energies

Article
A Performance Prediction Method for Pumps as
Turbines (PAT) Using a Computational Fluid
Dynamics (CFD) Modeling Approach
Emma Frosina *, Dario Buono and Adolfo Senatore
Department of Industrial Engineering, University of Naples Federico II, Via Claudio, 21-80125 Naples, Italy;
[email protected] (D.B.); [email protected] (A.S.)
* Correspondence: [email protected]; Tel.: +39-081-768-32-85

Academic Editor: Bjørn H. Hjertager

Received: 18 October 2016; Accepted: 6 January 2017; Published: 16 January 2017

Abstract: Small and micro hydropower systems represent an attractive solution for generating
electricity at low cost and with low environmental impact. The pump-as-turbine (PAT) approach
has promise in this application due to its low purchase and maintenance costs. In this paper, a new
method to predict the inverse characteristic of industrial centrifugal pumps is presented. This method
is based on results of simulations performed with commercial three-dimensional Computational
Fluid Dynamics (CFD) software. Model results have been first validated in pumping mode using data
supplied by pump manufacturers. Then, the results have been compared to experimental data for a
pump running in reverse. Experimentation has been performed on a dedicated test bench installed in
the Department of Civil Construction and Environmental Engineering of the University of Naples
Federico II. Three different pumps, with different specific speeds, have been analyzed. Using the
model results, the inverse characteristic and the best efficiency point have been evaluated. Finally,
results have been compared to prediction methods available in the literature.

Keywords: energy saving; PAT; urban hydraulic network; numerical modeling

1. Introduction
Electricity generation presents many issues and is studied with different techniques in order
to reduce its production cost and environmental impact. Conventional production with fossil fuels
presents problems associated with the high cost, rapid depletion and detrimental environmental
effects of these fuels. Renewable energy is probably the best solution for environmental issues, and
many solutions have been developed since the last century, such as hydropower, hydrogen, fuel cells,
biofuels, and solar power generation.
Among the renewable sources, small hydropower represents a very attractive source of energy
generation. In many countries, small and micro hydropower systems are an important means
of electricity generation. An efficient solution, from the point of view of energy efficiency, is the
adoption of a turbine, but the purchase and maintenance costs of turbines make their implementation
economically unattractive, especially for small hydropower [1–6].
Reverse-running centrifugal pumps (also called pumps as turbines or PAT) are a solution for
generating and recovering power in small and micro hydropower situations. Pumps are relatively
simple machines, inexpensive (compared to a hydraulic turbines), and readily available worldwide.
It has been estimated that the capital payback period of a reverse-running pump in the range of
5–50 kW is less than two years [7,8]. Moreover, the use of PAT could be suitable because manufacturers
of turbines worldwide are less numerous than pump producers, the market for turbines is smaller
compared to pumps, and pumps are mechanically simple and require less maintenance. Moreover,

Energies 2017, 10, 103; doi:10.3390/en10010103 www.mdpi.com/journal/energies

Energies 2017, 10, 103 2 of 19

an integral pump and electric motor can be purchased for use as a turbine and generator set; pumps
are available in a wide range of heads and flows and in a large number of standard sizes. Generally,
pumps have short delivery times, spare parts (such as seals and bearings) are easily available and the
installation can be done using standard pipes and fittings.
The use of a pump running in reverse mode to generate electricity is not new; the first
applications started almost 80 or 90 years ago and many theoretical and experimental studies have been
done [2–6,9,10]. Much research is still being conducted, especially to predict the operating conditions
and the efficiency of centrifugal pumps running in reverse [11].
The selection of a proper PAT for an existing site represents a critical issue because pump
manufacturers do not supply the characteristic of the pump running in reverse. Many methods have
been used to predict the inverse characteristic of a pump, based on numerical models, experiments, or
theoretical procedures [4–10].
This research has demonstrated that these methods can be used only for a limited set of pumps.
None of them, in fact, allows prediction for the reverse running conditions for all geometries and over
a wide range of pump specific speeds. Several studies, based on a modeling approach with CFD code,
are available and generally show good correspondence with the available experimental data [4,10].
A study [4] carried out with a computational model of a PAT is based on the concept called “flow
zone”. The flow regime within a PAT is divided into four major flow regions (volute casing, impeller,
casing outlet and draft tube). A comparison has been made between the experimental and numerical
results of a single stage end suction centrifugal pump that was operated in turbine mode at a speed
of 800 rpm. CFD predictions of the hydraulic parameters were in good correspondence with the
experimental results, but deviations (within 5% to 10%) have been found at certain load regions.
Nautiyal et al. [5,6] carried out a study on the application of CFD and its limitations for PAT
using cases reported by previous researchers [4,9,10]. The study reported that CFD analysis was an
effective design tool for predicting the performance of centrifugal pumps in turbine mode and for
identifying the losses in turbo-machinery components such as the draft tube, impeller and casing,
but there was some deviation between the experimental results and the CFD modeling results.
Barrio et al. [11] carried out a numerical investigation on the unsteady flow in commercial centrifugal
pumps operating in direct and reverse mode with the help of CFD code. The results of their simulation
were in good correspondence with the experimental results. The study revealed that in the reverse
mode, the flow only matched the geometry of the impeller at nominal conditions; re-circulating fluid
regions developed at low flow rates (near the discharge side of the blades) and high flow rates (near
the suction side).
Many correlations based on theoretical approaches are available to predict the performance of a
PAT. Several researchers (Stepanoff, Childs, Sharma, Wong, Williams, Alatorre-Frenk, and others) have
presented correlations for predicting the performance of a pump-as-turbine [5]. These correlations
were based upon either pump efficiency or specific speed. However, deviations of more than 20%
have been found between the experimental and predicted reverse operation of standard pumps [12].
The objective of these correlations is to calculate the best efficiency point (BEP) of pumps for operation
in turbine mode by using the pump operation data provided by the manufacturer.
In 1962, Childs [13] presented a PAT prediction method based on the efficiency of the pump.
A similar approach was then presented by McClaskey and Lundquist [14] and Lueneburg and
Nelson [15] in 1976 and 1985, respectively.
Hancock [16] stated that for most pumps the turbine BEP lies within 2% of the pump mode BEP.
Grover and Hergt [17,18] proposed a PAT prediction method based on specific speed for the turbine
mode (obtained similar to the specific speed for a pump). Grover’s method is applicable for the turbine
mode specific speed range between 10 and 50 [17]. A comparison between experimental results and
the methods proposed by the above researchers show relatively large deviations; therefore, the use of
these formulae must be confined to an approximate selection of PATs.
Energies 2017, 10, 103 3 of 19

