Cavitation Instabilities and Rotordynamic Effects in Turbopumps and Hydroturbines Turbopump and Inducer
Cavitation Instabilities and Rotordynamic Effects in Turbopumps and Hydroturbines Turbopump and Inducer
Cavitation Instabilities and Rotordynamic Effects in Turbopumps and Hydroturbines Turbopump and Inducer
Lucad'Agostino
MariaVittoriaSalvetti Editors
Cavitation
Instabilities and
Rotordynamic Effects
in Turbopumps and
Hydroturbines
Turbopump and Inducer Cavitation,
Experiments and Design
International Centre
for Mechanical Sciences
CISM International Centre for Mechanical
Sciences
Volume 575
Series editors
The Rectors
Friedrich Pfeiffer, Munich, Germany
Franz G. Rammerstorfer, Vienna, Austria
Elisabeth Guazzelli, Marseille, France
Executive Editor
Paolo Serani, Udine, Italy
The series presents lecture notes, monographs, edited works and proceedings in the
eld of Mechanics, Engineering, Computer Science and Applied Mathematics.
Purpose of the series is to make known in the international scientic and technical
community results obtained in some of the activities organized by CISM, the
International Centre for Mechanical Sciences.
Cavitation Instabilities
and Rotordynamic Effects
in Turbopumps
and Hydroturbines
Turbopump and Inducer Cavitation,
Experiments and Design
123
Editors
Luca dAgostino Maria Vittoria Salvetti
Civil and Industrial Engineering Civil and Industrial Engineering
Department Department
Pisa University Pisa University
Pisa Pisa
Italy Italy
v
vi Preface
applications. Next, Ceccio and Mkiharju review the traditional and novel experi-
mental methods for the investigation of hydrodynamic cavitation, and illustrate
some of their recent developments in the use of ionizing radiation as a means to
visualize cavitating flows. Tsujimoto presents the experimental results, theoretical
analyses and numerical simulations he and his collaborators have been using for
their pioneering work on cavitation-induced instabilities and rotordynamic effects in
inducers, turbopumps and hydroturbines. The fundamental aspects of the prelimi-
nary hydraulic design of inducers and centrifugal turbopumps is discussed by
dAgostino et al., which illustrate the development and experimental validation
of their reduced-order models capable of jointly providing the geometrical deni-
tion and performance prediction of these machines. The last three contributions
concern the numerical simulation of cavitating flows. The article by Goncalves
presents a comparison of different turbulence and cavitation models, with and
without thermal effects, in unsteady RANS simulations, while the contribution by
Salvetti focuses on numerical simulation of cavitating flows in complex geometry.
Finally, Saurel et al. present a flow model for the numerical simulation of interfacial
flows with phase transition and its applications to cavitating nozzle flows.
The monograph is addressed to doctoral/postdoctoral students, researchers,
scientists, scholars, and professionals interested in perfecting their knowledge and
understanding of cavitating flow phenomena and research occurring in a wide range
of applications in aerospace, mechanical, hydraulic, naval and chemical engineer-
ing, applied mechanics, applied mathematics, industrial chemistry, and applied
physics.
The editors are especially grateful to Prof. Steve Ceccio, Prof. Eric Goncalves,
Prof. Richard Saurel, and Prof. Yoshinobu Tsujimoto for their contributions to the
present volume. The editors are also very thankful to Prof. B. Schrefler, Secretary
General, Prof. A. Soldati, Deputy Secretary General, and to Prof. P. Serani,
Executive Editor, as well as to the staff of the CISM, for their support and assistance
in the organization of the Course and in the publication of the present volume.
vii
viii Contents
The term cavitation was rst introduced by R. E. Froude to indicate the change of
phase occurring within a liquid due to local pressure reductions generated by
dynamic actions. In this respect cavitation is closely related to boiling, where phase
changes are due instead to the increase of temperature as a consequence of heat
transfer to the liquid. In both cases the phase change ideally occurs in the presence
of some nucleating agents when the vapor pressure of the liquid locally exceeds its
static pressure.
From the physical point of view cavitation represents a complex phenomenon
because it involves the simultaneous dynamic and thermodynamic interaction of
two phases, it is greatly affected by the uncertain concentration and susceptibility of
nucleating agents, it manifests itself in a variety of topological forms, and can be
strongly coupled with the fully wetted flow of the liquid phase. Cavitation
phenomena in hydraulic machines are, therefore, only partially understood and
2 Cavitation Nuclei
Early studies have shown (Plesset 1969; Knapp et al. 1970) that the maximum
tensile strength that pure liquids can theoretically sustain according to thermody-
namic considerations is much larger than observed in practice. The highest tensions
attained in experiments with extremely pure water at room temperature are on the
order of a few hundred atmospheres. On the other hand, values comparable to the
vapor pressure at the liquid temperature are usually observed in cavitation exper-
iments on technical liquids. It has therefore been postulated that the tensile strength
of liquids is considerably reduced by the presence of weak spots for the onset of
liquid rupture, generically called nuclei
It has long been recognized that small bubbles (Fox and Herzfeld 1954; Plesset
1969) and gas pockets contained inside crevices and cracks of a liquidsolid
interface (Harvey et al. 1947) act as cavitation nuclei (Holl 1970). The thermo-
dynamic equilibrium of bubbles and/or pockets of gas is influenced by the content
of dissolved gas in the liquid, which therefore represents an important parameter in
the study of cavitation inception. Local fluctuations of the liquid density of ther-
mostatistical nature or due to the interaction with high energy radiation are also
thought to act as cavitation nuclei, although their role is quite negligible in most
practical applications where the effects of the abundant supply of weaker nuclei
from other sources usually prevail.
Cavitation nuclei can be suspended in the fluid stream or located at the flow
boundaries. The relative importance of the cavitation susceptibility of surface and
free stream nuclei often determines the form of cavitation developing in a specic
flow conguration. There are indications that the effects of free stream nuclei
An Introduction to Cavitation in Inducers and Turbopumps 3
usually prevail in laboratory tests on reduced-scale models and that this situation
may sometimes be reversed in full-scale prototype operation.
The concentration of nuclei which become unstable at a given level of tension
synthetically indicated as liquid qualityrepresents an important aspect for
predicting the inception and later development of cavitation, as well as for deducing
scaling laws capable of extending the results obtained from model tests to full-scale
operation.
Since the size of nuclei can be measured more easily than their susceptibility by
means of optical, acoustical, and electrical methods (OHern et al. 1988), in cavi-
tation research the liquid quality is usually expressed in terms of the nuclei con-
centration density distribution:
dN Ro
nRo =
dRo
Usually the inception of cavitation manifests itself with the sporadic formation of
isolated voids in the low pressure regions of the flow when the pressure drops
below some critical value, and is associated with the emission of sharp noise bursts
due to the sudden collapse of the cavities as they are convected in a region of higher
pressure (Holl 1969). Acoustic detection of these bursts is routinely used for the
identication of cavitation inception conditions.
Cavitation occurs in different forms as a consequence of various aspects and
conditions of the liquid flow. In the absence of separation, it generally takes place
either at or close to the boundaries where the minimum pressure occurs, depending
on which among surface and freestream nuclei are more susceptible to cavitate. On
the other hand, when the flow is separated, cavitation takes place far from the
boundaries in the turbulent wake (cloud cavitation) or in the core of shedding
vortices (vortex cavitation).
When free stream nuclei are more susceptible than surface nuclei, cavitation
develops in the formation of bubbles traveling with the liquid. This form of cavi-
tation is characterized by cavities originating in the low pressure regions of the flow
and collapsing further downstream as they are convected in higher pressure regions.
When, on the other hand, surface nuclei are more susceptible to cavitate than free
stream nuclei, cavitation develops either in a thin recirculation region generated by
the separation of the liquid flow (attached cavitation), or as a wake of cavities
sequentially released from localized nuclei on the boundary surface (spot cavita-
tion). When developing in regions where the flow is laminar, attached cavities have
a quasi-steady glossy appearance in their upstream portion (sheet cavitation) and
become increasingly unsteady and fuzzy further downstream as a consequence of
the RayleighTaylor destabilization of the liquid/vapor interface. On the other hand,
cloud cavitation leads to the formation of large intermittent clusters of bubbles in
the fully separated wake of the flow. The latter forms of cavitation are collectively
indicated as developed cavitation and, in particular, as partial cavitation when the
cavities collapse before reaching the trailing edge of the cavitating body, and
supercavitation when a further drop of the pressure leads to the formation of a
cavity extending downstream beyond the trailing edge of the body. The above
classication should, however, only be considered as indicative since in many
practical cases several forms of cavitation can occur simultaneously and variously
interact.
cavitation inception occurs as soon as the static pressure p in the flow locally
drops below the vapor pressure pV , corresponding to the liquid temperature;
pressure changes in the flow are solely due to dynamic (inertial) actions.
Then the fundamental parameter for cavitation scaling is the Euler number (also
indicated as cavitation number):
po pV
=
L Uo2 2
where L is the liquid density and the subscript o indicates the relevant reference
values of the flow pressure p and velocity U. By comparison of the Euler number
with the pressure coefcient:
p po
Cp =
L Uo2 2
which in the stated assumptions is equal for geometrically similar steady flows,
cavitation is therefore predicted when:
pmin pV Cpmin
(or, equivalently, the Net Positive Suction Head NPSH = NPSP g, where g is the
acceleration of gravity) in order to dene the required total pressure pt1 of the inlet
flow to the machine at nominal flow rate for operation under acceptable cavitation
conditions (typically at 3% head degradation).
Although more informative for engineering purposes, both NPSP and NPSH are
dimensional numbers, unsuited for similarity considerations. Hence, from dimen-
sional analysis, either the design values of ,
pto pV
= cavitation index
L Uo2 2
or:
p
V
SS = 4
suction specific speed
pt1 pV L 3
are more efciently used for scaling the cavitation performance of geometrically
similar machines rotating at speed and displacing the volumetric flux V . Two
6 L. dAgostino et al.
geometrically similar flows are therefore considered to be equivalent for the pur-
pose of cavitation (i.e., cavitation is predicted to occur in similar form and extent)
when they have the same values of , or SS , as applicable.
The Euler theory of cavitation scaling provides in rst approximation an ade-
quate correlation of cavitation phenomena between full-scale and model flows.
However, in specic cases violation of the underlying assumptions as a conse-
quence of:
(1) changes of the fully wetted flow eld due to viscosity, gravity, and com-
pressibility, including boundary layer development, separation, transition, tur-
bulence, reattachment, fully separated shear and vortical flows, turbulent
pressure fluctuations;
(2) changes of the geometric similarity, including surface roughness;
(3) changes of the cavity growth, due to inertia, heat transfer, surface tension,
liquid viscosity and compressibility, transport phenomena between dissolved
and noncondensable gas, susceptibility and concentration of cavitation nuclei.
can lead to signicant discrepancies, collectively indicated as cavitation scaling
effects (Arndt 1981). In particular, the available evidence indicates that the latter
set 3) of scaling effects usually prevails in developed cavitation.
Scaling effects from various sources often overlap and influence each other,
making it very difcult to recognize their ultimate causes and to predict their
intensity. The literature on cavitation scaling effects is therefore quite vast and
intricate. Not all of it is, however, of immediate relevance to the analysis of the
steady and unsteady performance of hydraulic turbomachinery, which is mostly
concerned with fully developed cavitation. Thus, here we limit our attention to this
form of cavitation, which is the decisive factor for the occurrence of major dif-
culties in the operation of high power density turbopumps.
Whatever the initial nature and size of cavitation nuclei may be, one can in rst
approximation visualize their growth into larger cavities (and their later collapse)
with reference to the idealized case of a spherical bubble in an unbounded
incompressible liquid subject to a change of its far eld pressure. In practice there
can be signicant departures from this simplied model, especially in the initial
stages of growth and in the nal stage of collapse, or as a consequence of various
causes such as the proximity of the boundaries, interactions with other bubbles,
flow shearing effects, instabilities of the cavity interface, etc. However, consider-
ation of spherical cavities greatly simplies the analysis, while still retaining the
most important aspects of the physical phenomena governing the interactions of the
bubbles with the surrounding liquid. Under these assumptions the bubble radius
An Introduction to Cavitation in Inducers and Turbopumps 7
3 2 R 2S pB pL
RR + R + 4L + =
2 R L R L
Here dots indicate Lagrangian derivatives with respect to time t, S is the surface
tension, L = L L is the kinematic viscosity of the liquid, pL is the far eld liquid
pressure and pB is the bubble internal pressure, which consists of the partial pres-
sures of vapor pV and noncondensable gas pG . When mass diffusion effects within
the bubbles are neglected, pV is constant and pG is expressed by the perfect gas
equation:
3
Ro TB
pG = pGo
R TBo
where pGo is the gas partial pressure at the initial radius Ro , TBo is the corresponding
temperature and TB is in general determined by the simultaneous solution of the
energy equation for the thermal contact of the bubble with the surrounding liquid.
In this case, if the initial state (indicated by the subscript o) is a mechanical
equilibrium state, then the corresponding pressure in the liquid is:
2S
pLo = pGo + pVo
Ro
shows that no stable mechanical equilibrium exists when the tension applied to the
bubble exceeds the critical value expressed by:
4S pLo pV 1 2
pV pL cr = 3 1+
3Ro 2S Ro
In this case, neglecting thermal and diffusive effects, the bubble growth rate is
determined by the inertia of the liquid and, when pV pL is constant, for large times
approaches the value:
s
2pV pL
R =
3L
8 L. dAgostino et al.
thermal effects originate from the evaporation rate necessary to balance the flux of
vapor entrained in the wake of the cavity. In the absence of entrainment and
diffusion of noncondensable, gas the pressure in a steady attached cavity would be
equal to the vapor pressure of the liquid at its bulk temperature and no thermal
effects would occur. Thus, in principle, a comprehensive scaling theory of thermal
effects should also account for the type of cavitation occurring in a specic
application. This, however, proved at rst exceedingly complex, and therefore
simplied scaling theories have initially been developed on the basis of averaged
homogeneous flow models, which do not account for any specic topological form
of cavitation.
The rst method to scale thermal effects in cavitating turbopumps has been
proposed by Stahl and Stepanoff (1956), and later by Stepanoff (1961, 1964). They
suggested to correlate the cavitation characteristics of different liquids using the
parameter B, dened as the vapor to liquid volume ratio consequent to the flashing
of various liquids subject to the same dynamic depression produced by the
operation of the machine:
L cpL T
B=
V qLV
Here T is the temperature drop due to the phase change and cpL is the specic
heat of the liquid. Implicit in this method is the assumption that phase changes take
place adiabatically (Acosta and Hollander 1959) and that the same dynamic
depression is produced in geometrically similar machines independently on the
nature of the cavitating liquid, thus neglecting both the dynamic aspects of cavi-
tation and its mutual interactions with the liquid flow. From the relative value of the
parameter B in various liquids it would then be possible to infer, for example,
whether a certain NPSH reduction is possible for a pump operating at design
conditions in a specic flow application. The temperature and vapor pressure drops
are related by the ClausiusClapeyron equation:
dpV qLV
pV = T = T
dT o T vV vL
bubble internal pressure. Since the effect of gas content quickly becomes negligible,
this implies a drop of the vapor pressure due to the inability of the heat transfer from
the liquid to the bubble surface to sustain the evaporation rate required to keep pace
with the increase of the bubble volume (thermal effects). This type of cavitation is
usually indicated as vaporous cavitation and can be viewed as an intermediate
regime between inertially controlled cavitation and boiling.
As previously mentioned, the asymptotic growth of the bubble radius is linear in
time when controlled by inertial effects and is proportional to the square root of
time when controlled by thermal effects with purely conductive heat transfer. This
situation is modied, but not fundamentally changed, by the simultaneous presence
of convection due to the relative motion of the bubble in the liquid. One important
consequence is that, the other conditions being the same, a bubbly mixture will
display a larger tensile strength when the growth of the cavities is controlled by
thermal rather than inertial effects. This phenomenon provides a qualitative inter-
pretation of the lower cavitation susceptibility and more gradual performance
breakdown commonly observed in pumps operating in travelling bubble cavitation
conditions with liquids closer to the boiling point (Chivers 1969).
In view of the above considerations, inertial effects tend to be important in the
initial stages of bubble growth, especially if the vapor pressure is small, but at some
later time thermal effects will inevitably prevail. In some situations, for example
when the vapor pressure is large and therefore the required evaporation rate is high,
inertial effects are never signicant and the bubble growth will be controlled by
thermal effects alone. Whether in a practical case the bubble growth will be con-
trolled by thermal effects rather than by inertial effects clearly depends on the
relative importance of the two phenomena during the largest portion of bubble
growth.
A quantitative description of this phenomenon clearly requires the determination
of the behavior of the bubble pressure pB through the simultaneous solution of the
RayleighPlesset equation and the energy equations for the two phases. The
resulting mathematical problem is quite complex due to its nonlinear nature and to
the strong coupling between mechanical and thermodynamic effects. This analysis
for a stream of bubbles travelling with velocity Uo in a liquid stream at temperature
TLo has been outlined by Brennen (1973). It is assumed that the pressure and
temperature of the vapor inside the bubble are uniform and that the temperature
boundary layer at the liquidvapor interface is small with respect to the bubble
radius (Plesset and Zwick 1954; Zwick and Plesset 1955). The drop of the liquid
temperature at the bubble surface TR = TR TLo is related to the corresponding
drop of the vapor pressure inside the bubble pV = pV pVo using the Clausius
Clapeyron equation. Then, by only retaining the prominent terms in the energy
equations and neglecting inertial effects in the momentum equation for the bubble
growth, it follows that vaporous cavitation will occur when
12 L. dAgostino et al.
3 1 2
Uo V 2q2 2
< < 1 2 LV V 2 = TL
L L cpL TLo L
In this equation L is the typical length of the reduced pressure region crossed by
the bubbles and TL is a very strong dimensional function of the flow temperature
TL (Brennen 1973) which represents the combined thermodynamic properties of the
liquid and the vapor with respect to the insurgence of vaporous cavitation.
Therefore, the above equation implicitly denes a transitional temperature of the
p
flow depending on the value of the parameter Uo3 V L. Vaporous cavitation will
occur when the flow temperature is higher than the above transitional temperature,
oth-erwise inertially dominated cavitation will take place. Assuming, for example,
Uo = 100 m s, L = 0.5 m, and V = 0.1, the transitional temperatures of water and
liquid oxygen, respectively, are 60 C and 70 K ca., while, on the other hand,
cavitation in liquid hydrogen would virtually always be of the vaporous type.
Experimental evidence also suggests the existence of a correlation between the
relative magnitude of thermal effects with respect to inertial effects and the topo-
logical form of cavitation. In particular, travelling bubble cavitation has been
consistently observed in connection with the predominance of thermal effects,
whereas attached cavitation is increasingly more likely when inertial effects prevail
(Salemann 1959; Sarosdy and Acosta 1961). Quantitative evaluation of thermal
effects in travelling bubble cavitation is difcult because the departure of the vapor
pressure from its unperturbed value depends on various uncertain factors, including
the size and the growth time of the bubble. As a rst orientation, the maximum
change of the vapor pressure according to Plesset (1957) is expressed by
where Rmax is the maximum radius reached by the individual bubble during its
growth time tmax . Due to the different physical characteristics of the problem, the
above analysis of thermal effects in travelling bubble cavitation is not directly
applicable to developed attached cavitation. In this case entrainment theories have
been proposed by various authors (Brennen 1969; Billet 1970; Holl et al. 1975,
Billet et al. 1978) on the assumption that the cavity behavior and the hydraulic
performance degradation in geometrically similar cavitating turbomachines are only
functions of the cavitation index based on the cavity pressure pC :
po pC
C =
2L Uo
1 2
CM Pe L qLV
T =
CA Nu V cpL
Cavitation occurs when the local pressure in the liquid stream drops below the
inception value as a consequence of a combined decrease of the inlet liquid pressure
and a simultaneous increase of the rotational speed and volume flux. When
investigating cavitation in turbomachines it is, however, more convenient to con-
sider the situation where the inlet pressure is quasi-statically reduced, while the
rotational speed and volume flow rate are kept constant. The resulting variation of
the head rise is the cavitation performance curve (or cavitation characteristics) of
the turbomachine. The development of cavitation in turbomachines can ideally be
divided in four steps: inception, usually rst appearing as a laminar separation
cavity at the leading edges of the blade tips, or as tip vortex cavitation; blade
cavitation, a more extensive form of cavitation in some cases occurring on alternate
blades only; developed cavitation, characterized by larger, periodically pulsating
cavities; cavitation breakdown, where the cavities greatly interfere with the flow
and produce a rapid drop of the turbomachines performance. This rather idealized
situation is complicated in practice by the simultaneous presence of separation, tip
clearance vortex cavitation, secondary and leakage flows, real fluid effects, and
other interfering phenomena. The steady cavitation problem in turbomachines
essentially consists in the prediction of head and efciency degradation and nal
breakdown as a function of the inlet pressure at constant flow rate and rotational
speed.
The theoretical analysis of steady cavitation in turbomachines has mostly relied
on the application of free streamline theory on simplied models of the cavitating
flow. A classic example is the work of Gongwer (1941) who used the theory of Betz
and Petersohn (1931) for flat plate cavitating cascades (essentially derived from the
corresponding theories for isolated hydrofoils) to predict the cavitation performance
of centrifugal impellers. The same approach has been applied to axial machines
and, more specically, to helical inducers by Stripling and Acosta (1962).
Neglecting radial velocity components, the flow in the inducer is assumed to occur
on cylindrical surfaces, which are then developed on a plane. Cavitation in the
resulting cascade of flat plate hydrofoils is then represented by a free streamline
model, where flow separates at the leading edges and reattaches downstream on
ctitious surfaces, parallel to the hydrofoils, whose only function is to guide and
stabilize the flow in the wake. At the end of these surfaces the flow undergoes a
sudden expansion, with the simultaneous occurrence of diffusion losses and pres-
sure recovery. The cavity is assumed to be much shorter than the blade passages
(partial cavitation), so that the uniform downstream flow is perfectly guided and not
affected by the occurrence of cavitation upstream. The height of the cavity relative
to the blade spacing determines the flow blockage and therefore the diffusion losses
in the wake, which crucially contribute to the performance degradation and
breakdown of the turbomachine. The solution clearly depends on the cavitation
number and on other geometric and kinematic parameters: inlet flow velocity, angle
of attack, blade angle and spacing. In their paper Stripling and Acosta extensively
document the variation of diffusion losses and cavity height, length, and aspect ratio
as functions of the cavitation number (or other equivalent cavity pressure param-
eter) for various values of the attack and blade stagger angles.
An Introduction to Cavitation in Inducers and Turbopumps 15
The above theory has been extended and modied by Stripling (1962) in order to
predict head rise performance and breakdown conditions of cavitating inducers as a
function of the inlet pressure. The results are rst expressed in terms of the NPSH
and all radially varying parameters are reduced to blade tip station. Head break-
down is assumed to occur when the cascade reaches supercavitation conditions. In
this situation the cavity length becomes innite and the cavitation number attains its
lowest possible value, which only depends on the angle of attack and on the blade
stagger angle. The maximum value of the cavitation number at breakdown, which
determines the suction performance of the inducer, is achieved when the angle of
attack is equal to one half the blade stagger angle measured from the circumferential
direction. Comparison with experimental results shows, however, that the theory
signicantly underestimates the inducer NPSH at breakdown. Stripling speculates
that this discrepancy is due to real fluid displacement effects, which effectively
reduce the available through flow area in the blade passages. Consequently he
introduces an empirical correction coefcient of the through flow area accounting
for the difference between the ideal and actual blockage of the cavity. Small
changes of this correction coefcient have a large influence on the suction per-
formance of the inducer. Stripling shows that the minimum NPSH at breakdown
and the blockage correction coefcient of a helical inducer can be expressed as a
function of a single parameter, which only involves the inducers flow coefcient,
the angle of attack and the blade angle. The minimum NPSH at breakdown and the
whole cavitation performance curve predicted by the modied theory are in sub-
stantial agreement with the experimental results from various helical inducers of
different geometric characteristics. In his paper Stripling also discusses the effects of
different leading edge shapes on the inducer cavitation characteristics. Theoretical
considerations (Stripling and Acosta 1962) suggest that superior suction perfor-
mance should be obtained when the blade leading edge is fully contained in the
cavity (i.e., when the suction side does not interfere with the separating streamline)
and the actual flow more closely follows the ideal free streamline model. The
experimental results presented by Stripling (1962) and relative to inducers with
various leading edge shapes conrm this conclusion, indicating that maximum
suction performance and better agreement with the theory are achieved when the
suction side, rather than the pressure side, is led in order to obtain a sharp corner
for xed separation to occur from the pressure side without any flow rotation
around the leading edge. The modied theory by Stripling (1962) proved to be
fairly successful in predicting the cavitation characteristics of helical inducers in
water and represents a very useful tool in inducer design analysis. However, caution
must be exercised in generalizing its results to liquids with important thermody-
namic effects, where cavitation is more homogeneous and the free streamline model
may be misleading. The same consideration holds for the extension of Striplings
theory to inducers of signicantly different geometries, such as aerodynamically
designed inducers with variable cross-sectional area and spiraling leading edges,
where the assumption of cylindrical flow is less applicable and tip vortex cavitation
may play a larger role with respect to cavitation on the blade suction surfaces.
16 L. dAgostino et al.
the various components and therefore vary widely in the specic applications. Apart
from the case of turbomachines operating in the positively sloped part of their
characteristic curve, the large majority of the global instabilities observed in tur-
bomachines are related to the interaction of the inertia and the compressibility of the
flow system or of some of its components, in the presence of an excitation source
capable to initiate the instability.
Excluding the case of acoustic phenomena (Abbott et al. 1963; Jeager 1963), in
liquid pumping systems unsteady cavitation, especially near rotating parts, often
represents the main source of fluid volume changes and can be responsible for the
onset of operational instabilities similar to rotating stall and surge phenomena
commonly observed in compressors. The rst type of unsteady cavitation is usually
indicated as rotating cavitation and consists in a blade to blade variation of the
extent of cavitation, which rotates with respect to the inducer (or the impeller) of the
machine. This type of instability involves signicant perturbations of the inlet
pressure and generates circumferential flow distortions similar to those observed
during rotating stall phenomena (Taylor et al. 1969; Kamijo et al. 1977; Brennen
1994; Hashimoto et al. 1997; Zoladz 2000; Pace et al. 2015), but is not associated to
time-dependent fluctuations of the output pressure and mass flow rate of the
machine. The second type of unsteady cavitation phenomena consists, instead, of
perturbations of volume fluxes and pressures at the input and output of the machine
as a consequence of axial fluctuations of the flow caused by simultaneous changes
in the volume of the cavities. This type of instability, called auto-oscillation or
surge, is generally more severe, since it often involves rather large and violent
fluctuations of the flow parameters with respect to their steady-state values (Sack
and Nottage 1965; Young et al. 1972; Hobson and Marshall 1979). In turbopumps
belonging to more complex systems, unsteady cavitation can also be strongly
coupled with other related components capable to provide positive feedback and
therefore sustain joint oscillations. A typical example is the POGO instability in
liquid propellant rockets (Fashbaugh and Streeter 1965; Rubin 1966; Zielke 1969).
Here the longitudinal vibrations of the rocket structure generate inertial pressure
fluctuations in the vertical lines of the fuel feed system, which can interfere with the
occurrence of cavitation in the turbopumps. The resulting unsteady perturbations of
the fuel and/or oxidizer fluxes to the rocket engine can in turn induce thrust fluc-
tuations capable to excite the structure and make the oscillations self-sustained.
The occurrence of cavitation-induced auto-oscillations depends greatly on the
specic characteristics of the pumping system in addition to the characteristics of
the pumping elements alone. If the conditions are favorable, auto-oscillations ini-
tiate when a critical value of the cavitation number is reached and can seriously
compromise the overall performance of a pumping system even before the occur-
rence of large scale performance breakdown due to chocking and flow blockage
effects (Braisted 1979). The cavitating element of the hydraulic system is respon-
sible for initiating the auto-oscillations, while the other components only influence
their development. The onset of instabilities is greatly facilitated by the presence of
the compliant volume due to the presence of the cavities, in conjunction with the
inherently unsteady nature of cavitation, which provides a natural excitation and
18 L. dAgostino et al.
X + 2X + X = 0
with and constant. By assuming solutions of the form est for the controlled
variable X as a function of time t, one is reduced to the determination of the roots of
the associated characteristic equation:
s2 + 2s + = 0
for the complex parameter s. Then, the transient response of the system is expressed
by:
h p i h p i
X t = c1 exp + 2 t + c2 exp 2 t
with the two constants c1 and c2 depending on the initial conditions. Thus, clearly,
the system will be statically and/or dynamically stable when either of the following
two conditions is satised:
(1) 0 < < 2 with > 0, leading to converging exponential growth;
(2) > 2 with > 0, leading to exponentially converging oscillations.
Viceversa, the system will be statically and/or dynamically unstable in any of the
complementary situations, i.e., when
(3) > 2 with < 0, leading to exponentially diverging oscillations;
(4) 0 < < 2 with < 0, leading to diverging exponential growth;
(5) < 0 with any , leading to diverging exponential growth.
Thus, in general, a system will be stable when both and are positive and will
be unstable when the reverse is the case. On the other hand, from the standpoint of
quasi-static considerations the system reduces to a rst order system with = 1 and
the condition for static stability simply requires > 0, i.e., that the system response
to any perturbation is of restoring type. Note that static stability ( > 0) is necessary
for dynamic stability (conditions 1 and 2), but that a statically stable system can still
be dynamically unstable (conditions 3 and 4).
The above approach can easily be applied to a simple liquid pumping consisting
of a pump, a pressurized storage vessel, a throttle and the necessary pipelines for
connection to the inlet and outlet constant pressure reservoirs. Assuming that the
liquid is inviscid and incompressible and upon linearization for small perturbations
from the steady-state equilibrium conditions, the following characteristic equation
is obtained (Greitzer 1981):
dpT 2 p1 Ao dpP dpT dpT dpP p1 Ao
s s+ =0
dm L V G L dm dm dm dm L VG L
Here p1 is the inlet liquid pressure, dpP dm and dpT dm are, respectively, the
slopes of the steady-state characteristic curves of the pump and of the throttle, VG is
20 L. dAgostino et al.
the volume of the pressurizing gas in the vessel, which behaves isoentropically with
specic heat ratio , and L is the effective inertial length of the liquid piping system
(including the pump and the inlet and outlet lines), expressed by
Z x2
dx
L = Ao
x1 A x
dpT dpP
<
dm dm
L1 dm 1
p1 =
A1 dt
pP pP LP dm 1
p2 p1 = p1 + m 1
p1 m A1 dt
VC VC
VC = p1 + m 1
p1 m
and, assuming G L :
dVC
m 2 m 1 = L
dt
where LP is the inertial length of the pump (from the inlet to the outlet section) and
VC is the cavitation volume. Finally, the linearized pressure drop in the throttle is
expressed by
L2 pT
p2 = m 2
A2 m
In addition to the steady-state slopes of the pump and throttle characteristics, the
above equation also involves the derivatives of the cavitation volume with respect
to the inlet pressure and mass flow. These parameters, respectively called com-
pliance ( VC p1 ) and mass flow gain factor ( VC m ), characterize the
behavior of cavitation in the pump with respect to changes of the independent inlet
variables p1 and m 1 . Since the pressure versus flow characteristics of the pump at
the conditions of interest and the pump compliance are both negative, the system
will become unstable when the mass flow gain factor exceeds a critical value:
pP LP
VC 1 + p1 + L1
> A1 pT
m L
L m 1
should be independent on the pump speed. The same analysis also indicates that the
natural frequency of the cavitation-induced flow oscillations should be proportional
to the pump speed, a result which seems to be in agreement with the experimental
observations of Braisted (1979), Brennen and Braisted (1980).
pt = Re pe it ; m t = Re m e it
by means of their complex amplitudes, here indicated by hats. Therefore, for the
generic component of the pumping system:
p2 p1 Z11 Z12 p1
=
m 2 m 1 Z21 Z22 m 1
where the indexes 1 and 2 refer to the inlet and outlet conditions. In another slightly
different representation the pressure and mass flow fluctuation amplitudes at dis-
charge are expressed as functions of the corresponding inlet values:
p2 Y Y12 p1
= 11
m 2 Y21 Y22 m 1
An Introduction to Cavitation in Inducers and Turbopumps 23
Y =Z +1
where 1 is the unitary matrix of order two. In general, the coefcients of the transfer
matrices Z and Y are complex and are expected to depend on the frequency, the
mean operating point and the extent of cavitation. In the limit for low frequencies
the above equations reduce to the steady-state operational characteristics.
Of particular interest is the case of a cavitating turbopump. In a more generalized
treatment also the rotational speed could be included in the modeling of the
unsteady flow through a turbopump, leading to 2 3 matrix representation.
However, usually the fluctuations of the pump rotational speed in response to
fluctuating loads on the impeller blades are insignicant, being effectively limited
by the large inertia of the rotating parts. The influence of the pump rotational speed
can only become important where there exists the potential for signicant inter-
actions between the performance of the pumping system and the operational speed
of the pump itself. Conceivably this situation may occur when the pump is driven
by a turbine powered by fluid bled from the pump output, a conguration that is
typical of low pressure turbopumps (prepumps) of liquid propellant rocket engine
feed systems, like in the Space Shuttle Main Engine. In general, Brennen and
Acosta (1976) have shown that angular speed changes have a negligible effect on
the dynamics of the pump as long as their relative value is small compared with the
relative change of the cavitation number. In cavitating turbopumps Z11 is the change
of the pressure rise versus inlet pressure, which is expected to be zero in the absence
of cavitation (or of a second gaseous component of the flow), since in this case the
pressure rise produced by the pump does not depend on the value of the inlet
pressure. Clearly, for very low frequencies Z11 reduces to the slope of the cavitation
performance curve versus the cavitation number of the pump. The element Z12 is
the change of the pressure rise versus inlet mass flow rate and therefore represents
the impedance of the pump, comprising a resistive and an inertial component At
low frequencies the pump resistance simply approaches the slope of the steady-state
characteristic curve of the pump. The elements Z21 and Z22 account for the effects of
the inlet pressure and mass flow rate on the cavitation volume of the pump and
therefore on the instantaneous difference between the inlet and outlet mass fluxes.
Hence, Z21 and Z22 ,respectively, represent the pump compliance and mass flow
gain factor. Neglecting the effects of liquid compressibility and the deformability of
the pump casing, Z21 and Z22 are only due to the occurrence of cavitation or to the
presence of a second gaseous component in the flow. Therefore, the pump com-
pliance and mass flow gain factor are expected to vanish either in fully wetted flow
conditions or in the limit for very low frequencies, when the cavitation volume (if
any) remains constant in time.
Similar considerations can be made for other typical elements of an hydraulic
system. For example, the transfer matrix of an incompressible flow in a rigid duct of
reference cross- sectional area A and inertial length L is given by:
24 L. dAgostino et al.
1 R + i AL
Y=
0 1
Once the transfer matrix of each element of a pumping system is known it is then
possible to determine by simple multiplication the transfer matrix of the whole flow
line.
