . * Q* (3600/1000) , fQtot . * Q* (3600/1000) , ’ k ’ ) ;
100 plot (Q* (3600/1000) , etaP , ’b ’ )
101 grid on
102 plot (Q_R * (3600/1000) ,M, ’ r * ’ , ’ markers ’ , 10)
103 t x t 2 = [ ’ \uparrow \ eta = ’ num2str ( round (M, 3 ) ) ] ;
104 t e x t ( 8 . 8 6 , 0 . 4 1 , txt2 , ’ FontSize ’ , 12)
105 t i t l e ( ’ \ eta − Q diagram f o r n_ {R}=1450 rpm ’ )
106 y l a b e l ( ’ \ eta ’ , ’ FontSize ’ , 18)
107 x l a b e l ( ’Q [m^3/h ] ’ )
108 legend ( ’ Error in \ eta and Q ’ , ’Pump e f f i c i e n c y curve ’ , ’ Location ’ , ’ southeast ’ )
D.3 Test velocity diagrams
1 clear a l l
2 close a l l
3 clc
4
5 %Rated BEP values from t e s t s

6 Q_RT = 0.0049863;
7 Q_RP = 0.0024869;
8 H_RP= 5 . 9 1 ;
9
10 %Pump input parameters

11 n_R=1450; %Rated r o t a t i o n a l speed
12 D1P=0.051; %I n l e t diameter
13 D2P=0.153; %Outlet diameter
14 B1P=0.0304; %I n l e t height
15 beta2P =29; %Blade angle
16 B2P=0.005; %Outlet height
17
18 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
19 %Parameters needed to plot v e l o c i t y diagrams
20
21 omega_R=(2 * pi * n_R ) / 6 0 ;
22 cm1P = Q_RP/ ( pi * B1P * D1P) ;
23 u1P = omega_R * (D1P/2) ;
24 u2P = omega_R * (D2P/2) ;
25 cm2P = Q_RP/ ( pi * B2P * D2P) ;
26 cu2P = u2P−(cm2P/ tan ( beta2P * pi /180) ) ;
27 %cu2P = ( 9 . 8 2 * H_RP) /u2P * 0 . 8
28
29 D1T = D2P ;
30 D2T = D1P ;
31 B1T = B2P ;
32 B2T = B1P ;
33 beta1T = beta2P ;
34 u1T = omega_R * (D1T/2) ;
35 cm1T = Q_RT/ ( pi * B1T *D1T)
36 cu1T=u1T−(cm1T/ tan ( beta1T * pi /180) ) ;
37 u2T = omega_R * (D2T/2) ;
38 cm2T = Q_RT/ ( pi * B2T *D2T)
39
40 alpha2P = atan (cm2P/cu2P ) * (180/ pi )

41 alpha1T = atan (cm1T/cu1T ) * (180/ pi )
42
43 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
44 %Points in order to plot v e l o c i t y diagrams

45
46 p1 = [0 0 ] ;
47 p2 = [−cu2P 0 ] ;
48 p3 = [−u2P 0 ] ;
49 p4 = [−cu2P cm2P ] ;
50 p5 = [ cu1T , 0 ] ;
51 p6 = [ u1T , 0 ] ;
52 p7 = [ cu1T , −cm1T ] ;
53 p10 = [0 , 0 ] ;
54 p11 = [−u1P , 0 ] ;
55 p12 = [ 0 , cm1P ] ;
56 p13 = [ u2T , 0 ] ;
57 p14 = [ 0 , −cm2T ] ;
58
59 %Pump outlet , turbine i n l e t

60 figure ( ) ;
61 subplot ( 2 , 1 , 1 )
62 hold on
63 plot ( [ p1 ( 1 ) p2 ( 1 ) ] , [ p1 ( 2 ) p2 ( 2 ) ] , ’b ’ , ’ LineWidth ’ , 1 )
64 plot ( [ p1 ( 1 ) p3 ( 1 ) ] , [ p1 ( 2 ) p3 ( 2 ) ] , ’b ’ , ’ LineWidth ’ , 1 ) ;
73 t i t l e ( ’ Actual v e l o c i t y diagrams : Pump o u t l e t and PAT i n l e t ’ )
74 y l a b e l ( ’m/ s ’ )
75 x l a b e l ( ’m/ s ’ )
76
77 %Turbine outlet , pump i n l e t