Finally, a large number of experimental studies can be used to evaluate the inverse characteristics.
These are often limited to the specific pumps tested, so that they cannot serve as a valid tool for pump
selection, but are very useful for tuning and validating theoretical and modeling analyses.
In this paper, authors present a methodology for obtaining the reverse characteristics of a pump,
starting from the results of three-dimensional CFD models. After a description of all prediction
methods available in the literature, in the third section the adopted modeling approach is described.
Three pumps have been studied and modeled using the three-dimensional CFD commercial code
PumpLinx® , developed by Simerics Inc.® (1750 112th Ave NE, Ste C250, Bellevue, WA, USA). In the
fourth section, the test bench layout is shown with all the transducers’ characteristics.
Numerical models have been validated with experimental data obtained on a dedicated test bench
installed in the Department of Civil Construction and Environmental Engineering of the University of
Naples Federico II. Simulations have been run in both direct (as pump) and reverse (as turbine) modes
with good accuracy.
In the fifth section, the models’ results have been used to predict the efficiency curves of the three
analyzed pumps in both modes. At the end of the paper, results of the proposed methodology have
been compared to the prediction methods described in the second section. Analysis has demonstrated
that the existing prediction methods underestimate or overestimate the real operation.
In this paper, the authors have shown only the first step of a research done on PAT. Other pumps
are under study using the same modeling approach in order to realize a macro database for the
prediction of pump performance as turbines. The final aim is the identification of a new prediction
method more accurate than the others already available in literature. Using the new prediction method,
a reduction of the experimentation will then be realized allowing an easy and fast choice of the PAT for
each application.

2. Literature Overview on Prediction Methods

A methodology to calculate the inverse characteristic of a commercial pump is presented in this
paper. The proposed approach is based on results of CFD modeling using a commercial code developed
to simulate centrifugal machines. Therefore, it is important to describe prediction methods already
available in literature. In fact, these methods, will be used in the last section of the paper to analyze
the proposed methodology and to discuss our results. The following equations summarize different
methods to predict the pump inverse characteristics. They are based on theoretical or experimental
analyses [13–24]. Stepanoff [19] calculates the head, flow rate at the BEP in reverse mode using the
efficiency, head and flow rate value at the BEP in direct mode. All relations between head and flow
rate are reported in Equation (1).

Ht 1 Qt
= = √1 ; ηT = ηP ; Nst = Ns η p
Hp ηp ; Qp ηp (1)

Alatorre-Frenk [20] calculates the head, flow rate and efficiency at the BEP in reverse mode
using the efficiency, head and flow rate value at the BEP in direct mode. Correlations are presented
by following:
Ht 1 Qt 0.85η p 5 +0.385
Hp = 0.85η p 5 +0.385 ; Qp = 2η p 9.5 +0.205
; ηt = η p − 0.03 (2)

The prediction method developed by Sharma [21] calculates the head and flow rate at the BEP in
reverse mode using the efficiency, head and flow rate value at the BEP in direct mode:

Ht 1 Qt 1
Hp = η p 1.2
; Qp = η p 0.8
; Pt = Pp ; ηt = η p (3)
Energies 2017, 10, 103 4 of 19

Schmiedl [22] as Sharma [21] calculates the head and flow rate at the BEP in reverse mode
using the efficiency, head and flow rate value at the BEP in direct mode. Correlations are reported
by following:
Ht 2.5 Qt 2.4 ηt
H p = −1.4 + η ; Q p = −1.5 + η 2 p = 1.158 − 0.265Nst
p p
(4)

Head and flow rate at the BEP in reverse mode can be evaluated with correlations of Grover [17]
using the specific speed value of the PAT at the BEP (Nst = Ns η p ). All relations between head and
flow rate are listed by following.

Ht


 Hp = 2.693 − 0.0229Nst
Qt
Qp = 2.379 − 0.0264Nst (5)
ηT

= 0.893 − 0.0466Nst


ηP

As Grover [17], knowing that Nst = Ns η p , Hergt [18] calculates the head and flow rate at the BEP
in reverse mode using the specific speed value of the PAT at the BEP:
Ht 6 Qt 1.6
Hp = 1.3 − Nst −3 ; Qp = 1.3 − Nst −5 (6)

Relations of Childs [13] evaluate the head and the flow rate at the BEP in reverse mode using the
efficiency, head and flow rate value at the BEP in direct mode. Equations are reported by following:

Ht 1 Qt 1
Hp = ηp ; Qp = ηp ; ηt = η p (7)

Moreover, Derakhshan and Nourbakhsh [24] introduced a method based on theoretical analysis
to evaluate the BEP of an industrial centrifugal pump. This method is based on the geometrical and
hydraulic characteristics of the pump in direct mode. The final formula to evaluate the turbine’s
maximum efficiency is:

Pnt γ × Qt × Ht − Pvt − Plt − Pet − Pit − Pmt

ηt = = (8)
γ × Qt × Ht γ × Qt × Ht

All of presented methods are based on different hypotheses. In Section 6 of this paper, all methods
will be compared with the results of the proposed modeling methodology. The comparison has
confirmed that each prediction method can be used only for a limited set of pumps. None of them, in
fact, allows the prediction in the reverse running conditions for all geometries and over a wide range
of pump specific speeds. In some cases, performance is underestimated or overestimated by as much
as 30%.

3. Simulation Model
Three different centrifugal pumps have been modeled in order to obtain the necessary data
to predict the performance of the pumps by the described procedures. The analyzed pumps are
commercial ones and have three different specific speeds. The main characteristics are summarized in
Table 1. It was decided to use pumps with different heads (from 3.9 to 60 m) and flow rates (from 45.4
to 148 m3 /s) to have different geometries and operating conditions to better test the prediction method.

Table 1. Pumps characteristics.

Impeller Delivery Outlet

Hbep (m) Qbep (m3 /h)
Diameter (mm) Diameter (mm)
(Ns 37.6) 190 80 39 148
(Ns 20.5) 200 70 60 45.4
(Ns 64.0) 120 80 3.9 54
Energies 2017, 10, 103 5 of 19

On the top side of Figure 1, the disassembled pump with a NS = 37.6 is shown. This pump is
aEnergies
shrouded one
2017, 10, 103with one- channel impeller and six blades and is called Pump 1. Starting from the
5 of 18
real geometry in a .step format, the fluid volume has been extracted. In Figure 1, the fluid-volume is
in greenin
colored while
green the solidthe
while impeller is in blueisand
solid impeller the solid
in blue rotor
and the in red.
solid Inin
rotor the same
red. way,
In the fluidway,
same volumes
fluid
of others of
volumes two pumps
others twohave been
pumps extracted
have and thenand
been extracted modelled using a 3D-CFD
then modelled approach.
using a 3D-CFD approach.

Figure 1.
Figure 1. Geometry
Geometry and
and fluid
fluid volume.
volume.