The stability analysis of the pumping system can then be carried out using
energy considerations, as proposed by Brennen and Braisted (1980). Since any
form of instability requires the supply of a certain amount of energy in order to
sustain the oscillations, Brennen and Braisted suggested to consider as an instability
parameter the net output change of the mechanical energy flux developed by the
system in an oscillation cycle:
Z t+T
m 2 p2 m 1 p1
E = dt
t L g
E 0
RefI1 Z12 + I2 g 0
where I1 and I2 are, respectively, the impedances of the inlet and outlet lines and Z12
is the only nonzero term of the transfer matrix of the pumping element. Thus, the
An Introduction to Cavitation in Inducers and Turbopumps 25
stability condition simply requires that the resistance of the pumping element must
be smaller than the combined resistances of the inlet and outlet lines. Violation of
this condition can lead to surge both in compressors (Pearson and Bowmer 1949;
Greitzer 1976a, b) and in pumps (Stepanoff 1948). In the cavitating case all of the
pump transfer matrix elements are in general different from zero and the condition
for stability becomes formally more complex (Braisted 1979):
I1 1 + Z11 + I2 1 + Z22 Z12 I1 I2 Z21
Re 0
1 + Z22 I1 Z21
The above two equations conrm that stability is a global characteristics of the
pumping system, involving the combined properties of each individual component.
It is also evident from the rst of these equations that only the pumping element(s)
can actively induce instability in a system and that increasing the (positive) resis-
tances of the remaining elements have a favorable stabilizing effect. In general, the
above equations indicate that the system stability can be improved by properly
modifying and/or combining the characteristics of its components. Attempts to this
effect have actually been made for example in order to suppress auto-oscillations in
the Space Shuttle Main Engine (Murphy 1969; Dorian 1977).
In general, the dynamic activity parameter is expected to depend on the cavi-
tation number, the mean operational point and the frequency of the auto-oscillation.
Specically, the observations of Braisted (1979) indicate that the dynamic activity
parameter goes through a maximum for a certain value of the oscillation frequency
and increases at low cavitation numbers. These considerations, therefore, suggest a
simple method for the determination of both the frequency of auto-oscillation and
of the cavitation number at which the auto-oscillation will occur. In particular,
given the mean operational conditions of the pumping system, one would expect
that auto-oscillations would initially develop with frequency corresponding to the
peak in the spectral distribution of the dynamic activity parameter and at the highest
value of the cavitation number for which the peak becomes positive. The analysis
also indicates that the auto-oscillation frequency is proportional to the pump rota-
tional speed for given values of the flow coefcient and of the cavitation number.
Another noteworthy result is that the stability of the system is most sensitive to
changes of the mass flow gain factor and, to a lesser degree, of the pump impedance
(Braisted 1979). Braisted and Brennen (1980) have used the above stability analysis
in conjunction with experimentally determined characteristics of the pumping
system and found reasonable in agreement with direct observation of the
auto-oscillations. Experimental data also conrm that the auto-oscillation frequency
tends to decrease with the cavitation number, as indicated by the stability analysis.
Thus, within the limits of a linearized analysis, the dynamic problem of a cav-
itating turbopump consists in the determination of the possibly complex elements of
its transfer matrix. Transfer matrices of cavitating pumps have been investigated
both theoretically (Brennen and Acosta 1973; Brennen and Acosta 1976; Brennen
1978a; Brennen and Braisted 1980) and experimentally (Ng 1976; Ng et al. 1976;
Ng and Brennen 1978; Brennen 1978b; Braisted 1979; Brennen et al. 1982;
26 L. dAgostino et al.
11 Conclusions
References
Abbot, H. F., et al. (1963). Measurements of auto-oscillation in hydroelectric supply tunnel and
penstock system. ASME Journal Basic Engineering, 85, 625630.
Acosta, A. J. (1974). Cavitation and fluid machinery. In Cavitation (pp. 386396), Conference
held at Heriot-Watt University, Edinburgh, Scotland, September 1974, London Institution of
Mechanical Engineers.
Acosta, A. J., and Hollander, A. (1959). Remarks on Cavitation in Turbomachines. California
Institute Technical Report No. 79.3, Pasadena, CA, USA.
Acosta, A. J., & Parkin, B. R. (1975). Cavitation inceptiona selective review. Journal of Ship
Research, 19, 193205.
Apfel, R. E. (1970). The role of impurities in cavitation-threshold determination. Journal of the
Acoustical Society of America, 48(5), 11791186, Part 2.
Arndt, R. E. A. (1981). Cavitation in fluid machinery and hydraulic structures. Annual Review of
Fluid Mechanics, 13, 273328.
An Introduction to Cavitation in Inducers and Turbopumps 29
Barr, R. A. (1967). Study of instabilities in high tandem row inducer pumps. Hydronautics Interim
Technical Report 703-1.
Bader, H. (1970). The hyperbolic distribution of particle sizes. Journal of Geophysical Research,
75(15), 28222830.
Betz, A., & Petersohn, E. (1931). Anwendung du Theorie der Frien Strahlen. In Ingenieur Archiv,
Band II.
Bhattacharyya, A. (1994). Internal Flows and Force Matrices in Axial Flow Inducers. Report
No. E249.18, California Institute of Technology, Pasadena, USA.
Bhattacharyya, A., Acosta, A. J., Brennen, C. E., & Caughey, T. K. (1997). Rotordynamic forces
in cavitating inducers. ASME Journal of Fluids Engineering, 199(4), 768774.
Billet, M. L. (1970). Thermodynamic efects on developed cavitation in water and freon 113. M.S.
thesis, Department of Aerospace Engineering, Pennsylvania State University, University Park,
PA, USA.
Billet, M. L., et al. (1978). Correlations of thermodynamic effects for developed cavitation. In
Polyphase flow in turbomachinery (pp. 271289). ASME Special Publications.
Billet, M. L., et al. (1981). Correlations of thermodynamic effects for developed cavitation. ASME
Journal of Fluids Engineering, 103, 534542.
Blake, F. G. Jr. (1949). The Onset of Cavitation in Liquids. In Harvard Acoustics Research
Laboratory, TM 12.
Braisted, D. M. (1979). Cavitation Induced Instabilities Associated with Turbomachines. Report
No. E184.2, California Institute of Technology, Pasadena, CA, USA.
Brennen, C. E. (1978a). Bubbly flow model for the dynamic characteristics of cavitating pumps.
Journal of Fluid Mechanics, 89, 223240.
Brennen, C. E. (1978b). On unsteady dynamic response of phase changes in hydraulic systems. In
Proceedings of International Seminar of Center for Heat and Mass Transfer in Two-Phase and
Chemical Systems, Dubrovnik, Yugoslavia, Hemisphere Publishing Corporation.
Brennen, C. E. (1969). The Dynamic balances of dissolved air and heat in natural cavity flows.
Journal of Fluid Mechanics, 37, 115127.
Brennen, C. E. (1994). Hydrodynamics of pumps. Oxford Engineering Science Series 44. Oxford
University Press.
Brennen, C. E. (1995). Cavitation and bubble dynamics. Oxford University Press.
Brennen, C. E. (2005). Fundamentals of multiphase flows. Cambridge University Press.
Brennen, C. E. (1973). The dynamic behavior and compliance of a stream of cavitating bubbles.
Transactions of the ASME Journal Fluids Engineering, 533541.
Brennen, C. E., & Acosta, A. J. (1973). Theoretical, quasi-static analysis of cavitation compliance
in turbopumps. Journal of Spacecraft, 10(3), 175179.
Brennen, C. E., & Acosta, A. J. (1976). The dynamic transfer function for a cavitating inducer.
ASME Journal of Fluids Engineering, 98, 182191.
Brennen, C. E., & Braisted, D. M. (1980). Stability of hydraulic systems with focus on cavitating
pumps. In Proceedings of IAHR Conference, Tokyo, Japan.
Brennen, C. E., et al. (1982). Scale effects in the dynamic transfer functions for cavitating inducers.
ASME Journal of Fluids Engineering, 104, 428433.
Childs, D. W. (1983). Finite length solutions for rotordynamic coefcients of turbulent annular
seals. ASME Journal of Lubrication Technology, 105, 437444.
Chivers, T. C. (1969). Cavitation in centrifugal pumps. In Proceedings of the Institution of
Mechanical Engineers, 184(2), Pt. l.
Colding-Jorgensen, J. (1980). Effect of fluid forces on rotor stability of centrifugal compressors
and pumps. In 1st Workshop on Rotordynamic Instability Problems in High Performance
Turbomachinery, Texas A&M University, TX, USA.
dAgostino, L., & Acosta, A. J. (1991a). A cavitation susceptibility meter with optical cavitation
monitoringpart one: design concepts. ASME Journal of Fluids Engineering, 113(2), 261269.
dAgostino, L., & Acosta, A. J. (1991b). A cavitation susceptibility meter with optical cavitation
monitoringpart two: experimental apparatus and results. ASME Journal of Fluids
Engineering, 113(2), 270277.
30 L. dAgostino et al.
dAgostino, L., & Acosta, A. J. (1991c). Separation and surface nuclei effects in a cavitation
susceptibility meter. ASME Journal of Fluids Engineering, 113(4), 695699.
Dorian, H. H. (1977). Space Shuttle POGO Prevention. SAE, 770969.
Ehrich, F. F. (1967). The influence of trapped fluids in high speed rotor vibration. ASME Journal
of Engineering for Industry, 806812.
Eisenberg, P. (1961). Mechanics of cavitation. In V.L. Streeter (Ed.), Handbook of fluid dynamics
(pp. 12.212.24). New York: McGraw-Hill.
Epstein, P. S., & Plesset, M. S. (1950). On the stability of gas bubbles in liquid-gas solutions.
Journal of Chemical Physics, 18, 15051509.
Etter, R. J. (1970). An Investigation on Tandem Row High Head Inducers. Hydronautics Technical
Report 703-9.
Farrel, E. C., & Fenwick, J. R. (1973). Pogo Instabilities Suppression Evaluation. NASA Report
CR-134500.
Fashbaugh, R. H., & Streeter, V. L. (1965). Resonance in liquid rocket engine systems.
Transactions of the ASME, Series D, 87(4).
Fox, F. E., & Herzfeld, K., F. (1954). Gas Bubbles with Organic Skin as Cavitation Nuclei.
Journal of the Acoustical Society of America, 26, 984.
Franc, J. -P., & Michel, J. -M. (2004). Fundamentals of cavitation. Kluwer Academic Publishers.
Franz, R. J. (1989). Experimental Investigation of the Effect of Cavitation on the Rotordynamic
Forces on a Whirling Centrifugal Pump Impeller. Ph.D. Thesis, California Institute of
Technology, Pasadena, USA.
Gelder, T. F., Ruggeri, R. S., & Moore, R. D. (1966). Cavitation Similarity Considerations Based
on Measured Pressure and Temperature Depressions in Cavitated Regions of Freon 114.
In NASA TN D-3509.
Greitzer, E. M. (1976a). Surge and rotating stall in axial flow compressorspart I: Theoretical
compression system model. ASME Journal of Engineering for Power, 98, 190198.
Greitzer, E. M. (1976b). Surge and rotating stall in axial flow compressorspart II: Experimental
results and comparison with theory. ASME Journal of Engineering for Power, 98, 198211.
Greitzer, E. M. (1981). The stability of pumping systems. ASME Journal of Fluids Engineering,
103, 193242.
Goggin, D. C. (1982) Field experiences with rub induced instabilities in turbomachinery. In 2nd
Workshop on Rotordynamic Instability Problems in High Performance Turbomachinery, Texas
A&M University, TX, USA.
Gongwer, C. A. (1941). A theory of cavitation flow in centrifugal pump impellers. Transactions of
the ASME, 63, 2940.
Grabowski, B. (1982). Shaft vibrations in turbomachinery excited by cracks. In 2nd Workshop on
Rotordynamic Instability Problems in High Performance Turbomachinery, Texas A&M
University, TX, USA.
Gunter, E. J., Springer, H., & Humphris, R. (1982). Influence of unbalance on the stability
characteristics of flexible rotor bearing systems. In ASME/Applied Mechanics, Bioengineering
and Fluids Engineering Conference (Vol. 55), Houston, TX, USA, AMD.
Harvey, E. N., McElroy, W. D., & Whiteley, A. H. (1947). On cavity formation in water. Journal
of Applied Physics, 18, 162.
Hashimoto, T., Yoshida, M., Watanabe, M., Kamijo, K., & Tsujimoto, Y. (1997). Experimental
study of rotating cavitation of rocket propellant pump inducers. AIAA Journal of Propulsion
and Power, 13(4), 488494.
Hobson, D. E., & Marshall, A. (1979). Surge in centrifugal pumps. In Proceedings of 6th
Conference on Fluid Machinery (Vol. 1), Hungarian Academy of Sciences, Budapest.
Holl, J. W. (1969). Limited cavitation. ASME, Cavitation State of Knowledge, 2663.
Holl, J. W. (1970). Nuclei and cavitation. ASME Journal of Basic Engineering, 92, 681688.
Holl, J. W., & Wislicenus, G. F. (1961). Scale effects on cavitation. ASME Journal of Basic
Engineering, 385398.
Holl, J. W. et al. (1975). Thermodynamic effect on developed cavitation. In: Cavity Flows, ASME
Fluids Engineering Conference (pp. 101109), Minneapolis, MN, USA.
An Introduction to Cavitation in Inducers and Turbopumps 31
Hord, J., Anderson, L. M., & Hall, W. J. (1972). Cavitation in Liquid Cryogens. I. Venturi.
In NASA CR-2054 (83 pp).
Hord. J. (1973a). Cavitation in Liquid Cryogens. II. Hydrofoil. In NASA CR-2156 (157 p).
Hord, J. (1973b). Cavitation in Liquid Cryogens. III. Ogives. In NASA CR-2242 (235 pp).
Hord, J., (1974). Cavitation in Liquid Cryogens, Volume IVCombined Correlations for Venturi,
Hydrofoil, Ogives and Pumps. In NASA CR-2448.
Jakobsen, J. K. (1964). On the mechanism of head breakdown in cavitating inducers. ASME
Journal of Basic Engineering, 86, 291304.
Jeager, C. (1963). The theory of resonance in hydropower systems: Discussion of incidents and
accidents occurring in pressure systems. ASME Journal of Basic Engineering, 85, 631640.
Kamijo, K., Shimura, T., Watanabe, M. (1977). Experimental Investigation of Cavitating Inducer
Instability.
Kimball, A. L., Jr. (1924). Internal friction theory of shaft whirling. General Electric Review, 27,
244251.
Knapp, R. T., Daily, J. W., & Hammit, F. G. (1970). Cavitation. New York: McGraw-Hill.
Lund, J. W. (1974). Stability and Damped Critical Speeds of a Flexible Rotor in Fluid-Film
Bearings. ASME Journal of Engineering for Industry, 509517.
Matsushita, O., et al. (1982). Rotor vibration caused by external excitation and rub. In 2nd
Workshop on Rotordynamic Instability Problems in High Performance Turbomachinery, Texas
A&M University, Texas.
Moore, R. D., & Ruggeri, R. S. (1968). Predictions of Thermodynamic Effects of Developed
Cavitation Based on Liquid Hydrogen and Freon 114 Data in Scaled Venturis. NASA TN
D-4899.
Moore, R. D. (1974). Prediction of pump cavitation performance. In B. Lakshminarayana, W.
R. Britsch & W. S. Gearhart (Eds.), Symposium on Fluid Mechanics, Acoustics and Design of
TurbomachineryPart II, NASA SP-304 (pp. 733755).
Mori, Y., Hijikata, K., & Nagatani, T. (1977). Fundamental Study of Bubble Dissolution in Liquid.
International Journal of Heat and Mass Transfer, 20, 4150.
Murphy, G. L. (1969). Pogo Suppression Analysis of the S-11 and S-IVB LOX Feed Systems.
Summary Report ASD-ASTN-1040, Brown Engineering Co., Huntsville, AL, USA.
Ng, S. L. (1976). Dynamic Response of Cavitating Turbomachines. PhD thesis, California Institute
of Technology, Pasadena, CA, USA.
Ng, S. L., Brennen, C. E., and Acosta, A.J. (1976). The Dynamics of Cavitating Inducer Pumps. In
Conference on Two-Phase Flow and Cavitation, Grenoble, France.
Ng, S. L., & Brennen, C. E. (1978). Experiments on the dynamic behavior of cavitating pumps.
ASME Journal of Fluids Engineering, 100, 166176.
OHern, T. J., dAgostino, L., & Acosta, A. J. (1988). Comparison of holographic and coulter
counter measurements of cavitation nuclei in the ocean. ASME Journal of Fluids Engineering,
110, 200207.
Pace, G., Valentini, D., Pasini, A., Torre, L., Fu, X., & dAgostino, L. (2015). Geometry Effects on
Flow Instabilities of Different Three-Bladed Inducers. ASME Journal of Fluids Engineering,
137(4)/011102-1, 041304.
Pasini, A., Torre, L., Cervone, A., & dAgostino, L. (2011a). Characterization of the rotordynamic
forces on tapered axial inducers by means of a rotating dynamometer and high-speed movies.
In Proceedings of WIMRC 3rd International Cavitation Forum 2011, University of Warwick,
United Kingdom.
Pasini, A., Torre, L., Cervone, A., & dAgostino, L. (2011b). Continuous spectrum of the
rotordynamic forces on a four-bladed inducer. ASME Journal of Fluids Engineering, 133(12).
Pearsall, L. S. (1972). Cavitation. London: Mills & Boon.
Pearson, H., & Bowmer, T. (1949). Surging of axial compressors. Aeronautical Quarterly, I, 195210.
Plesset, M. S. (1957). Physical effects in cavitation and boiling. In Proceedings of 1st Symposium
on Naval Hydrodynamics (pp. 297323), Washington, D.C., USA.
Plesset, M. S. (1969). The tensile strength of liquids. In ASME Fluids Engineering and Applied
Mechanics Conference, (pp. 1525), Evanston, Illinois.
32 L. dAgostino et al.
Plesset, M. S., & Prosperetti, A. (1977). Bubble dynamics and cavitation. Annual Review of Fluid
Mechanics, 9, 145185.
Plesset, M. S., & Zwick, S. A. (1954). The growth of vapor bubbles in superheated liquids.
Journal of Applied Physics, 5(4), 493500.
Rosenmann, W. (1965). Experimental Investigations of Hydrodynamically Induced Shaft Forces
With a Three Bladed Inducer. In Proceedings of ASME Symposium on Cavitation in Fluid
Machinery.
Rubin, S. (1966). Longitudinal instability of liquid rockets due to propulsion feedback (POGO).
Journal of Spacecraft and Rockets, 3(8), 11881195.
Ruggeri, R. S., & Moore, R. D. (1969). Method for Prediction of Pump Cavitation Performance
for Various Liquids, Liquid Temperatures and Rotative Speeds. In NASA TN D-5292,
Washington, D.C., USA.
Sack, L. E., & Nottage, H. B. (1965). System oscillations associated to cavitating inducers. ASME
Journal of Basic Engineering, 87, 917924.
Salemann, V. (1959). Cavitation and NPSH requirements of various liquids. ASME Journal of
Basic Engineering, 81, 167180.
Sarosdy, C. R., & Acosta, A. J. (1961). Note on observations of cavitation in different fluids.
ASME Journal of Basic Engineering, 3, pp. 399400.
Stahl, H. A., & Stepanoff, A. J. (1956). Thermodynamic aspects of cavitation on centrifugal
pumps. ASME Transactions, 78, 16911693.
Stepanoff, A. J. (1948). Centrifugal and axial pumps; theory, design and application. Wiley.
Stepanoff, A. J. (1961). Cavitation in centrifugal pumps with liquids other than water. ASME
Transactions, Series A, Journal of Engineering for Power, 83(1), 7990.
Stepanoff, A. J. (1964). Cavitation properties of liquids. ASME Journal of Engineering for Power,
86, 195200.
Stripling, L. B. (1962). Cavitation in turbopumps, part 2. ASME Journal of Basic
Engineering, 84(3), 329.
Stripling, L. B., & Acosta, A. J. (1962). Cavitation in turbopumpspart 1. ASME Journal Basic
Engineering, 84, 326338.
Taylor, W. E., et al. (1969). Systematic Two-Dimensional Cascade Tests, Volume IDouble
Circular Arc Hydrofoil. In NASA CR-72498.
Torre, L., Pasini, A., Cervone, A., Pecorari, L., Milani, A., & dAgostino, L. (2010).
Rotordynamic forces on a three bladed inducer. In Proceedings of Space Propulsion
Conference, San Sebastian, Spain.
Torre, L., Pasini, A., Cervone, A., & dAgostino, L. (2011a). Continuous spectrum of the
rotordynamic forces on a four-bladed inducer. In ASME/JSME/KSME Joint Fluids Engineering
Conference, Hamamatsu, Japan.
Torre, L., Pasini, A., Cervone, A., & dAgostino, L. (2011b). Experimental Characterization of the
Rotordynamic Forces on Space Rocket Axial Inducers. ASME Journal of Fluids Engineering,
133(10).
Torre, L., Cervone, A., Pasini, A., & dAgostino, L. (2011c). Experimental characterization of
thermal cavitation effects on space rocket axial inducers. ASME Journal of Fluids Engineering,
133(11).
Vaage, R. D., et al. (1972). Investigation of Characteristics of Feed System Instabilities. Final
Report MCR-72-107, Martin Marietta Corporation, Denver, CO, USA.
Valentini, D., Pace, G., Pasini, A., Torre, L., & dAgostino, L. (2015). Influences of the operating
conditions on the rotordynamic forces acting on a three-bladed inducer under forced whirl
motion. Journal of Fluids Engineering, 137(7), 071304.
Wachel, J. C., & Szenasi, F. R. (1980). Field verication of lateral-torsional coupling effects on
rotor instability in centrifugal compressors. In 1st Workshop on Rotordynamic Instability
Problems in High Performance Turbomachinery, Texas A&M University, TX, USA.
Wagner, R. G. (1971). Titan II Engine Transfer Function Test Results. Report No. TOR-0059
(6471)-9, Aerospace Corporation, El Segundo, CA, USA.
An Introduction to Cavitation in Inducers and Turbopumps 33
Young, W. E., et al. (1972). Study of Cavitating Inducer Instabilities. Pratt and Whitney
Report PWA FR-5131.
Zielke, W., et al. (1969). Forced and self-excited oscillations in propellant lines. ASME Journal of
Basic Engineering, 91, 671677.
Zika, V. J. (1984). Correlation of cavitating centrifugal pumps. ASME Journal of Fluids
Engineering, 106, 141146.
Zoladz, T. (2000). Observations on rotating cavitation and cavitation surge from the development
of the fastrac engine turbopump. In 36th AIAA/ASME/SAE/ASEE Joint Propulsion Conference,
Huntsville, AL, USA.
Zwick, S. A., & Plesset, M. S. (1955). On the dynamics of small vapor bubbles in liquids. Journal
of Mathematics and Physics, 33, 308330.
Experimental Methods for the Study
of Hydrodynamic Cavitation
Abstract A review of traditional and novel experimental methods for the investi-
gation of hydrodynamic cavitation is presented. The importance of water quality is
discussed, along with its characterization and management. Methods for the direct
and indirect experimental determination of cavitation inception are presented. Along
with traditional optical visualization, methods of measuring developed cavitation
are described, including point and surface electrical probes, optical bubble probes,
acoustic measurements, and indirect measurements of noise and vibration. Recent
developments in the use of ionizing radiation as a means to visualize cavitating ows
are also discussed.
1 Introduction
S.L. Ceccio ()
Department of Naval Architecture and Marine Engineering, University of Michigan,
Ann Arbor MI, USA
e-mail: [email protected]
S.A. Mkiharju
Department of Mechanical Engineering, University of California, Berkeley, CA, USA
e-mail: [email protected]
design before its manufacture at full scale. Experimental testing has also been used
to illuminate the basic processes of cavitating ows, usually through the examination
of canonical ows that may be used to study specic cavitating ow processes, such
as ows with variable area ducts and Venturis, and over headforms and hydrofoils.
The goal of this chapter is to review some experimental methods that have been
successfully employed to study hydrodynamic cavitation (e.g., cavitation produced
by owing liquids). However, many of the methods are also useful for examination of
cavitation produced by acoustic elds as well as other gasliquid multiphase ows,
including boiling ows. And some are also applicable to gassolid and liquidsolid
ows as well. The focus will be on the experimental methods that are available to
researchers as they examine the ow processes that lead to cavitation inception, its
development, and its eect on system performance, and its erosive potential rather
than the test facility itself. But, a brief summary is provided here.
Experimental examination of hydrodynamic cavitation is often performed in a
dedicated test facility. These can be broadly classied into ow loops and towing
tanks. In the former, a prime mover delivers liquid (usually trough ow condition-
ers and a contraction) to the inlet of a test section where the cavitating ow will be
examined. The ow is then returned (usually after passing through a diuser where
the pressure rises) to the prime mover to continue recirculation. Examples of modern
cavitation tunnels are the Grand Tunnel Hydrodynamique (GTH) in France (Lecof-
fre et al. 1987), the Large Cavitation Channel (LCC) in the U.S.A. (Etter et al. 2005),
and the Hydrodynamics and Cavitation Tunnel (HYKAT) in Germany (Wetzel and
Arndt 1994a; Wetzel and Arndt 1994b). Turbomachinery may be tested with closed
recirculating ow loops as well, as described by Avellan et al. (1987). Cavitation may
also be studied by towing a test article in a stationary liquid, and facilities have been
developed to allow for the ambient pressure over the free surface to be varied such
that cavitation can form under a variety of ow conditions, such as the Depressurized
Wave Basin in The Netherlands (Van der Kooij and De Bruijn 1984). The Interna-
tional Towing Tank Committee (ITTC) oers a catalog of many of these facilities,
and Brandner et al. (2007) describe the recent design of a modern cavitation tunnel.
The capabilities and quality of the test facility is, of course, of central impor-
tance to the conduct of experimental investigations of cavitating ows. Many of the
criteria used to assess a cavitation test facility are identical to those used for any
aerodynamic or hydrodynamic test apparatus, including the uniformity and quality
of the freestream ow, including the level of freestream turbulence, and the preci-
sion and range over which ow speed and pressure may be xed. Many of the design
requirements and approaches of subsonic wind tunnels presented by Rae and Pope
(1984) apply equally to conventional water tunnels.
The acoustic characteristics of a cavitation ow facility may also be important
to manage, as the acoustic emission of cavitation may be an important aspect of the
testing program. The recently developed cavitation ow facility of the Japan Defense
Agency is a modern example of a channel developed with these acoustic considera-
Experimental Methods for the Study of Hydrodynamic Cavitation 37
tions (Sato et al. 2003). The tunnel was designed to reduce the amount of background
noise and reverberation to improve the signal-to-noise ratio of for both noncavitat-
ing and cavitating ows. Finally, an important consideration for cavitating ow is
management of the freestream water quality, which will be discussed below.
The inception and development of cavitation can be strongly related to the amount
of free and dissolved gas within the cavitating liquid. Liquid that is supersaturated
with dissolved gas and has many large free gas nuclei would be considered weak,
and cavitation may form where the liquid pressure falls below vapor pressure. Con-
versely, liquid that is undersaturated with dissolved gas and has few free gas nuclei
can sustain pressures below vapor pressure (e.g., can be in tension), and is considered
strong. Determination and control of the liquids cavitation water quality or sus-
ceptibility is an important consideration for many experimental studies of cavitating
ows. Discussions of the importance of water quality with regard to the conduct and
interpretation of cavitation experiments are found in Lecore and Bonnin (1979),
Kuiper (1985), Gindroz and Billet (1998), Arndt (2002), and Atlar (2002).
Cavitation inception occurs when a reduction in the liquid pressure results in a
local pressure below the vapor pressure, and the liquid begins to change phase into
vapor. Homogeneous nucleation process can occur in the bulk of the liquid as a result
of inclusions that naturally form due to the random motion of the liquid molecules
(Brennen 1995). Homogeneous nucleation typically requires a signicant level of
liquid tension, and ultraclean water can sustain tensions of over 30 MPa at room
temperature (Mrch 2007). Yet, for many practical situations, cavitation incepts as
a result of heterogeneous nucleation whereby nucleation sites within the bulk of
the uid or at solid boundaries grow when exposed to suciently strong tension.
The characterization of the uids susceptibility to nucleation is an important aspect
of many cavitation studies, especially those concerned with inception. In turn, the
susceptibility of the ow is related to both the free and dissolved gas content as well
as the nature of potential surfaces and ow-borne nucleation sites.
Water quality can also aect developed cavitation. For example, the presence of
many freestream nuclei can lead to the suppression of sheet cavitation through the
formation of traveling bubbles upstream of the cavity separation location (Li and
Ceccio 1996; Keller 2001), and diusion of dissolved gas into a developed tip vortex
can signicantly alter its core size (Gowing et al. 1995). As a consequence, it is
incumbent upon the cavitation research engineer to adequately characterize and, if
possible, manage the facilitys water quality before conducting experimental studies
and scale testing.
38 S.L. Ceccio and S.A. Mkiharju
Henrys law states that the equilibrium amount of a dissolved gas in a liquid at a given
temperature is related to the partial pressure of the gas. When the gas concentration is
at equilibrium, this is the saturated condition. Then, a reduction in the liquid pressure
would result in supersaturation, and outgassing can occur. Likewise, if the pressure of
the liquid is increased, the undersaturated liquid can dissolve more free gas. Hence,
as a saturated liquid ows into regions of low pressure and cavitates, it is possible
that signicant amounts of outgassing may accompany any vaporization. Moreover,
the level of gas saturation will inuence the stability of free gas nuclei, which will be
discussed below. The total dissolved gas content can be determined using a van Slyke
apparatus developed for measurement of blood gas (Simoni et al. 2002). The van
Slyke apparatus uses a vacuum placed over the liquid sample to produce outgassing,
and it is quite accurate. Traditionally, the vacuum was created by the movement of a
mercury manometer, which has led to its replacement by other methods that do not
require the manual manipulation of mercury reservoirs.
For many practical applications, the cavitating liquid is water, and the dissolved
gases are the main components of air, molecular nitrogen, and oxygen. But, other
dissolved gases may be of interest, especially noble gases (Rooze et al. 2013). Mea-
surement of dissolved gas content can be achieved with a Total Dissolved Gas Pres-
sure (TDGP) probe. A sample of the liquid is placed in a probe beneath a headspace
of gas at a known pressure that is separated from the liquid by a permeable mem-
brane. Over a period the transfer of dissolved gas into or out of the headspace will
change the gas pressure, and this change can then be related to the original dissolved
gas concentration of the liquid sample. In order to relate the pressure change to the
gas concentration, the chemical composition of the dissolved gases must be known,
and some systems combine the probe with a separate instrument to sample and char-
acterize the composition of the gas released into the headspace. For air dissolved in
water, it is often assumed that the ratio of the dissolved gases track the ratio of nitro-
gen and oxygen in air at standard conditions (Yu and Ceccio 1997; Lee et al. 2016). In
fact, in many test facilities, only the dissolved oxygen is measured, and it is assumed
that the level of oxygen saturation parallels the overall dissolved air concentration.
Dissolved oxygen (DO) probes employ a measurement technique similar to that of
pH meters. Two electrodes are suspended in a liquid electrolyte, which is separated
from the test sample by a semipermeable membrane. A low DC voltage is applied
between the electrodes within the electrolyte, and when oxygen molecules from the
test liquid cross the membrane, the magnitude of current between the electrodes will
change.
Experimental Methods for the Study of Hydrodynamic Cavitation 39
Measurement of the free gas content is more dicult. The intent is to characterize
the nucleation sites in the freestream liquid, and a variety of methods have been pro-
posed and evaluated (Lecore and Bonnin 1979; Oldenziel 1982; dAgostino and
Acosta 1991; Ceccio et al. 1995; Billet 1985; Pham et al. 1999). Freestream nuclei
are any inclusion in the uid that will cavitate when exposed to a sucient tension
(as opposed to surface nucleation sites which reside on ow boundaries). An ideal
nucleus is a clean gas bubble, and the nucleation characteristics of such a bubble
can be readily predicted. However, a wider variety of nucleation sites exist in the
ow, including gas pockets on the surface of particles and bubbles that have signi-
cant surface contamination. Nevertheless, it is useful to review the basic nucleation
process of a clean bubble in order to give us a reference to compare with practically
observed nuclei.
Consider a nucleus that is a clean gas bubble with a radius RN that contains vapor
and noncondensable gas. The pressure inside the bubble, PB = PV + PG is the sum of
the partial pressures of the vapor and noncondensable gas, respectively. This pressure
is balanced by the liquid pressure on the bubble surface, P , such that
2S
P = PV + PG ,
RN
where S is the surface tension. When a nucleus experiences a drop in the surround-
ing liquid pressure, the radius may grow quasi-statically from its initial equilibrium
radius to a larger equilibrium radius. However, if a tension is applied below a criti-
cal value, PV = PC P , the radius will grow unboundedly. This critical tension is
given by
4S 2S
< PV P <
3RN RN
(Brennen 1995). Note that since the uid is in tension, PV > P , and that the static
pressure is in fact negative. While not all nuclei are clean bubbles, the uids nuclei
content is often reported as a distribution of nuclei with a given critical radius, RC .
This is the radius that corresponds to the required critical pressure, PC , needed to
produce explosive growth of the nucleus. Therefore, the nuclei content of a liquid is
typically reported as a Nuclei Number Density Distribution (NNDD) as a function of
RC , where NNDD(RC ) has units of [Number]/[Length]4 . Then, the number of nuclei
over a range of critical radius RC is given by NNDD(RC )RC . Alternatively, if the
bin size of the distribution is xed, the Nuclei Number Distribution (NND) with
units of [Number]/[Length]3 can be presented, as shown in Fig. 1 (Chang et al.
2011). The typical range of nuclei critical radii in test facilities spans 1 micron <
RC < 500 micron, and the critical tensions range from 0 > PC > 100 kPa. Nuclei
concentrations can range widely, with order 1 per cubic centimeter for the smallest
nuclei and order 10 per cubic meter for the largest (Gindroz 1998).
40 S.L. Ceccio and S.A. Mkiharju
Measurement of the nuclei distribution can be accomplished with both direct and
inferred means. In direct methods, a sample of the ow is exposed to a known tension,
resulting in the cavitation of nuclei that incepts at (or above) that critical pressure.
These devices are collectively known as Cavitation Susceptibility Meters (CSMs),
as rst discussed by Schiebe (1972). The most common CSM consists of a simple
venture through which the liquid is passed. The pressure in the throat is measured or
inferred, the occurrence of single cavitation events are detected optically, electrically,
and/or acoustically. CSMs of this type have been developed by Oldenziel (1982),
Chahine and Shen (1986), dAgostino and Acosta (1991), Chambers et al. (2000), for
example. During typical operation, a sample of the freestream liquid is drawn from
the ow facility and delivered to the CSM inlet. Care must be taken to ensure that
the sampling process does not signicantly alter the nuclei distribution itself. Then,
the ow is passed through the CSM. Changing the ow rate then varies the throat
pressure, and the number of cavitation inception events is counted at each level of
throat tension. Knowledge of the ow rate and throat tension yields the concentration
of nuclei that incept at or above the given tension. This data is then converted into
the nuclei number distribution.