78 subplot ( 2 , 1 , 2 )
79 hold on
86 t i t l e ( ’ Actual v e l o c i t y diagrams : Pump i n l e t , PAT o u t l e t ’ )
87 y l a b e l ( ’m/ s ’ )
88 x l a b e l ( ’m/ s ’ )
D.4 Error function
1 function [ f x ] = xErr ( x , a )
2 Sx=var ( x ) ;
3 t s = t i n v ( [ a/2 1−a / 2 ] , length ( x ) −1) ;
4
5 f x = t s ( 2 ) * ( Sx/ s q r t ( length ( x ) ) ) ;
6
7 end
APPENDIX E
Risk assessment
93
Scanned by CamScanner
APPENDIX E. RISK ASSESSMENT 94




Fulltext

Uploaded by

Copyright:

Available Formats

Fulltext

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fulltext

Uploaded by

Copyright:

Available Formats

Pump as Turbine

Symmetry Prediction Method for Pump as

Master of Energy and Environmental Engineering

Norwegian University of Science and Technology

Øyvind Albert and Sondre Skjoldli

2.1 Different prediction methods investigated by Williams (1994)[1]. . . . . . . . . . . . 7

3.1 Test-Pump 2 impeller geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.1 Comparison of test and estimated values in pump operation. . . . . . . . . . . . . . 31

B.1 Component errors in the calibration of an instrument. . . . . . . . . . . . . . . . . . 63

3.1 PAT test set-up UDSM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.1 H −Q characteristic and efficiency curve from pump testing. . . . . . . . . . . . . . 30

Preface and acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

A Paper written for CRHTVIII’18 47

B Calibration and Uncertainty Analysis 63

B.4 RPM sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

2.1 The concept of a Pump as Turbine

2.1.1 Advantages and disadvantages of a PAT

2.2 Earlier work

Table 2.1: Different prediction methods investigated by Williams (1994)[1].

Name of method/investigator Based on Head correction factor h̃ Discharge correction factor q̃

Grover Specific speed 2.693 − 0.0229N st 2.379 − 0.0264N st

2.3 Pump and turbine performance

2.4 Symmetry method

2.4.1 Design equations and assumptions

As the prediction method is based on a symmetrical transition between parameters in pump

u 1,t c u1,t − u 2,t c u2,t

2.4.2 Producing the PAT’s H −Q curve

2.5 Experimental preparation

3.1 Experimental setup UDSM

Table 3.1: Test-Pump 2 impeller geometry.

Parameter Value Unit

Figure 3.1: PAT test set-up UDSM.

until UDSM receives the necessary calibration apparatus.

Figure 3.3: a) Insertion of permanent magnets into the rotor - b) stator.

3.2 Pedrollo FG 32/160B

Parameter Value Unit

Figure 3.4: The pump impeller of Pedrollo FG 32/160B.

3.3 Experimental setup NTNU

Ø50 Turbine operation

Flow sensor Pump operation Torque and

Figure 3.5: Illustration of test rig constructed at NTNU, designed in Creo.

3.4 Data acquisition

3.4.1 Pump operation

3.4.2 Pump as Turbine

3.5 Calibration and uncertainty

Table 3.3: Total uncertainty for the experiment parameters.

Quantity Calibration Uncertainty Mean Measured Uncertainty Total Mean Uncertainty

4.1 Pump testing

Figure 4.1: H −Q characteristic and efficiency curve from pump testing.

Table 4.1: Comparison of test and estimated values in pump operation.

Parameter Test value Estimated (given) value Unit

4.2 PAT testing

Figure 4.2: H −Q characteristics, both predicted and acquired through testing.

Figure 4.3: Pump as turbine efficiency curve.

4.2.1 Velocity diagrams

Test velocity diagrams: Pump outlet, PAT inlet

Test velocity diagrams: Pump inlet, PAT outlet

Figure 4.4: Velocity diagrams from test results.

4.3 Potential improvements