The study has been approached using the commercial code PumpLinx®. PumpLinx® is a three-
The study has been approached using the commercial code PumpLinx® . PumpLinx® is a
dimensional CFD software developed by Simerics Inc. (1750 112th Ave NE, Ste C250, Bellevue, WA
three-dimensional CFD software developed by Simerics Inc. (1750 112th Ave NE, Ste C250, Bellevue,
98004, USA) [25–29]. It numerically solves the fundamental conservation equations of mass,
WA 98004, USA) [25–29]. It numerically solves the fundamental conservation equations of mass,
momentum and energy and includes robust models of turbulence and cavitation.
momentum and energy and includes robust models of turbulence and cavitation.
The fluid volume of each pump has then been meshed with the PumpLinx® grid generator using
The fluid volume of each pump has then been meshed with the PumpLinx® grid generator using
a body-fitted binary tree approach. These grids have been demonstrated to be extremely accurate and
a body-fitted binary tree approach. These grids have been demonstrated to be extremely accurate and
efficient [25–29]. In fact, the parent-child tree architecture allows for an expandable data structure
efficient [25–29]. In fact, the parent-child tree architecture allows for an expandable data structure with
with reduced memory storage, the binary refinement is optimal for transitioning between different
reduced memory storage, the binary refinement is optimal for transitioning between different length
length scales and resolutions within the model, the majority of cells are cubes, and, since the grid is
scales and resolutions within the model, the majority of cells are cubes, and, since the grid is created
created from a volume, it can tolerate inaccurate CAD surfaces with small gaps and overlaps. It is
from a volume, it can tolerate inaccurate CAD surfaces with small gaps and overlaps. It is important
important to underline that, cells are hexagonal not deformed therefore the skewness is zero [30].
to underline that, cells are hexagonal not deformed therefore the skewness is zero [30].
Figure 2 shows the binary tree mesh of three pumps under study: Pump 1 (Ns = 37.6), Pumps 2
Figure 2 shows the binary tree mesh of three pumps under study: Pump 1 (Ns = 37.6), Pumps 2
(Ns = 20.5) and Pump 3 (Ns = 64).
(Ns = 20.5) and Pump 3 (Ns = 64).
It is important to underline that using the binary tree approach in the boundary layer can easily
increase the grid density on the surface without excessively increasing the total cell count. In this way,
the grid has been subdivided and cut to conform it to the surface in regions of high curvature and
small details [27]. Impellers and rotors fluid volumes of each model have been meshed separately.
A maximum cell size of 0.025 has been chosen, where no cell in the volume can have a cell side larger
than the maximum cell size. The minimum cell size has been fixed at 0.0001. The minimum cell size
is a parameter used to limit how small cells can be in attempting to resolve the geometry using the
general mesher. No cell in the volume can have a cell side smaller than the minimum cell size. The cell
size on surfaces has also been fixed at 0.00625. This parameter is used to control the size of the cells
for all surfaces of a mesh volume. Using a pure Eulerian approach the mesh of rotors is deformed by
squeezing and expanding the cells. The rotor mesh is colored in green in Figure 2.
Energies 2017, 10, 103 6 of 19
Energies 2017, 10, 103 6 of 18

Figure
Figure 2.
2. Binary
Binary tree
tree mesh
mesh for
for three
three pumps.
pumps.

It is important to underline that using the binary tree approach in the boundary layer can easily
A mesh sensitivity, as well known, is fundamental in studying any new problem with a CFD solver.
increase the grid density on the surface without excessively increasing the total cell count. In this
The target is to allow excellent accuracy (in comparison with experimental tests) and computational
way, the grid has been subdivided and cut to conform it to the surface in regions of high curvature
efficiency. Pumps fluid volumes have been meshed increasing and decreasing the already described
and small details [27]. Impellers and rotors fluid volumes of each model have been meshed
parameters: maximum cell size, minimum cell size and the cell size on surfaces. For each case the best
separately. A maximum cell size of 0.025 has been chosen, where no cell in the volume can have a cell
compromise between results accuracy and computational time has been found and by following all
side larger than the maximum cell size. The minimum cell size has been fixed at 0.0001. The minimum
mesh characteristic are listed:
cell size is a parameter used to limit how small cells can be in attempting to resolve the geometry
using
PUMPthe 1 (Ngeneral mesher.
S = 37.6): No cell in
Total number of the volume
cells: 851.673can have a cell side smaller than the minimum cell
size. The cell size on surfaces has also been fixed at 0.00625. This parameter is used to control the size
Total number of faces: 3.383.745
of the cells for all surfaces of a mesh volume. Using a pure Eulerian approach the mesh of rotors is
deformed by squeezing Total
andsimulation
expandingtime:
the8.9 h as
cells. Pump,
The rotor9mesh
h as PAT
is colored in green in Figure 2.
PUMP A 2mesh
(NS =sensitivity,
20.5): Totalas well known,
number of cells:is1.039.450
fundamental in studying any new problem with a CFD
solver. The target is to allow excellent accuracy (in comparison with experimental tests) and
Total number of faces: 3.926.412
computational efficiency. Pumps fluid volumes have been meshed increasing and decreasing the
Total simulation
already described parameters: maximumtime:cell
4.2 h as Pump,
size, minimum4.8 hcell
as PAT
size and the cell size on surfaces.
For
PUMP each case
3 (N S =the best
64): compromise
Total number between
of cells: results
324.596 accuracy and computational time has been found
and by following all mesh characteristic are listed:
Total number of faces: 3.348.318
PUMP 1 (NS = 37.6): Total number of cells: 851.673
Total simulation time: 4.5 h as Pump, 4.8 h as PAT
Total number of faces: 3.383.745
Total simulation time: 8.9 h as Pump, 9 h as PAT
Energies 2017, 10, 103 7 of 19

Simulations have been run with an Intel(R) Xeon(R) CPU 2.66 GHz (two processors). It is
important to underline that models are transient, and during simulations there is a simultaneous
treatment of moving and stationary fluid volumes. In particular, each volume connects to the others via
an implicit interface. Mismatched grid interfaces can identify overlap areas and match them without
interpolation. These faces, during the simulation process, are treated no differently than an internal
face between two neighboring cells in the same grid domain.
Using the models described it has been possible to study the internal fluid dynamics of each
pump working in the direct (as pump) and reverse (as turbine) modes. A mature turbulence model
has been implemented. It has been demonstrated that for these applications one of the more accurate
model to study turbulence is the k–ε model and RNG k–ε model. Other turbulence models, such as
LES and RNG k–ε model, might also provide good results but the adoption of higher order turbulence
models would have increased the computational time with no relevant improvement of the results.
Authors have also applied the presented strategy confirming the solution accuracy [25–29,31–33] in
other analyses.
Therefore, in this research, the k-epsilon model has been used providing a good accuracy and
computationally efficiency. This model is used by the adopted CFD code since it has been available for
more than a decade and has been widely demonstrated to provide good engineering results for a wide
range of applications [29,31–33]. The standard k–ε model, used for the simulations presented in this
paper is based on the following two equations [25–29,31,32]:
Z 0 Z 0 Z 0  Z 0
∂ µt
ρkdΩ + ρ((v − vσ ) × n)kdσ = µ+ (∇k × n)dσ + ( Gt − ρ)dΩ (9)
∂t Ω( t ) σ σ σk Ω

R0 R0 R 0 µt
 R 0 2

∂
∂t Ω(t) ρdΩ + σ ρ(( v − vσ ) × n )εdσ = σ µ+ σε (∇ × n)dσ + Ω c1 Gt εk − c2 ρ εk dΩ (10)

where c1 = 1.44, c2 = 1.92, σk = 1, σε = 1.3, where σk and σε are the turbulent kinetic energy and the
turbulent kinetic energy dissipation rate Prandtl numbers.
The turbulent kinetic energy, k, is defined as [25–29,31–33]:

1 0 0

k= v ·v (11)
2
where v’ is the turbulent fluctuation velocity, and the dissipation rate, ε, of the turbulent kinetic energy
is defined as [25–29]:
µ 
ε = 2 Sij0 Sij0 (12)
ρ
In which the strain tensor is [29,31]:

∂u0 j
!
1 ∂u0 i
Sij0 = + (13)
2 ∂x j ∂xi

with ui ’ (i = 1, 2, 3) being components of v’.