One limitation of Venturi-based CSMs is the limited volume of the throat. If the
nuclei distribution is too dense, multiple bubbles may simultaneously form in the
throat, creating a blockage. A solution is to use a centerbody Venturi (Lavigne 1991),
as shown in Fig. 2. Now the high-tension region is an annulus of uid around the cen-
terbody, increasing the volume of liquid that is in tension. Keller (1987) developed
Experimental Methods for the Study of Hydrodynamic Cavitation 41
a CSM that used a swirling ow passed through a vortex to create the region of low
pressure.
While the operation of a CSM can be somewhat cumbersome, it is a device that
directly measures the number of cavitation nuclei and their critical tension. It is
important to note that not all nuclei are clean gas bubbles. In fact, particulates with
small gas pockets on their surface can readily act as nuclei, and free gas bubbles
can be coated by an organic skin, eectively modifying their interfacial properties.
(Mrch 2007). Hence, a measurement of the size of the nucleus in the ow may not
necessarily yield an accurate measure of its critical tension.
record the light scattered from a particle passing through the probe volume made
by two crossed laser beams (Tanger and Weitendorf 1992). PDA systems can more
readily determine the radius and velocity of the presumably spherical nucleus pass-
ing through the control volume. Care must be taken to relate the measured event
rate to the actual nuclei density, since the eective measurement volume may not
be easily determined given that nuclei may not pass directly through the measure-
ment volume, for example. And, it is important to discriminate between bubbles and
particles as they pass through the probe volume.
A direct optical measurement of the nuclei distribution can be made using holog-
raphy (Katz et al. 1984; Katz and Sheng 2010). Holographic imaging can yield the
absolute nuclei distribution in a volume of uid, and with proper resolution, can be
used to distinguish between bubbles, particulates, and other contaminants. Hence,
holography is often used as the calibration standard for other nuclei measurement
systems. Holographic systems have been used to measure nuclei distributions both
in the laboratory and in the environment (Katz and Acosta 1981; OHern et al. 1985).
Kawanami et al. (2002) used laser holography to study the structure of a cloud shed
from a hydrofoil and estimated the bubble size distribution. Holography is not typ-
ically used as a routine nuclei measurement method, but recent advances in both
camera technology and digital processing have made its everyday use more feasible.
Characterization of the freestream dissolved and free gas content can be an essen-
tial component of any experimental test and evaluation eort. Moreover, it may be
advantageous to actively manage these quantities through the addition or removal
of dissolved gas and freestream nuclei. Besides ltering, the most basic method to
control the water quality is through degassing the bulk of the test liquid, and deaer-
ation is a common practice to reduce both the free and dissolved gas during testing.
Typically, the dissolved gas concentration will be reduced to levels below 50%.
Control of the water quality can be improved via the active management of the free
and dissolved gas content. The major ow facilities that have implemented these sys-
tems include the Grand Tunnel Hydrodynamique in France (Lecore et al. 1987) and
the Australian Maritime College Cavitation Tunnel (Brandner et al. 2007). Figure 3
shows an image of latter facility that highlights the means of gas control. The facility
is equipped with bulk degassing systems to control the average dissolved gas concen-
tration. Additionally, small gas bubbles can be controllably injected directly into the
ow upstream of the test section, while both small and large gas bubbles are removed
downstream of the test section via gravity separation, coalescence, and resorption. In
this way, the nuclei distribution can be prescribed and maintained during an experi-
ment.
Experimental Methods for the Study of Hydrodynamic Cavitation 43
Fig. 3 A schematic diagram of the water quality management systems of the Australian Mar-
itime College Cavitation Tunnel (Brandner et al. 2007). Nuclei can be injected upstream of the test
section, and the tunnel is designed to remove gas by separation, coalescence, and resorption
Cavitation inception occurs when cavitation is rst observed in the ow, and the
determination of inception conditions is important characterization of the ow itself
as well as an important consideration for the scaling of the performance of model
hydraulic systems. Inception usually occurs when the rst freestream or surface
nuclei encounter sucient tension in the ow eld to cavitate. Since a distribu-
tion of nuclei exists in the ow, and the pressure eld producing the tension can
often have important contributions from ow unsteadiness, inception is usually a
stochastic process. Therefore, the average ow conditions under which inception is
determined is, many cases, subjective. Moreover, the extent of cavitation chosen as
necessary to call inception can vary widely. In some cases, such as the characteriza-
tion of naval systems, only a minimal amount of cavitation is required to call incep-
tion, and the cavitation may not be easily visible to the naked eye. Conversely, limited
cavitation may not be of practical interest to the operators of industrial hydraulic sys-
tems, and inception would be called only when the amount of cavitation begins to
alter the performance of the device. And nally, proper use of inception observations
in the scaling of model hydraulic systems to full scale is also of vital importance.
Acosta and Parkin (1975) and Rood (1991) review the basic elements of cavitation
inception for a variety of cavitation forms.
44 S.L. Ceccio and S.A. Mkiharju
Since the presence of incipient cavitation will many times be accompanied by emis-
sion of sound from the cavitating nuclei, the detection of inception is often accom-
plished through acoustic means. Hydrophones can be placed directly within the ow
eld (Ran and Katz 1994), within the cavitating test article itself (Ceccio and Bren-
nen 1991; Kuhn de Chizelle et al. 1995) or in acoustically coupled chambers that
are separated from the ow by an acoustic window (Choi and Ceccio 2007). Since
the acoustic impedance of acrylic (3.1 106 Pa s / m3 ) is only about twice that of
Fig. 4 Schematic diagram of the quiescent laser-induced cavitation bubble experiment (a) and a
bubblevortex interaction experiment (b). A single laser pulse is used to create a cavitation bubble
in the bulk of the uid from Oweis et al. (2004). The acoustic emission of the bubble is captured
with a hydrophone within the ow (a) and in an external chamber through an acoustic window (b)
Experimental Methods for the Study of Hydrodynamic Cavitation 45
Fig. 5 Images and sound traces of a growing and collapsing vortex cavitation bubble in the sec-
ondary vortex producing an a acoustic pop and b a chirp. The broadband acoustic pulse was
abrupt lasting approximately 1 ms (Chang and Ceccio 2011)
46 S.L. Ceccio and S.A. Mkiharju
and the sounds recorded from individual bubbles could be a pulse (pop) or a
periodic tonal burst (chirp). The hydrophone array can be seen in the image, and,
in this case, an array of hydrophones was used to localize the sound source (Chang
and Dowling 2009).
As the cavitation develops, the amount of sound emitted and the cavitations
eect on the overall ow begins to increase. Hence, measurement of vibration and
changes to the system performance (e.g., lift coecient, ow coecient, eciency)
can be used to call inception. See, for example, Arndt (1981), McNulty and Pearsall
(1982), Shen and Dimotakis (1989), and Koivula (2000). Indirect methods of incep-
tion detection are often calibrated against visual observations when optical access
to the incepting ow is available. Escaler et al. (2006) report on a comprehensive
study that illustrates how cavitation inception can be detected using measurements
of structural vibrations, acoustic emissions, and hydrodynamic pressures measured
in turbomachinery where optical access may be limited or unavailable.
Direct visual observations of incepting nuclei are often used to call inception. Tradi-
tionally, the ow is illuminated with stroboscopic lighting, and a human observer is
tasked with determining which conditions have produced detectable and sustained
cavitation. The availability of high-speed video systems have enabled the digital
recording and analysis of the incepting ow, making inception calls with the naked
human eye less common. However, at the rst moments of inception, the cavitation
bubbles may be quite small and dicult to locate; and they may occur infrequently.
Such limited event rate cavitation inception is dicult to discern with visual detec-
tion alone, and the camera systems can be synchronized with acoustic detection sys-
tems (see, for example, Gopalan et al. (1999) and Chang and Ceccio (2011)). If the
location of inception is known a priori, then focused light scattering can be used to
detect the onset and rate of bubble formation. Keller (1972) developed such a system
by directing a focused light source into the inception region of a headform and then
into a photodetector. The presence of the bubble in the measurement volume would
block the light to produce a signal.
Many optically based methods that have been developed for uid measurements can
also be eectively employed to study cavitating ows. General reviews of optical
methods are provided by Goldstein (1996) and Tropea et al. (2007), and these meth-
ods and their applications are wide raging and varied. With this in mind, this section
will concentrate on the use of optical methods in cavitating ows.
Experimental Methods for the Study of Hydrodynamic Cavitation 47
The need to study the dynamics of cavitation has stimulated the development of high-
speed imaging. From the early work of Benjamin and Ellis (1966), Kling and Ham-
mitt (1972), and Lauterborn and Bolle (1975), high-speed photography has played
a key role in understanding bubbly dynamics and cavitation. A recent review by
Thoroddsen et al. (2008) provides a good summary of the history and recent methods.
As the resolution and frame rate of high-speed digital imaging systems has improved,
the ability for detailed examination of cavitating ows has signicantly improved.
Frame rates of order 1 Khz with spatial resolution of 103 103 pixels are now com-
monly available, and cameras with much higher imaging speeds are commercially
available. Ultrahigh-speed imaging systems have been created as well, with frame
rates in excess of 1 million per second. Such systems can resolve the ne details
of cavitation bubble dynamics, as shown by Lauterborn and Hentschel (1985), Ohl
et al. (1995), and Obreschkow et al. (2006).
The study of high-speed bubble dynamics often requires the controlled creation
of single or multiple bubbles. In many cases cited above, a laser pulse is focused
to create the bubble in the imaging region of the camera. The bubble creation can
then be synchronized with the imaging system. This technique can be used to con-
trollably place nuclei into the freestream ow as well, as shown by Choi and Ceccio
(2007). Figure 6 presents a time series from high-speed imaging of a laser-induced
bubble from Ohl et al. (1998), including a schematic of the setup and an example
of an aspherically collapsing bubble with detected luminescence. And, Fig. 7 shows
images of a vortex cavitation bubble formed after a laser-induced nucleus was cre-
ated within the upstream core of the vortex.
Fig. 6 High-speed imaging of a laser-induced bubble from Ohl et al. (1998). A schematic of the
setup used to create the laser-induced bubbles is shown in along with images of an aspherically
collapsing bubble with detected luminescence
LDV can be a useful method to examine cavitating ows. Kubota et al. (1989)
used conditional sampling of the LDV for velocity measurements in a ow around
a shedding partial cavity, as shown in Fig. 8. More recently, Roth et al. (2002) used
similar methods to examine the cavitating ow in a fuel injector. In these cases,
LDA was used to measure the velocity of the liquid ow around and outside the
Experimental Methods for the Study of Hydrodynamic Cavitation 49
Fig. 7 Images of a vortex cavitation bubble created by a laser-induced nucleus; the images were
used to compute the length and average radius of the bubble as a function of position within the
Venturi. Also shown is the corresponding acoustic signal detected from a hydrophone Choi and
Ceccio (2007)
Fig. 8 Local measurements of ow velocity acquired with Laser Doppler Velocimetry were con-
ditionally sampled and correlated to the measured surface pressure during the shedding of cloud
cavitation on a hydrofoil (Kubota et al. 1989)
Fig. 9 The image processing steps employed by Foeth et al. (2006) to determine the ow eld
around a developed cavitation forming on a twisted hydrofoil
52 S.L. Ceccio and S.A. Mkiharju
Fig. 10 Examination of the inuence of cavitation on a shear layer using PIV (Iyer and Ceccio
2002). The setup and an example image of the cavitating shear layer are shown in (a) and (b),
and the mean ow and vorticity proles are shown in (c) for increasing levels of cavitation (void
fraction) in the shear layer
Experimental Methods for the Study of Hydrodynamic Cavitation 53
As the void fraction of the cavitating ow begins to exceed a few percent, the bubbly
ow becomes opaque, and optical methods such as LDV and PIV begin to fail due
to multiple scattering of the incident and reected light. For very high void fraction
ows, alternative techniques must be used, as described below.
Cavitating ows with high void fraction often occur near solid boundaries, and it is
therefore possible to place instruments close to or ush against the cavitating sur-
face. The most common ush-mounted instruments measure the static and dynamic
pressure. These measurements can be combined with local or average acceleration
and forces on the test article to help elucidate the underlying ow. For example,
Kjeldsen et al. (2000) measured the static and dynamic pressures on the surface of a
cavitating hydrofoil, along with the time-varying lift force. Callenaere et al. (2001)
employed dynamic and ultrasonic transducers to measure the reentrant ow beneath
a partial cavity, and Le et al. (1993) employed arrays of dynamic pressure trans-
ducers to examine the unsteady pressures developed by partial cavitation. Escaler
et al. (2006) illustrated how a range of nonoptical methods can be used to detect
and quantify cavitation in turbomachinery, including external vibrations and noise
signatures.
Since the gas and liquid phases of the cavitating ow have quite dierent electrical
properties, measurements of the local or average mixture impedance can be used to
infer the void fraction of the ow, and a review of these general methods is pro-
vided by Ceccio and George (1996). Measurement of the mixture impedance can
be accomplished with ush-mounted electrodes or intrusive probes, albeit the lat-
ter may excessively perturb the cavitating ow. The electrodes can be made large
enough to measure the bulk-averages void fraction, or small enough to measure the
local void fraction or the passage of individual gas pockets. A high-frequency alter-
nating current can be applied as the probing signal, making the temporal response
of the transducers very rapid. However, as the path lines of the applied current are
strongly aected by the presence of the gas phase, it is not always possible to x the
measurement volume of the probe. And, the relationship between the measure bulk
impedance and the void fraction may not be straightforward. For bubbly ows, a
mixture relationship can be developed that can relate the bulk impedance to the void
fraction and electrical properties of the liquid and gas components (Hewitt 1978;
54 S.L. Ceccio and S.A. Mkiharju
Fig. 11 The near-surface gas-phase velocity beneath a partial cavity measured with ush-mounted
electrical impedance probes (George et al 2000a). The electrode locations on the hydrofoil (a), sam-
ple voltage signal transduced from the probes (b), and the gas-phase velocity determined through
cross-correlation of the signals (c)
George et al. 2000b). But such relationships will generally fail for highly stratied
ows.
Flush-mounted impedance probes have been used successfully to quantify cavi-
tating ows. Examples include the work of Ceccio and Brennen (1991) and Kuhn de
Chizelle et al. (1995) for the study of traveling bubble cavitation; and Ceccio and
Brennen (1992), Pham et al. (1999), and George et al (2000a) for the study of partial
cavitation. Cross-correlation of signals from multiple electrodes can be used to deter-
mine the gas-phase interface velocity, as reported by George et al (2000a) (Fig. 11).
Intrusive conductivity probes with one or more electrode pairs have also been devel-
oped for gasliquid ows, as discussed by Wu and Ishii (1999), Lucas and Mishra
(2005), and Elbing et al. (2013).
Experimental Methods for the Study of Hydrodynamic Cavitation 55
Fiber optic probes can be used to detect the presence of bubbles and gasliquid inter-
faces via the dierence in the index of refraction between the liquid and gas. The end
of the ber is placed in the ow, and light is directed toward the sharpened tip. If
the tip is immersed in the pure liquid, the majority of the light will be transmitted
into the uid. But, if gas is present at the tip, the light will internally reect within
the ber and return to its source to be detected. Fiber optic probes have been suc-
cessfully employed to measure bubble size populations, average void fraction, phase
speed, and interfacial area density, and a review is provided by Boyer et al. (2002)
and Chang et al. (2003). The performance of optical and conductivity probes for
measurement of bubbly ow was compared by Le Corre et al. (2003). The use of
intrusive optical probes in cavitating ow has been limited due to the probes delicate
construction and their ability to perturb the ow, but such probes were successfully
used to study the dynamics of partial cavitation by Stutz and Reboud (2000) and
Stutz (2003). Figure 12 presents the probe employed by Stutz and Reboud (2000) to
measure the bubbly ow within a partial cavity, the time traces from two probes that
can be used to measure the local volume fraction and phase speed, and an exam-
ple data set showing the average volume fraction within the cavity. As in the case
of electrical impedance probes, care must be taken to carefully determine how the
signal transduced from the probe relates to the ow quantity of interest (e.g., bubble
size and velocity) and what the inuence volume of the probe may be. Single point
probes have shown the best results when they are oriented in a ow with a strong
rectilinear velocity and when the probe tip is small compared to the bubbles to be
measured (Cartellier 1992; Mkiharju et al. 2013).
Images of a high void fraction bubbly cavitating ow were acquired by Coutier-
Delgosha et al. (2006) by traversing an endoscope within a partial cavity ow. They
were able to demonstrate that the bubbly ow within the cavity often consists of
highly distorted gas bubbles, as shown in Fig. 13.
The use of X-ray and gamma-ray densitometry and tomography for the study of
multiphase ows has been well established, and reviews are provided by Kumar et al.
(1997), George et al. (2000b), and Heindel (2011). These methods have also been
applied for the study of high void fraction cavitating ows.
The underlying principal of these systems relies the material and density depen-
dent attenuation coecient of high-energy photons as they pass through the multi-
phase mixture. When a beam of high-energy X-ray photons, for example, encounters
the mixture, a fraction of the photons passes through without scattering or absorp-
tion, and this fraction depends on the mixtures attenuation coecient, , density, ,
and thickness, x. For a beam encountering a domain with N distinct materials, the
56 S.L. Ceccio and S.A. Mkiharju
Fig. 12 The ber optic probe used by Stutz and Reboud (2000) to measure ow within a partial
cavity (a), example the time traces from two probes that can be used to measure volume fraction
and phase speed (b), and a plot showing the average volume fraction within the cavity (c)
Experimental Methods for the Study of Hydrodynamic Cavitation 57
Fig. 13 Images of the bubbly ow within a partial cavity obtained by traversing an endoscope into
the bubbly mixture (Coutier-Delgosha et al. 2006)
The transmitted beam intensities IL , IG , and IM are values obtained for the pure
liquid, pure gas, and mixture, respectively. For densitometry, average mixture void
fraction is determined along a known linear beam path, while tomography involves
the reconstruction of the two- or three-dimensional spatial distribution of attenuation
through many measurements of linear path-averaged attenuation.
Stutz and Legoupil (2003) used X-ray densitometry to nonintrusively measure
void fraction in a partial cavity. The setup consisted of a single row of 24 detectors
that could acquire void fraction proles along a line at a rate of 1000 samples per
second. The measurements were compared with optical probes, and it was found
that the maximum void fraction for the case of periodic shedding was about 25%.
58 S.L. Ceccio and S.A. Mkiharju
Fig. 14 A time series of X-ray densitometry-based images of a shedding partial cavity illustrating
the presence of a condensation shock (Ganesh et al. 2016)
Experimental Methods for the Study of Hydrodynamic Cavitation 59
Coutier-Delgosha et al. (2007) used the same diagnostic setup to measure void frac-
tion proles on a plano-convex foil. They reported a maximum averaged void frac-
tion values of close to 60% with instantaneous values exceeding 85%. These systems
provided the time- and phase-averaged void fraction averaged across the span of the
cavity. Mkiharju et al. (2013) recently developed an cinemagraphic X-ray densito-
metry system that measures two-dimensional void fraction ow elds of gasliquid
ows, and this system was used to examine the dynamic void fraction within shed-
ding partial cavities with frame rates up to 1000 per second Ganesh et al. (2016).
Figure 14 shows X-ray densitometry images of the partial cavity forming at the apex
of a wedge revealing the presence of a condensation shock. While the spatial and tem-
poral resolution of these radiation-based techniques is presently nonideal for many
cavitating ows, rapid technological advances (e.g., in X-ray detectors) will make
these nonintrusive techniques increasingly useful.
Acknowledgements The rst author is grateful to Prof. Salvetti and Prof. dAgostino for their
invitation to the Udine summer course on cavitating ows. The authors wish to acknowledge the
support of the U.S. Oce of Naval Research and the U.S. Naval Sea Systems Command for their
ongoing support of their research.
References
Acosta, A. J., & Parkin, B. R. (1975). Cavitation inception-a selective review. Journal of Ship
Research, 19(4), 193205.
Adrian, R. J., & Westerweel, J. (2011). Particle image velocimetry. Cambridge University Press.
Arndt, R. E. (1981). Cavitation in uid machinery and hydraulic structures. Annual Review of Fluid
Mechanics, 13(1), 273326.
Arndt, R. E. (2002). Cavitation in vortical ows. Annual Review of Fluid Mechanics, 34(1), 143
175.
Atlar, M. (2002). Final report of the specialist committee on water quality and cavitation. In Pro-
ceedings of the 23rd ITTC.
Avellan, F., Henry, P., & Ryhming, I. L. (1987). A new high speed cavitation tunnel for cavitation
studies in hydraulic machinery. Proceedings of international cavitation research facilities and
techniques (Vol. 57, pp. 4960). Boston: ASME Winter Annual Meeting.
Balachandar, S., & Eaton, J. K. (2010). Turbulent dispersed multiphase ow. Annual Review of
Fluid Mechanics, 42, 111133.
Benjamin, T. B., & Ellis, A. T. (1966). The collapse of cavitation bubbles and the pressures thereby
produced against solid boundaries. Philosophical Transactions of the Royal Society of London
A: Mathematical, Physical and Engineering Sciences, 260(1110), 221240.
Billet, M. L. (1985). Cavitation nuclei measurementsa review. Forum: Proceedings of ASME
Cavitation and Multiphase Flow.
Boyer, C., Duquenne, A. M., & Wild, G. (2002). Measuring techniques in gas-liquid and gas-liquid-
solid reactors. Chemical Engineering Science, 57(16), 31853215.
Brandner, P. A., Lecore, Y., & Walker, G. J. (2007). Design considerations in the development
of a modern cavitation tunnel. Proceedings of the 16th Australasian uid mechanics conference
(pp. 630637).
Brennen, C. E. (1995). Cavitation and bubble dynamics. Cambridge University Press.
Callenaere, M., Franc, J. P., Michel, J., & Riondet, M. (2001). The cavitation instability induced by
the development of a re-entrant jet. Journal of Fluid Mechanics, 444, 223256.
60 S.L. Ceccio and S.A. Mkiharju
Cartellier, A. (1992). Simultaneous void fraction measurement, bubble velocity, and size estimate
using a single optical probe in gas-liquid two-phase ows. Review of Scientic Instruments,
63(11), 54425453.
Ceccio, S. L. (2010). Friction drag reduction of external ows with bubble and gas injection. Annual
Review of Fluid Mechanics, 42, 183203.
Ceccio, S. L., & Brennen, C. E. (1991). Observations of the dynamics and acoustics of travelling
bubble cavitation. Journal of Fluid Mechanics, 233, 633660.
Ceccio, S. L., & Brennen, C. E. (1992). Dynamics of attached cavities on bodies of revolution.
Journal of Fluids Engineering, 114(1), 9399.
Ceccio, S. L., & George, D. L. (1996). A review of electrical impedance techniques for the mea-
surement of multiphase ows. Journal of Fluids Engineering, 118(2), 391399.
Ceccio, S. L., Gowing, S., & Gindroz, B. (1995). A comparison of csm bubble detection methods.
In Proceedings of A.S.M.E. symposium on cavitation and gas-liquid ows in uid machinery.
FED (Vol. 226, pp. 4350).
Chahine, G. L., & Kalumuck, K. M. (2003). Development of a near real-time instrument for nuclei
measurement: The abs acoustic bubble spectrometer. Proceedings of A.S.M.E./J.S.M.E. 4th joint
uids summer engineering conference (pp. 183191).
Chahine, G. L., & Shen, Y. T. (1986). Bubble dynamics and cavitation inception in cavitation sus-
ceptibility meters. Journal of Fluids Engineering, 108(4), 444452.
Chambers, S. D., Bartlett, R. H., & Ceccio, S. L. (2000). Hemolytic potential of hydrodynamic
cavitation. Journal of Biomechanical Engineering, 122(4), 321326.
Chang, K. A., Lim, H. J., & Su, C. B. (2003). Fiber optic reectometer for velocity and fraction
ratio measurements in multiphase ows. Review of Scientic Instruments, 74(7), 35593565.
Chang, N., Ganesh, H., Yakushiji, R., & Ceccio, S. L. (2011). Tip vortex cavitation suppression by
active mass injection. Journal of Fluids Engineering, 133(11301), 111.
Chang, N. A., & Ceccio, S. L. (2011). The acoustic emissions of cavitation bubbles in stretched
vortices. The Journal of the Acoustical Society of America, 130(5), 32093219.
Chang, N. A., & Dowling, D. R. (2009). Ray-based acoustic localization of cavitation in a highly
reverberant environment. Journal of the Acoustical Society of America, 125(5), 30883100.
Choi, J., & Ceccio, S. L. (2007). Dynamics and noise emission of vortex cavitation bubbles. Journal
of Fluid Mechanics, 575, 126.
Coutier-Delgosha, O., Devillers, J. F., Pichon, T., Vabre, A., Woo, R., & Legoupil, S. (2006). Inter-
nal structure and dynamics of sheet cavitation. Physics of Fluids (1994-present), 18(1), 017103.
Coutier-Delgosha, O., Stutz, B., Vabre, A., & Legoupil, S. (2007). Analysis of cavitating ow struc-
ture by experimental and numerical investigations. Journal of Fluid Mechanics, 578, 171222.
dAgostino, L., & Acosta, A. J. (1991). A cavitation susceptibility meter with optical cavitation
monitoringpart one: Design concepts. Journal of Fluids Engineering, 113(2), 261269.
Dular, M., Bachert, R., Stoel, B., & irok, B. (2005). Experimental evaluation of numerical sim-
ulation of cavitating ow around hydrofoil. European Journal of Mechanics-B/Fluids, 24(4),
522538.
Duraiswami, R., Prabhukumar, S., & Chahine, G. L. (1998). Bubble counting using an inverse
acoustic scattering method. Journal of the Acoustical Society of America, 104(5), 26992717.
Durst, F. (1982). Reviewcombined measurements of particle velocities, size distributions, and
concentrations. Journal of Fluids Engineering, 104(3), 284296.
Elbing, B. R., Mkiharju, S. A., Wiggins, A., Perlin, M., Dowling, D. R., & Ceccio, S. L. (2013).
On the scaling of air layer drag reduction. Journal of Fluid Mechanics, 717, 484513.
Escaler, X., Egusquiza, E., Farhat, M., Avellan, F., & Coussirat, M. (2006). Detection of cavitation
in hydraulic turbines. Mechanical Systems and Signal Processing, 20(4), 9831007.
Etter, R. J., Cutbirth, J. M., Ceccio, S. L., Dowling, D. R., & Perlin, M. (2005). High reynolds
number experimentation in the U.S. Navys William B. Morgan large cavitation channel. Mea-
surement Science and Technology, 16(9), 17011709.
Foeth, E. J., Van Doorne, C. W. H., Van Terwisga, T., & Wieneke, B. (2006). Time resolved PIV
and ow visualization of 3D sheet cavitation. Experiments in Fluids, 40(4), 503513.
Experimental Methods for the Study of Hydrodynamic Cavitation 61
Ganesh, H., Mkiharju, S. A., & Ceccio, S. L. (2016). Bubbly shock propagation as a mechanism
for sheet-to-cloud transition of partial cavities. Journal of Fluid Mechanics, 802, 3778.
George, D. L., Iyer, C. O., & Ceccio, S. L. (2000a). Measurement of the bubbly ow beneath partial
attached cavities using electrical impedance probes. Journal of Fluids Engineering, 122(1), 151
155.
George, D. L., Torczynski, J. R., Shollenberger, K. A., OHern, T. J., & Ceccio, S. L. (2000b). Val-
idation of electrical-impedance tomography for measurements of material distribution in two-
phase ows. International Journal of Multiphase Flow, 26(4), 549581.
Gindroz, B. (1998). Cavitation nuclei and cavitation inception of marine propellers: State of the art
at the dawn of the 21st century. J.S.M.E. International Journal Series B, 41(2), 464471.
Gindroz, B., & Billet, M. L. (1998). Inuence of the nuclei on the cavitation inception for dierent
types of cavitation on ship propellers. Journal of Fluids Engineering, 120(1), 171178.
Goldstein, R. (1996). Fluid mechanics measurements. CRC Press.
Gopalan, S., & Katz, J. (2000). Flow structure and modeling issues in the closure region of attached
cavitation. Physics of Fluids, 12(4), 895911.
Gopalan, S., Katz, J., & Knio, O. (1999). The ow structure in the near eld of jets and its eect
on cavitation inception. Journal of Fluid Mechanics, 398, 143.
Gowing, S., Brianon-Marjollet, L., Frechou, D., & Goderoy, V. (1995). Dissolved gas and nuclei
eects on tip vortex cavitation inception and cavitating core size. In Proceedings of 5th interna-
tional symposium on cavitation (pp. 173180).
Hassan, Y. A., Blanchat, T. K., Seeley, C. H., & Canaan, R. E. (1992). Simultaneous velocity mea-
surements of both components of a two-phase ow using particle image velocimetry. Interna-
tional Journal of Multiphase Flow, 18(3), 371395.
Heindel, T. J. (2011). A review of X-ray ow visualization with applications to multiphase ows.
Journal of Fluids Engineering, 133(074001), 116.
Hewitt, G. F. (1978). Measurement of two phase ow parameters. Academic Press.
Iyer, C. O., & Ceccio, S. L. (2002). The inuence of developed cavitation on the ow of a turbulent
shear layer. Physics of Fluids, 14(10), 34143431.
Katz, J., & Acosta, A. (1981). Observations of nuclei in cavitating ows. Proceedings of I.U.T.A.M.
symposium on mechanics and physics of bubbles in liquids (pp. 123132).
Katz, J., & Sheng, J. (2010). Applications of holography in uid mechanics and particle dynamics.
Annual Review of Fluid Mechanics, 42, 531555.
Katz, J., Gowing, S., OHern, T., & Acosta, A. (1984). A comparative study between holographic
and light-scattering techniques of microbubble detection. Proceedings of I.U.T.A.M. symposium
on measuring techniques in gas-liquid two-phase ows (pp. 4166).
Kawanami, Y., Kato, H., Yamaguchi, H., Maeda, M., & Nakasumi, S. (2002). Inner structure of
cloud cavity on a foil section. J.S.M.E. International Journal Series B Fluids and Thermal Engi-
neering, 45(3), 655661.
Keller, A. P. (1972). The inuence of the cavitation nucleus spectrum on cavitation inception, inves-
tigated with a scattered light counting method. Journal of Fluids Engineering, 94(4), 917925.
Keller, A. P. (1987). A vortex-nozzle cavitation susceptibility meter in routine application in cavita-
tion inception measurements. In Proceedings of euromech colloquium 222-unsteady cavitation
and its eects.
Keller, A. P. (2001). Cavitation scale eectsempirically found relations and the correlation of
cavitation number and hydrodynamic coecients. In Proceedings of fourth international sym-
posium on cavitation-CAV2001.
Kjeldsen, M., Arndt, R. E. A., & Eertz, M. (2000). Spectral characteristics of sheet/cloud cavita-
tion. Journal of Fluids Engineering, 122(3), 481487.
Kling, C. L., & Hammitt, F. G. (1972). A photographic study of spark-induced cavitation bubble
collapse. Journal of Fluids Engineering, 94(4), 825832.
Koivula, T. (2000). On cavitation in uid power. In Proceedings of 1st FPNI-PhD symposium,
Hamburg (pp. 371382).
62 S.L. Ceccio and S.A. Mkiharju
Kubota, A., Kato, H., Yamaguchi, H., & Maeda, M. (1989). Unsteady structure measurement of
cloud cavitation on a foil section using conditional sampling technique. Journal of Fluids Engi-
neering, 111(2), 204210.
Kuhn de Chizelle, Y., Ceccio, S. L., & Brennen, C. E. (1995). Observations and scaling of travelling
bubble cavitation. Journal of Fluid Mechanics, 293, 99126.
Kuiper, G. (1985). Reections on cavitation inception. In Proceedings of A.S.M.E. cavitaiton and
multiphase ow forum, FED-23.
Kumar, S. B., Dudukovic, M. P., Chaouki, J., Larachi, F., & Dudukovic, M. P. (1997). Computer
assisted gamma and x-ray tomography: Applications to multiphase ow systems. Non-invasive
monitoring of multiphase ows (pp. 47103).
Laberteaux, K. R., & Ceccio, S. L. (2001a). Partial cavity ows. part 1. cavities forming on models
without spanwise variation. Journal of Fluid Mechanics, 431, 141.
Laberteaux, K. R., & Ceccio, S. L. (2001b). Partial cavity ows. part 2. cavities forming on test
objects with spanwise variation. Journal of Fluid Mechanics, 431, 4363.
Lauterborn, W., & Bolle, H. (1975). Experimental investigations of cavitation-bubble collapse in
the neighbourhood of a solid boundary. Journal of Fluid Mechanics, 72(2), 391399.
Lauterborn, W., & Hentschel, W. (1985). Cavitation bubble dynamics studied by high speed pho-
tography and holography: Part one. Ultrasonics, 23(6), 260268.
Lavigne, S. (1991). Le venturi analyseur de germes. Journees DRET Cavitation.
Le, Q., Franc, J. P., & Michel, J. M. (1993). Partial cavities: Pressure pulse distribution around
cavity closure. Journal of Fluids Engineering, 115(2), 249254.
Le Corre, J.-M., Hervieu, E., Ishii, M., & Delhaye, J.-M. (2003). Benchmarking and improvements
of measurement techniques for local-time-averaged two-phase ow parameters. Experiments in
uids, 35(5), 448458.
Lecore, Y., & Bonnin, J. (1979). Cavitation tests and nucleation control. In Proceedings of
A.S.M.E. international symposium on cavitation inception (pp. 141145).
Lecore, Y., Chantrel, P., & Teiller, J. (1987). Le grand tunnel hydrodynamique (GTH): Frances
new large cavitation tunnel for naval hydrodynamics research. In Proceedings of A.S.M.E. inter-
national symposium on cavitation research facilities and techniques (pp. 1318).
Lee, I.-H., Mkiharju, S., Ganesh, H., & Ceccio, S. L. (2016). Scaling of gas diusion into limited
partial cavities. Journal of Fluids Engineering.
Leger, A. T., & Ceccio, S. L. (1998). Examination of the ow near the leading edge of attached
cavitation. part 1. detachment of two-dimensional and axisymmetric cavities. Journal of Fluid
Mechanics, 376, 6190.
Leger, A. T., Bernal, L. P., & Ceccio, S. L. (1998). Examination of the ow near the leading edge of
attached cavitation. part 2. incipient breakdown of two-dimensional and axisymmetric cavities.
Journal of Fluid Mechanics, 376, 91113.
Li, C. Y., & Ceccio, S. L. (1996). Interaction of single travelling bubbles with the boundary layer
and attached cavitation. Journal of Fluid Mechanics, 322, 329353.
Lindken, R., & Merzkirch, W. (2002). A novel piv technique for measurements in multiphase ows
and its application to two-phase bubbly ows. Experiments in Fluids, 33(6), 814825.
Lucas, G. P., & Mishra, R. (2005). Measurement of bubble velocity components in a swirling gas-
liquid pipe ow using a local four-sensor conductance probe. Measurement Science and Tech-
nology, 16(3), 749758.