The turbulent viscosity µt is calculated by [16,23,24,32,33]:

k2
µt = ρCµ (14)

with Cµ = 0.09.
The turbulent generation term Gt can be expressed as a function of velocity and the shear stress
tensor as [16,23,24]:
∂u0 i
Gt = −ρu0 i u0 j (15)
∂x j
 ′
= −ρ ′ ′ (15)

where τ′ = ρ ′ ′ is the turbulent Reynolds stress, which can be modelled by the Boussinesq
hypothesis [15,16,23,24,32,33]:
Energies 2017, 10, 103 2 8 of 19
τ′ = μ + − ρ + δ (16)
3

whereWith
τ0 ij the built
= ρu 0 u0models, simulations have first been run comparing the results in the direct and
i j is the turbulent Reynolds stress, which can be modelled by the Boussinesq
hypothesis [15,16,23,24,32,33]: results for Pump 1 at 2900 rpm are shown in Figure 3; in Figure 3a
reverse working modes. Model
the pressure distribution at the BEP in the 0–9 bar pressure range is presented. The fluid properties
!
of water have been used in all the simulations. ∂ui ∂u j 2

∂uk

0
Figure 3b shows the velocity τ ij = µ t
vectors + − ρk + δij range (0–32 m/s) is(16)
∂x j in the
∂xi fluid3volume;∂xthe
k
velocity the
same for the direct and reverse modes. In this picture, it is possible to visualize the flow evolution
inside thethe
With machine and thesimulations
built models, acceleration/deceleration
have first been of runthe fluid. Both
comparing thefigures
results confirm that and
in the direct the
velocityworking
reverse is highermodes.
in the reverse
Modelmode. results for Pump 1 at 2900 rpm are shown in Figure 3; in Figure 3a
Similarly,
the pressure Figure 4 shows
distribution at the the
BEPpressure
in the 0–9 distributions
bar pressure for pump
range is 2presented.
(NS = 20.5)The
andfluid
pump 3 (NS = 64)
properties of
at 2900 rpm.
water have been used in all the simulations.

(a)

(b)

Figure 3.
Figure 3. Model results
Model for Pump
results 1 (Ns1= (Ns
for Pump 37.6)=at37.6)
2900 at
rpm. (a) rpm.
2900 Pressure(a)distribution; (b) velocity
Pressure distribution;
vectors.
(b) velocity vectors.

Figure 3b shows the velocity vectors in the fluid volume; the velocity range (0–32 m/s) is the same
for the direct and reverse modes. In this picture, it is possible to visualize the flow evolution inside the
machine and the acceleration/deceleration of the fluid. Both figures confirm that the velocity is higher
in the reverse mode.
Similarly, Figure 4 shows the pressure distributions for pump 2 (NS = 20.5) and pump 3 (NS = 64)
at 2900 rpm.
Energies 2017, 10, 103 9 of 19
Energies 2017, 10, 103 9 of 18
Energies 2017, 10, 103 9 of 18

(a) (b)
(a) (b)
Figure 4. Pressure distribution inin
thethe
fluid volume of pumps 2 and 3 at 2900 rpm. rpm.
(a) Ns = 20.5;
Ns =(b) Ns
Figure 4. Pressure distribution in the fluid volume of pumps 2 and 3 at 29002900
Figure 4. Pressure distribution fluid volume of pumps 2 and 3 at rpm. (a) Ns(a)
= 20.5; 20.5;
(b) Ns
= 64.Ns = 64.
(b)
= 64.
For the direct mode, CFD models have been validated using the data supplied by the pump
For the direct mode, CFD models have been validated using the data supplied by the pump
manufacturers. In Figure 5, the head vs. flow rate plots (as the blue curves) are shown. Across the
manufacturers. In InFigure
Figure5,5,the
thehead
headvs.
vs.flow
flow rate
rate plots
plots (as(as
thethe blue
blue curves)
curves) areare shown.
shown. Across
Across the
range of flow rates (30–207) m3/h, the 3
head varies from 47 m to 3 m. In the plots in Figure 5, the model
the range
range of flow
of flow ratesrates (30–207)
(30–207) m /h,m
3 the/h, thevaries
head head from
varies47from
m to 47 mIntothe
3 m. 3 m. In in
plots the plots 5,inthe
Figure Figure
model 5,
results are shown in red. The comparison in Figure 5 demonstrates the accuracy of the adopted
the model
results are results
shownarein shown
red. Theincomparison
red. The comparison
in Figure in Figure 5 demonstrates
5 demonstrates the accuracy
the accuracy of the
of the adopted
methodology; in fact, the percentage error is always less than 4% while for many points the error is
adopted methodology;
methodology; in percentage
in fact, the fact, the percentage error isless
error is always always
thanless
4%than
while4%forwhile
many forpoints
manythe points
errorthe
is
near zero.
near zero.
error is near zero.

Figure 5. Model validation: Comparison between manufacturer’s data and model results.
Figure 5. Model validation: Comparison between manufacturer’s data and model results.
Figure 5. Model validation: Comparison between manufacturer’s data and model results.
4. Ns 37.6 Pump Model Validation with Experimental Data
4. Ns 37.6 Pump Model Validation with Experimental Data
4. NsOnce
37.6 Pump Model
the model hadValidation with Experimental
been validated in the pumping Data
mode, it was decided to also validate it in
Once the model had been validated in the pumping mode, it was decided to also validate it in
the reverse
Once themode,
modeltohad
assess
beenwhether
validatedthe model
in the reproduces
pumping mode, itthe turbine
was mode
decided well.
to also Because
validate it in the
the reverse mode, to assess whether the model reproduces the turbine mode well. Because the
proposed methodology
reverse mode, is based the
to assess whether onlymodel
on thereproduces
results of the
the CFD model
turbine modeunder
well.reverse
Because conditions,
the proposedthe
proposed methodology is based only on the results of the CFD model under reverse conditions, the
validation under reverse conditions was necessary to confirm the entire methodology.
validation under reverse conditions was necessary to confirm the entire methodology.
Energies 2017, 10, 103 10 of 19

Energies 2017, 10, 103

methodology is based only on the results of the CFD model under reverse conditions, the validation 10 of 18
under reverse conditions was necessary to confirm the entire methodology.
The
The model
model of of centrifugal
centrifugal Pump
Pump 11 has has been
been validated
validated with
with data
data from
from anan experiment
experiment performed
performed
on a dedicated test bench of the Department of Civil Construction and Environmental
on a dedicated test bench of the Department of Civil Construction and Environmental Engineering Engineering
of the University
of the University of Naples Federico
of Naples Federico II. The bench
II. The bench enables
enables testing
testing aa centrifugal
centrifugal pump
pump running
running in in
reverse mode. The aim of this activity was to further validate the simulation
reverse mode. The aim of this activity was to further validate the simulation model under reverse model under reverse
conditions.
conditions. TheThe test
test bench
benchreproduces
reproducesaafull-scale
full-scalehydraulic
hydraulicnetwork,
network,mademade upupofof four
four nodes
nodes (Figure
(Figure 6).
6). An external pump increases the water pressure to simulate the behavior of
An external pump increases the water pressure to simulate the behavior of a real urban network whilea real urban network
while
an air an air chamber
chamber stabilizes
stabilizes therate.
the flow flow The
rate.tested
The tested
pumppump has been
has been installed
installed in onein node
one node
where where
two
two pressure-reducing valves (PRV) regulate the water flow rate and the pressure
pressure-reducing valves (PRV) regulate the water flow rate and the pressure at the inlet and outlet at the inlet andof
outlet of the pump. The pressure is regulated with a valve installed at the network
the pump. The pressure is regulated with a valve installed at the network inlet, while the flow rate inlet, while the
flow
at therate at the
outlet outlet with
is varied is varied with a valve.
a butterfly butterfly valve.
Valves areValves are controlled
remotely remotely controlled with electronic
with electronic actuators
actuators
controlledcontrolled by a dedicated
by a dedicated homemade homemade software with
software program, program, with anPLC
an external external PLC [34–40].
[34–40].

Figure 6. Test
Figure 6. Test bench—water
bench—water grid
grid of
of the
the Department
Department of
of Civil Construction and
Civil Construction and Environmental
Environmental
Engineering of the
Engineering of the University
University of
of Naples
Naples Federico
Federico II.
II.

The electric motor of the pump is linked to an inverter and the produced electrical power is
connected to the urban power grid. In the node, two pressure transducers, P1 and P2 (Burkert® model
8314), and a flow meter Q (Siemens® mag 500) have been installed. All test bench data have been
acquired by a homemade acquisition system. Furthermore, a 360-tooth encoder has been installed on
the electrical motor to acquire the shaft speed.
Energies 2017,10,
Energies 2017, 10,103
103 11
11 of
of 19
18

The hardware system is based on a data acquisition board NI DAQ Card (12-bit ADC converter
The electric motor of the pump is linked to an inverter and the produced electrical power is
resolution, 16 input channel, two 24 bit counters), a 68-pin shielded desktop connector block (NI TBX-
connected to the urban power grid. In the node, two pressure transducers, P1 and P2 (Burkert® model
68). NI LabView performs a homemade software.
8314), and a flow meter Q (Siemens® mag 500) have been installed. All test bench data have been
The pressure is acquired with Buckert transducers. Sensors are installed upstream and
acquired by a homemade acquisition system. Furthermore, a 360-tooth encoder has been installed on
downstream of the PAT, as it is shown in Figure 6. The characteristics of the pressure transducers are
the electrical motor to acquire the shaft speed.
as follows:
The hardware system is based on a data acquisition board NI DAQ Card (12-bit ADC converter
• Ceramic
resolution, 16 technology
input channel, two 24 bit counters), a 68-pin shielded desktop connector block (NI
• 0–10 bar pressure range a homemade software.
TBX-68). NI LabView performs
• ±0.25% accuracy
The pressure is acquired with Buckert transducers. Sensors are installed upstream and
• 2 ms response
downstream time as it is shown in Figure 6. The characteristics of the pressure transducers are
of the PAT,
as follows:
The electric motor shaft speed is acquired with a BAUMER BHK 16.05A.360-I2-5 incremental
encoder, with 360 teeth, while, as already said, the flow rate is measured with an electromagnetic
• Ceramic technology
Siemens mag 500 transducer, (accuracy 0.25%, response time 1 s). The sample frequency was 1 Hz.
• 0–10 bar pressure range
The experiments have been performed only in steady-state conditions, varying the water flow
• ±0.25% accuracy
rate and the pressure at the inlet of the pump, for different shaft speeds. In particular, the flow rate
•has been
2 ms varied
response time 8 and 21 L/s, and the shaft speed between 300 and 2200 rpm. During the
between
test, pressure at the inlet and outlet of the pump have been acquired at a sample frequency of 1 Hz.
The electric motor shaft speed is acquired with a BAUMER BHK 16.05A.360-I2-5 incremental
As stated above, the tests have been done in steady-state conditions, running the pump in
encoder, with 360 teeth, while, as already said, the flow rate is measured with an electromagnetic
reverse mode. The flow rate, the pressures at the inlet and outlet of the pump and the shaft rpm were
Siemens mag 500 transducer, (accuracy 0.25%, response time 1 s). The sample frequency was 1 Hz.
measured. In Figure 7, all results of the experimental campaign are shown.
The experiments have been performed only in steady-state conditions, varying the water flow
To examine the PAT performance, the total head (m) versus shaft rpm is reported, varying the
rate and the pressure at the inlet of the pump, for different shaft speeds. In particular, the flow rate has
water flow rate for all the examined conditions. Results confirm what is known from the literature:
been varied between 8 and 21 L/s, and the shaft speed between 300 and 2200 rpm. During the test,
the PAT head increases with the rpm and with the flow—rate, and it can be easily noted that, for the
pressure at the inlet and outlet of the pump have been acquired at a sample frequency of 1 Hz.
tested conditions, the head varies between 0.1 and 1.8 m.
As stated above, the tests have been done in steady-state conditions, running the pump in reverse
In Figure 8, the whole validation of the simulation model is presented: it shows that the model
mode. The flow rate, the pressures at the inlet and outlet of the pump and the shaft rpm were measured.
reproduces the experimental data well with very small differences between the experimental and the
In Figure 7, all results of the experimental campaign are shown.
model results for all the running conditions that were analyzed.

(a)

Figure 7. Cont.
Energies 2017, 10, 103 12 of 19
Energies 2017, 10, 103 12 of 18

Energies 2017, 10, 103 12 of 18

(b)
Figure
Figure 7.
7. Experimental
Experimental results.
results. (a)
(a) Head;
Head; (b)
(b) Efficiency.
Efficiency.