Mkiharju, S. A., Gabillet, C., Paik, B.-G., Chang, N. A., Perlin, M., & Ceccio, S. L. (2013). Time-
resolved two-dimensional x-ray densitometry of a two-phase ow downstream of a ventilated
cavity. Experiments in Fluids, 54(7), 121.
McNulty, P. J., & Pearsall, I. S. (1982). Cavitation inception in pumps. Journal of Fluids Engineer-
ing, 104(1), 99104.
Mrch, K. A. (2007). Reections on cavitation nuclei in water. Physics of Fluids, 19(072104), 17.
Obreschkow, D., Kobel, P., Dorsaz, N., De Bosset, A., Nicollier, C., & Farhat, M. (2006). Cavitation
bubble dynamics inside liquid drops in microgravity. Physical Review Letters, 97(094502), 14.
Experimental Methods for the Study of Hydrodynamic Cavitation 63
OHern, T. J., Katz, J., & Acosta, A. J. (1985). Holographic measurements of cavitation nuclei in
the sea. In Proceedings of A: S.M.E. cavitation and multiphase ow forum.
Ohl, C. D., Philipp, A., & Lauterborn, W. (1995). Cavitation bubble collapse studied at 20 million
frames per second. Annalen der Physik, 507(1), 2634.
Ohl, C. D., Lindau, O., & Lauterborn, W. (1998). Luminescence from spherically and aspherically
collapsing laser induced bubbles. Physical Review Letters, 80(2), 393396.
Oldenziel, D. M. (1982). A new instrument in cavitation research: the cavitation susceptibility
meter. Journal of Fluids Engineering, 104(2), 136141.
Oweis, G. F., Choi, J., & Ceccio, S. L. (2004). Dynamics and noise emission of laser induced
cavitation bubbles in a vortical ow eld. Journal of the Acoustical Society of America, 115(3),
10491058.
Pham, T. M., Larrarte, F., & Fruman, D. H. (1999). Investigation of unsteady sheet cavitation and
cloud cavitation mechanisms. Journal of Fluids Engineering, 121(2), 289296.
Rae, W. H., & Pope, A. (1984). Low-speed wind tunnel testing, 2nd edition. Wiley.
Ran, B., & Katz, J. (1994). Pressure uctuations and their eect on cavitation inception within
water jets. Journal of Fluid Mechanics, 262, 223263.
Rood, E. P. (1991). Reviewmechanisms of cavitation inception. Journal of Fluids Engineering,
113(2), 163175.
Rooze, J., Rebrov, E. V., Schouten, J. C., & Keurentjes, J. T. (2013). Dissolved gas and ultrasonic
cavitation-a review. Ultrasonics Sonochemistry, 20(1), 111.
Roth, H., Gavaises, M., & Arcoumanis, C. (2002). Cavitation initiation, its development and link
with ow turbulence in diesel injector nozzles. S.A.E. Technical Paper, (2002-01-0214).
Sato, R., Mori, T., Yakushiji, R., Naganuma, K., Nishimura, M., Nakagawa, K., & Sasajima, T.
(2003). Conceptual design of the ow noise simulator. In Proceedings of joint A.S.M.E./J.S.M.E.
4th joint uids summer engineering conference (pp. 129133).
Schiebe, F. R. (1972). Measurement of the cavitation susceptibility of water using standard bodies.
St. Anthony Falls Laboratory Project Report 118, University of Minnesota.
Shen, Y. T., & Dimotakis, P. E. (1989). Viscous and nuclei eects on hydrodynamic loadings and
cavitation of a naca 66 (mod) foil section. Journal of Fluids Engineering, 111(3), 306316.
Simoni, R. D., Hill, R. L., & Vaughan, M. (2002). The measurement of blood gases and the
manometric techniques developed by donald dexter van slyke. Journal of Biological Chemistry,
277(27), e16.
Straka, W. A., Meyer, R. S., Fontaine, A. A., & Welz, J. P. (2010). Cavitation inception in quiescent
and co-ow nozzle jets. Journal of Hydrodynamics, Series B, 22(5), 813819.
Stutz, B. (2003). Inuence of roughness on the two-phase ow structure of sheet cavitation. Journal
of Fluids Engineering, 125(4), 652659.
Stutz, B., & Legoupil, S. (2003). X-ray measurements within unsteady cavitation. Experiments in
Fluids, 35(2), 130138.
Stutz, B., & Reboud, J. L. (2000). Measurements within unsteady cavitation. Experiments in Fluids,
29(6), 522545.
Tanger, H., & Weitendorf, E. A. (1992). Applicability tests for the phase doppler anemometer for
cavitation nuclei measurements. Journal of Fluids Engineering, 114(3), 443449.
Thoroddsen, S. T., Etoh, T. G., & Takehara, K. (2008). High-speed imaging of drops and bubbles.
Annual Review of Fluid Mechanics, 40, 257285.
Tropea, C., Yarin, A. L., & Foss, J. F. (Eds.). (2007). Springer handbook of experimental uid
mechanics. Springer Science and Business Media.
Van der Kooij, J., & De Bruijn, A. (1984). Acoustic measurements in the NSMB depressurized
towing tank. International Shipbuilding Progress, 31(353), 1325.
Vogel, A., & Lauterborn, W. (1988). Time-resolved particle image velocimetry used in the investi-
gation of cavitation bubble dynamics. Applied Optics, 27(9), 18691876.
Wetzel, J. M., & Arndt, R. E. A. (1994a). Hydrodynamic design considerations for hydroacoustic
facilities: Part I- ow quality. Journal of Fluids Engineering, 116(2), 324331.
64 S.L. Ceccio and S.A. Mkiharju
Wetzel, J. M., & Arndt, R. E. A. (1994b). Hydrodynamic design considerations for hydroacoustic
facilities: Part II- pump design factors. Journal of Fluids Engineering, 116(2), 332337.
Wosnik, M., Schauer, T. J., & Arndt, R. E. (2003). Experimental study of a ventilated supercavitat-
ing vehicle. In Proceedings of fth international symposium on cavitation (pp. 14).
Wu, Q., & Ishii, M. (1999). Sensitivity study on double-sensor conductivity probe for the measure-
ment of interfacial area concentration in bubbly ow. International Journal of Multiphase Flow,
25(1), 155173.
Yu, P. W., & Ceccio, S. L. (1997). Diusion induced bubble populations downstream of a partial
cavity. Journal of Fluids Engineering, 119(4), 782787.
An Introduction to Flow-Induced
Instabilities in Rocket Engine Inducers
and Turbopumps
Abstract The article reviews the main forms of flow-induced instabilities detected
in the liquid propellant turbopumps systems of modern rocket engines, with special
reference to rotating stall, rotating cavitation, cavitation surge and higher order
surge modes, illustrating their characteristics, origin and damage potential.
1 Introduction
Propellant feed turbopumps are a crucial component of all liquid propellant rocket
engines because of the severe limitations associated with the design of high power
density, dynamically stable machines capable of meeting the extremely demanding
suction, pumping and reliability requirements of modern space transportation sys-
tems. Figure 1 illustrates the power/weight ratio of the main propellant feed tur-
bopumps used by US space rocket engines powering the Mercury, Gemini, Apollo,
and Space Shuttle programs in the two decades from the early 60s to the 80s. The
tenfold increase of the power density clearly indicates that todays rocket propellant
feed turbopumps belong to a very special family of machines with respect to earlier
more conventional land- or sea-based machines.
Such high power/weight ratios are obtained by running the impeller at the
maximum allowable rotational speed, in order to reduce the sizeand therefore the
weightof the machine. However, the resulting higher velocity of the flow also
reduces the pressure on the suction sides of the blades, creating the conditions for
the development of cavitation where the flow pressure drops below the vapor
Fig. 1 Power/weight
evolution of propellant feed
turbopumps in the US space
rocket engines (Rockwell
International, personal
communication)
pressure of the liquid. Indeed, operation under limited cavitation, and sometimes at
supercritical speeds, is usually tolerated in rocket propellant feed turbopumps in
order to maximize their performance, exposing the machine to the risk of devel-
oping dangerous cavitation-induced fluid dynamic and rotordynamic instabilities.
Cavitation usually represents the most stringent fluid mechanical constraint to
the increase of the power density of hydraulic machinery and in particular the major
source of degradation of the suction performance, efciency, useful life, and reli-
ability of turbopumps (Stripling and Acosta 1962; Stripling 1962). Its occurrence is,
in fact, connected to the insurgence of undesirable phenomena ranging from ef-
ciency deterioration to operational irregularities, mechanical erosion, noise, vibra-
tions, as well as fluid mechanical and/or dynamical instabilities of the machinery
exposed to the cavitating flow. This is particularly true in feed systems of cryogenic
liquid propellant rocket engines, where propellant storage under saturation condi-
tions greatly facilitates the occurrence of cavitation. More specically, in space
propulsion applications cavitation can provide the flow compliance and excitation
for triggering dangerous fluid mechanic and/or rotordynamic instabilities (Sack and
Nottage 1965; Natanzon et al. 1974; Brennen and Acosta 1973; Brennen and
Acosta 1976; Ng and Brennen 1978; Braisted 1979; dAgostino et al. 1998; Cer-
vone et al. 2005; Cervone et al. 2006; dAgostino and Salvetti 2007; dAgostino
2013; Pace et al. 2015) or even, through the coupling with thrust generation, of the
entire space vehicle (POGO auto-oscillations, Rubin 1966).
The most critical rotordynamic instability in high-performance turbopumps is
the development of self-sustained lateral motions (whirl) of the impeller under the
action of destabilizing forces of mechanical and fluid dynamic origin (Ehrich and
Childs 1984). The role of rotordynamic fluid forces in promoting rotordynamic
instabilities by signicantly modifying, in conjunction with cavitation, the dynamic
properties of the impeller, and therefore the critical speeds of the whole machine,
has long been recognized (Rosenmann 1965). However, because of their com-
plexity and the experimental difculties in their measurement, so far rotordynamic
fluid forces in cavitating inducers and turbopumps have received relatively little
An Introduction to Flow-Induced Instabilities in Rocket 67
(based on the reference pressure pref , the vapor pressure pV , and the tip-speed
dynamic pressure 12L 2 rT2 ) drops below the inception value, which essentially
depends on the geometry and operational conditions of the machine. Next, more
extensive cavitation develops, starting in the leakage flow at the blade tips and
progressively extending on the suction sides of the blades toward the hub. The
bubbly, cloudy, or attached nature of blade cavitation depends on the dominant
location and susceptibility of the nuclei responsible for the inception of cavitation,
as well as on the thermodynamic state of the liquid and the nature of the flow
(laminar/turbulent, separated/unseparated, etc.). Finally, especially at high blade
loads, cavitation can also appear in the separated, recirculating and slowly
co-rotating backflow that develops at the inlet of the inducer near the side wall of
the casing as a consequence of the combined effects of tip blade and leading edge
flow leakage. Figure 3 illustrates the typical progression of cavitation in turbopump
inducers as a function of the NPSH (Net Positive Suction Head) or, in nondi-
mensional form, of the inlet cavitation number.
68 L. dAgostino et al.
Fig. 3 Evolution of cavitation in the SSME liquid hydrogen turbopump inducer as a function of
the NPSH (Rockwell International, personal communication)
2 Rotating Stall
He measured stall propagation velocities between 0.45 and 0.6 times the impeller
rotational speed, and observed that is was somewhat affected by the occurrence of a
limited amount of cavitation.
A useful approximate criterion in compressors predicts the onset of rotating stall
when the maximum of the total head rise curve is approached as the flow coefcient
decreases. As a consequence, a number of authors postulated that also rotating stall
in inducers could be associated to areas of positive slope of the pumping perfor-
mance curve (see for example Kamijo et al. 1977). This is, however, no more than
an approximation (Greitzer 1981). More sophisticated stall criteria typically take
into consideration the evolution of the displacement thickness on the blade suction
side. In particular, following this approach, Leiblein (1965) has been able to
demonstrate the correlation of stall with a simple integral parameter of the flow, the
diffusion factor, predicting the occurrence of stall on typical blade proles when its
value exceeds about 0.6.
3 Rotating Cavitation
At low Euler numbers, cavitating inducers and pump impellers can exhibit a phe-
nomenon similar to rotating stall, known as rotating cavitation. Even the two
phenomena look similar, signicant differences exist. Rotating stall occurs at
locations of the head-flow characteristic where the slope of the curve is positive and
therefore unstable. Rotating cavitation, on the other hand, is normally observed in
the zone of negative slope at cavitation numbers between 2 and 3 times larger than
the breakdown value. Even if subsynchronous (or backward rotating) rotating
cavitation has sometimes been observed, usually it is supersynchronous (forward
rotating with respect to the impeller blades) with frequencies between 1.1 and 1.2
times the pumps rotational speed, as indicated in Fig. 5 (from Brennen 1994) with
reference to the inducer tested by Kamijo et al. (1977).
Fig. 6 Schematic representation of the asymmetric cavitation (left) and the alternate blade
cavitation (right)
Fig. 7 Left suction performance curve of a 3-bladed inducer at the design flow coefcient. Right
waterfall plot of the power spectra of mid-chord pressure fluctuations, for the same flow conditions
(Shimagaki et al. 2006)
Backflow vortex cavitation was rst detected and identied by Tsujimoto et al.
(1997) in inducers operating at flow coefcients lower than design conditions and
cavitation numbers either close to breakdown or signicantly higher (34 times the
breakdown value). As illustrated in Fig. 8, it is a rotating instability of the inducer
backflow with the formation of multiple cells of axial vorticity (5 cells in Tsujimoto
et al. 1997), which slowly rotate at an angular speed less than 0.2 times the inducer
rotating speed.
The occurrence of a backflow vortex instability with 8 cells rotating at about
32% of the impeller speed has been documented by Zoladz (2000) with reference to
the tests of the Fastrac turbopump. Imamura et al. (2003) showed that backflow
Fig. 8 Schematic
representation of the
backflow vortex cavitation
instability (Tsujimoto et al.
1997)
An Introduction to Flow-Induced Instabilities in Rocket 75
Fig. 9 Left Suction performance and rotor vibration as a function of the cavitation number, at two
different flow coefcients (95% and 98% of the operating conditions), for the liquid hydrogen
inducer of the Japanese LE-7 engine. Right schematic representation of the change of slope in the
performance curve (Tsujimoto and Semenov 2002)
76 L. dAgostino et al.
near breakdown. The rotating nature of the phenomenon was later conrmed by
flow visualization by means of a high-speed camera (Cervone et al. 2007).
In a series of experiments on several inducers intended for use in liquid
hydrogen Uchiumi and Kamijo (2008) showed that rotating stall cavitation
occurred when the flow incidence on the leading edges of the inducer was larger
than about 43% of the tip blade angle measured from the tangential direction.
5 Cavitation Surge
Fig. 10 Schematic of stable and unstable characteristic curves of a pumping system (Brennen
1994)
An Introduction to Flow-Induced Instabilities in Rocket 77
Indeed in cavitating pumps violent oscillations of the pressure and flow rate of the
entire pumping system have routinely been observed when the cavitation number is
decreased at values for which the head rise begins to be affected, as documented by
Rosenman (1965), Sack and Nottage (1965), Natanzon et al. 1974, Brennen and
Braisted (1980), among others. This surge phenomenon, named cavitation
auto-oscillation in the original classication by Brennen (1994), has been later
indicated as cavitation surge by other researchers. It can lead to very large
fluctuations across the pumping system and can also cause substantial axial forces
on the shaft, on the order of 20% of the steady axial thrust on the impeller. Brennen
and Braisted (1980), and later Yamamoto (1991), showed that the frequency of
cavitation surge oscillations is roughly proportional to the pumps rotating speed
and the square root of the cavitation number.
With reference to space propulsion applications, the onset of cavitation surge is
especially dangerous in liquid propellant rocket feed systems because the resulting
flow rate fluctuations, by modulating the engine thrust, can trigger self-sustained
POGO oscillations of the entire space vehicle, a global instability characterized by
extremely dangerous large-amplitude fluctuations of the thrust (NASA 1970).
A violent cavitation surge instability at very low frequency (<10 Hz) was
observed by Hashimoto et al. (1997a) on a 3-bladed model of the liquid oxygen
inducer of the Japanese LE-7 engine operating near breakdown conditions and at
flow coefcients close to the nominal design value. The authors called this insta-
bility low cycle oscillations and reported a signicant interaction with steady
uneven cavitation and backflow vortices. A similar surge mode, but at higher
frequency (about 18 Hz), was detected on a similar 3-bladed inducer by Tsujimoto
et al. (1997), who interpreted it as the result of a resonance between the vibration
mode of the system at 18 Hz and the oscillations of the blade cavities due to the
occurrence of rotating cavitation.
A surge mode instability at about 2 Hz was detected by Furukawa et al. (2001)
in water tests on a 2-bladed inducer with a relatively high tip blade angle (about 11
degrees) operating near breakdown conditions at flow coefcients signicantly
lower than the nominal design value (Fig. 11). The onset pressure of this instability
mode is reported to coincide with the value for which the cavities on the blades
reach the inlet of the blade channels.
The dependence of the frequency of cavitation surge on the Euler number is
probably the consequence of a strong connection between this form of instability
and rotating cavitation, as documented by Zoladz (2000) in connection with the
tests carried out on the LOX turbopump of the Fastrac engine. If is the pump
rotational speed, frc is the frequency of rotating cavitation instability and Z is the
number of blades of the inducer, the frequency f of the cavitation surge mode is
reported to be equal to:
f = Z frc
78 L. dAgostino et al.
Fig. 11 Left suction performance curve of a 2-bladed inducer at = 0.017, including the eld of
existence of the surge mode instability. Right waterfall plot of the power spectrum of outlet
pressure fluctuations on the same inducer, under the same test conditions (Furukawa et al. 2001)
Free oscillation modes with frequencies signicantly higher than the inducer
rotational speed are generically indicated as higher order instabilities and can
involve unstable flow motion in the streamwise (higher order surge modes) and
azimuthal (higher order rotating modes) directions. They have been detected and
investigated in detail only in recent years and are universally recognized as
extremely important in the design and characterization of space rocket inducers
because of their potential for coupling with the lower critical speeds of the tur-
bopump or with the flutter frequencies of the inducer blades, possibly leading to
catastrophic failures of the machine. As an example, the failure of the Japanese H-II
rocket in 1999 is thought to have been caused by the fatigue rupture of the inducer
blades under the resonant excitation of an unstable high-order cavitation instability
An Introduction to Flow-Induced Instabilities in Rocket 79
(NASDA 2000a, b). Furthermore, several recent flight tests on important compo-
nents of the Space Shuttle main engine have suffered from vibration problems at
frequencies compatible with the characteristic frequencies of high-order rotating
cavitation instabilities observed in the propellant feed turbopumps.
High-order streamwise and azimuthal instabilities were rst predicted on theo-
retical grounds by Tsujimoto (2001) as unstable modes resulting from the appli-
cation of his 2D linear cavity model originally developed for the investigation of
rotating cavitation. He anticipated the occurrence of these modes with
system-independent frequencies between 4.2 and 4.7 times the inducer rotating
speed and at Euler numbers equal or larger than about four times the breakdown
value, just above those characteristic of conventional rotating cavitation. The author
also suggested that the high-order azimuthal mode might be caused by the modu-
lation of the inlet backflow structure, also in consideration that it was predicted to
occur even without cavitation on the blades.
Experimental evidence of the occurrence of higher order instabilities was rst
provided by Fujii et al. (2002) in a test campaign on a 3-bladed inducer similar to
the LOX inducer of the LE-7 turbopump, whose main results are illustrated in
Fig. 12. The spectral analysis of the pressure fluctuations, recorded at the inlet and
outlet cross-sections of the inducer when operating at nearly nominal flow and
variable inlet pressure, indicated the presence, together with conventional rotating
cavitation denoted by 1 in the gure, of two additional higher order instabilities:
A higher order surge (denoted by 2 in the gure) with a frequency about 5 times
the rotational speed of the inducer at cavitation numbers higher than for con-
ventional rotating cavitation.
A higher order rotating cavitation (denoted by 3 in the gure) at frequencies and
cavitation numbers close to those of the higher order surge oscillations. It was
Fig. 12 Waterfall plots of the power spectra of inlet and outlet pressure fluctuations in a 3-bladed
inducer at flow coefcient = 0.080 ([28] Fujii et al. 2002)
80 L. dAgostino et al.
also found that the cavity size fluctuation for this instability mode is smaller than
for conventional rotating cavitation. Furthermore, while in conventional rotating
cavitation the outlet pressure fluctuations are signicantly smaller than those at
the inlet, for the higher order rotating cavitation mode the situation is reversed.
In the same paper the authors also indicate that the above instabilities could be
eliminated almost completely by means of the introduction of a suitable step in the
design of the inducer casing. Finally, the detection of a higher order surge insta-
bility with the same features observed by Fujii et al. (2002) was conrmed by
Tsujimoto and Semenov (2002) also in full-scale tests on the liquid hydrogen
turbopump of the LE-7 engine.
Figure 13 summarizes the results of similar tests conducted in the U.S. by
Subbaraman and Patton (2006) on a 4-bladed inducer for the characterization of
higher order instabilities at frequencies up to ten times the rotational speed of the
impeller.
The plot in the gure clearly shows the occurrence of higher order surge cavi-
tation (HOSC) and higher order rotating cavitation (HORC) at nearly nominal flow
conditions and cavitation numbers equal to or higher than about four times the
breakdown value. The HORC and HOSC frequencies only differ by about one tenth
of the inducer rotational speed and, not surprisingly, the HORC component displays
both forward and backward running modes. These higher order instabilities also
produced many nonlinear modulations and harmonics over the whole range of
experimental observation, including a component at very low frequency indicated
as Low Freq Cav-Induced in the above plot. The most intense harmonic
Fig. 13 2D Color plot of the power spectra of the pressure fluctuations versus the cavitation
number in a 4-bladed inducer, at a flow coefcient equal to 1.02 times the design value
(Subbaraman and Patton 2006)
An Introduction to Flow-Induced Instabilities in Rocket 81
In the past generation the technology and knowledge base of liquid propellant feed
turbopumps has been completely revolutionized. The development of the propul-
sion systems for the Space Shuttle in the 1970s and of the Japanese launchers in the
1980s and early 1990s have represented corner points in the quest for higher
performance and more detailed insight on the hydrodynamics of the flow eld
through the machine. As a consequence of these development efforts the design of
turbopumps for space propulsion applications has rapidly evolved and their power
density has raised by more than one order of magnitude with the successful
introduction of faster, lighter and often supercritical machines. A large number of
new types of fluid dynamic instabilities, mostly induced by the occurrence of
cavitation, have been identied both theoretically and experimentally, and eld
experience has conrmed their potentially dangerous nature. In addition, operation
at supercritical speeds with non-negligible cavitation has signicantly increased the
importance of rotordynamic instabilities and the need for more accurate prediction
of the critical speeds and their dependence on the properties of the flow.
Today cavitation phenomena in hydraulically operating turbomachinery cannot
be safely predicted by theoretical or numerical methods because of the complexity
of the physical phenomena involved, the imperfections and approximations of the
physical models, and the insufcient temporal and spatial resolution of current
numerical simulation tools. Hence experimentation still plays an essential role for
technology progress in this eld. Operational and economic limitations also clearly
indicate that detailed experiments can only be effectively carried out on turbopumps
under fluid dynamic and thermal cavitation similarity.
82 L. dAgostino et al.
References
Bhattacharyya, A. (1994). Internal Flows and Force Matrices in Axial Flow Inducers. Report
No. E249.18, California Institute of Technology, Pasadena, USA.
Bhattacharyya, A., Acosta, A. J., Brennen, C. E., & Caughey, T. K. (1997). Rotordynamic forces
in cavitating inducers. ASME Journal of Fluids Engineering, 199(4), 768774.
Braisted, D. M. (1979). Cavitation Induced Instabilities Associated with Turbomachines. In Report
No. E184.2, California Institute of Technology, Pasadena, CA, USA.
Brennen, C. E. (1994). Hydrodynamics of pumps. Oxford University Press, Oxford Engineering
Science Series 44.
Brennen, C. E., & Acosta, A. J. (1973). Theoretical, quasi-static analysis of cavitation compliance
in turbopumps. Journal of Spacecraft, 10(3), 175179.
Brennen, C. E., & Acosta, A. J. (1976). The dynamic transfer function for a cavitating inducer.
ASME Journal of Fluids Engineering, 98, 182191.
Brennen, C. E., & Braisted, D. M. (1980). Stability of hydraulic systems with focus on cavitating
pumps. In Proceedings of IAHR Conference, Tokyo, Japan.
Callenaere, M., Franc, J. P., & Michel, J. M. (1998, April 710). Influence of cavity thickness and
pressure gradient on the unsteady behaviour of partial cavities. In 3rd International Symposium
on Cavitation, Grenoble, France.
Cervone, A., Testa, R., Bramanti, C., Rapposelli, E., & dAgostino, L. (2005). Thermal Effects on
Cavitation Instabilities in Helical Inducers. AIAA Journal of Propulsion and Power, 21(5),
893899.
An Introduction to Flow-Induced Instabilities in Rocket 83
Cervone, A., Torre, L., Bramanti, C., Rapposelli, E., & dAgostino, L. (2006). Experimental
characterization of cavitation instabilities in a two-bladed axial inducer. AIAA Journal of
Propulsion and Power, 22(6), 13891395.
Cervone, A., Bramanti, C., Torre, L., Fotino, D., & dAgostino, L. (2007). Setup of a high-speed
optical system for the characterization of flow instabilities generated by cavitation. ASME
Journal of Fluids Engineering, 129(7), 877885.
Cervone, A., Pace, G., Torre, L., Pasini, A., Bartolini, S., Agnesi, L., & dAgostino, L. (2012,
February 27March 2). Effects of the leading edge shape on the performance of an axial three
bladed inducer. In 14th International Symposium on Transport Phenomena and Dynamics of
Rotating Machinery, ISROMAC-14, Honolulu, HI, USA.
dAgostino, L. (2013, September 1922). On the hydrodynamics of rocket propellant engine
inducers and turbopumps. In 6th International Conference on Pumps and Fans with
Compressors and Wind Turbines (IPCF 2013), Beijing, China, IOP Publishing, IOP
Conference Series: Materials Science and Engineering 52 (2013) 012004.
dAgostino, L., dAuria, F., & Brennen, C. E. (1998). A three-dimensional analysis of
rotordynamic forces on whirling and cavitating helical inducers. ASME Journal of Fluids
Engineering, 120, 698704.
dAgostino, L., & Salvetti, M. V. (Eds.). (2007). Fluid dynamics of cavitation and cavitating
turbopumps. In CISM Courses and Lectures No. 496, International Centre for Mechanical
Sciences. Vien and New York: Springer.
Ehrich, F., & Childs, S. D. (1984, May). Self-excited vibrations in high performance
turbomachinery. In Mechanical Engineering (pp. 6679).
Franc, J. P. (2001, June 2023). Partial cavity instabilities and re-entrant jet. In CAV2001,
International Symposium on Cavitation, Pasadena, CA, USA.
Franz, R. J. (1989). Experimental Investigation of the Effect of Cavitation on the Rotordynamic
Forces on a Whirling Centrifugal Pump Impeller. Ph.D. Thesis, California Institute of
Technology, Pasadena, USA.
Fuji, A., Azuma, S., Yoshida, Y., & Tsujimoto, Y. (2002, February 1014). Higher order rotating
cavitation in an inducer. In Proceedings of ISROMAC-9The 9th International Symposium on
Transport Phenomena and Dynamics of Rotating Machinery, Honolulu, Hawaii, USA.
Furukawa, A., Ishizaka, K., & Watanabe, S. (2001, June 2023). Experimental estimate of helical
inducer blade forces in cavitation surge condition. In CAV2001, International Symposium on
Cavitation, Pasadena, California USA.
Furukawa, A., Ishizaka, K., & Watanabe, S. (2002). Experimental study of cavitation induced
oscillation in two bladed inducers. In Space Launcher Liquid Propulsion: 4th International
Conference on Space Launcher Technology, Liege, Belgium.
Greitzer, E. M. (1981). The stability of pumping systems. ASME Journal of Fluids Engineering,
103, 193242.
Hashimoto, T., Yoshida, H., Funatsu, S., Hishimoto, J., Kamijo K., & Tsujimoto Y. (1997a).
Rotating cavitation in three and four-bladed inducers. In 33th AIAA/ASME7SAE/ASEE Joint
Propulsion Conference and Exhibit, Seattle, USA.
Hashimoto, T., Yoshida, H., Watanabe, M., Kamijo, K., & Tsujimoto, Y. (1997b). Experimental
study on rotating cavitation of rocket propellant pump inducers. Journal of Propulsion and
Power, 13(4).
Horiguchi, H., Arai, S., Fukutomi, J., Nakase, Y., & Tsujimoto, Y. (2003b, July 610). Quasi
three-dimensional analysis of cavitation in an inducer. In 4th ASME-JSME Joint Fluids
Engineering Conference, Honolulu, HIa, USA.
Imamura, H., Kurokawa, J., Matsui, J., & Kikuchi, M. (2003, November 14). Suppression of
cavitating flow in inducer by J-Groove. In CAV2003, International Symposium on Cavitation,
Osaka, Japan.
Kamijo, K., Shimura, T., & Watanabe, M. (1977). Experimental investigation of cavitating
inducer instability. ASME Paper 77/WA-FE-14.
84 L. dAgostino et al.
Kang, D., Cervone, A., Yonezawa, K., Horiguchi, H., Kawata, Y., & Tsujimoto, Y. (2007). Effect
of blade geometry on tip leakage vortex of inducer. In The 9th Asian International Conference
on Fluid Machinery, Jeju, South Korea.
Kawanami, Y., Kato, H., Yamaguchi, H., Tagaya, Y., & Tanimura, M. (1997). Mechanism and
control of cloud cavitation. Journal of Fluids Engineering, 119, 788795.
Kimura, T., Yoshida, Y., Hashimoto, T., & Shimagaki, M. (2006). Numerical simulation for
unsteady cavitating flow in a turbopump inducer. In CAV2006, 6th International Symposium
on Cavitation, Wageningen, The Netherlands.
Kubota, A., Kato, H., Yamaguchi, H., Maeda, M. (1989). Unsteady structure measurement of
cloud cavitation on a foil section using conditional sampling technique. Journal of Fluids
Engineering, 111, 204210.
Lieblein, S. (1965). Experimental flow in two-dimensional cascades. In Aerodynamic Design of
Axial Flow Compressors, NASA SP-36, 101149.
Murai, H. (1968). Observations of cavitation and flow patterns in an axial flow pump at low flow
rates. Memoirs of the Institute of High Speed Mechanics, 24(246), Tohoku University.
NASA. (1970). Prevention of Coupled Structure-Propulsion Instability. NASA SP-8055, Space
Vehicle Design Criteria Manuals.
NASDA. (2000a). Report No. 94, May 2000.
NASDA. (2000b). Report No. 96, June 2000.
Natanzon, M. S., et al. (1974). Experimental investigation of cavitation induced oscillations of
helical inducers. Fluid Mechanics Soviet Research, 3(1), 3845.
Ng, S. L., & Brennen, C. E., (1978). Experiments on the dynamic behavior of cavitating
pumps. ASME Journal of Fluids Engineering, 100, 166176.
Ohta, T., & Kajishima, T. (2008, February 1722). Transition of different unsteady cavitating flows in
2D cascade with flat blades. In Proceedings of ISROMAC-12The 12th International
Symposium on Transport Phenomena and Dynamics of Rotating Machinery, Honolulu, HI, USA.
Pace, G., Valentini, D., Pasini, A., Torre, L., Fu, X., dAgostino, L. (2015). Geometry effects on
flow instabilities of different three-bladed inducers. ASME J. Fluids Engineering, 137(4)/
011102-1, 041304.
Pasini, A., Torre, L., Cervone, A., & dAgostino L. (2010). Rotordynamic forces on a four-bladed
inducer. In Proceedings of 46th AIAA/ASME/SAE/ASEE Joint Propulsion Conference,
Nashville, USA.
Pasini, A., Torre, L., Cervone, A., & dAgostino, L. (2011a). Characterization of the rotordynamic
forces on tapered axial inducers by means of a rotating dynamometer and high-speed movies.
In Proceedings of WIMRC 3rd International Cavitation Forum 2011, University of Warwick,
United Kingdom.
Pasini, A., Torre, L., Cervone, A., & dAgostino, L. (2011b). Continuous spectrum of the
rotordynamic forces on a four-bladed inducer. ASME J. Fluids Engineering, 133(12).
Rosenmann, W. (1965). Experimental investigations of hydrodynamically induced shaft forces
with a three bladed inducer. In Proceedings of ASME Symposium on Cavitation in Fluid
Machinery.
Rubin, S. (1966). Longitudinal Instability of Liquid Rockets Due to Propulsion Feedback (POGO).
Journal of Spacecraft and Rockets, 3(8), 11881195.
Sack, L. E., & Nottage, H. B. (1965). System oscillations associated to cavitating inducers. ASME
Journal of Basic Engineering, 87, 917924.
Sakoda, M., Yakushiji, R., Maeda, M., & Yamaguchi, H. (2001, June 2023). Mechanism of cloud
cavitation generation on a 2-D hydrofoil. In CAV2001, International Symposium on Cavitation,
Pasadena, California, USA.
Sergeant, S. R., Sorensen, K. P., & McGlynn, R. D. (2008, February 1722). Measurement of
cavitation induced strain on a 3 bladed inducer. In Proceedings of ISROMAC-12The 12th
International Symposium on Transport Phenomena and Dynamics of Rotating Machinery,
Honolulu, Hawaii, USA.
Semenov, Y. A., Fujii, A., & Tsujimoto, Y. (2004). Rotating choke in cavitating turbopump
inducer. ASME Journal of Fluids Engineering, 126, 8793.
An Introduction to Flow-Induced Instabilities in Rocket 85
Shimagaki, M., Hashimoto, T., Watanabe, M., Hasegawa, S., Nakamura, N., & Shimura, T.
(2006). Unsteady pressure fluctuations in an inducer. JSME International Journal of Fluids and
Thermal Engineering, Series B, 49(3), 806811.
Shimiya, N., Fujii, A., Horiguchi, H., Uchiumi, M., Kurokawa, J., & Tsujimoto, Y. (2006).
Suppression of cavitation instabilities in an inducer by J-Groove. In CAV2006, 6th
International Symposium on Cavitation, Wageningen, The Netherlands.
Shimura, T., Yoshida, M., Kamijo, K., Uchiumi, M., & Yasutomi, Y. (2002, February 1014).
Cavitation induced vibration caused by rotating-stall-type phenomenon in LH2 turbopump. In
Proceedings of ISROMAC-9The 9th International Symposium on Transport Phenomena and
Dynamics of Rotating Machinery, Honolulu, Hawaii, USA.
Stripling, L. B. (1962). Cavitation in turbopumps, part 2. ASME Journal of Basic Engineering, 84
(3), p. 329.
Stripling, L. B., & Acosta, A. J. (1962). Cavitation in turbopumpspart 1. ASME Journal of Basic
Engineering, 84, 326338.
Subbaraman, M., & Patton, M. (2006). Suppressing higher-order cavitation phenomena in axial
inducers. In CAV2006, 6th International Symposium on Cavitation, Wageningen, The
Netherlands.