To examine the PAT performance, the total head (m) versus shaft rpm is reported, varying the
water flow rate for all the examined conditions. Results confirm what is known from the literature: the
PAT head increases with the rpm and with the flow—rate, and it can be easily noted that, for the tested
conditions, the head varies between 0.1 and 1.8 m.
In Figure 8, the whole validation of the simulation
(b) model is presented: it shows that the model
reproduces the experimental data well with very small differences between the experimental and the
Figure 7. Experimental results. (a) Head; (b) Efficiency.
model results for all the running conditions that were analyzed.

Figure 8. Model validation.

5. Model Results
After the validation phase in pump and turbine mode, simulation models have been used to
predict the efficiency curves of the three analyzed pumps. Then, all the simulations have been
performed to obtain the data necessary to evaluate the inverse characteristics.
The specific head ψ can be evaluated as:

Figure 8.
Figure 8.ψModel validation.
= validation.
Model (17)

Results capacity φ depends by the flow Q and the impeller diameter D:

The specific
5. Model
After the validation phase in pump and φ turbine
= mode, simulation models have been used
(18)to
predict the efficiency curves of the three analyzed pumps. Then, all the simulations have been
performed to obtain
The specific the data
power necessary
is defined as: to evaluate the inverse characteristics.
The specific head ψ can be evaluated as:

ψ= (17)
Energies 2017, 10, 103 13 of 19

5. Model Results
After the validation phase in pump and turbine mode, simulation models have been used to
predict the efficiency curves of the three analyzed pumps. Then, all the simulations have been
performed to obtain the data necessary to evaluate the inverse characteristics.
The specific head ψ can be evaluated as:

gH
ψ= (17)
n2 D 2
The specific capacity ϕ depends by the flow Q and the impeller diameter D:

Q
ϕ= (18)
nD3
The specific power is defined as:
P
π= (19)
ρn3 D5
While the efficiency can be evaluated as:

P
η= (20)
ρQH

where H (m), Q (m3 /s), and P (W) are the head, flow rate and power, respectively. The rotational speed
is n (RPS) and D (m) is the impeller diameter. In the reverse mode simulation, the boundary conditions
in pump and PAT mode were the same (declared data in pump mode). The boundary conditions in
reverse mode are summarized in Table 2.

Table 2. Boundary conditions.

Boundary Conditions Pump 1 Pump 2 Pump 3

Outlet pressure 1.9 bar 1.9 bar 1.9 bar
Inlet Volumetric Flow 90/210 m3 /h 30/85 m3 /h 48/90 m3 /h
Tin 293.15 K 293.15 K 293.15 K
Psat 2886 Pa 2886 Pa 2886 Pa

The specific head, the specific power and the efficiency have been evaluated for both pump and
turbine mode and for all the studied pumps. These are plotted versus specific capacity in Figure 9.
It is clear that at high capacity the specific head in reverse mode is always higher than in direct
mode. In reverse mode, pumps have a larger power range than in direct mode. The trends are quite
different and the curves arising at low flow rate and at higher capacities have a higher power value
than in direct mode. In pump mode, the efficiency has a typical “bell shape” while in PAT mode its
profile resembles that of a Francis turbine: it increases at low flow rates and reaches a maximum value
at high flow rates. Moreover, the pump with the low specific speed works with low flow rates but
high heads in direct mode. In reverse mode at high flow rates, the head is higher than the direct-mode
value. The pump with the high specific speed works at high flow rates but low heads. In reverse
mode, working with the same flow rates, the maximum head is also higher than the direct-mode value.
For all pumps, the maximum value is lower than the direct-mode value. For low specific speed pumps,
this maximum value is approximately equal to the direct-mode value, while for high specific speed
pumps, it is lower.
high heads in direct mode. In reverse mode at high flow rates, the head is higher than the direct-
mode value. The pump with the high specific speed works at high flow rates but low heads. In reverse
mode, working with the same flow rates, the maximum head is also higher than the direct-mode
value. For all pumps, the maximum value is lower than the direct-mode value. For low specific speed
Energies 2017, 10, 103 14 of 19
pumps, this maximum value is approximately equal to the direct-mode value, while for high specific
speed pumps, it is lower.

Figure 9. Specific head, specific power and efficiency.

Figure 9. Specific head, specific power and efficiency.

In conclusion, in Table 3, the BEP values are summarized:

Table 3. Boundary conditions.

Pump 1 Pump 2 Pump 3

Boundary
Conditions Direct Reverse Direct Reverse Direct Reverse
Mode Mode Mode Mode Mode Mode
Head (m) 39 61 45.4 67 3.9 4.6
Capacity (m3 /s) 0.041 0.05 0.017 0.021 0.015 0.022
Power (kW) 20.5 19.98 10.01 10.24 1.05 0.75
Efficiency 0.787 0.663 0.743 0.741 0.543 0.487

6. Comparison of Prediction Methods

After the evaluation of the inverse characteristics, the results of the proposed methodology have
been compared to the prediction methods available in literature. To this end, all the previously
discussed methods have been applied to the three analyzed pumps. Some methods predict only the
head and flow rate, while others also predict power and efficiency. In Table 4, all the results are shown
for Pump 1.
The flow rates calculated using the methods of Sharma and of Hergt and Schmiedl are very close
to those of the proposed CFD methodology. Grover’s and Alatorre-Frenk’s methods overestimate
this value while the other methods underestimated the value by a margin of 10%–15%. All methods
underestimate the head value, except for Grover’s. These values diverge with errors of up to 30%.
Evaluating the percentage deviation as:

predicted value − c f d value

percentage deviation = × 100 (21)
c f d value

In Figure 10 the deviations between the predictions of these methods and the simulated data are
plotted vs. the pump specific speeds for all three pumps.
this value while the other methods underestimated the value by a margin of 10%–15%. All methods
underestimate the head value, except for Grover’s. These values diverge with errors of up to 30%.

Table 4. Comparison for pump 1.

Energies 2017, 10, 103 Methods H (m) Q (m3/s) P (kW) η 15 of 19
Model results 61.42 0.05 19.98 0.663
Stepanoff 49.55 0.0463 22.5 0.787
Alatorre-Frenk 4. Comparison0.091
Table 42.79 for pump 1. 28.92 0.757
Sharma 51.99 0.0498 20.5 0.807
Schmiedl
Methods 43.94
H (m) 0.0514
3
Q (m /s) 16.82
P (kW) 0.759
η
Grover 78.59 0.0657 36.92 0.729
Model results
Hergt 61.42
41.90 0.05
0.0508 19.98
- 0.663
-
Stepanoff
Childs 49.55
49.55 0.0463
0.0522 22.5
19.96 0.787
0.787
Alatorre-Frenk
D&N 42.79
58.56 0.091
0.0411 28.92
18.17 0.757
0.769
Sharma 51.99 0.0498 20.5 0.807
Schmiedl 43.94 0.0514 16.82 0.759
Evaluating the percentage deviation as:
Grover 78.59 0.0657 36.92 0.729
predicted value − value
percentage deviation
Hergt 41.90 = 0.0508 - × 100 - (21)
Childs 49.55 0.0522 value 19.96 0.787
In Figure 10 the D&N
deviations between the predictions
58.56 0.0411 of these methods
18.17 and the simulated data are
0.769
plotted vs. the pump specific speeds for all three pumps.

Energies 2017, 10, 103 15 of 18

(a)

(b)
Figure
Figure 10.
10. Prediction
Prediction methods
methods comparison.
comparison. (a)
(a) Head
Head Percentage
Percentage Error;
Error; (b)
(b) Capacity
Capacity Percentage
Percentage Error.
Error.

It easy to observe that in some cases the deviation is very high. To evaluate the Derakhshan and
It easy to observe that in some cases the deviation is very high. To evaluate the Derakhshan and
Nourbakhsh efficiency [24], the CFD model results were used as shown in Figure 11. In this figure,
Nourbakhsh efficiency [24], the CFD model results were used as shown in Figure 11. In this figure,
the relative velocity magnitude (m/s) is shown and the angle between the relative and absolute
the relative velocity magnitude (m/s) is shown and the angle between the relative and absolute velocity
velocity is highlighted in red, at BEP conditions and in pump mode.
is highlighted in red, at BEP conditions and in pump mode.
 Figure 10. Prediction methods comparison. (a) Head Percentage Error; (b) Capacity Percentage Error.

It easy to observe that in some cases the deviation is very high. To evaluate the Derakhshan and
Nourbakhsh efficiency [24], the CFD model results were used as shown in Figure 11. In this figure,
the relative velocity magnitude (m/s) is shown and the angle between the relative and absolute
Energies 2017, 10, 103 16 of 19
velocity is highlighted in red, at BEP conditions and in pump mode.

Figure
Figure 11.
11. Relative
Relative velocity
velocity magnitude.
magnitude.

7. Conclusions
7. Conclusions
In this paper a methodology to predict the inverse characteristic of a centrifugal pump has been
In this paper a methodology to predict the inverse characteristic of a centrifugal pump has been
presented. This methodology is based on the results of a three-dimensional simulation model built
presented. This methodology is based on the results of a three-dimensional simulation model built with
with a commercial CFD code. Three industrial pumps have been analyzed, with different specific
a commercial CFD code. Three industrial pumps have been analyzed, with different specific speeds.
speeds. First, the simulation models have been validated with data supplied by the pump
First, the simulation models have been validated with data supplied by the pump manufacturers. Then
manufacturers. Then the results of an experimental campaign have been used to validate a model
the results of an experimental campaign have been used to validate a model simulating the pump
simulating the pump working in reverse conditions.
working in reverse conditions.
Starting from the CFD model results, the specific head, capacity, power and efficiency have been
evaluated and the best efficient point of all the analyzed pumps was found. Furthermore, several
prediction methods have been applied to the tested pumps and their predicted values were compared
with those of the proposed methodology. Some methods (e.g., Childs’ method) are not in accord while
others (e.g., Stepanoff’s method) show small relative differences.

Acknowledgments: This research has been supported by the Department of Industrial Engineering of the
University of Naples “Federico II”. We appreciate the technical contribution from Gennaro Stingo and Giuseppe
Iovino of the Department of Industrial Engineering and from colleagues of the Department of Civil Construction
and Environmental Engineering of the University of Naples “Federico II”.
Author Contributions: Emma Frosina, Dario Buono and Adolfo Senatore conceived and designed the experiments;
Dario Buono performed the experiments; Emma Frosina and Dario Buono analyzed the data; Adolfo Senatore
contributed reagents/materials/analysis tools; Emma Frosina wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.

Nomenclature
PRV Pressure Reducing Valve
PAT Pump as Turbine
CFD Computational Fluid Dynamic
BEP Best Efficiency Point
Hp Pump head
Qp Pump flow rate
Pp Pump power
Energies 2017, 10, 103 17 of 19

ηp Pump overall efficiency

Ht Turbine head
Qt Turbine flow rate
Pt Turbine power
ηt Turbine overall efficiency
ψ Specific head
ϕ Specific capacity
π Specific power
η Efficiency
Ns Pump specific speed
Nst Turbine specific speed
z Impeller’s blade number
Pit Power losses due to leakage
Pvt Volute power losses
Pet Kinetic energy losses
Pnt Turbine net power
Pit Hydraulic losses of the impeller in turbine mode
n Surface normal
k Turbulence kinetic energy
p Pressure (Pa)
Q Flow rate (m3 /h)
rpm Revolution per minute
U Initial velocity
u Velocity component (m/s)
u0 Component of v’
v Velocity vector
v0 Turbulent fluctuation velocity
µ Fluid viscosity (Pa-s)
ρ Fluid density (kg/m3 )
τ
e Shear stress tensor
c1 , c2 Constant
σk Turbulent kinetic energy
σε Turbulent kinetic energy dissipation
S’ij Strain tensor
µt Turbulent viscosity
Gt Turbulent generation
τij Turbulent Reynolds stress
ε Turbulence dissipation
Ω Control volume

References
1. Williams, A. Pumps as Turbines: A User’s Guide; Intermediate Technology Publications: London, UK, 1995.
2. Arriaga, M. Pump as turbine—A pico-hydro alternative in Lao People’s Democratic Republic. Renew. Energy
2010, 35, 1109–1115. [CrossRef]
3. Derakhshan, S.; Nourbakhsh, A. Experimental study of characteristic curves of centrifugal pumps working
as turbines in different specific speeds. Exp. Therm. Fluid Sci. 2008, 32, 800–807. [CrossRef]
4. Williams, A.A.; Rodrigues, A.; Singh, P.; Nestmann, F.; Lai, E. Hydraulic analysis of a pump as a turbine
with CFD and experimental data. In Proceedings of the IMechE Seminar, Computational Fluid Dynamics for
Fluid Machinery, London, UK, 18 November 2003.
5. Nautiyal, H.; Varun, A.K. Reverse running pumps analytical, experimental and computational study: A
review. Renew. Sustain. Energy Rev. 2010, 14, 2059–2067. [CrossRef]
6. Nautiyal, H.; Varun, K.A.; Yadav, S. 2010 CFD analysis on pumps working as Turbines. Hydro Nepal 2010, 6,
35–37.
7. Carravetta, A.; Del Giudice, G.; Fecarotta, O.; Ramos, H. PAT design strategy for energy recovery in water
distribution networks by electrical regulation. Energies 2013, 6, 411–424. [CrossRef]
8. Carravetta, A.; Del Giudice, G.; Fecarotta, O.; Ramos, H. Pump as Turbine (PAT) design in water distribution
network by system effectiveness. Water 2013, 5, 1211–1225. [CrossRef]
9. Derakhshan, S.; Nourbakhsh, A.; Mohammadi, B. Efficiency improvement of centrifugal reverse pumps.
ASME J. Fluids Eng. 2009, 131. [CrossRef]
Energies 2017, 10, 103 18 of 19