Torre, L., Pasini, A., Cervone, A., Pecorari, L., Milani, A., & dAgostino, L. (2010).
Rotordynamic forces on a three bladed inducer. In Proceedings of Space Propulsion
Conference, San Sebastian, Spain.
Torre, L., Pasini, A., Cervone, A., & dAgostino, L. (2011a). Continuous spectrum of the
rotordynamic forces on a four-bladed inducer. In ASME/JSME/KSME Joint Fluids Engineering
Conference, Hamamatsu, Japan.
Torre, L., Pasini, A., Cervone, A., & dAgostino, L. (2011). Experimental characterization of the
rotordynamic forces on space rocket axial inducers. ASME Journal of Fluids Engineering, 133
(10).
Tsujimoto, Y. (2001, June 2023). Simple rules for cavitation instabilities in turbomachinery. In
CAV2001, International Symposium on Cavitation, Pasadena, California, USA.
Tsujimoto, Y., Kamijo, K., & Yoshida, Y. (1993). A theoretical analysis of rotating cavitation in
inducers. Journal of Fluids Engineering, 115.
Tsujimoto, Y., Yoshida, Y., Maekawa, Y., Watanabe, S., & Hashimoto, T. (1997). Observations of
oscillating cavitation of an inducer. ASME Journal of Fluids Engineering, 119, 775781.
Tsujimoto, Y., Watanabe, S., & Horiguchi H. (1998, November 16). Linear analyses of cavitation
instabilities of hydrofoils and cascades. In Proceedings of US-Japan Seminar: Abnormal Flow
Phenomena in Turbomachinery, Osaka, Japan.
Tsujimoto, Y., & Semenov, Y. A. (2002). New types of cavitation instabilities in inducers. In
Space Launcher Liquid Propulsion: 4th International Conference on Space Launcher
Technology, Liege, Belgium.
Uchiumi, M., Kamijo, K., Hirata, K., Konno, A., Hashimoto, T., & Kobayasi, S. (2003).
Improvement of inlet flow characteristics of LE-7A liquid hydrogen pump. Journal of
Propulsion and Power,19(3).
Uchiumi, M., & Kamijo, K. (2008, February 1722). Occurrence range of a rotating-stall-type
phenomenon in a high head liquid hydrogen inducer. In Proceedings of ISROMAC-12The
12th International Symposium on Transport Phenomena and Dynamics of Rotating Machinery,
Honolulu, HI, USA.
Valentini, D., Pace, G., Pasini, A., Torre, L., & dAgostino, L. (2015). Influences of the operating
conditions on the rotordynamic forces acting on a three-bladed inducer under forced whirl
motion. Journal of Fluids Engineering, 137(7), 071304.
Wade, R. B., & Acosta, A. J. (1966). Experimental observations on the flow past a plano-convex
hydrofoil. Journal of Basic Engineering, 87, 273283.
Watanabe, S., Yokota, K., Tsujimoto, Y., & Kamijo, K. (1999). Three-dimensional linear analysis
of rotating cavitation in inducers using an annular cascade model. ASME Journal of Fluids
Engineering, 121, 866871.
86 L. dAgostino et al.
Yoshinobu Tsujimoto
Since the flow under alternate blade cavitation is steady, the cause of alternate blade
cavitation can be found from steady calculations. So, steady flow calculations were
made on two pitches of a four-bladed inducer. The commercial software,
ANSYS-CFX11.0, was used for the simulation of the cavitating flow through the
inducer. Three-dimensional Reynolds averaged NavierStokes (RANS) equations
were solved by nite volume method with hybrid unstructured grids. The k-
turbulence model was used. The cavitation model is based on the homogenous
multiphase flow framework of the CFD solver taking into the dynamics of cavi-
Y. Tsujimoto ()
Osaka University, Osaka, Japan
e-mail: [email protected]
s
curve
0.14 Cal.
exposed to the strong axial flow and the cavity on this blade is suppressed, as shown
in Fig. 4c, due to the decreased incidence angle to the blade. The radial component
of the disturbance flow in the meridional plane is much smaller than the axial
component. This is one of the reasons why the two-dimensional analysis can predict
the cavitation instabilities correctly.
90 Y. Tsujimoto
Void Fraction
0.085Ut
1.0
/ s /s /s
2.0 2.0 2.0
Blade1
Void Fraction
1.0 Blade2 Blade1 0.1Ut
r/Rt Blade1
0.0
1.0
0.9
0.8
0.7
0.6
0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2
z/Dt
(a) (b) (c) (d) (e) (f)
Fig. 5 Disturbance velocity and void fraction in meridional plane. = 0.06: a s = 0.2,
b s = 0.9 = 0.04: c s = 0.2 , d s = 0.9, e s = 1.2, f s = 1.9
Three-dimensional Simulation of Cavitation Instabilities 91
2.0
Blade2
1.0
Blade1
0.0
-0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 z/Dt
16.05Rev. 16.65Rev. 17.30Rev. 17.90Rev. 18.55Rev. 19.15Rev. 19.80Rev. 20.40Rev. 21.05Rev.
Fig. 6 Velocity eld in a cylindrical plane at r Rt = 0.98 under rotating cavitation, = 0.078,
= 0.04
92 Y. Tsujimoto
Void Fraction
1.0 0.1Ut
r/Rt Blade2 Blade1
0.0
1.0
0.9
0.8
0.7
0.6
0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2
16.05Rev. 16.65Rev. 17.3Rev. 17.90Rev. 18.55Rev. 19.15Rev.
r/Rt
1.0
0.9
0.8
0.7
0.6
0.0 0.1 0.2 0.0 0.1 0.2 0.0 0.1 0.2
z/Dt
19.80Rev. 20.40Rev. 21.05Rev.
c
0.3
0
/ s
3.0
blade3
2.0
blade2
1.0
blade1
0.0
z/Dt
-0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2 -0.2 0.2
16.05Rev. 16.65Rev. 17.3Rev. 17.90Rev. 18.55Rev. 19.15Rev. 19.80Rev. 20.40Rev. 21.05Rev.
interaction of the flow towards the cavity trailing edge with the leading edge of the
next blade.
As mentioned above, in real three-dimensional flow with signicant tip cavita-
tion, there exists a local flow near the cavity trailing edge in which the incidence
angle to the next blade is decreased. Cavitation instabilities such as alternate blade
cavitation, rotating cavitation, and cavitation surge start to occur when this local
flow starts to interact with the leading edge of the next blade. Figure 9 shows the
Three-dimensional Simulation of Cavitation Instabilities 93
Circumferential location, /s
Circumferential location, /s
0.6 0.6 0.6
Cavity
relative velocity eld around the tip cavity, obtained by experiments. As shown in
(b), the flow penetrates into the cavity and does not round about the cavity in the
same way as for the calculated result shown in Fig. 3. By subtracting the
non-cavitating velocity vector shown in (a) from the cavitating flow vector shown
in (b), the disturbance flow vector is obtained and shown in (c). The disturbance
eld shown in (c) is similar to that in Fig. 4 with outward flow near the cavity
leading edge and the flow towards the cavity trailing edge. Thus, the existence of
disturbance flow eld predicted by CFD was conrmed by experiments.
In many cases, the tip cavitation is more signicant than blade surface cavitation.
This means that cavitation instabilities can be avoided by designing the inducers so
that the interaction of the tip cavity with the next blade can be avoided. Actually
several inducers were designed based on this guideline and it was conrmed by
experiments that cavitation instabilities are successfully suppressed (Kang et al.
2009b).
Since it was found in the last section that cavitation instabilities occur due to the
interaction of the flow near the cavity trailing edge with the leading edge of the next
blade, it might be possible to avoid cavitation instabilities by avoiding the inter-
action by trapping the tip leakage flow by a circumferential groove on the casing.
94 Y. Tsujimoto
This method has been used for suppressing rotating stall for compressors (Hah et al.
2008). This section describes about the application of this method for inducers to
illustrate the usefulness of CFD for developments (Kang et al. 2010).
First, we numerically investigated the effect of groove geometries on the tip leakage
vortex and found the optimum geometries to pack the tip leakage vortex in the
circumferential groove.
Figure 10 shows the geometry of circumferential groove tested. The width of
groove is 19.5 mm and the upstream edge is located 3.5 mm upstream of the
leading edge of blade tip. The depth of groove is 10 mm for the impeller with
diameter 149.8 mm. Table 1 shows the static pressure coefcient s obtained from
the calculation. With the grooved casing, the pressure coefcient is about 5% lower
than that with the straight casing.
Fig. 10 Geometry of
circumferential groove
straight casing and the separating point of tip leakage vortex cavity is moved
downstream.
To clarify the effect of cavity, the velocity disturbance caused by the cavity was
evaluated by subtracting the velocity of non-cavitating flow from that of cavitating
flow. The disturbance velocity vectors and void fraction distributions in z, and
r, z-planes are shown in Fig. 12. With the straight casing, the disturbance flow near
the cavity trailing edge has an axial component which reduces the incidence angle
to the next blade. It has been shown in the last section that the axial disturbance
flow causes alternate blade cavitation, cavitation surge, and rotating cavitation.
With the grooved casing, the disturbance flow is weaker and no positive axial flow
is found near the leading edge of the blade. This is preferable to suppress cavitation
instabilities.
Figures 13 and 14 show the velocity vectors and void fraction distributions at
d = 1.0 and = 0.04 in r-z planes. With the straight casing, the tip leakage vortex
Groove 0.1Ut
0
r/Rt r/Rt
1.0 1.0
207 207 1
0.9 0.9
198 198
0.8 0.8
187 187
0.7 0.7
r/Rt r/Rt
147 147
1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7
core appears near the leading edge at blade tip shown in Fig. 13a, and then it extends
upstream as is increased. The cavity appears not in the tip leakage vortex but in the
lower pressure region around the junction of blade suction surface and the casing.
With the grooved casing, the tip leakage vortex cavity develops only in the cir-
cumferential groove. The vortex is enhanced by the jet at = 170 from the clearance
between the pressure surface of the blade and the rear wall of the groove. At = 110 ,
a weak cavity is observed on the pressure surface. This is caused by the interaction
with the tip leakage vortex cavity. However, the head s = 0.1214 at = 0.04 is not
signicantly lower than the head s = 0.121 under non-cavitating condition. Inward
radial velocity is observed in the upstream edge of the circumferential groove, sup-
pressing the tip leakage vortex cavity on the suction side of blade tip.
Figure 15 shows the blade pressure distributions at r Rt = 0.98. At = 0.10 a
cavity appears on the suction surface in 90 < < 135 with the straight casing but
no cavity is found with the grooved casing. With the grooved casing, negative blade
loading is found near the leading edge and the blade loading is moved downstream.
Thus, we can control the blade loading by the groove. At = 0.04 the cavitating
region is extended to 90 < < 175 with the straight casing. With the grooved
casing, the effect of cavitation number is smaller although a cavity appears in
100 < < 120 on the pressure surface. The pressure on the pressure surface is
0
r/Rt r/Rt r/Rt r/Rt
1.0 1.0 1.0 1.0
0.9 0.9 0.9 0.9 1
0 0.1 0.2 z/D 0 0.1 0.2 z/D 0 0.1 0.2 z/D 0 0.1 0.2 z/D
t t t t
Fig. 13 Disturbance velocity and void fraction in meridional plane, straight casing, d = 1,
= 0.04
0.2Ut
0
r/Rt r/Rt r/Rt r/Rt
1.0 1.0 1.0 1.0
1
0.9 0.9 0.9 0.9
0.8 0.8 0.8 0.8
0.7 0.7 0.7 0.7
0 0.1 0.2 z/D 0 0.1 0.2 z/D 0 0.1 0.2 z/D 0 0.1 0.2 z/D
t t t t
Fig. 14 Disturbance velocity and void fraction in meridional plane, grooved casing, d = 1,
= 0.04
Three-dimensional Simulation of Cavitation Instabilities 97
Suction side with the straight casing Suction side with the straight casing
Pressure side with the straight casing Pressure side with the straight casing
Suction side with the grooved casing Suction side with the grooved casing
Pressure side with the grooved casing Pressure side with the grooved casing
0.2 Groove 0.2 Groove
Pressure, (p-p1 ) / ( Ut2)
0.1 0.1
0.0 0.0
0.3
0.2
0.1
lower and the cavity appears caused by the interaction with the tip leakage vortex
from the preceding blade.
Figure 16 shows the non-cavitating performance curves obtained from the exper-
iments and the computation. With the grooved casing, the static pressure coef-
cients at lower flow coefcients are lower than that with the straight casing.
However, the decrease of static pressure coefcient is acceptable at the design flow
coefcient, d = 0.078.
98 Y. Tsujimoto
Figure 17 compares the suction performance curve with and without the groove. With
the straight casing (Tsujimoto et al. 1997), the test was conducted at 3000 rpm with
orice plates at the inlet, while the test was conducted at 4500 rpm without the orice
plates for the case with the grooved casing. The larger scatter with the straight casing
is caused by cavitation instabilities with some effects of bubbles from the orice
plates. We nd that the suction performance is somewhat improved by the groove.
Figures 18, 19 and 20 show the spectra of inlet pressure fluctuations measured by
the pressure transducers flush mounted 44 mm upstream of the blade leading edge
at the tip. The horizontal axis shows the frequency, the vertical axis shows the
magnitude of the pressure fluctuation dened as p Ut2 ) and the depth axis
shows the cavitation number . N is the frequency 50 Hz of the impeller rotation.
The phase difference of pressure fluctuations measured at two circumferential
0.15 0.15
s
0.14 0.14
Static pressure coefficient,
0.13 0.13
0.12 0.12
0.11 0.11
0.10 0.10
0.09 0.09
0.08 / d=0.975 0.08 / d=0.93
0.07 / d=1.000 0.07 / d=0.96
0.06 / d=1.025 0.06 / d=1.11
0.05 0.05
0 0.02 0.04 0.06 0.08 0.10 0 0.02 0.04 0.06 0.08 0.10
Cavitation number, Cavitation number,
locations 90 apart are shown for typical components. The negative/positive phase
difference means that the pressure pattern rotates in the same/opposite direction of
the impeller rotation.
In 0 < f < 150 Hz, with the circumferential groove, rotating cavitation, cavitation
surge, and asymmetric cavitation were successfully suppressed at all flow rates
except for weak instabilities observed at d = 1.1.
In 150 < f < 500 Hz, instabilities with higher frequency components are
observed. A higher frequency component with 4.8N was observed in the initial
engine ring tests and water tests on liquid hydrogen turbopump inducer for HII
rocket (Tsujimoto and Semenov 2002). In this case, the phase difference was 0.
The higher frequency components are important since they can be closer to the
natural frequencies of blade bending mode. For this reason, extensive studies have
been made to avoid the higher frequency components. In the present tests, we nd
components around 5N, 7 8N, and 10N. The phase difference is different from
100 Y. Tsujimoto
case to case but can be assorted into groups around zero (surge mode), 90 (one
cell rotating forward) and + 90 (one cell rotating backward). The cause of these
higher frequency modes is discussed later.
Figures 21 and 22 compare the pictures of cavities at the design flow coefcient at
d = 1.0 with and without the groove. We observe the following differences.
1. With the straight casing, the cavity occupies the entire tip region while the
cavity mostly occurs in the groove with the grooved casing. This agrees with the
numerical results shown in Figs. 13 and 14.
2. With the straight casing, the backflow vortex cavitation extends more upstream
than the case with the groove. With the grooved casing, the backflow vortex
cavitation extends only slightly from the upstream edge of the groove.
Groove
Although not shown, it was found that the increase of cavity volume with the
decrease of flow rate was much smaller with the grooved casing.
Figure 23 shows the pictures of cavities from high-speed video to examine the
propagation of backflow vortex cavities for the case with the grooved casing, at
d = 1.0 and = 0.045. The propagation speed v is measured from the pictures
and number of vortices Nv was evaluated by counting the passage of vortices. The
results are shown in Table 2 where n is the impeller speed.
For the case without groove, the backflow vortex structure has been studied
extensively (Yokota et al. 1998) and it was found that the vortex structure is caused
by the roll-up of the shear layer between the straight main flow and swirling
backflow. It was found that the number of vortices is determined from the stability
of vortices and the number decreases with the decrease of flow rates caused by the
inward shift of the radial location of the vortices. It was also found that both the
velocity and the number of vortices fluctuate largely with time. So, the values in
Table 2 are only representative values.
To conrm the suppression of rotating cavitation and to clarify the cause of higher
frequency components, unsteady calculations were made on full passage with the
grooved casing, at d = 0.9 and = 0.05.
Figure 24 shows the spectrum of inlet pressure fluctuation at the same location
with the experiments. As expected, we do not have signicant component around
N = 50 Hz, suggesting that the rotating cavitation has been suppressed. In addition
to the blade passing component and its modulations around 3N = 150 Hz we
observe a component with 5.44N.
Figure 25 shows the three-dimensional representation of iso-pressure surface of
c = p pv Ut2 = 0.083 and the pressure distributions in the planes of r Rt = 0.87
0.0020 3N
2.72N
(70deg) (118deg) 5.44N
0.0015 (-37.5deg)
0.0010
0.0005
0.0000
0 100 200 300 400 500
Frequency, f (Hz)
50 1 0.04
0 2
1
20
1 -0.01
2
350
320 1
290 2
0 260 2 0
It is known that rotating modes occur due to rotorstator interaction (Tyler and
Sofrin 1962). We rst consider a rotor with R blades r = 1, . . . R rotating with an
angular speed of and a stator with S blades s = 1, . . . S as shown in Fig. 26.
We represent the pressure component caused by the interaction of a stator blade
s = 1 with R rotor blades (r = 1, . . . R) by
where m and n are the harmonic order in and t. Then the pressure pattern caused
by the interaction of a stator blade s = q with R rotor blades r = 1, . . . R can be
represented by
2 2
pm, n , t = am, n cos m
s=q
q 1 nR t q 1 2
S S
S
2
pm, n , t = am, n cos m nRt m nR q 1
q=1 S
3
S am, n cosm nRt m = nR + kS
=
0 m nR + kS
We consider the case when the stator is also rotating and represent the stator
speed by v and the rotor speed by n . We introduce an absolute circumferential
coordinate * = + v t. By putting = * v t and = n v , we obtain
pm, n * , t = S am, n cos m* nRn v + mv t 4
m = nR + kNv 5
= nRn v + mv 6
Fig. 27 Spectra of pressure fluctuation in a four-bladed inducer, from Subbaraman and Patton,
2006
frequency components are caused by the interaction of the backflow vortices with
the blades. The discrepancy might be caused by the unsteady nature of the backflow
vortices; the number of vortices and the propagation velocity changes irregularly
with time.
Figure 27 shows the spectra of pressure fluctuation in a four-bladed inducer
measured at 0.03Dt axial distance downstream from the blade leading edge at the
tip (Subbaraman and Patton 2006). We have a case of 0 = 2 4 1 8 for
m = nR + kNv . For this case, the estimated frequency with v = 0.15 is
= 2 4 1 0.15 n = 6.8. Another case is 0 = 3 4 1 12 with the fre-
quency = 3 4 1 0.15 n = 10.2. The frequencies are 4/3 times the fre-
quencies for the three-bladed inducer shown in Table 4. Figure 27 includes the
components close to these frequencies.
3 Conclusions
It has been shown that CFD has been developed to the level that three-dimensional
unsteady cavitating code can be used to simulate cavitation instabilities such as
rotating cavitation and cavitation surge. These cavitation instabilities are caused by
the interaction of local flow near cavitation trailing edge with the leading edge with
the next blade. This shows that cavitation instabilities can be avoided by avoid-
ing the interaction. As an example, an avoidance method using circumferential
Three-dimensional Simulation of Cavitation Instabilities 107
groove on the casing was shown. This lead to the explanation of higher frequency
oscillation which may cause structural resonance.
Qualitative simulation of cavitation instabilities is rather easy as compared with
the simulation of instabilities caused by flow separation since the pressure around
cavity is kept nearly constant at the saturated vapor pressure while the back pressure
in separated flow region largely depends on the shear stress. The flow disturbance
near the cavity trailing edge occurs through the continuity relation, and not through
the momentum relation in which the shear stress plays an important role. However,
the precise prediction of onset range of cavitation instabilities is still difcult since it
depends also on the shear stress. So, further progress in CFD technique is needed in
order to examine the occurrence of cavitation instabilities at the design stage of
machines.
References
Hah, C., Mueller, M., & Schiffer, H. (2008). Aerodynamic study of circumferential grooves in a
transonic axial compressor. ASME Fluid Engineering Conference, 55232, 18.
Horiguchi, H., Watanabe, S., & Tsujimoto, Y. (2000). A Linear stability analysis of cavitation in a
nite blade count impeller. ASME Journal of Fluids Engineering, 122(4), 798805.
Kang, D., Yonezawa, C., Horiguchi, H., Kawata, Y., & Tsujimoto, Y. (2009a). Cause of
Cavitation instabilities in three-dimensional inducer. International Journal of Fluid Machinery
and Systems, 2(3), 206214.
Kang, D., Watanabe, T., Yonezawa, K., Horiguchi, H., Kawata, Y., & Tsujimoto, Y. (2009b).
Inducer Design to Avoid Cavitation Instabilities. International Journal of Fluid Machinery and
Systems, 2(4), 439448.
Kang, D., Arimoto, Y., Yonezawa, K., Horiguchi, H., Kawata, Y., Hah C., & Tsujimoto, Y.
(2010). Suppression of cavitation instabilities in an inducer by circumferential groove and
explanation of higher frequency components. International Journal of Fluid Machinery and
Systems, 3(2)
Subbaraman, M., & Patton, M. (2006). Suppressing higher order cavitation phenomena in axial
inducers. Sixth International Symposium on Cavitation CAV2006 pp. 113.
Tsujimoto, Y., Yoshida, Y., Maekawa, M., Watanabe, S., & Hashimoto, T. (1997). Observations
of oscillating cavitations of an inducer. ASME Journal of Fluids Engineering, 119(4), 775781.
Tsujimoto, Y. and Semenov, Y. (2002). New types of cavitation instabilities in inducers. In 4th
International Conference on Launcher Technology Space Launcher Liquid Propulsion
pp. 110.
Tyler, J. M. and Sofrin, T. G. (1962). Axial compressor noise studies. In Transactions Society of
Automotive Engineers 70, pp. 309322.
Watanabe, S., Sato, K., Tsujimoto, T., & Kamijo, K. (1999). Analysis of rotating cavitation in a
nite pitch cascade using a closed cavity mode and a singularity method. ASME Journal of
Fluids Engineering, 121(4), 834840.
Yokota, K., Kurahara, K., Kataoka, D., & Tsujimoto, Y. (1998). AS study of swirling backflow
and vortex structure in the inlet of an inducer. Journal of the Japan Society of Mechanical
Engineers, 64622, 5158.
Rotordynamics of Turbopumps
and Hydroturbines
Yoshinobu Tsujimoto
1 Introduction
Y. Tsujimoto ()
Osaka University, Osaka, Japan
e-mail: [email protected]
self-excited vibration of the rotor. Generally, this occurs when the shaft speed
exceeds a certain critical speed and the amplitude does not decrease even if the
speed becomes much larger than the critical speed. This is quite different from
forced vibrations. The frequency is maintained as one of the natural frequencies of
the shaft system. These circumstances are shown in Fig. 1b, which is partly
modied from a gure in Ehrich and Childs article (1984).
Rotordynamic forces are found in various parts of turbomachinery such as
hydrodynamic bearings, annular or labyrinth seals and impellers. In the rst part of
this note we focus on rotordynamic impeller forces simply because the author has
not been involved in other forces. Textbooks such as Vance (1988), Ehrich (1992),
and Childs (1993) cover other rotordynamic forces and rotordynamic analysis.
Brennens book (1994) focuses more on the fluid dynamic explanations of the
forces. However, there are certain cases when self-excited vibration occurs even
when the rotational speed is lower than the critical speed. These cases are treated in
the second part of this note.
Fig. 2 Experimental apparatus for the measurements of shaft vibrations at supercritical speed
Fig. 4 Effects of critical speed/shaft speed ratio, * , on self-excited and forced vibrations
the frequencies of forced vibration caused by impeller and diffuser stalls are kept
nearly constant, and that their magnitude increases as the natural frequency, * ,
approaches their own frequencies, 0.08 for impeller rotating stall,
0.05 for diffuser rotating stall. The frequency of self-excited vibration is
identical to the natural frequency * and it occurs when * < 0.5. In other
words, self-excited vibration occurs when the rotational frequency, , exceeds
twice the natural frequency, 2* . This is caused by the rotordynamic forces on the
impeller, which is the main focus of the present note.
Rotordynamics of Turbopumps and Hydroturbines 113
F = jjfeit + = fei 1
M = jjmeit + = mei 2
The displacement of the shaft = jjeit caused by the force F * and the moment
*
M applied on the shaft by the rotor can be represented as follows, by using the
stiffness coefcients k and km of the shaft.
= F * k + i M * km 3
If we consider the inertia ma and the damping force on the rotor c, the
momentum equation of the rotor for the lateral vibration can be given by
F * = F ma c 4
If we assume that the angular displacement of the rotor is so small that the inertia
moment and damping can be neglected, we can write
M* = M 5
where
ma + c + k fe ei* = 0 7
If we put = et in Eq. (7), and assume that c and fe are sufciently small, we
obtain
r
c kI kR
= p i 8
2ma 2 ma kR ma
where kR = k fe cos* and kI = fe sin* . Equation (8) shows that the whirl
frequency is given by
r
kR
= R = 9
ma
p
The forward whirl with R = kR ma grows when
c kI
p > 0
2ma 2 ma kR
That is,
1 c
fet > or jjfet > cjjR 10
2ma R 2ma
This means that the whirl radius jj grows when the equivalent tangential force
jjfet = jjfft + k km mn g becomes larger than the damping force cjjR . Equa-
p
tion (10) also shows that the backward whirl with R = kR ma occurs when
jjfet < cjjR . The contribution of the tangential force ft was shown by Shoji
and Ohashi et al. (1981). Present result shows that the whirl instability can occur
also when the normal moment mn has the same sign as the whirl angular velocity .
Rotordynamics of Turbopumps and Hydroturbines 115
We should stress here that the radial force, fen , has the effect of simply modifying
the shaft effective stiffness but that the tangential force, fet , has the effect of pro-
moting the whirling instability in the direction of fet .
Generally, the rotordynamic forces (fn , ft ) are functions of the whirl velocity
ratio, , where and are whirl and impeller rotational frequencies. For the
case of Fig. 6a or b, forward or backward whirl instability will occur when
R < cr 11
where R is the shaft natural frequency of Eq. (9) and cr is the critical whirl
frequency ratio at which
f t = j j R 12
For the forward whirl, Eq. (12) shows that the instability occurs when the
rotational frequency, , exceeds a critical speed, cr :
> cr = R cr 13
where R and L are representative radial and axial length dimensions of the device.
is used to represent jj hereafter. For centrifugal impellers, the impeller outlet
radius, r2 , and breadth, b2 , are used for R and L.
In this section we consider only the whirling forces F. The fluid force moment will
be discussed in the next section.
116 Y. Tsujimoto
As the most fundamental case, Ohashi and Shoji (1987) measured the rotordynamic
forces on a 2-D centrifugal impeller executing a circular motion with a constant
radius at various whirl speed ratios, , in a vaneless space as shown in Fig. 7.
The forces on the outer surfaces of frond and back shrouds have been subtracted
by using the data of a dummy impeller without flow channel. Figure 8 shows the
dependences of Fn and Ft on the whirl speed ratio, , at ve different flow rates.
This gure shows that
1. Rotordynamic forces are basically stabilizing with Ft < 0 for > 0 and Ft > 0
for < 0 except for the cases of sf < 0.6, where sf is the incidence free
flow coefcient.
2. At smaller flow rates with sf < 0.6, the tangential force becomes positive for
0 < < 0.5. Thus, the impeller forces become destabilizing for forward whirl
at smaller flow rates.
The radial component is explained as the total of centrifugal force on the added
mass, Ma , for lateral acceleration, and the KuttaJoukovsky force on the impeller
circulation due to the whirl velocity, . This is expressed by
Fn = Ma r22 b2 2 16
where is the head coefcient and is the hydraulic efciency. This result is
plotted in Fig. 8 as simple theory.
Another type of 2-D flow analysis has been made by Tsujimoto and Acosta
(1987). In this model the analysis is simplied by assuming an innite number of
impeller vanes. However, the effects of vaneless diffuser and of impeller loss are
taken into account, instead. Figure 9 shows the results. At the shock-free flow
coefcient * = 1.0 the rotordynamic force is stabilizing. However, at reduced
flow rates, a destabilizing region appears at small positive whirl speed ratio. Large
changes of forces occur when the flow coefcient and the whirl speed ratio
approach the onset flow coefcients and the stall propagation velocity ratios of
impeller and vaneless diffuser rotating stalls.
Experiments at Caltech (Chamieh et al. 1985: Jery et al. 1985) were started with a
3-D impeller (Impeller X) combined with a volute casing (Volute A). The measured
rotordynamic forces were signicantly larger than the results shown in Fig. 8 and a
destabilizing region was found for 0 < < 0.5 even at the design point. To nd
out the reason of the differences, a two-dimensional impeller (Impeller Z) was
tested in Volute A (Franz and Arndt 1986). The results are shown in Fig. 10.
The magnitude of the forces is equivalent to that shown in Fig. 8 but a desta-
bilizing region is found at 0 < < 0.3 even near the design flow coefcient
d = 0.092. This is caused by the interaction with the volute.
To conrm this, a 2-D linear inviscid flow analysis assuming an innite number
of impeller vanes was made by representing the effect of volute by a singularity
method (Tsujimoto et al. 1988a). The results are shown in Fig. 11.
We observe reasonable agreements with the 2-D impeller experiments and the
destabilizing region 0 < < 0.3 is well simulated by the model. It was also
conrmed by the analysis that the destabilizing region disappears when the volute is
neglected or when the normal velocity disturbance on the volute is allowed. It was
also found by the calculation that the radial location of the volute tongue rt should
Fig. 11 Comparison of theoretical results with experiments on 2-D (Impeller Z) and 3-D
(Impeller X) impellers in Volute A
Rotordynamics of Turbopumps and Hydroturbines 119
Fig. 12 Effects of the diffuser inlet/impeller outlet radius ratio r3 r2 at the design flow rate,
diffuser vane angle = 11
satisfy rt r2 > 1.6 to substantially remove the destabilizing region. These facts show
that the destabilizing region is caused by the interaction with the volute.
A similar 2-D inviscid flow analysis was made to examine the effects of the
interaction with vaned diffusers (Tsujimoto et al. 1988). Figure 12 shows the effects
of the diffuser inlet/impeller outlet radius ratio, r3 r2 at the design flow rate.
A destabilizing tangential force occurs at 0 < < 0.5 for the cases with realistic
values of r3 r2 . This is the cause of the self-excited vibration shown in Fig. 4.
The impeller forces can be calculated from the pressure distributions around the
inlet and outlet of the impeller, momentum flux at the inlet and outlet of the
impeller, and the change of the momentum of the fluid in the impeller. It was found
that the component due to the pressure distribution around the impeller is the largest
and determines the rotordynamic forces.
The impeller forces under the conditions of strong interaction with a volute or a
vaned diffuser can be explained as follows. We consider the cases when the angle of
the volute or the diffuser vanes measured from the circumferential direction is very
small. In this case the fluctuation of radial absolute velocity is strongly regulated by
the existence of the volute or the vane surfaces. The radial velocity fluctuation can
be divided into the components caused by
(a) whirling motion,
(b) dislocation of the steady flow around the impeller represented by a source and a
vortex,
(c) Nonuniform relative flow through the impeller flow passage.
120 Y. Tsujimoto
Under the condition with the strong interaction, the component (c), i.e., the
nonuniform flow through the impeller, occurs so as to cancel the components
(a) and (b). Then the pressure distribution around the impeller is established such as
to balance the inertia due to the fluctuation of the flow through the impeller. By
integrating the pressure distribution around the impeller, we obtain the following
rotordynamic forces
Ft = M r2 1 17
Fn = M r2 1 18
where and are the flow and head coefcients of the impeller, and
Z r2
r2
M= dr
r1 r sin2 r
is the inertial length of the impeller flow channel (r is the local vane angle from
the tangent). The results of Eqs. (17) and (18) are shown in Fig. 13.
We have seen that the rotordynamic forces on a 3-D impeller are signicantly larger
than those on a 2-D impeller (Fig. 11). In order to determine the rotordynamic
forces on the shroud, experiments were made by Guinzburg et al. (1994), using the
apparatus shown in Fig. 14.
The leakage flow was generated by using an external pump. The forces were
normalized by Eqs. (14) and (15) using the axial length of the leakage path for L.
The leakage flow rate, Q, was normalized as = Q 2r22 H, where H is the
clearance between the rotor and casing. Figure 15 shows the effect of leakage flow
rate. The dimensional magnitude of the rotordynamic forces is the same order as
that of 3-D impeller and a destabilizing region is found at small positive whirl. The
Rotordynamics of Turbopumps and Hydroturbines 121
Fig. 14 Experimental
apparatus for the
measurement of shroud forces
Fig. 15 Effects of leakage flow rate, = 0.0254 cm, H = 0.14 cm, at 1000 rpm
normal force signicantly increases as the leakage flow increases. It was also found
by giving inlet swirl, that the tangential force and the destabilizing region is sig-
nicantly increased by adding a swirl at the inlet, in the direction of the rotor
rotation (Fig. 16). It was also found that the magnitude of the force is inversely
proportional to the clearance.
For the calculation of the shroud forces a bulk flow model proposed by Childs
(1989) was examined. In this model, dependent variables are averaged over the
clearance and the averaged 2-D flow in the variable clearance is calculated taking
account of the empirical shear stress on the wall. Figure 16 shows an example of
calculated results. Reasonable agreement is obtained for < 0.5 without inlet
swirl = 0, but the calculated results are too sensitive to the inlet swirl. The large
changes of the forces in > 0.5 are called resonance but this was not
observed throughout the experiments.
Other examples showing the importance of the shroud forces are given in
Figs. 17 and 18 (Ohashi et al. 1991). In Fig. 18, the results with and without the
spacer are shown, together with the case of a spacer having radial grooves to reduce
the swirl of leakage flow. The forces are larger for the case with the spacer, i.e., with
smaller clearance. More important is the fact that a destabilizing region appears at
122 Y. Tsujimoto
Fig. 16 Comparison with bulk flow calculation, = 0.0118 cm, H = 0.14 cm, at 1000 rpm
0 < < 0.5 with the spacer. The tangential force and hence the destabilizing
region is decreased by using the grooves on the spacer.
This is explained as follows. The tangential velocity of the leakage flow at the
clearance approaches one-half of the tangential velocity of the rotor, from the
balance of the tangential stresses exerted by the rotor and the casing. Then, the
swirling flow will push the rotor in the direction of the whiling motion if the whirl
speed ratio is smaller than the tangential velocity ratio 0.5 of the swirling
leakage flow. This is the reason why the destabilizing region was found for the
smaller clearance. This explanation also applies to the destabilizing forces on
journal bearings and seals. If we have grooves on the casing, the tangential velocity
ratio of the leakage flow will be decreased. This is the reason why the upper limit of
the destabilizing region was decreased for the case with the grooves.