10. Natanasabapathi, S.R.; Kshirsagar, J.T. Pump as Turbine—An Experience with CFX 5.6; Kirloskar Bros. Ltd.,
Corporate Research and Engineering Division: Pune, India, 2004.
11. Barrio, R.; Fernández, J.; Parrondo, J.; Blanco, E. Performance prediction of a centrifugal pump working
in direct and reverse mode using Computational Fluid Dynamics. In Proceedings of the International
Conference on Renewable Energies and Power Quality (ICREPQ‘10), Granada, Spain, 23–25 March 2010.
12. Chapallaz, J.M.; Eichenberger, P.; Fischer, G. Manual on pumps used as turbines. In Friedr Vieweg Sohn
Verlagsgesellschaft, Braunschweig; Informatica International, Inc.: Braunschweig, Germany, 1992.
13. Childs, S.M. Convert pumps to turbines and recover HP. Hydro Carbon Process. Pet. Refin. 1962, 41, 173–174.
14. McClaskey, B.M.; Lundquist, J.A. Can You Justify Hydraulic Turbines? Hydrocarb. Process. 1976, 55, 163–166.
15. Lobanoff, V.S.; Ross, R.R. Hydraulic power recovery turbines. In Centrifugal Pumps: Design and Applications;
Gulf Publishing Company: Houston, TX, USA, 1985; pp. 246–282.
16. Sanjay, V.J.; Rajesh, N.P. Investigations on pump running in turbine mode: A review of the state-of-the-art.
Renew. Sustain. Energy Rev. 2014, 30, 841–868.
17. Grover, K.M. Conversion of Pumps to Turbines; GSA Inter Corp.: Katonah, NY, USA, 1980.
18. Williams, A. The turbine performance of centrifugal pumps: A comparison of prediction methods.
J. Power Energy 1994, 208, 59–66. [CrossRef]
19. Stepanoff, A.J. Centrifugal and Axial Flow Pumps; John Wiley: New York, NY, USA, 1957.
20. Alatorre-Frenk, C. Cost Minimization in Micro Hydro Systems Using Pumps-Asturbines. Ph.D. Thesis,
University of Warwick, Coventry, UK, 1994.
21. Sharma, K. Small Hydroelectric Project-Use of Centrifugal Pumps as Turbines; Technical Report; Kirloskar Electric
Co.: Bangalore, India, 1985.
22. Schmiedl, E. Serien-Kreiselpumpen im Turbinenbetrieb; Pumpentagung: Karlsruhe, Germany, 1988.
23. Lewinsky, K.; Heinz, P. Pumpen als Turbinen fur Kleinkraftwerke. Wasserwirtschaft 1987, 77, 531–537.
24. Derakhshan, S.; Nourbakhsh, A. Theoretical, numerical and experimental investigation of centrifugal pumps
in reverse operation. Exp. Therm. Fluid Sci. 2008, 32, 1620–1627. [CrossRef]
25. Frosina, E.; Senatore, A.; Buono, D.; Santato, L. Analysis and Simulation of an Oil Lubrication Pump for
the Internal Combustion Engine. In Proceedings of the Fluids Engineering Systems and Technologies,
San Diego, CA, USA, 15–21 November 2013.
26. Frosina, E.; Senatore, A.; Buono, D.; Olivetti, M. A Tridimensional CFD Analysis of the Oil Pump of a High
Performance Engine. In Proceedings of the SAE 2014 World Congress and Exhibition, Detroit, MI, USA,
8–10 April 2014.
27. Frosina, E.; Buono, D.; Senatore, A. Modeling methodology to study the internal fluid-dynamic of a gas filter.
Int. Rev. Model. Simul. 2015, 8, 533. [CrossRef]
28. Frosina, E.; Senatore, A.; Buono, D.; Stelson, K.A.; Wang, F.; Mohanty, B.; Gust, M.J. Vane pump power
split transmission: Three dimensional computational fluid dynamic modeling. In Proceedings of the
ASME/BATH 2015 Symposium on Fluid Power and Motion Control, Chicago, IL, USA, 12–14 October 2015.
29. Frosina, E.; Senatore, A.; Buono, D.; Stelson, K.A. A Mathematical Model to Analyze the Torque Caused by
Fluid-Solid Interaction on a Hydraulic Valve. ASME J. Fluids Eng. 2016, 138, 061103. [CrossRef]
30. Automated Meshing with Binary Tree. Available online: https://www.simerics.com/download/automated-
meshing-with-binary-tree/ (accessed on 11 January 2017).
31. Frosina, E.; Buono, D.; Senatore, A.; Costin, I.J. A Simulation Methodology Applied on Hydraulic Valves for
High Fluxes. Int. Rev. Model. Simul. 2016, 9, 217. [CrossRef]
32. Frosina, E.; Buono, D.; Senatore, A.; Stelson, K.A. A modeling approach to study the fluid dynamic forces
acting on the spool of a flow control valve. J. Fluids Eng. 2016, 139, 11103–11115. [CrossRef]
33. Pellegri, M.; Vacca, A.; Frosina, E.; Buono, D.; Senatore, A. Numerical analysis and experimental validation
of Gerotor pumps: A comparison between a lumped parameter and a computational fluid dynamics-based
approach. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 2016, 1989–1996, 203–210. [CrossRef]
34. Pugliese, F.; De Paola, F.; Fontana, N.; Giugni, M.; Marini, G. Experimental characterization of two pumps as
turbines for hydropower generation. Renew. Energy 2016, 99, 180–187. [CrossRef]
35. Su, X.; Huang, S.; Zhang, X.; Yang, S. Numerical research on unsteady flow rate characteristics of pump as
turbine. Renew. Energy 2016, 94, 488–495. [CrossRef]
36. De Marchis, M.; Milici, B.; Volpe, R.; Messineo, A. Energy saving in water distribution network through
pump as turbine generators: Economic and environmental analysis. Energies 2016, 9, 877. [CrossRef]
Energies 2017, 10, 103 19 of 19

37. Wang, T.; Kong, F.; Chen, K.; Duan, X.; Gou, Q. Experiment and analysis of effects of rotational speed on
performance of pump as turbine. Trans. Chin. Soc. Agric. Eng. 2016, 32, 67–74.
38. Frosina, E.; Senatore, A.; Buono, D.; Monterosso, F.; Olivetti, M.; Arnone, L.; Santato, L. A tridimensional
CFD analysis of the lubrication circuit of a non-road application diesel engine. In Proceedings of the 11th
International Conference on Engines and Vehicles, ICE 2013, Naples, Italy, 15–19 September 2013.
39. Venturini, M.; Alvisi, S.; Simani, S. Energy potential of pumps as turbines (PATs) in water distribution
networks. In Proceedings of the 28th International Conference on Efficiency, Cost, Optimization, Simulation
(ECOS) and Environmental Impact, Pau, France, 29 June–3 July 2015.
40. Barbarelli, S.; Amelio, M.; Florio, G. Predictive model estimating the performances of centrifugal pumps
used as turbines. Energy 2016, 107, 103–121. [CrossRef]

© 2017 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).