The characters of rotordynamic forces on centrifugal impellers can be summa-
rized as follows:
(1) Rotordynamic forces on a 2-D impeller are basically stabilizing at the design
flow coefcient. It can become destabilizing for forward whirl at smaller flow
rate caused by rotating stalls in the impeller or the diffuser.
Rotordynamics of Turbopumps and Hydroturbines 123
Fig. 18 Influence of
clearance conguration on the
rotordynamic forces on 3-D
impeller shown in Fig. 17
(2) Rotordynamic forces on a 2-D impeller can become destabilizing for forward
whirl caused by the interantion with volute or vaned diffuser placed down-
stream of the impeller.
(3) Rotordynamic forces on a 3-D impeller can become signicantly larger than
those on 2-D impeller, caused by the forces on the front shroud of the impeller.
In addition, the shroud forces can cause forward whirl in the range < 0.5.
The destabilizing region can be decreased by increasing the clearance between
the shroud/casing, or by decreasing the tangential flow velocity by grooves on
the casing.
(4) Although not shown in the present text, rotordynamic forces promoting forward
whirl has been measured also for an open type compressor impeller (Yoshida
et al. 1999). The magnitude of the rotordynamic forces is in the same order as
the closed type impellers as shown in Fig. 18. It was found that the destabi-
lizing force is caused by the pressure on the hub, caused by the change in the
thickness of the flow cannel due to the whirling motion (Hiwata and Tsujimoto,
2002).
124 Y. Tsujimoto
A rotordynamic problem was found in the fuel turbopump for the main engine
LE7A of HIIA rocket and it was suppressed successfully by a small design mod-
ication (Motoi et al. 2003). The rotational speed of this pump is 42,200 rpm
= 703 Hz, which is between third and fourth critical speeds. The spectra of
shaft vibration are shown in Fig. 19 for the LE7 and LE7A turbopumps. For the
LE-7 turbopump for HII rocket, no signicant frequency peak is found. For the
LE-7A turbopump, two peaks are found at 250 Hz and near 500 Hz. These fre-
quencies correspond to the rst and the third critical frequencies of the rotor.
The vibration modes of the LE7A turbopump are shown in Fig. 20. For the rst
and the third modes, for which the frequency peak is found in the data in Fig. 19, a
large amplitude is found at the turbine end (to the right). From this result, it was
conjectured that the vibration is caused by the forces on the turbine.
cr1 cr1
Fig. 19 Spectra of shaft vibration of turbopumps for LE7 and LE7A engines
Fig. 21 Evaluation of tangential forces from nonuniform driving force due to eccentricity
For axial flow turbines, it is well known as steam whirl that forward whirl is
promoted by the difference of the driving forces caused by the difference of the tip
clearance due to the eccentricity. Figure 21 explains the tangential rotordynamic
forces resulting from the nonuniform driving forces.
It is expected that we have a larger driving torque with smaller tip clearance. The
torque with the minimum tip clearance at = 0 is represented by T1 and the average
torque with the mean tip clearance by T0 . If we integrate the tangential force
component dF cos = T R cosd 2 at the mean radius R by assuming a
sinusoidal dependence on as shown in Fig. 21, the total tangential force can be
evaluated to be
Ft = T1 T0 T0 T0 2R 19
This shows that Ft > 0 and forward whirl is promoted for axial turbines. Thomas
(1958) represents the tangential force by
Ft = T0 DL, 20
where is the eccentricity, D the mean diameter of the rotor, L the length of the
turbine blades and is a nondimensional empirical factor. To determine the values
of the turbine torques T0 and T1 with nominal and minimum tip clearance,
numerical calculations of the flow were made by assuming uniform tip clearance
corresponding each case. It was found that the value of the empirical factor for the
expression of Eq. (20) is = 0.56 for the present case.
To provide sufcient damping to the rotor system, the direction of the injection
of the seal liquid (high-pressure liquid hydrogen) to the seal between the turbine
and the pump was changed as shown in Fig. 22, in such a way as to decrease the
mean tangential flow velocity in the seal. Then the sub-synchronous vibration was
successfully suppressed as shown in Fig. 23. This is because the mean tangential
velocity of the leakage flow is decreased by the injection of the seal liquid in the
direction opposite to the rotation of the rotor and hence the destabilizing force on
the seal is decreased.
126 Y. Tsujimoto
Seal
flow
flow injection
roatation roatation
n
tio
jec
in
w
flo
cr1 cr1
Severe flexural vibration of the rotor shaft of a Francis turbine generator shown in
Fig. 24 was experienced in its test operation and reported by Tomita and Kawamura
(1965).
The mechanism of vibration is explained as follows (Ohashi 1991). Figure 25
shows the leakage flow model. As long as the runner remains in the center of the
casing, the leakage is uniform around the runner. Once the runner deviates from the
Rotordynamics of Turbopumps and Hydroturbines 127
center of the casing by y1 , the uniformity will be broken and the resulting change of
the leakage flow, represented by q in Fig. 25 will occur in the back chamber. When
the runner vibrates, the variational flow must change its direction accordingly, thus
causing the variation of the pressure in the chamber. As the rst step of under-
standing, the rotor is modeled by a plate with an entrance orice and the leakage
flow is modeled by a one-dimensional flow as shown in Fig. 26. We assume that
the leakage flow rate q is proportional to the leakage flow velocity U in the entrance
orice and the width B and the height y, i.e., q = BUy. The pressure distribution
128 Y. Tsujimoto
Fig. 26 One-dimensional
flow model
This equation shows clearly that the pressure distribution caused by the accel-
eration of the leakage flow causes negative damping for the vibration of the plate.
Then, a lateral vibration model of an overhung rotor is proposed as shown in
Fig. 27. The upper end is rigidly supported and a bearing is placed at y2 . The lateral
force on the rotor is represented by Ph = Af y1 and the moment on the rotor by
Mh = Am y1 . Here, the influence factors i, j giving the lateral displacement at the
node i due to the force at j, and i, j giving the lateral displacement at the node i due
to the moment at j are introduced. Then, the equation of motion can be expressed by
y1 = 11 My1 + Af y1 + 12 cy2 + 11 Am y1
22
y2 = 21 My1 + Af y1 + 22 cy2 + 21 Am y1
From these equations, it can be shown that the lateral vibration grows when
11
c221 < Af + Am 211 23
11
This result shows that fluid moment on the rotor can cause lateral vibration
through structural coupling.
Thus, the possible mechanism of the vibration has been shown by the simplied
models. However, rotors generally exhibit whirling and precession motion and it is
Rotordynamics of Turbopumps and Hydroturbines 129
not clear under what conditions the self-excited vibration occurs. The rotordynamic
analysis presented in Sect. 3 was developed to obtain the criterion of the whirl
instability caused by whirl moments.
As shown in Fig. 28, the vibration of an overhung rotor shown in (a) can be
decomposed to the whirling motion shown in (b) and the precession motion shown
in (c). When the overhung is large compared with the lateral displacement, the
displacement due to precession motion can be neglected as compared with
the displacement due to whirling motion. This assumption was made also in the
130 Y. Tsujimoto
Fig. 29 Schematic of
experimental facility
analysis in Sect. 3. The fluid force moment due to the leakage flow was measured
on the facility shown in Fig. 29.
The backshroud of the runner is modeled by a disk set close to the casing. The
forced whirl motion was given by a special bearing system supporting the shaft and
the leakage flow was given by an external pump. Figure 30 shows the
whirl
moment normalized by the reference moment M0 = jjjj C2 vl RT where4
C2 = 4 mm is the axial clearance, vl is the mean leakage flow velocity in the radial
clearance C1 = 1 mm and RT = 149.5 mm is the radius of the disk (Song et al.
2010a, 2010b). The moment was measured by a force balance attached to the shaft
and also evaluated by the unsteady pressure distribution on the casing. The results
agree qualitatively. The moment was evaluated also numerically using a bulk flow
model (Childs 1989) in which the flow in the clearance is averaged over the
clearance and the effect of wall stress is taken into account. Although quantitative
difference is found, the tendency can be reproduced by the model. The rotordy-
namic analysis in Sect. 3 shows that the whirl instability occurs when the normal
moment Mn has the same sign as the whirl frequency . The result in Fig. 30 shows
that the moment can destabilize both forward and backward whirl at any whirl
frequency. This is quite different from the rotordynamic forces which destabilize
only forward whirl with the frequency less than a certain percentage of the rotor
frequency, as discussed in Sect. 4.
To demonstrate the moment whirl, free vibration tests were carried out using a
facility similar to that shown in Fig. 2. The length of the main shaft 1000 mm is
signicantly larger than the seal radial clearance C1 = 1 mm of the disk with the
diameter of 299 mm. The maximum axial displacement of the disk 0.15 mm =
Rotordynamics of Turbopumps and Hydroturbines 131
Fig. 30 Fluid force moments obtained by the force sensor, the unnsteady pressure, and the
computation
n
n
n
30 Hz n
Fig. 31 Spectrum of shaft vibration, Leakage flow rate 9 l/min, natural frequency
n 2 = 2.35 Hz
150 mm/1000 is signicantly smaller than the axial clearance between the disk and
the back casing, C2 = 4 mm. This veries the assumption that the tilting motion of
the disk can be neglected as compared with the lateral motion, which was made in
the simplied analysis in Sect. 3 as Eq. (5).
The spectrum of shaft vibration with the leakage flow rate of 9 l/min and the
natural frequency of the vibration n 2 = 2.35 Hz is shown in Fig. 31 for two
different initial condition of the shaft vibration.
The results of the test with zero initial velocity without displacement are shown
in (a). The rotational speed and the frequency of shaft vibration are normalized
by the natural frequency n of the shaft. The amplitude of natural frequency
132 Y. Tsujimoto
component with becomes larger when the rotational speed exceeds 30 rpm or
n = 0.21. The direction of whirl was found to be the same as the disk rotation
and the amplitude was the maximum value allowed by the restrictor on the shaft.
We can observe the rotational frequency component with n = n although the
magnitude is much smaller.
The results with the initial velocity in the direction of forward whirl with
maximum displacement are shown in (b). Similar results are obtained also with the
backward initial velocity. By giving the initial velocity, the vibration with natural
frequency with n = 1 occurs at all rotational speed n of the disk. The
direction of whirl is the same as the initial velocity.
These results show that the moment due to the leakage flow causes both forward
and backward whirl, at all rotational speed of the disk. This is consistent with the
rotordynamic moment shown in Fig. 30a and is quite different from the rotor whirl
due to rotordynamic forces discussed in Sect. 4, in which only forward whirl occurs
when the rotational frequency is sufciently higher than the rst critical speed. The
whirl direction is the same as that given as the initial condition. When no initial
velocity is given, impeller rotation triggers forward whirl if the rotational speed is
sufciently high. This may be caused by the rotordynamic forces due to the
interaction of the swirling flow induced by the disk rotation with the uneven
clearance due to displacement.
Similar tests were conducted also for outward flow, with inner and outer seals. It
was conrmed that the self-excited vibration occurs when the seal is placed at the
entrance of the leakage flow, outer seal with inward flow and inner seal with
outward flow. However, exit seal can be destabilizing if the seal is designed so that
the seal clearance is increased with the deflection with which the clearance of
normal design is decreased.
Two examples of vibrations at higher frequency than the rotational frequency are
introduced here. In the Deriaz pump (Ruud 1976) shown in Fig. 32, violent
counter-rotation shaft whirl with the frequency of 3.37 Hz occurred during the
shutdown transient from the nominal rotating speed of 120 rpm or 2 Hz, after
breaking the siphon downstream and the power shutoff. Even when the shaft is
stopped, the vibration at 5 Hz was observed although the amplitude is decreased.
This is quite different from the rotordynamic problems in high speed centrifugal
pumps discussed in Sect. 4 where forward whirl is expected at the rotational speed
higher than the rst critical speed. Since a rotating pressure pattern was observed in
the space between the head cover and the top of the impeller when it is lled with
water, air was injected into the space from the ports (a)(e) in Fig. 32. Injection
from a and e helped to mitigate the vibration. At nominal speed of 120 rpm, it
Rotordynamics of Turbopumps and Hydroturbines 133
is expected that there is sufcient pumping action to remove water from this space
but when the rotational speed decreased during shut down, the pumping action is
decreased and the space is flooded resulting in the vibration. Thus, it is quite
possible that the vibration was caused by the moments caused by the leakage flow.
Next example is the vibration of a hydraulic turbine runner (Crawford and Ruud
1967) around 11.58 Hz, at the rotational speed of 200 rpm or 3.33 Hz. It is stated
The typical whirl was in the direction of rotation, while counter-rotational whirl
was relatively rare. The turbine has a xed plate behind the runner, as shown in
Fig. 33. In the space between the xed plate and the runner, outward flow occurs
due to the centrifugal force on the swirling flow in the space driven by the runner.
An orice is placed at the inner radius of the xed plate. If the runner moves to the
right, the clearance is decreased and the pressure in the space would decrease and
results in the self-excited vibration as explained in Sect. 5. However, the authors of
the paper focus on the disturbance of flow due to the change in the thickness of the
gap between the xed plate and the runner and the gyroscopic effect of the runner.
We need to make quantitative analysis to determine which of the mechanisms is
more important. Air injection was used also for this case to mitigate the vibration.
134 Y. Tsujimoto
6 Conclusion
Rotordynamic forces on pump impellers and turbine runners are reviewed. In many
cases where the leakage flow swirl is important, rotordynamic forces causes for-
ward whirl at supercritical speeds where the rotor is running more than twice the
rst critical speed. For this case the destabilizing region can be reduced by reducing
the swirl velocity. However, when the change of the leakage flow due to dis-
placement is important, the fluid force moment can cause rotor whirl even when the
rotor speed is signicantly lower than the rst critical speed. The instability can be
avoided by placing the seal at the exit of the leakage flow or changing the
opening/closing convention of the seal associated with the deflection of the rotor.
So far it was difcult to predict the rotordynamic forces by CFD since it is
usually small as compared with steady force component. However, it is becoming
possible to predict the forces by careful application of CFD. Further progress is
expected in this eld making use of CFD.
References
Tsujimoto, Y., Acosta, A. J., & Yoshida, Y. (1988b). A theoretical study of fluid forces on a
centrifugal impeller rotating and whirling in a vaned diffuser. NASA CP, 3026, 307322.
Vance, J. M. (1988). Rotordynamics of turbomachinery. New York: Wiley.
Yoshida, Y., Tsujimoto, Y., Murakami, Y., & Ito, T. (1992). Self excited and forced vibrations of
centrifugal impeller operating at supercritical speed. In Proceedings of 4th ISROMAC
(pp. 1123).
Yoshida, Y., Tsujimoto, Y., Ishii, N., Ohashi, H., & Kano, F. (1999). The rotordynamic forces on
an open-type centrifugal compressor in whirling motion. ASME Journal of Fluids Engineering,
1212, 259265.
On the Preliminary Design and
Performance Prediction of Centrifugal
TurbopumpsPart 1
The present work has been supported by the European Space Agency under Contract
No. 40001025856/10/NL/SFe. The authors would like to express their gratitude to Dr. Giorgio
Saccoccia of ESA-ESTEC for his constant and friendly encouragement.
1 Introduction
Finally, Euler methods use similar numerical approaches to the solution of the
ideal fluid equations, without necessarily requiring the flow to be irrotational.
Clearly, all inviscid methods are inherently uncapable to account for real fluid
effects and dissipative phenomena such as turbulence, boundary layers, separation,
flow reversal, and secondary flows. In order to address these aspects of turboma-
chinery flows, viscous methods must be used, which can be generally divided into
distributed loss, boundary layer, and NavierStokes methods.
Distributed loss methods are based on the idea of correcting the inviscid flow
models by accounting for dissipative effects in complex turbomachinery flows on
an averaged basis, without detailed consideration of the specic mechanisms and
locations where energy dissipation actually occurs (Ainley and Mathieson 1951;
Dunham and Came 1970; Horlock and Marsh 1971; Bosman and Marsh 1974;
Kacker and Okapuu 1982). The appropriate intensity of dissipative effects strongly
depends on the operational conditions and must be either deduced from experi-
mental observations or, in more sophisticated models, estimated from the analysis
of the various sources of dissipation. Clearly, a sizable amount of information is
missed in this intrinsically phenomenological approach. Therefore, distributed loss
methods are of little or no use when accurate results are required as, for instance, in
direct optimization of turbopump design where the sensitivity of the loss model to
minor changes of the controlling parameters is indispensible. However, when
supported by adequate experimental data, these models represent an economic and
rapid way to include in rst approximation the global effects of energy dissipation
in turbomachinery analyses.
Boundary layer methods are based on Prandtls original intuition that viscous
effects in unseparated flows at high Reynolds numbers are conned to relatively
thin layers of the fluid adjacent to the solid surfaces, while the rest of the flow is
virtually inviscid. Therefore, the flow in the two regions can be studied with sep-
arate approaches and the solutions matched together in order to obtain an
approximate description of the entire flow eld. Boundary layers have been studied
extensively and a number of reliable boundary integral methods exist for their
efcient evaluation in the steady 2D case (White 2006; White and Christoph 1972).
The effects of surface roughness, free stream turbulence and, with some uncertainty,
turbulent transition can also be included. The problem of steady viscous/inviscid
coupling between the two flow regions can either be neglected in the case of weak
interaction, or otherwise treated iteratively in order to satisfactorily match the inner
and outer flow solutions (Carter 1979; Le Balleur 1981; Whiteld et al. 1981).
Boundary layer methods in two dimensions are relatively simple and, in the absence
of flow separation, accurate and computationally efcient. On the other hand,
extensions to three-dimensional and/or unsteady cases are more difcult and
uncertain because of the lack of reliable experimental correlations for solving the
integral boundary layer equations and because of the complexity of the
computations.
NavierStokes methods solve the viscous flow equations, and therefore represent
the most general and comprehensive approach to the analysis of turbopump flows.
Current methods typically employ either the Reynolds-Averaged NavierStokes
140 L. dAgostino et al.
The observation that none of the above methods satisfactorily matches these
requirements justies the search for more efcient alternatives. In this context, the
development of accurate 3D, closed-form, reduced-order models capable of jointly
predicting the geometry and performance of radial impellers and turbopumps is of
particular interest to rocket engineers in their search for effective, rapid, and pos-
sibly accurate tools for the preliminary design of these machines. However, no such
model has been proposed so far, mainly due to the difculty of adequately
describing the 3D floweld through the impeller. Hence, even today turbopump
designers still refer to simple rules of thumb or the general indications of specic
manuals (Douglass 1973) for the preliminary denition of their machines.
In rocket engine turbopumps the attainment of high power/weight ratios is
invariably obtained by running the impeller at the maximum allowable speed and
lowest shaft torque. Operation under limited cavitation conditions with lighterbut
also more flexibleshafts is therefore tolerated, exposing rocket propellant feed
turbopumps to the onset of dangerous fluid dynamic and rotordynamic instabilities
(dAgostino 2013a, b). However, the development of cavitation inside the blade
channels should possibly be avoided, because the highly compressible and rever-
berating nature of the flow would greatly promote the onset of potentially lethal
self-sustained flow instabilities at frequencies susceptible to become resonant with
the flutter oscillations of the impeller blades. The occurrence of this phenomenon in
the inducer of the LE-7 engine has actually been identied as the likely cause of the
catastrophic failure of the Japanese H-II launcher in November, 1999 (NASDA
2000a, b). It is therefore advisable to design the impellers, both axial and radial, for
cavitation to concentrate on the initial part of the blades at all operational condi-
tions. This can be realized if, for any positive value of the inlet flow incidence, the
fully wetted flow pressure on the suction sides of the blades is minimum at the
leading edges, and monotonically increasing further downstream. Hence, in par-
ticular, the local value of the blade lift vanishes identically at zero incidence, when
the stagnation point is located at the leading edge.
These considerations provided the fundamental basis for the selection and jus-
tication of the floweld assumptions adopted by dAgostino and his collaborators
in the development of their closed-form methods for jointly dening the geometry
and performance of high-head radial inducers (dAgostino et al. 2008a, b) and
mixed-flow impellers (dAgostino et al. 2011) optimized for operation under lim-
ited cavitation conditions. Specically, they rst showed that in helical inducers the
condition of minimum flow pressure at the leading edges, when associated with the
preservation of radially uniform axial flow velocity through the rotor (as commonly
required in axial turbomachines), results in a functional relation between the local
values of the blade angle and hub radius. Later, with a similar approach dAgostino
et al. (2011) demonstrated that, when applied to mixed-flow centrifugal impellers,
the same condition links together the axial variations of the blade pitch and
backsweep angle.
The present formulation of the preliminary design of mixed-flow turbopumps
represents therefore the natural extension of the earlier model developed by some of
the authors for the simultaneous geometry denition and noncavitating performance
142 L. dAgostino et al.
The present model jointly describes the flow and geometry of the three main
components of a radial machine: the impeller, the diffuser, and the volute, as
schematically indicated in Fig. 1.
The radial impeller transfers energy to the working fluid as a consequence of its
rotational speed and the hydrodynamic forces developing on its blades. The
geometry of its flow channels is dened by the intersection of N blades, with
variable helical and backsweep angles h and , and the rotational surfaces gen-
erated by the axial proles of the hub and tip radii, rH and rT . The exit flow of the
impeller is then collected by a vaneless diffuser with constant axial width bD , and
nally guided into the discharge line by a single-spiral volute with elliptical cross
sections, continuously varying in the azimuthal direction from a straight segment at
the tongue to a circular cross section with radius R4 at the exit.
As mentioned in the introduction, the flow through the machine is considered as
incompressible, inviscid, and irrotational. For relatively large values of the blade
On the Preliminary Design and Performance Prediction 143
solidity the 3D velocity eld inside the blade channels is approximated as the
superposition of a fully guided axisymmetric flow and a 2D axial vorticity cor-
rection on each impeller cross-sectional orthogonal to the centerline. The merid-
ional component of the fully guided axisymmetric flow is chosen as the velocity
eld of an incompressible axisymmetric stagnation flow with stagnation plane
located at the station (st), as illustrated in Fig. 1. Perfect mixing is assumed to take
place at the exit of the impeller, so that the flow in the diffuser can be considered
steady, axisymmetric, and axially uniform. Finally, steady axisymmetric flow is
also assumed in the volute at design flow rate. Consistently with the ideal nature of
the present model, mixing losses at the entrance of the diffuser and the volute are
neglected, even if they could be evaluated without explicit consideration of real
flow effects.
In the stated assumptions, the ideal flow through the impeller is held by the conti-
nuity and irrotationality equations for the velocity u = u + u, sum of the (rotational)
fully guided axisymmetric flow eld u, which generates the volumetric flux through
the machine and most of its total head rise, and a 2D cross-sectional slip velocity
correction u (Fig. 2), which does not contribute to the flowrate but is necessary to
satisfy the irrotationality condition and decreases the fully guided head rise.
Fully Guided Flow and Geometry of the Impeller. With reference to Fig. 3,
the azimuthal velocity component of the fully guided flow is expressed by
Fig. 2 Schematic of the 2D cross-sectional slip flow in the impeller blade channels
Fig. 3 Helical angle in the meridional plane (left), backsweep angle in the axial cross-sectional
plane (center), and velocity triangle in the blade-to-blade plane (right)
in terms of the radial and axial velocities u and w , where the angle , measured from
the axial direction, includes the effects of the axial variations of the helical pitch Ph
and of the logarithmic backsweep angle of the blades:
2r d r 2r
tan = +r ln tan =
Ph dz rbs P
In this expression rbs z is the radius where the azimuthal coordinate bs (as-
sociated to the axial variation of ) vanishes (Fig. 3) and P is an equivalent helical
pitch of the blades accounting for the combined axial changes of Ph and .
The relatively narrow range of variation of the radial coordinate
rH r rmax = minfrT , r2 g in the blade channels justies the approximation
p
ln r ln r = ln rH rmax , so that P is only function of the axial coordinate z.
Hence, in particular, the equivalent pitch of the blades reduces to P Ph z when
choosing rbs r .
The 2D slip velocity components on each axial cross section of the impeller are
most synthetically expressed and solved in terms of a scalar axisymmetric stream
function r , , z in the rotating cylindrical coordinates r = r, = t, z = z.
On the Preliminary Design and Performance Prediction 145
Moreover, in the proposed approximation the axial variation of the slip flow is
neglected, so that z 0. Imposition to the assumed velocity eld u = u + u of
the irrotationality condition along the radial, azimuthal, and axial coordinates yields
w dw
4 tan = 0
z P dz
2
d w
=0
dz2
1 1 2 w dw
r + 2 2 = 2 4 + tan
r r r r P dz
and
1 dw 1 w 1
ur = r = r
2 dz 2 z1 zst
which, in turn, dene the proles of the hub and tip radii rH z and rT z as the
corresponding streamlines through rH1 , z1 and rT1 , z1 .
The blade surfaces are dened by the azimuthal coordinate B resulting from the
combined effects of the helical pitch and backsweep:
Z z
2 r
B = h + bs = B1 dz ln tan
z1 Ph rbs
where B1 (possibly function of r) is the azimuthal position of the blade at the inlet
section z = z1 . The adoption of this blade shape yields a forced vortex flow design of
the impeller with
2r 1 dw
v = v r = w + r tan r
P 2 dz
It is worth noting that the radial component of the vorticity equation implies that
the azimuthal components of the fully guided flow velocities do not depend on the
axial coordinate:
146 L. dAgostino et al.
w dw v v
4 tan = 0 = =0
z P dz z z
as a function of the axial schedule of the blade pitch Pz (or viceversa). For
simplicity, in the present analysis a cubic variation of 1 P has been assumed, with
nite inlet pitch (P1 ) and vanishing rst and second axial derivatives of tan at
the blade leading edge (z = z1 );
innite pitch (1 Pst = 0) at the stagnation plane z = zst .
The rst condition assures a second-order smooth transition of the blading into a
helical surface of assigned pitch at the impeller eye, while the second has been
imposed to control the angle of the blade root w.r.t. the hub surface at the rotor exit.
Although these conditions seemed suitable for rst validation of the proposed
model, they have not been optimized and further renements can possibly lead to
better results in the generation of more efcient and realistic impeller bladings.
Impeller Slip Flow. The Poissons boundary value problem for the stream
function r, of the slip flow on each axial cross section of the blade channels
(see Fig. 4), together with the impermeability condition = 0 on the boundaries, is
transformed in a rectangular domain, where it can be solved in closed form by
standard spectral methods (Hildebrand 1976).
In particular, using as comparison functions the (orthogonal) eigenfunctions of the
corresponding homogeneous problem for the Laplaces equation, the solution writes
+ + lnr rH 2n 1N Bk
= Cm, n sin n sin
m=1 n=1 lnrmax rH 2
with
In the stated assumptions, the steady, axisymmetric, axially uniform flow in the
diffuser is held by the continuity and momentum equations. In order to satisfy the
condition of axially uniform flow in the diffuser, perfect mixing of the impeller
discharge flow is assumed to occur at the diffuser inlet, with flow velocity equal to
the mass-averaged velocity at rotor discharge, station (2). The radial and azimuthal
components of the velocity inside the diffuser can then be computed by radially
integrating the continuity and azimuthal momentum equations in cylindrical
coordinates. The resulting flow inside the diffuser is characterized by constant axial
velocity and log spiral streamlines with flow angle (D ) with respect to the radial
direction in a cross-sectional plane. Finally, the flow streamlines through the hub
and tip surfaces of the impeller at its discharge station (2) dene the lateral surfaces
of the diffuser in the meridional plane.
With reference to Fig. 5, the geometry of the spiral volute is designed for alignment
of the tongue with the flow leaving the diffuser at nominal operational conditions
and smooth transition to azimuthal flow discharge at the exit section, station (4).
The meridional cross sections of the volute are assumed to be segments of
ellipses of width b3 = bD at the diffuser exit radius r3 , transitioning from a (de-
generate) straight segment at the tongue ( = 0) to a segment of a circle of radius R4
at the exit of the volute (station 4, = 2), as schematically illustrated in Fig. 6.
In the assumption of axisymmetric flow in the volute at design flow conditions,
integration of the azimuthal momentum equation results in a free vortex distribution
of the azimuthal velocity component. The volumetric flux can then be computed on
each meridional cross section of the volute and equated to the radial inlet flow rate
from the diffuser. If, in addition, the external radius rV of the volute is assigned, the
above condition fully denes the geometry of the volute as a function of the
azimuthal angle . Here, in particular, the following expression for
148 L. dAgostino et al.
1 + 1
rV = r3 exp
tan D 2 1 +
with
1
2
= 1
tan D lnrV4 r3
has been used, corresponding to a smooth variation of the spiral angle V of the
volute from D at the tongue ( = 0) to zero at the exhaust cross section ( = 2).
The average flow velocities in the volute have been approximated as
A3 rV4
tan v = u3 and u = v tan V
A 4 rV
On the Preliminary Design and Performance Prediction 149
Finally, in the assumption of a perfect mixing in the outlet duct, the flow at the
discharge section
(station 5) of the machine is taken uniform with velocity
u5 = V R24 .
3 Turbopump Performance
In the present work the pumping performance of the machine has been evaluated
neglecting all sources of energy dissipation, consistently with the ideal flow
assumption originally introduced to solve for the machine geometry and floweld.
Clearly, the major contributions to fluid dynamic losses (typically arising from
viscous and secondary flow effects, turbulent mixing, and flow incidence at the
leading edges of wetted surfaces) play a crucial role for realistic performance
prediction and optimization of turbopumps. Their inclusion is illustrated in detail in
a companion paper of the present volume (dAgostino et al. 2017), where it is
shown to lead to excellent predictions of the measured performance of the machine
over a wide range of operation above and below design conditions.
Under the stated assumptions, the impeller is the only element affecting the total
pressure of the flow through the machine. Hence, in the reference frame r, , z
rotating with the impeller the (steady) pressure of the (absolutely) irrotational flow
in the blade channels is obtained from the Bernoullis equation:
1 1
p+ u u r r = pt0
2 2
where u = u r is the relative velocity and pt0 is the total pressure on the
centerline (r = 0) at the upstream station (0).
In the absence of losses the total pressure of the flow downstream of the impeller
is uniform. Therefore, using the Eulers equation, the pressure in the diffuser and
the volute is computed from the relevant flow velocity u by means of
1
p+ u u = pt0 + r2 v2 = pt5
2
dependence between the axial schedules of their backsweep and helical pitch
angles. The relevant conservation equations and assumptions also parametrically
determine the shape of the remaining components (diffuser and volute) and the total
head rise of the machine. Both the geometry and performance of the turbopump
have therefore been dened in terms of a reduced number of controlling parameters.
4 Model Discussion
In order to illustrate the versatility of the proposed design approach, three examples
of pump geometries, indicated as A, B, and C, and representative of mixed-flow
machines with different shapes and number of blades have been generated for
suitable inputs of the free design parameters of the model. The 3D renderings of the
three pump samples A, B, and C and the meridional cross sections of their impellers
are shown in Figs. 7, 8, and 9, while Table 1 summarizes their main geometrical
characteristics.
Hub, Tip
0.15
0.1
r (m)
0.05
0
0 0.02 0.04 0.06 0.08
z (m)
Fig. 7 3D rendering of the impeller of the rst sample of centrifugal pump geometry (A, left) and
meridional cross section of the impeller (right)
Hub, Tip
0.1
r (m)
0.05
0
0 0.02 0.04 0.06
z (m)
Fig. 8 3D rendering of the second sample of centrifugal pump geometry (B, left) and meridional
cross section of the impeller (right)
On the Preliminary Design and Performance Prediction 151
r (m)
0.05
0
0 0.02 0.04 0.06 0.08 0.1
z (m)
Fig. 10 Comparison 1
between the predicted
noncavitating performance of Geometry A
the sample turbopump 0.8 Geometry B
geometries A, B, and C Geometry C
0.6
0.4
0.2
0
0 0.05 0.1 0.15 0.2 0.25 0.3
In the present inviscid flow approximation, the deviation of the flow leaving the
impeller w.r.t. the exit angle of the blades is the major source of head degradation in
centrifugal turbopumps. In particular, flow deviation at the impeller outlet is mainly
due to the slip flow effect generated by the irrotationality condition and the
imperfect guidance of the flow in the blade channels, especially at higher loads
(Peterson and Hill 1992). Of these two effects, the former is actually predicted by
the present model and can be assessed here, while the latter is rather small in high
solidity bladings like those typical of centrifugal impellers with large backsweep
angles, and in rst approximation can be neglected.
In order to preliminarily assess the performance of the ideal flow model before
inclusion of fluid dynamic losses and nal validation against experimental data
from relevant turbopump geometries, the predicted slip factors have been compared
to the results of the theoretical/semi-empirical methods reported in literature for
centrifugal pumps (Stodola 1927; Busemann 1928; Wiesner 1967; Dixon 1978;
Ferguson 1963; Wislicenus 1947).
In particular, as shown in Figs. 11 and 12, the comparison has been focused on
the influence of the main parameters affecting the slip factor, such as the number of
blades and the exit backsweep angle. The results indicate that the model usually
slightly overestimates the slip factor at low number of blades and small exit
backsweep angle. However, it is worth noticing that the geometry obtained by the
model is more complex than the innitely thin, logarithmic spiral blades used by
Busemann to compute his theoretical slip factors. Moreover, the approximate for-
mula proposed by Stodola provides a reasonable rst approximation of the more
exact results of Busemann, but underestimates the slip factor when applied to radial
impellers with a small number of blades. Both of these methods do not account,
even indirectly, for viscous effects:
Fig. 11 Comparison 1
between the slip factor
computed from the model and
the available semi-empirical 0.8
formulas for radial impellers
based on the hub and tip 0.6
slip factor
Fig. 12 Comparison 1
between the slip factor
computed from the model and
the available semi-empirical 0.8
formulas for radial impeller
geometries based on geometry 0.6
slip factor
C with variable number of
blades
0.4
Model
Busemann
0.2
Stodola
Wiesner
0
5 10 15 20
number of blades
AB + BB 2 tan 2
B = Busemann
1 2 tan 2
0.63 N
S = 1 Stodola
1 2 tan 2
p 0.7
cos 2 N
W = 1 Wiesner
1 2 tan 2
On the other hand, the approach proposed by Wiesner is based on empirical slip
factor data. It provides better agreement with the experimental measurements and
implicitly accounts for the additional contributions of viscous effects. Figures 11
and 12 show that the predictions of the present model well reproduce the trend of
Wiesners results but, not surprisingly, are systematically higher because of the
neglect of viscous effects. The introduction of the fluid viscosity decreases the value
of the slip factor as a consequence of the acceleration and the additional deviation
of the relative flow at the rotor exit induced by blockage and asymmetric boundary
layer displacement effects on the two sides of the impeller blades. Both of these
effects produce in fact the systematic reduction of the azimuthal component of the
absolute flow velocity at the exit of the impeller, and therefore of the head
developed by the machine, justifying the observed discrepancy between Wiesners
semi-empirical data and the theoretical results of the present inviscid flow model.
5 Conclusions
The present theoretical model proved to be capable of rapidly and efciently pro-
viding quantitative indications for the geometry denition, the 3D floweld
description, and the prediction of the ideal noncavitating pumping characteristics of
154 L. dAgostino et al.
References
Adler, D., & Krimerman, Y. (1978). The complete 3-dimensional calculation of the compressible
flow eld in turbo impellers. Journal of Mechanical Engineering Science, 20, 149.
Ainley, D. G., & Mathieson, G. C. R. (1951). A method of performance estimation for axial flow
turbines. British ARC, R&M, 2974, 923952.
Bosman, C., & Marsh, H. (1974). An improved method for calculating the flow in turbomachines,
including a consistent loss model. Journal of Mechanical Engineering Science, 16, 2331.
Bosman, C., & El-Shaarawi, M. A. I. (1976). Quasi-three-dimensional numerical solution of flow
in turbomachines. ASME Paper 76-FE-23.
Brennen, C. E. (1994). Hydrodynamics of pumps. Oxford University Press.
Busemann, A. (1928). Das Frderhhenverhltnis Radialer Kreiselpumpen mit Logarithmisch-
Spiraligen Schaufeln. Weinheim: WILEY-VCH Verlag GmbH & Co. KGaA.
Carter, J. E. (1979). A new boundary layer inviscid interaction technique for separated flow. AIAA
Paper 791450.
Chung-Hua, W. (1952). A general theory of three-dimensional flow in subsonic and supersonic
turbomachines of axial. NACA, TN: Radial and Mixed Flow Types. 2604.
Cervone, A., Pace, G., Torre, L., Pasini, A., Bartolini, S., Agnesi, L., & dAgostino, L. (2012).
Effects of the leading edge shape on the performance of an axial three bladed inducer. In 14th
international symposium on transport phenomena and dynamics of rotating machinery,
ISROMAC-14, Honolulu, HI, USA.
dAgostino, L., Pasini, A., Valentini, D., Pace, G., Torre, L., & Cervone, A. (2012). A reduced
order model for optimal centrifugal pump design. In 14th international symposium on transport
phenomena and dynamics of rotating machinery, ISROMAC-14, February 27thMarch 2,
Honolulu, HI, USA.
dAgostino, L. (2013a). Turbomachinery developments and cavitation. In STO-AVT-LS-206, paper
NBR 121, VKI Lecture series on fluid dynamics associated to launcher developments, von
Karman institute of fluid dynamics, Rhode-Saint-Gense, Belgium, April 1517.
On the Preliminary Design and Performance Prediction 155
White, F. M., & Christoph, G. H. (1972). A simple theory for the two-dimensional compressible
boundary layer. ASME Journal of Basic Engineering, 94, 636642.
Whiteld, D. L., et al. (1981). Calculation of turbulent boundary layers with separation and
viscous-inviscid interaction. AIAA Journal, 19, 13151322.
Wiesner, F. J. (1967). A review of slip factors for centrifugal impellers. ASME Journal of
Engineering for Power, 89, 558576.
Wislicenus, G. F. (1947). Fluid mechanichs of turbomachinery. New York: McGraw Hill.
On the Preliminary Design and
Performance Prediction of Centrifugal
TurbopumpsPart 2
Abstract The ideal flow model for the preliminary design and performance pre-
diction of radial turbopumps presented in the companion paper of the present
volume (dAgostino et al. 2017) is here interfaced with the calculation of the
boundary layers inside the blade channels and other major forms of flow losses,
with the aim of developing an effective tool for rapid parametric optimization of the
machine performance and geometry under appropriate design constraints, such as
assigned values of the specic speed, flow coefcient, and blade solidity.
A mixed-flow turbopump, with a six-bladed impeller, a vaneless diffuser, a
single-spiral volute and nondimensional performance characteristics similar to those
typically used in liquid propellant rocket engine feed systems, has been designed,
parametrically optimized and manufactured in accordance with the indications of
the present model. The pumping and suction performance of the machine has been
determined in a series of tests in the cavitating pump rotordynamic test facility
(CPRTF). Under fully wetted flow conditions the measured pumping characteristics
of the machine (hydraulic head and efciency as functions of the flow coefcient)
proved to be in excellent agreement with the model predictions, thus successfully
conrming the validity of the proposed model as an effective tool for rapid and
efcient design of high-performance centrifugal turbopumps.
The present work has been supported by the European Space Agency under Contract
No. 40001025856/10/NL/SFe. The authors would like to express their gratitude to Dr. Giorgio
Saccoccia of ESA-ESTEC for his constant and friendly encouragement.
1 Introduction
The main fluid dynamic losses to be considered for more realistic performance
prediction and optimization of mixed-flow centrifugal turbopumps typically arise
from flow incidence at the leading edges of the impeller blades and the volute
tongue, from viscous and secondary flow effects in the impeller, and from turbulent
mixing in the diffuser and the volute, as schematically indicated in Fig. 1.
160 L. dAgostino et al.
Their contributions have been included in the present model and are illustrated in
detail in this Section.
Notice that, in order to provide effective indications for parametric optimization
of the machine, the evaluation of fluid dynamic losses must be sensitive to specic
and relatively small changes of the geometry of the various components. Standard
empirical loss correlations, such as those commonly used for preliminary assess-
ment of viscous, slip flow, and turbulent mixing effects in centrifugal impellers, are
therefore inadequate to this purpose, and have therefore been replaced here by more
detailed and realistic loss models.
2.1 Impeller
Incidence losses due to the sudden change of the flow direction at the leading edge
of the inducer blades are expressed in terms of the nondimensional equivalent
length Leq = Dch 3 (function of the turning angle , in degrees, evaluated on
the mean streamline) by means of:
Leq 1 2
ptI1 = f V ,
Dch 2 1
where is the density of the liquid, V1 is the relative flow velocity at the mean inlet
radius and the friction factor f is obtained from Moodys chart using the Reynolds
number based on the hydraulic diameter Dch and the relative roughness of the blade
channel.
Viscous effects developing on the surfaces of the blade channels generate fric-
tional losses, flow blockage and, as a consequence of the asymmetric growth of the
boundary layers on the pressure and suction sides of the blades, deviation of the
flow at the impeller discharge. Moreover, the effective flow cross-section is also
reduced by the blade thickness.
On the Preliminary Design and Performance 161
In order to account for the influence of these aspects on the pumping perfor-
mance of the machine, the evolution of the boundary layers has been evaluated
along the four streamlines passing through the mid-points of the blade channel sides
at the impeller exit. The results so obtained have been considered as representative
of the average viscous effects on the corresponding surfaces, as schematically
illustrated in Fig. 2.
Consequently, the flow blockage at the impeller exhaust has been approximated
by the following expression:
" !
1 2r 2 *
2r 2 *
2ps
BBL2 * 2ss + *T2
Ach H2 N cos 2 N cos 2
#
* b2 *T2 *2ps b2 *H2
+ 2ss + ,
cos 2 cos 2
where *2ss , *2ps and *H2 , *T2 are respectively, the displacement BL thicknesses on
the suction and pressure sides of the blades and on the impeller hub and tip, while
b2 is the exhaust axial blade height, N is the blade number and Ach is the nominal
channel area. Then the flow velocity at the impeller exit, corrected for blockage
effects, is:
u2
u2b =
1 B2
where B2 = BB2 + BBL2 is the total blockage at the impeller exit r = r2 , sum of the
contributions generated by the blade thickness and the boundary layers.
162 L. dAgostino et al.
The flow deviation generated by the boundary layer growth at the impeller
exhaust mainly affects the azimuthal velocity and is expressed by:
!
1 d*ss d*ps
2 = sin ss2 sin ps2 ,
2 dx 2 dx
2
where x is the curvilinear coordinate along the streamline and is the angle between
the rotational axis and the streamline in the meridional plane. With these correc-
tions, the relative azimuthal velocity at the impeller discharge becomes:
v2b = u2b tan tan 1 v2b u2b + 2
Similarly, with the same approach used for the evaluation of viscous blockage
effects (Fig. 2), the total pressure loss at the impeller exhaust due to boundary layer
friction is expressed by:
2 2
2r2 UH2 2r2 UT2
ptBL2 H2 2ss + T2 2ps
Ach N cos 2 2 Ach N cos 2 2
2
2ss bI2 T2 Uss2 b U 2
+ + 2ps I2 H2 ps2 ,
Ach cos 2 2 Ach cos 2 2
where 2ss , 2ps and H2 , T2 are the momentum boundary layer thicknesses on the
suction and pressure sides of the blades and on the impeller hub and tip and U is
the streamwise velocity of the free stream in the relative frame.
The streamlines for the evaluation of the boundary layers along the impeller
blade channels have been determined by numerical integration of the relevant
ordinary differential equations:
dr u
= p
dx u2 + v2 + w2
d v
= p
dx r u2 + v2 + w2
dz w
= p
dx u2 + v2 + w2
using the relative velocity eld obtained from the ideal flow model and the
appropriate (nal) conditions at the impeller discharge (Fig. 3).
The governing equations for the laminar incompressible boundary layers on the
impeller blades have then been obtained by writing the relative continuity and
momentum equations:
On the Preliminary Design and Performance 163
Fig. 3 Relative streamlines of the inviscid flow on the impeller blade channels for turbulent BL
computations
u = 0
1 h T i
u u = p + u + u 2 u + 2 r
r0 ux uy
+ r0 =0
x y
ux ux 1 p 2 ux
uy + ux = + v 2 + 2 r sin cos ,
y x x y
1 p
0 + 2 r0 cos
y
where r0 is the radial distance from the axis, is the meridional angle of the
boundary layer surface from the axial z-direction and is the angle of the boundary
layer free streamline from the meridional plane.
The pressure can now be eliminated by differentiating the steady Bernoullis
equation:
164 L. dAgostino et al.
p 1
+ U U 2 r r = constant
2
1 p dU dr0 dU
U 2 r U 2 r0 sin cos
x dx dx dx
1 p dr
= 2 r 2 r0 cos ,
y dy
ux ux dU 2 ux
uy + ux =U +v 2
y x dx y
(independent on ).
Finally, the equations for the turbulent boundary layers on the impeller blades
can be obtained by standard Reynolds averaging in the form:
r0 ux r0 uy
+ =0
x y
ux ux dU 1 t
ux + uy =U +
x y dx y
where bars indicate the mean flow properties, primes the turbulent fluctuations and
t = ux y ux uy is the turbulent shear stress (equal to w at the wall).
These equations have been solved by extending the method proposed by White
(1974) to axisymmetric congurations, as required for application to mixed-flow
impellers. The basic idea of Whites method consists in eliminating uy with the
continuity equation and integrating the boundary layer momentum equation from
the blade surface y = 0 to the freestream y = . The velocity prole is approx-
imated by the logarithmic law of the wall:
1
u+ ln y + + 5.5 + wy + ,
0.4
ux uy yv*
u+ = ; v+ = ; y+ =
v* v* v
On the Preliminary Design and Performance 165
p
with frictional velocity v* = w and an outer wake wx, y + = 0.6y + x, linear
+
in y and in the pressure-gradient correlation parameter (Mellor 1966):
U dU
x =
w v* dx
33 H , d 1
=
dx* G, Re dx* U*
D
Here ReD = 2r22 v is the Reynolds number based on the impeller tip speed r2
and diameter 2r2 ,
Z +
G, = u + 2 dy +
0
and:
Z Z !
+ y+
u + u + u + +
H , = u +
+
dy dy +
0 y 0
+ * + +
= 2r2 ; * = 2r2 and = 2r2
U* ReD U* ReD U* ReD
166 L. dAgostino et al.
Their evolution along the blade channels of the VAMPIRE pump operating at
nominal conditions is illustrated in Fig. 4. The results are similar for , * and ,
which in turbulent boundary layers are nearly proportional to each other. Speci-
cally, the development of the boundary layer is predicted to proceed rather regularly
on the suction side of the blades and, even if not relevant to the present case of
unshrouded impeller as indicated later, also on the casing surface at the blade tip
radius. On the other hand, the evolution of , * and is slightly more irregular on
the pressure side of the blades and on the hub surface where, however, flow turning
effects tend to oppose the separation of the flow. The regular growth of the
boundary layer is indicative of a favorable pressure distribution and is especially
signicant on the suction side of the blades, where the flow is generally more
susceptible to separation. As a whole, the results of Fig. 4 fully conrm the validity
of the proposed method for the denition of hydrodynamically efcient impeller
geometries.
By supporting the blades on both sides, impeller shrouds can be quite useful to
reduce bending stresses and flutter instabilities. But they also add to the manu-
facturing complexity and weight of impellers, thus reducing their critical speeds.
Besides, the narrow recirculating leakage flow between their outer surface and the
machine casing increases ventilation losses and generates spurious and uncertain
rotordynamic effects. For these reasons, whenever possible unshrouded impellers
On the Preliminary Design and Performance 167
are typically favored in space applications (Droege et al. 2000). In these impellers
the boundary layers developing on the impeller casing are continuously wiped out
by passing blades. Therefore, an approximate but practical way to account for the
absence of the shroud in the present model consists in simply not considering the
boundary layers on the inner surface of the impeller casing.
The above approach allows for the quantitative evaluation of viscous effects
(blockage, deviation and losses) taking place in the flow through centrifugal
impellers as functions of their geometry and operating conditions. Perfect mixing is
assumed to occur at rotor discharge at the inlet cross-section of the diffuser. The
resulting total pressure losses have then been be evaluated as the mass averaged
integral difference between the dynamic pressures of the mean floweld at the
diffuser inlet and the nonuniform velocity distribution at the impeller discharge
station, that is:
Z
N 2 N 2
ptM2 z = v2 v2S d
4 0
where v2 is the local azimuthal component of the vorticity correction and vs is the
azimuthal slip velocity.
2.2 Diffuser
Friction on the side walls and the sudden increase of the flow passage area at the
exit of the impeller are the main sources of fluid dynamic losses in vaneless dif-
fusers. Frictional losses can be evaluated along the nominal streamlines (consisting
in log-spirals with constant angle from the radial direction) by means of the
expression:
1 2 2 r3 r3 r2
ptF3 = u3 + v 3 f D
2 2bD r2 cos
where bD is the width of the diffuser and the friction factor fD depends on the flow
Reynolds number based on the hydraulic diameter DHdiff = 2bD . Finally, diffusion
losses due to the sudden variation of the flow passage area from the impeller
discharge to the diffuser can be estimated using the following expression:
1 2 1 2
2 bI2
ptD2 = ub2 u2 = ub2 u2 2
2 2
2 2 bD
where ub2 and u2 are the average values of the radial velocity at the impeller exit
with and without corrections for blockage effects.
168 L. dAgostino et al.
2.3 Volute
The flow losses occurring in the volute typically include radial diffusion losses at
the volute entrance, incidence losses at the tongue, frictional losses on the walls,
azimuthal diffusion losses in the volute and mixing losses in the exhaust duct. The
volute floweld is relatively complex, being intrinsically 3D, unsteady, and, in most
cases, turbulent. Accurate prediction of the resulting losses would in principle call
for the simultaneous simulation of the entire flow through the turbopump, quite an
ambitious task, computationally very expensive and time-consuming. As antici-
pated in the rst part of this publication (dAgostino et al. 2017), such simulations
are indeed capable of evaluating the flow properties in the volute with spatial
resolution dependent on the size of the computational grid (Van den Braembussche
2006), but do not provide any direct indication for the selection of the machine
geometry, which must be assigned as a known input to the computations.
On the other hand, the present model is intended as a rapid and effective tool for
simultaneous and accurate denition of the geometry of the machine and prediction
of its fluid dynamic performance, as required in the preliminary phase of the design.
For most efcient denition of the model it is worth noting that total pressure losses
in the volutes of radial turbopumps are usually dominant, especially at off-design
conditions, with respect to the contributions from other components. This is
because the separated/swirling nature of the flow in the volute promotes large-scale
turbulent mixing and because the geometry of the volute can only be optimally
adapted to the impeller for one single value of the flow coefcient. Based on these
considerations, only incidence losses at the tongue and mixing losses in the volute
have been taken into account in the present model. Available evidence demon-
strates that this approximation is capable of providing accurate predictions of the
magnitude and trends of flow losses occurring in the volute of the test turbopump,
thus supporting the conclusion that it can be safely applied to similar machines
designed in accordance with the indications of the present model.
The total pressure loss due to the flow incidence at the tongue of the volute has
been computed as:
1 n o
ptLVi tLVi u23 ju3 j cos3 T 2
2
corresponding to the dissipation of a constant portion tLVi (here equal to 0.9) of the
kinetic energy of the approach velocity component normal to the orientation T of
the tongue w.r.t. the azimuthal direction T = 3des at design conditions).
For simplicity, mixing losses in the volute have been included only when the
incoming flow velocity is faster than at the exit cross-section and correspond to
losing a constant portion tLVd (here equal to 0.9) of the associated dynamic pres-
sure change:
On the Preliminary Design and Performance 169
1
ptLVd = tLVd u23 u24
2
Finally, frictional losses along the walls of the volute have been neglected
because of their relatively small magnitude.
With the above results, the pumping performance of the machine becomes:
pt ptE2 ptLOSS
= = ,
2 r22 2 r22
where the total pressure increase computed from Euler equation includes the effects
of flow blockage and deviation generated by the BLs developing along the blade
channels:
ptE2 = r2 v2b
and the total pressure loss comprises all of the sources of dissipation considered
above:
Finally, the hydraulic efciency of the machine has been evaluated by means of
the expression:
Qpt
h =
where Q is the volumetric flow rate, pt is the total pressure increase, is the
rotational speed and is the torque applied to the pump shaft.
3 Experimental Apparatus
The Test Facility. The experimental activity reported in the present paper has been
carried out in the Cavitating Pump Rotordynamic Test Facility (CPRTF, Fig. 5).
This facility is specically designed for characterizing the performance of cav-
itating and noncavitating turbopumps in a wide variety of alternative congurations
(axial, radial or mixed flow, with or without an inducer (Cervone et al. 2012; Pace
et al. 2012; Rapposelli et al. 2002). It operates in water at temperatures from room
170 L. dAgostino et al.
P (1 bar FS)
p (5 bar FS)
Fig. 5 The cavitating pump rotordynamic test facility (left) and cut-out of the test section showing
the pump assembly and the locations of the pressure transducer (right)
A series of tests has been conducted on the VAMPIRE pump with the aim of
experimentally validating the proposed model for preliminary design and noncav-
itating performance prediction of centrifugal pumps.
The model has been generalized in order to account for the results of a parallel
test campaign aimed at investigating the influence of thermal cavitation effects on
the rotordynamic whirl forces on the VAMPIRE pump with different values of the
tip clearance (Pace et al. 2014, 2016; Valentini et al. 2016). The presence of a tip
clearance in unshrouded impellers signicantly increases the complexity of the flow
structure. Since the detailed treatment of this aspect is too demanding in terms of
additional complexity and/or computational cost of the model, a simplied
approach has been adopted to approximately describes its influence on the tur-
bopump performance.
Tip clearance changes have long been known to affect both the head and flow
coefcients of the machine, as recently conrmed by some of the authors in the case
of inducers (Torre et al. 2011). Specically, the presence of blade tip clearance in
unshrouded rotors involves three major effects since it:
affects the secondary flow in the blade channels;
induces the occurrence of tip leakage flows as a consequence of the pressure
difference on the two sides of the blades;
modies the intermittent scraping caused by the blade passage on the
boundary layers developing on the inner surface of the impeller casing (Koshide
and Nielson 1973).
With reference to Fig. 7, the operating point of the test pump is dened by the
intersection between its head characteristic and the load curve of the facility, which
can be considered constant for given setting of the silent throttle valve and
On the Preliminary Design and Performance 173
Fig. 7 Noncavitating
operating points (solid Low clearance
circles) of a turbopump
respectively operating with
higher (subscript h) and lower
(subscript l) impeller blade tip High clearance
clearances at constant settings l
of the test facility
h
Loading curve
losses =a 2
h l
approximately quadratic in the flow rate at Reynolds numbers higher than 106 based
on the impeller tip diameter and speed. Then, all other operating conditions being
the same, the relation between the head and flow coefcients of the machine for two
values (subscripts l for lower and h for higher) of the blade tip clearance becomes:
r
l
l = h
h
This equation allows for scaling the noncavitating performance of the machine
for different tip clearances once the associated head change is known. To this
purpose it is worth noting that, as reported in the literature (Torre et al. 2011;
Koshide and Nielson 1973; Brennen 1994), the noncavitating head coefcient is
approximately linear in the relative tip clearance c% , dened as the ratio between
the tip clearance c and the average height of the impeller blades ba .
With reference to Fig. 8 (where the head coefcient has been nondimensional-
ized with its nominal design value 0 , the interpolating straight line for the
VAMPIRE pump has been evaluated from the experimental results for two different
tip clearances (squares). For any given value of the relative clearance, the corrected
head coefcient is then obtained from this correlation and the corresponding flow
coefcient from the above equation. In turn, these coefcients have been used to
modify the pumping characteristics predicted by the model in order to account for
the actual value of the blade tip clearance.
The relative tip clearances of the reference turbopump congurations used for
the denition of the above correction procedure are c% = 5.6% and 11.2%, corre-
sponding to 1 and 2 mm blade tip clearances, respectively. Each experimental point
has been obtained by averaging the readings of the differential pressure and flow
rate acquired over 2 s at 5,000 sps (samples per second). In order to attain fully
developed turbulent flow conditions, rotational speed of the pump and the water
174 L. dAgostino et al.
Fig. 8 Normalized
nondimensional head of the 1
VAMPIRE turbopump as a
function of the relative tip 0.8
clearance. The linear
correlation of the diagram 0.6 c
= -0.7413 + 1.0493
0
(solid line) has been obtained
ba
/
from two experiments at 0
different blade tip clearances 0.4
(solid squares)
0.2
0
0 0.05 0.1 0.15
c
%
D
= 0.092
0.5 0.8
0.45
0.4 0.7
0.35
0.3 0.6
h
Theo. @ c % = 0.9%
0.25
Exp. @ c % = 0.9%
h Exp. @ c % = 11.2%
0.1 0.4
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
/ D
Fig. 9 Normalized comparison between the experimental and predicted (theoretical) noncavitat-
ing performance of the VAMPIRE pump operating at = 1500 rpm and Re = 5.01 106 for two
different values of the blade tip clearance (left scale). The gure also shows the corresponding
hydraulic efciencies (right scale)
= 0.092
D
50
c =0.9%
40 %
c =11.2%
%
30
=0.858%
0.9%
20
=1.942%
11.2%
%
10
e
)/
0
e
- T
-10
(
-20
Theo. Exp .
-30 c% = mean %
Exp .
-40
-50
0 0.5 1 1.5
/
D
Fig. 10 Relative deviation between the predicted and experimental head performance of the
VAMPIRE pump for two values of the relative blade tip clearance
176 L. dAgostino et al.
5 Conclusions
References
Ashihara, K., & Goto, A. (2011). Turbomachinery blade design using 3-d inverse design method,
CFD and optimization algorithm. ASME Turbo Expo, 2001.
Brennen, C. E., (1994). Hydrodynamics of pumps. Oxford University Press.
Busby, J., Batton, W., Furst, R., & Ashvin, H. (2008). High suction specic speed lox pump:
design, analysis and testing. In 44th AIAA/ASME/SAE/ASEE joint propulsion conference and
exhibit, Hartford, (CT), USA.
Cervone, A., Pace, G., Torre. L., Pasini, A., & dAgostino, L. (2012). Effects of the leading edge
shape on the performance of an axial three bladed inducer. In 14th international symposium on
transport phenomena and dynamics of rotating machinery, Honolulu, Hawaii, USA.
Choi, K. J., Kim, J. H., & Kim, K. Y. (2010). Design optimization of circumferential casing
grooves for a transonic axial compressor to enhance stall margin. ASME Turbo Expo 2010.
On the Preliminary Design and Performance 177
dAgostino, L., Pasini, A., Valentini, D., Pace, G., Torre, L., & Cervone, A. (2012). A reduced
order model for optimal centrifugal pump design. In 14th international symposium on
transport phenomena and dynamics of rotating machinery, ISROMAC-14, February 27th
March 2, Honolulu, HI, USA.
dAgostino, L. (2013a). Turbomachinery developments and cavitation. In STO-AVT-LS-206,
Paper NBR 12-1, VKI lecture series on fluid dynamics associated to launcher developments,
von karman institute of fluid dynamics, Rhode-Saint-Gense, Belgium, April 1517.
dAgostino, L. (2013b). On the hydrodynamics of rocket propellant engine inducers and
turbopumps. In 6th international conference on pumps and fans with compressors and wind
turbines (IPCF 2013), Sep. 1922, Beijing, China.
dAgostino, L., Valentini, D., Pasini, A., Torre, L., & Pace, G. (2017). On the preliminary design
and performance prediction of centrifugal turbopumpsPart 1. In this volume, CISM Courses
and Lectures No. 1408. In L. dAgostino & M. V. Salvetti (Eds.), International centre for
mechanical sciences. Vien and New York: Springer.
Droege, A. R., Williams, R. W., & Garcia, R. (2000). Unshrouded impeller technology
development status. NASA Technical Reports Server.
Huppertz, A., Flassig, P. M., Flassig, R. J., & Swoboda, M. (2007). Knowledge-based 2d blade
design using multi-objective aerodynamic optimization and a neural network. ASME Turbo
Expo.
Kim, J. H., Choi, K. J., & Kim, K. Y. (2010). Surrogate modeling for optimization of a centrifugal
compressor impeller. International Journal of Fluid Machinery and Systems, 3(1), 2938.
Kim, J.-H., Jin, C.-H., & Kim, K.-Y. (2011). ASME-JSME-KSME joint fluids engineering
conference performance improvement of a mixed-flow pump by optimization techniques. Japan:
Hamamatsu.
Koshide, R. K., & Nielson, C. E. (1973). Study of blade clearance effects on centrifugal pumps.
NASA CR-120815 R-8806.
Laskshminarayana, B. (1985). Fluid dynamics and heat transfer of turbomachinery. New York:
Wiley.
Mellor, G. L. (1966). The effects of pressure gradients on turbulent flow near a smooth wall.
Journal of Fluid Mechanics, 24(2), 255274.
Oh, H. W., & Kim, K. Y. (2011). Conceptual design optimization of mixed-flow pump impellers
using mean streamline analysis. Proceedings of the Institution of Mechanical Engineers, Part
A: Journal of Power and Energy, 215(1), 133138.
Pace, G., Pasini, A., Torre, L., Valentini, D., & dAgostino, L. (2012). The cavitating pump
rotordynamic test facility at ALTA S.p.A.: Upgraded capabilities of a unique test rig. In Space
propulsion conference, Bordeaux, France, 2012.
Pace, G., Valentini, D., Torre, L., Pasini, A., & dAgostino, L. (2014). Experimental
characterization of rotordynamic forces acting on space turbopumps. Space propulsion 2014,
May (pp. 1922). Germany: Cologne.
Pace G., Valentini D., Pasini A., Torre L., Hadavandi R., & dAgostino L. (2016). Inducer and
centrifugal pump contributions to the rotordynamic fluid forces acting on a space turbopump. In
ISROMAC 2016, international symposium on transport phenomena and dynamics of rotating
machinery, Honolulu, Hawaii, USA, April 1015.
Pasini, A., Torre, L., Cervone, A., & dAgostino, L. (2011). Characterization of the rotordynamic
forces on tapered axial inducers by means of a rotating dynamometer and high-speed movies.
In WIMRC 3rd international cavitation forum 2011, University of Warwick, UK, 4th6th July
2011.
Rai, M. M., & Madyavan, N. K. (2000). Aerodynamic design using neural networks. AIAA
Journal, 38(1), 173182.
Rapposelli, E., Cervone, A., & dAgostino, L. (2002). A new cavitating pump rotordynamic test
facility. In AIAA joint propulsion conference and exhibit, Indianapolis, USA.
Stripling, L., & Acosta, A. (1962). Cavitation in turbopumpspart 1. ASME Journal of Basic
Engineering, 84, 326338.
178 L. dAgostino et al.
Torre, L., Pasini, A., Cervone, A., Pace, G., Miloro, P., & dAgostino, L. (2011). Effect of tip
clearance of the performance of a three-bladed axial inducer. AIAA Journal of Propulsion and
Power, 27(4), 890898.
Valentini, D. (2011). Sviluppo di un Modello per la Progettazione di Turbopompe Centrifughe, M.
S. Thesis, Universit di Pisa, Pisa, Italy, October 10, 2011.
Valentini, D. (2015). Modelling and testing of chemical propulsion rocket subsystems. Ph.D.
Thesis, Universit di Pisa, Pisa, Italy, December 14, 2015.
Valentini, D., Pace, G., Pasini, A., Torre, L., Hadavandi, R., & dAgostino, L. (2016).
Fluid-induced rotordynamic forces on a whirling centrifugal pump. In ISROMAC 2016,
international symposium on transport phenomena and dynamics of rotating machinery,
Honolulu, Hawaii, USA, April 1015.
Van den Braembussche, R. A. (2006). Flow and loss mechanism in volutes of centrifugal pumps.
In Design and analysis of high speed pumps. RTO-EN-AVT-143, Paper 12 (pp. 12-112-26).
Neuilly-sur-Seine, France.
Visser, F. C., Dijkers, R. J. H., & op de Woerd, J. G. H. (2000). Numerical flow-eld analysis and
design optimization of a high-energy rst-stage centrifugal pump. Computing and Visualiza-
tion in Science, 3(12), 103108.
White, F. M. (1974). Viscous fluid flow (1st ed.). New York: McGraw-Hill Inc.
Numerical Simulation of Cavitating Flows
with Dierent Cavitation and Turbulence
Models
Eric Goncalves
1 Introduction
Several physical and numerical models have been developed to investigate cavitating
ows within the framework of averaged two-phase model. For the averaged model,
there are dierent approaches according to the assumptions made on the local ther-
modynamic equilibrium and the slip condition between phases. A hierarchy of mod-
els exists, with the numbers of equations ranging from seven to three only. The full
non-equilibrium two-uid models with relaxation procedures have been tested on
inviscid high-speed applications (see for example Petitpas et al. 2009; Zein et al.
2010), whereas one-uid models have been massively used for industrial applica-
tions and turbomachineries.
An important class of reduced models is formed by the ve-equation models, in
which velocity equilibrium and pressure equilibrium are considered. The archetype
ve-equation model is that of Kapila et al. (2001). It is composed of four conservation
laws: two for masses, one for the mixture momentum and one for the mixture energy.
It is completed by an equation for a non-conservative quantity describing the ow
E. Goncalves ()
LEGI Laboratory, University of Grenoble, Grenoble, France
e-mail: [email protected]
E. Goncalves
Institut Pprime, UPR 3346 CNRS, ENSMA, Futuroscope, Poitiers, France
topology, usually the void ratio. Such a model has been used for inviscid high-speed
cavitating applications and cavitation pocket in fuel injector nozzles (Saurel et al.
2008, 2009).
By assuming the thermal equilibrium between phases, a four-equation model can
be expressed. A very popular formulation has been developed to simulate turbu-
lent cavitating ows (Merkle et al. 1998; Kunz et al. 2000; Senocak and Shyy 2002;
Singhal et al. 2002; Venkateswaran et al. 2002; Vortmann et al. 2003; Hosangadi and
Ahuja 2005; Utturkar et al. 2005; Wu et al. 2005; Zhang et al. 2008; Ji et al. 2013).
It is composed by three conservation laws for mixture quantities (mass, momen-
tum, energy) plus a mass equation for the vapour or liquid density including a cav-
itation source term. The main diculty is related to the formulation of the source
term and the tunable parameters involved for the vaporisation and condensation
processes (dierent sets of parameters are presented in Utturkar et al. 2005). More-
over, the thermodynamic coherence was not attested and that constitutes a major
problem (Goncalves and Patella 2011). Recently, another formulation for the mass
transfer between phases was proposed using a void ratio transport-equation model.
The mass transfer was closed assuming its proportionality to the divergence of the
homogeneous velocity eld (Goncalves 2013, 2014; Goncalves and Charriere 2014).
Another popular model devoted to ebullition problems uses a mass fraction equation
with a relaxation term (Homogeneous Relaxation Model). The source term involves
a relaxation time, that is the time for the system to regain its thermodynamic equilib-
rium state. This time is very dicult to determine and is estimated from experimental
data (Downar-Zapolski et al. 1996; Bilicki et al. 1996; Barret et al. 2002).
With the assumption of complete thermodynamic equilibrium between phases
(local temperature, pressure and free Gibbs enthalpy equality between phases), we
obtain the 3-equation models or homogeneous equilibrium models (HEM). Vapor-
isation or condensation processes are assumed to be instantaneous. An equation of
state (EOS) is necessary to close the system. Dierent closure relations (tabulated
EOS or combination of pure phase EOSs) that link the pressure to the thermody-
namic variables have been proposed (Delannoy and Kueny 1990; Saurel et al. 1999;
Schmidt et al. 1999; Clerc 2000; Edwards and Franklin 2000; Ventikos and Tzabiras
2000; Shin et al. 2003; Liu et al. 2004; Vuyst et al. 2005; Schmidt et al. 2006; Sini-
baldi et al. 2006; Ihm and Kim 2008; Goncalves and Patella 2009; Goncalves et al.
2010b; Hu et al. 2011).
In addition, the turbulence modelling plays a determinant role in the capture of
unsteady behaviours. Cavitation sheets that appear on solid bodies are characterized
by a closure region, which always uctuates with the existence of a re-entrant jet and
the propagation of pressure waves. Moreover, compressibility eects on turbulence
are involved. These eects and interactions with two-phase structures are not yet
well known and understood. For usual applications, the Reynolds decomposition is
often used with an averaged statistical processing resulting in the RANS equations
for the mean ow quantities. The limitation of the turbulent viscosity evaluated with
transport-equation turbulence models (through the Boussinesq assumption) is a key
point to capture realistic cavitation sheets. Dierent methods have been investigated
to limit or to correct standard turbulence models. One of the most popular limiter
Numerical Simulation of Cavitating Flows . . . 181
was proposed by Reboud to reduce the turbulent viscosity (Reboud et al. 1998), and
has successfully been used by dierent authors (Coutier-Delgosha et al. 2002; Chen
et al. 2006; Zhou and Wang 2008; Srinivasan et al. 2009; Goncalves 2011; Ahn and
Kwon 2013).
The nal goal of the present work is to compare various 3- and 4-equation mod-
els for the simulation of realistic turbulent cavitating ows involving unsteady sheet
cavities. A comparison with incompressible numerical simulation is also proposed.
Moreover, the turbulence modelling is investigated, especially the compressibility
eects and the eddy viscosity limitation. The considered test cases are the following.
First, 1D inviscid rarefaction tubes are computed allowing a comparison of models
without the turbulence model inuence. Second, two Venturi geometries are consid-
ered involving unsteady partial cavities for which the dynamic behaviour is dierent
(an aperiodic quasi-stable sheet and a self-sustained periodic sheet). A good model
should be able to simulate both dynamics. Finally, another Venturi case is studied
for which the running uid is freon R-114 in order to compare cavitation models
including thermal eects.
This paper is organized as follows. We rst review the theoretical formulation
including cavitation models, equation of state and mass transfer formulations. Tur-
bulence equations are detailed followed by the description of the 1-uid solver. The
preliminary studies carried out in inviscid test cases are presented. This is followed
by sets of results on two turbulent Venturi ows and discussions.
The homogeneous mixture approach is used to model two-phase ows. The phases
are assumed to be suciently well mixed and the disperse particle size are su-
ciently small thereby eliminating any signicant relative motion. The phases are
strongly coupled and moving at the same velocity. In addition, the phases are
assumed to be in thermal and mechanical equilibrium: they share the same tem-
perature T and the same pressure P. The evolution of the two-phase ow can be
described by the conservation laws that employ the representative ow properties as
unknowns just as in a single-phase problem.
To obtain the one-uid equations, we use the phase average of a phase quantity
k dened by Ishii and Hibiki (2006):
1
k = d (1)
Tk k
where Tk is the time period of phase k. The mass-weighted average or Favre average
is commonly used for compressible ows. In the following, () and () will denote
a phase average and a Favre average, respectively. Fluctuations with respect to the
average will be () and () , respectively. Using this denition and starting from the
182 E. Goncalves
instantaneous conservation equations, the equations for each phase can be expressed.
The equation for mixture quantities are obtained by the summation of the separate
equations of the phase quantities.
We introduce k the void fraction or the averaged fraction of presence of phase k.
The density , the center of mass velocity u and the internal energy e for the mixture
are dened by
= k k (2)
k
ui = k k
uk,i (3)
k
e = k k
ek (4)
k
The mass gain per unit volume due to phase change (evaporation or condensation)
is noted k = k (uI uk ).nk I where I is a Dirac distribution having the dierent
interfaces as a support, uI the interface velocity and nk the vector normal to the
interface directed outward from phase k.
The 3-equation system consists in mixture balance laws for density, momentum and
total energy. In order to simplify the formulation, we present below the inviscid one-
dimensional equations, expressed in conservative variables w = (, u, E):
u
+ = 0 (5)
t x
(u) (u2 + P)
+ = 0 (6)
t x
(E) (uH)
+ = 0 (7)
t x
In the present study, we used the convex stiened gas EOS for the pure phases (see
Metayer et al. 2004):
Numerical Simulation of Cavitating Flows . . . 183
where = Cp Cv is the heat capacity ratio, Cp and Cv are thermal capacities, q the
energy of the uid at a given reference state and P is a constant reference pressure.
The speed of sound c is given by
P + P
c2 = = ( 1)Cp T (11)
Stiened gas EOS On the basis of the stiened gas EOS for each pure phase, an
expression for the pressure can be deduced from the mechanical equilibrium assump-
tion (Saurel et al. 2008). These expressions are available in all possible uid states,
function of the void ratio = v and the vapour mass fraction Y = Yv = v :
1 1
= + (16)
c2wallis v c2v l c2l
Sinusoidal EOS A sinusoidal law (Delannoy and Kueny 1990; Goncalves and
Patella 2009) is considered for the mixture. This law is characterized by its maxi-
mum slope 1c2baro . The quantity cbaro is an adjustable parameter of the model, which
can be interpreted as the minimum speed of sound in the mixture.
184 E. Goncalves
When the pressure is smaller than Pvap + P, the following relationship applies:
( )
sat sat
c2baro Arcsin (1 2)
l v
P() = Pvap + (17)
2
where P represents the pressure range of the law and, for a void ratio value of 0.5,
the pressure is equal to the saturation pressure Pvap at the reference temperature.
This law introduces a small non-equilibrium eect on the pressure. The cavitation
phenomenon is assumed to be isothermal and thermodynamic eects are neglected.
The void ratio is computed with saturation values of densities as follows:
sat
l
= (18)
sat
v
sat
l
The model consists in three conservation laws for mixture quantities (mass, momen-
tum and total energy) and an additional equation for the void ratio. It is obtained
from a reduction of a 5-equation model (Goncalves 2013) assuming the liquid is at
its saturation state. The expression for the void ratio equation is
2 2
c2 c2 cv + cl
+ div (V) = 2 + div V + 1
l l v v
v (20)
t l cl v c2v l c2l v c2v
1 + 1 +
=K = 1I the interfacial density
where v is the mass transfer between phases and V the velocity vector. The term K
involves the speed of sound of pure phases ck and it reects the eects of changes in
volume of each phase. Without mass transfer the system is hyperbolic, eigenvalues
are (u-cwallis , u, u, u+cwallis ).
Numerical Simulation of Cavitating Flows . . . 185
By assuming that the mass transfer is proportional to the divergence of the veloc-
ity, it is possible to build a family of models in which the mass transfer is expressed
as Goncalves (2013)
( )
l v c2
v = 1 div V (21)
l v c2wallis
When heat and mass transfer eects are involved in the ow, the sound speed c
decreases to the thermodynamic equilibrium one (Petitpas et al. 2009). This limit
speed is evaluated with the assumption of local thermodynamic equilibrium: equal-
ities of pressure, temperature and free enthalpy.
With this generic form for the mass transfer, we remark that all models in which
the mixture speed of sound is the Wallis one cannot produce void ratio during the
phase transition (that is the case of most of models proposed in the literature). The
void ratio is only modied through the term K div V.
Using dierent mixture EOS it is possible to built dierent models. Two models
were built using the stiened gas mixture equation and the sinusoidal EOS.
sat
l
= (22)
sat
v
sat
l
The mass and void ratio equations are not two independent equations. The compati-
bility of the system is not clear.
First model A formulation derived from the model proposed by Hosangadi and
Ahuja (2005) can be expressed as:
With this formulation, the void ratio value can be higher than one. A limiter is
added in the solver to clip the void ratio into its physical domain of evolution.
Second model The second model, implemented in the OpenFOAM solver (Weller
et al. 1998), is derived from the Kunz model (Kunz et al. 2000). It is based on the
incompressible RANS equations for an homogeneous mixture. The phase change is
modelled using a transport-equation for the liquid volume fraction l that reads
l
+ uj l = v (24)
t xj
v = mv + mc (25)
with
( )
v 2 max P Pvap ; 0
mc = C ( ) (26)
l v c t Llim max P Pvap ; 0.01 Pvap
( )
mv = Cv 1 v min P Pvap ; P0 (27)
l v U2 t
2 l
Cc , Cv , U and t are constant set by the user, whereas P0 and Llim are included
to avoid non- physical values. Usually U is set to the freestream value, and t
represents a relaxation time not well dened in the literature. For the present com-
putations, the following values are specied:
Cc = 10 ; Cv = 8000 ; t = 0.005
We detailed the formulation in the case of thermo-sensitive uid and non isothermal
thermodynamic path.
hl ql h qv h q(Y)
T(, h, Y) = = v = (28)
Cpl Cpv Cp (Y)
Cp (Y) = YCpv + (1 Y)Cpl (29)
We assume that the vaporisation pressure varies linearly with the temperature:
dP
Pvap (T) = Pvap (Tref ) + (T Tref ) (30)
dT
The constant quantity dPdT is evaluated with a thermodynamic table. The speed
of sound in the mixture can be expressed as Goncalves (2014):
( )
1 v hv l hl dv dl
C1 = + + (1 ) (31)
1 dP
(l v ) dT dT dT
[ ]
1 v l
c2 = (h hl ) (32)
C1 (l v ) v
Enthalpies of pure phase hl and hv are computed with the mixture temperature T.
Due to numerical problems, the derivatives of densities with the temperature are not
taken into account in this study.
A modied sinusoidal relation (Goncalves and Patella 2010) can be considered for
the mixture. When the pressure is smaller than Pvap (T) + P, the following relation-
ship applies:
( )
sat sat
c2baro Arcsin (1 2)
l v
P(, T) = Pvap (T) + (33)
2
where P represents the pressure range of the law and, for a void ratio value of 0.5,
the pressure is equal to the saturation pressure Pvap (T) at the local temperature T.
This temperature is evaluated using the relation (29). sat
l
and sat
v
are pure phases
densities at saturation evaluated at the reference temperature Tref . The quantity cbaro
is a parameter of the model, it is no more the minimum sound speed in the mixture.
As previously we assume that the vaporisation pressure varies linearly with the
temperature and follows the relation (30). The speed of sound in the mixture can be
evaluated (Goncalves and Patella 2010):
v l
(h
(l v ) v
dP
hl ) dT + Cp c2T
2
c = dP
(34)
Cp dT
188 E. Goncalves
( ) ( )
P P c2
c2T = = = baro (35)
s T 2 (1 )
The classical form for the homogeneous mixture balance equations are Ishii and
Hibiki (2006):
( )
ui
+ = 0 (37)
t xj
Numerical Simulation of Cavitating Flows . . . 189
( )
( )
ui uj + Pij ijv + ijt
( )
ui + = (38)
t xj xj
[ ]
[ ] (ijv + ijt )uj Qvi Qti
(E + P)ui
(E) + = (39)
t xj xj
where E is the mixture total energy. The mixture total stress tensor ij (with viscous
and turbulent contributions) is evaluated using the Stokes hypothesis, Newtons law
and the Boussinesq assumption. The mixture total heat ux vector Qi is obtained
from the Fourier law involving a turbulent thermal conductivity t with the constant
Prandtl number hypothesis.
[ ]
ui uj 2 ul 2
ij = ijv
+ ijt
= ( + t ) ( + ) + kij
xj xi 3 xl ij 3
( ) T t Cp
Qi =Qvi + Qti = + t with t = (40)
xi Prt
where t is the mixture eddy viscosity, k the mixture turbulent kinetic energy (TKE),
Cp the mixture thermal capacities and Prt the turbulent Prandtl number set to 1.
In pure liquid, the viscosity is determined by an exponential law and, in pure
vapour, the viscosity follows the Sutherland law. The mixture viscosity is dened as
the arithmetic mean of the liquid and vapour viscosities (uctuations of viscosity are
neglected) (Ishii and Hibiki 2006):
The mixture TKE equation Dierent authors have studied the turbulent kinetic
energy equation for a gas-liquid ow. The mixture equation can be obtained by the
summation of the separate equations of the phase turbulent quantities with homoge-
neous mixture assumptions (Besnard and Harlow 1988; Elghobashi and Abou-Arab
1983; Kataoka and Serizawa 1989; Lance et al. 1984). The expression of the mixture
TKE equation is
k kul
+ = Pk + + M + D + (45)
t xl
u
u
k,i k,i
where k = k , Pk is the mixture production term, is the mixture pressure-
k
2
dilatation term, is the mixture TKE dissipation rate which can be split into a
solenoidal part s and a dilatational part d , M is the mixture turbulent mass ux, D
is the mixture turbulent and viscous diusions. The last term is specic to multi-
phase ows. It involves the mass transfer due to phase change and is usually set to
zero.
In multiphase ow, even if pure phases are assumed to be incompressible, the
divergence of the uctuating phase velocity is not zero, caused by the presence
of interfaces. Indeed, with this assumption, the divergence of the uctuating phase
velocity is
uk,i u
k,i 1
= = uk .nk I (46)
xi xi k
4
Therefore, the dilatational dissipation rate d = k k sk,jj sk,ll is not zero. And
3 k
uk,i
the pressure-dilation term = k Pk is instantaneously present but null in
k
xi
mean as Pk = 0.
The modelling of these additional terms is very complex to realize in turbulent
cavitating ows due to the lack of experimental data for turbulent quantities and
DNS of cavitation ows. Only closure relations developed for aerodynamic high-
speed ows have been tested.
TKE dissipation rate equation for the mixture Kataoka and Serizawa (1989)
derived a transport equation for the mixture turbulent dissipation rate assuming that
both phases were considered as incompressible. This equation put in evidence that
the turbulent dissipation is aected by interfacial transport terms. Nevertheless, no
theoretical or experimental results regarding these terms are available in case of two-
Numerical Simulation of Cavitating Flows . . . 191
phase ows with phase change. Usually, it is used the same equation for than in
single-phase ows.
Various popular turbulence models are considered: the Smith k model (KL)
(Smith 1994), the one-equation SpalartAllmaras model (SA) (Spalart and Allmaras
1994), the Wilcox k model (Wilcox 1988) and the Jones-Launder k model
(KE) (Jones and Launder 1972). For comparisons with the OpenFOAM solver, the
Menter k SST model (Menter 1994) is used, assuming the validity of the Brad-
shaw assumption in a two-phase turbulent boundary layer.
Compressible terms have been included in both the k and k models.
A k model in compressible form We include the compressible terms , M and
d into the Jones- Launder k model:
[ ( ) ]
k k
+ kuj + t = Pk s + d + + M
t xj k xj
[ ( ) ]
s t s 2
+ s uj + = c1 s Pk c2 f2 s + E
t xj xj k k
The low Reynolds number term E, the damping functions f and f2 and all constants
follow the JonesLaunder formulation (Jones and Launder 1972).
The additional terms and d are modelled according to Sarkar propositions
(Sarkar 1992):
= 2 Pk Mt + 3 s Mt2
d = 1 s Mt2 (47)
where Mt = 2ck is the turbulent Mach number dened with the mixture speed of
sound c and i are constants to calibrate.
The turbulent mass ux M is closed with the Jones formulation (Jones 1979):
t P
M = (48)
2 p xi xi
Turbulence models always leads to the generation of stable cavities, because very
strong turbulent eddy viscosity t inside the cavity avoids the re-entrant jet develop-
ment and unsteadiness formation. The link to compressibility eects on turbulence
is not clear. In cavitating ows, supersonic regime is reached in the mixture area,
because of the drastic diminution of the speed of sound. The detailed mechanisms
of the interaction between turbulent ows and cavitation have not yet been clearly
revealed especially for the phenomena occurring at small scales.
As a remedy to limit the turbulent viscosity, one can use an eddy viscosity limiter
in the mixture area. The most famous limiter is the one proposed by Reboud et al.
(1998), which has shown its eciency in the simulation of sheet cavities (Coutier-
Delgosha et al. 2002; Chen et al. 2006; Zhou and Wang 2008; Srinivasan et al. 2009;
Goncalves 2011; Ahn and Kwon 2013). Other corrections can be used, for example
the Shear Stress Transport (SST) correction proposed by Menter (1994) to reduce the
eddy viscosity in case of positive pressure gradient and a variant of the latter based
on realizability constraints (Durbin 2009). Applications to cavitating simulation can
be found in Decaix and Goncalves (2012).
Reboud correction Reboud proposed an arbitrary limiter by introducing a function
f () in the computation of the turbulent viscosity for the k model:
k2
t = f ()C with f () = v + (1 )n (l v ) (54)
where n is a parameter xed to 10.
This correction can be extended to other turbulence models with the same func-
tion f ().
Numerical Simulation of Cavitating Flows . . . 193
where F2 is a blending function that tends to zero outside the boundary layer, and
is the vorticity.
The evolution of the ratio u v k was recently measured by Aeschlimann and
Barre (2009) in the case of a cavitating mixing layer. The authors showed that
this ratio decreased signicantly and continuously when cavitation developed. They
observed a factor 2 between the non-cavitating regime and the most severe cavitating
case. Yet, the evolution of this ratio for cavitating ows involving sheet cavities is
unknown.
Realizability constrains By replacing the vorticity with the stress tensor S in the
SST formula, a correction is derived based on the realizability principle (Durbin
2009). The correction involves a parameter c varying between 0 and 1 and set to 0.3.
The Scale-Adaptive Simulation rst appeared in the formulation of the KE1E one-
equation model for the eddy viscosity derived from the standard k model (Menter
1997). This model provided a dynamical behaviour similar to a Detached Eddy Sim-
ulation (DES) model but without an explicit grid dependence, allowing the devel-
opment of a turbulent spectrum in the detached regions. Later, Menter and Egorov
(2004) showed that the exact transport equation for the turbulent length scale l as
derived by Rotta actually introduces the second derivative of the velocity eld. They
proposed a model for this term which introduces the von Karman length-scale Lvk .
194 E. Goncalves
This term, referred to as SAS, was inserted in the existing k SST model and they
calibrated it for the decay of a homogeneous isotropic turbulence. Various test cases
(Menter 2010a, b; Davidson 2006) have showed the ability of the SST-SAS model to
resolve turbulent structures down to the limit grid and have conrmed that the SAS
term reduces the turbulent eddy viscosity and predicts resolved uctuations much
larger than standard models.
It is possible to include scale-adaptive terms into the k and SpalartAllmaras
turbulence models following the works of Menter and Egorov. Details of the for-
mulation are given in Decaix and Goncalves (2012, 2013b). The SAS term is only
activated in the mixture area when > 0.
For the modelling of ow close to the wall, a two-layer wall law approach is used as
follows:
u+ = y+ if y+ < 11.13
1
u+ = ln y+ + 5.25 if y+ > 11.13
(57)
u yU w
u+ = ; y+ = ; U =
U w w
where = 0.41 is the von Karman constant and the subscript w is used for a wall
value.
With regard to the turbulent transport-equation models, the production of k is
computed according to the formulation proposed by Viegas and Rubesin (1983).
The value of in the rst cell is computed with the linear relation l = y.
For the one-equation SpalartAllmaras model, the transported quantity is calcu-
lated using the models closure relations, the velocity prole and a mixing-length
formulation for the eddy viscosity. More details concerning the wall law approach
are given in Goncalves and Decaix (2012).
We assume that wall functions are similar in a two-phase ow and in a single-
phase ow. For unsteady ows, the existence of a wall law is assumed to be valid at
each instant. In order to investigate the validity of these assumptions, comparisons
were proposed with a thin boundary layer approach (TBLE). This approach is based
on the integration of a simplied averaged set of partial equations derived from the
NavierStokes equations:
[ ]
ui ui uj 1 dP ui
+ + = ( + t ) (58)
t xj dxi y y
A ne one-dimensional grid is embedded between the rst grid point and the wall
(see Fig. 1), and the turbulent boundary-layer equations are discretized and solved in
Numerical Simulation of Cavitating Flows . . . 195
the embedded mesh. The integration is performed using a Newton algorithm for
the wall shear stress w . In the present computations, the number of nodes in the
embedded grid is 30. More details can be found in Goncalves and Decaix (2012).
The major diculty in modelling multiphase turbulence is the wide range of length
and time scales on which turbulent mixing occurs. The Direct Numerical Simulation
(DNS) approach, with no modelling, resolves all the scales present in the ow. How-
ever, it is not feasible for practical engineering problems involving high Reynolds
numbers. The large-eddy simulation (LES) is an intermediate way between DNS
and RANS in terms of the fraction of the resolved scales. In LES, the large-scale
turbulent motions are resolved while the eects of the smaller ones are modelled.
This approach allows a signicant decrease in the computational cost over direct
simulation and captures more dynamics than a simple RANS model.
A rst point is to identify the scales at which the governing equations are to be
applied: micro-scales (scales which are small enough to describe individual bub-
ble shapes), meso-scales (which are comparable to bubble sizes) and macro-scales
(which entail enough bubbles for statistical representation). When LES is applied at
a micro-scale, ltering of turbulent uctuations needs to be combined with interface
tracking methods (for example Liovic and Lakehal (2012)). When LES is applied at
a macro-scale, the interface resolution is not considered.
Dierent authors have studied the 1-uid ltered equations for incompressible
gasliquid ows (Labourasse et al. 2007; Vincent et al. 2008; Toutant et al. 2009;
Larocque et al. 2010; Bois et al. 2010). The scale separation is mathematically
obtained by applying a convolution product using a large-scale-pass lter function
G. For a quantity , the ltered variable is dened as = G and the Favre ltered
variable: = . Let Xk be the phase indicator function (Xk = 1 in phase k and 0
elsewhere). The ltered phase indicator function Xk = GXk = k can be interpreted
as a ltered volume fraction of phase k.
196 E. Goncalves
It is assumed that the ltering operator commutes with time and spatial deriva-
tives. The mass transfer is assuming to be proportional to the velocity divergence
through a constant C. Applying the lter G to the governing system leads to the
ltered 1-uid equations system:
( )
ui
+ = 0 (59)
t xj
( )
ui ui u j + Pij ( )
+ = div 2 S + 2S 2S uu (60)
t xj
( v
)
H u i + Q
E i (
S 2 S
u + 4u S + H
u Hu
)
+ = div 2Su 2 V
t xj
u ( )
+ V. C i = u C div u (61)
t xj
u = V.Xv V. ; u = V V
S = S S ; S = S S
Hu = HV V
H ; Su = SV SV
( )
uu = V V V V
The subgrid terms must be formally modelled. A part of theses terms is present in
standard single-phase compressible LES equations (the subgrid stress tensor), while
the subgrid interfacial transport u is specic to two-phase ows. The magnitude
of the dierent subgrid terms was a priori evaluated in the case of phase separa-
tion ows and turbulence bubble interaction (Labourasse et al. 2007; Vincent et al.
2008). Following these studies, the inuence of u is highly dependent on the ow
congurations and/or on the chosen two-phase description. Yet, in cavitating ows,
the inuence and the hierarchy of all these terms have never been investigated.
Following the works of Bois et al. (2010) about the concept of continuous LES in
two-phase ows, a supplementary term appears in the viscous stress tensor formu-
lation due to the mass transfer and the jump relations:
( )( )
2 1 1 2
ij = 2Sij Skk ij + v 2 ij ni I (62)
3 v l 3
The numerical simulations are carried out using an implicit compressible solver for
multi-domain structured meshes. This solver is based on a cell-centred nite-volume
discretization.
For the mean ow, the convective ux density vector on a cell face is computed
with the Jameson scheme (Jameson et al. 1981). The viscous terms are discretized
by a second-order space-centred scheme. For the turbulence transport equations, the
upwind Roe scheme (Roe 1981) is used to obtain a more robust method. The second-
order accuracy is obtained by introducing a ux-limited dissipation.
For low Mach number applications, a well-known problem concerns the stiness
on the solution convergence. In this situation, the dominance of convection terms
renders the system sti and compressible solvers converge slowly. To overcome this
diculty, a preconditioning method is necessary. The physical acoustic waves are
replaced by pseudo-acoustic modes that are much closer to the advective velocity,
reducing the stiness and enhancing the convergence. The method is based on the
modication of the derivative term by a premultiplication with a suitable precon-
ditioning matrix. In order to simplify the formulation, we present below the one-
dimensional formulation. With the primitive variables W = (P, u, e, ) the precondi-
tioned Euler equations can be expressed as
W W
P1 + Ae = 0 (63)
e t x
We use the preconditioning matrix based on the Turkel approach (Guillard and
Viozat 1999; Turkel 1987):
2 0 0 0 u c2 0 0
0 1 0 0 1 u 0 0
Pe = Ae =
0 0 1 0 ; 0 P u 0
0 0 0 1 0 K 0 u
is a parameter[ on the
( order of)the]Mach number expressed by Choi and Merkle
(1993): 2 = min max M 2 , M 2
,1
198 E. Goncalves
This form implies that there is no preconditioning used in transonic and super-
sonic ow regions (in the mixture). When 2 = 1, the preconditioning matrix
becomes the identity matrix and the system returns to its classical non-preconditioned
form. Moreover, for a very small ow velocity, 2 is not allowed to be less than a
given percentage of the freestream velocity, determined by the coecient .
The eigenvalues of the preconditioned system are
[ ]
1
u; u ; = u(1 + 2 ) ( 2 1)2 u2 + 4 2 c2 (64)
2
For the variables w = [, u, E, ], the method involves the preconditioning
w 1 W
matrix P1
c = P and the Jacobian matrix of the convective uxes Ac .
W e w
Expressions of matrices are given in Goncalves and Charriere (2014).
Time integration is achieved using the dual time stepping approach and a low-cost
implicit method consisting in solving, at each time step, a system of equations arising
from the linearization of a fully implicit scheme. The derivative with respect to the
physical time is discretized by a second-order formula. The main advantage of this
method is that the storage of the Jacobian matrix is completely eliminated, which
leads to a low-storage algorithm (Goncalves and Patella 2009).
With the preconditioned method, the dissipation matrices are modied and the
preconditioned matrix P1 c remains in the formulation. By judiciously exploiting the
idempotence propriety of the matrix M, it is possible to preserve a low-cost system
where matrix operations and matrix-vector products can be easily computed.
For the turbulence transport equations, the diusive ux Jacobian matrix is
replaced by its spectral radius. The source term needs special treatment (Merci et al.
2000). Only the negative part of the source term Jacobian matrix is considered and
replaced by its spectral radius. The system obtained is solved with a line-alternated
Jacobi relaxation algorithm.
The numerical treatment of the boundary conditions is based on the use of the pre-
conditioned characteristic relations of Euler equations. The number of variables to
impose at boundaries is given by the number of characteristics directed into the
domain of interest. The characteristic relations obtained for the preconditioned sys-
tem, in two-dimensional ows, are
Numerical Simulation of Cavitating Flows . . . 199
c2 (c s ) + (Pc Ps ) = 0 (65)
Vtc Vts = 0 (66)
( ) K( ) = 0
c s c s
(67)
(+ Vn )(Pc Ps ) + 2 c2 (Vnc Vns ) = 0 (68)
( Vn )(Pc Ps ) + 2 c2 (Vnc Vns ) = 0 (69)
The variables with superscript c denote the variables to be computed at the boundary.
Variables with superscript s denote the variables obtained by the current numerical
scheme. Vt and Vn are the tangential and the normal component of the mean velocity,
respectively.
At inow, we impose the stagnation pressure Pi , the stagnation temperature Ti ,
the direction of the velocity and the initial values of the void ratio. The pressure is
evaluated with the relation (69) and all variables can be evaluated at the boundary.
At outow, the static pressure is imposed. The conservative variables are com-
puted with four characteristic relations (65)(68).
To investigate and to compare models various test cases are considered as follows:
1D inviscid tube cases
These cases involve a large rarefaction area in a tube which leads to the apparition
of cavitation. The ability of cavitation models to reproduce the physical phenom-
enon without viscosity and turbulence problems can be evaluated.
Cavitation sheet in Venturi geometries in turbulent ow
Two Venturi congurations are studied for which the divergent angle is dierent.
Partial sheet cavities appear along the solid wall downstream the Venturi throat.
Two dynamics behaviour were experimentally observed: a transitional aperiodic
phenomenon and a self-sustained periodic one.
For the study of thermal eects, a Venturi conguration in which the running uid
is freon R-114.
For this case, a steady cavitation sheet develops downstream the throat of the Ven-
turi.
Fig. 2 Expansion waves and pressure evolution in the tube, extracted from Zein et al. (2010)
density l =1150 kg/m3 . The temperature of water is Tref = 355 K. A weak volume
fraction of vapor = 0.01 is initially added to the liquid. The initial discontinuity
is set at 0.5 m, the left velocity and the right velocity have the same value but an
opposite sign. Two velocities are tested: u = 2 m/s and u = 100 m/s. The vapour
pressure Pvap (Tref ) is set to 51000 Pa at the initial temperature.
The solution involves two expansion waves. As gas is present, the pressure cannot
become negative. To maintain positive pressure, the gas volume fraction increases
due to the gas mechanical expansion and creates a pocket. Liquid water is expanded
until the saturation pressure is reached then evaporation appears and quite small
amount of vapor is created. The solution with phase transition is composed of four
expansion waves. The extra two expansion waves correspond to the evaporation
fronts (see Fig. 2).
These cases were computed in Zein et al. (2010) using a two-uid model in which
instantaneous relaxation processes towards equilibrium are included for the temper-
ature and the Gibbs free energy. The mesh contains 5000 cells. The time step is set
to 107 s. The value of cbaro is set to 1.31 m/s for the 4-equation sinus model as
presented in Goncalves (2013).
Fig. 3 Venturi design and probe locations, 4 divergence angle (left) and 8 angle (right)
Pinlet Pvap
as: inlet = 2
, where Pvap is the vapour pressure at 20 C and Pinlet , Uinlet
0.5Uinlet
are the pressure and velocity respectively at the reference section upstream of the
Venturi. The ow conditions are given in Table 1.
With these previous parameters and according to experimental observations
(Barre et al. 2009; Aeschlimann et al. 2013), cavitation sheets developed from the
Venturi throat (Fig. 3). The obtained cavity length is ranging from 70 to 85 mm for
case 1 and having a relatively stable aspect (see Fig. 4). The attached cavity length
corresponding to the end of the re-entrant jet is around 30 mm. For this case, no
periodic cycles with large shedding were observed.
For case 2, a typical self-oscillation behaviour was observed with quasi-periodic
vapour clouds shedding and the maximum cavity length (before the break-o of the
cavity) was 45 5 mm. Figure 4 shows an instantaneous photograph of the cav-
ity with a large structure shedding. The cloud shedding frequency was about 45 Hz
leading to the classical Strouhal number based on the cavity length of about 0.3. The
divergent part was equipped with eight probe holes to take various measurements
such as the instantaneous pressure, local void ratio and velocity in the sheet cavity.
Specic probe locations are presented in Fig. 3 for the two tested cases.
Numerical data Both grids used for simulations are a H-type topology. For the non-
cavitating regime, computations are started from an uniform ow-eld using a local
time step. For the unsteady cavitating regime, computations are performed with the
dual time stepping method and are started from the non-cavitating numerical solu-
tion. All numerical values are obtained by a time-averaged treatment. For case 2, a
202 E. Goncalves
direct Fourier transformation (DFT) of the vapour volume signal was performed to
evaluate the frequency of the phenomenon.
The freon R-114 experimental facility of the CREMHyG is a closed loop operating
with a reference pressure, obtained by pressurizing a tank with nitrogen gas. The
cavitation tunnel was designed to simulate cavitating ows developing on the blades
of space turbopump inducers (Stutz 1996; Fruman et al. 1999). The loop is tted
with a test section having the shape of a two-dimensional Venturi, characterized by
a divergence angle of 4 (similar to the previous case 1). This geometry is equipped
with three probing holes to take various measurements. Optical probes and micro-
thermocouples are used to evaluate the local void ratio and the wall temperature,
respectively. Flow conditions and experimental parameters are given in Table 2. With
these parameters, a cavity length around 80 mm was obtained, with a relatively stable
aspect.
The freon R-114 allows to study the thermodynamic eect in ambient conditions.
It provides the same temperature depression T = 1.22 K in comparison with the
liquid hydrogen at Tref = 22 K.
All cavitating simulations are steady computations, which are started from the
non-cavitating numerical solution. The H-type grid contains 251 nodes in the ow
direction and 77 in the orthogonal direction.
Table 2 Flow conguration for the Venturi case with freon R-114
Inlet velocity Vinlet 14.4 m/s
Inlet pressure Pinlet 265300 Pa
Reference temperature Tref 293 K
Vapour pressure at Tref 181100
Cavitation parameter inlet 0.55
Reynolds number ReLref 18.4 106
Numerical Simulation of Cavitating Flows . . . 203
We present here only results obtained on the rst Venturi (4 divergence angle). All
unsteady simulations were performed with the same cavitation model: the 3-equation
sinus model.
A qualitative view of the cavity sheet is plotted in Fig. 5 with the contours of the den-
sity gradient modulus (Schlieren-like visualization). On the left, the simulated cavity
using the SpalartAllmaras model with the Reboud correction is in good agreement
with the experimental visualization. The numerical sheet presents an attached cav-
ity up to the abscissa x = 0.03 m and downstream a recirculating area with small
cavitation clouds. The re-entrant jet is well captured. On the contrary, the solution
obtained without the Reboud limiter presents an extended attached cavity (up to the
abscissa x = 0.06 m) followed by a very small recirculation. The re-entrant is not
developed.
The eect of the eddy viscosity limiter is shown in Fig. 6 where are plotted the
proles of the ratio t at station 3 (the re-entrant jet reaches this station). On the
left, we can clearly observed the drastic reduction of t using the Reboud limiter in
comparison with the computation without correction. In the sheet cavity, the ratio
is close to zero. On the left, a similar observation is done using the KE model and
0.05
0.02
0.04
0.03
0.01
y (m)
y (m)
0.02
0
0.01
0
0 0.02 0.04 0.06 0.08
x (m) 0 0.025 0.05 0.075 0.1
x (m)
Fig. 5 Contour of the density gradient: with (left) and without (right) the Reboud limiter
0.015 0.015
SA Reboud station 3 KE
SA KE Reboud
KE Realizable
0.01 0.01
Y(m)
Y(m)
0.005 0.005
0 0
0 25 mut / mu 50 75 0 25 mut50
/ mu 75 100
the Reboud limiter. Moreover, the use of realizability constrains allows similarly to
strongly reduce the eddy viscosity and to make possible the re-entrant jet develop-
ment.
The Reboud limiter can be associated to dierent transport-equation models with
the same value of the parameter n = 10. It clearly improves the behaviour of the
model and the prediction of the re-entrant jet. Figure 7 shows the time-averaged lon-
gitudinal velocity proles (on the right) and void ratio proles (on the left) for the
experiments and computations using dierent turbulence models. Four models are
tested: SpalartAllmaras, k , k and k SST and three limiters are consid-
ered: Reboud, realizability and SST correction. For all simulations, the re-entrant jet
is well developed and the intensity of the recirculation is correctly simulated except
with the KE Reboud model. For the void ratio proles, discrepancies appear between
models especially for the near-wall value. All models over-estimate the void ratio
value at wall. The thickness of the cavity is well predicted by all simulations. More
comparisons can be found in Goncalves (2011).
Using the Wilcox k turbulence model, we observed a strong inuence of the
value of the parameter n. Contours of the density gradients are illustrated in Fig. 8
for two values of the parameter n. On the left, with n = 4, we obtained a sheet cavity
in correct agreement with the experiments. Using the reference value n = 10, a large
shedding of two-phase structures is simulated and the re-entrant jet is too extended.
Moreover, the calibration of n depends on the conguration. For the second Venturi
0.01
EXPERIMENT 0.01
SA Reboud EXPERIMENT
KL Reboud SA Reboud
0.008 0.008 KL Reboud
KE Reboud
KE Realizable KE Reboud
KWSST KE Realizable
0.006 0.006 KWSST
Y (m)
Y (m)
0.004 0.004
0.002 0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 5 10
alpha u (m/s)
0.05 0.05
0.04 0.04
0.03 0.03
y (m)
y (m)
0.02 0.02
0.01 0.01
0 0
case (8 of divergence angle), the value n = 4 is not adapted, whereas the value n =
10 allows to correctly simulate the sheet cavity dynamic (see Decaix and Goncalves
2013a).
The SST correction can be associated to other turbulence models. For exam-
ple, we simulated the Venturi case using a KL-SST model. The value of the ratio
u v k = c is a key point. The original value is c = 0.3. Yet, we have no idea of this
value in cavitating ows. We performed three simulations by varying the parameter
c from 0.3 to 0.1. Using the original value c = 0.3, we obtained a similar result in
comparison with a computation without correction. The instantaneous density gradi-
ent contours are plotted in Fig. 9 for c = 0.2 (on the left) and c = 0.1 (on the right).
The intermediate value c = 0.2 allows to simulate a sheet cavity in correct agree-
ment with the experimental data. Using the smallest value c = 0.1, a large shedding
of two-phase structures is computed, which is not observed in the experiments.
The time-averaged velocity and void ratio proles at station 3 are plotted in
Fig. 10. The three simulations with the KL-SST model are presented and also a KL
realizable computation with c = 0.2. Using the value c = 0.3, the re-entrant jet is
not computed at station 3 and the void ratio is close to 1. The decrease of the con-
stant allows the development of the re-entrant jet. More comparisons can be found
in Decaix and Goncalves (2012).
0.01 0.01
EXP EXP
KL-SST : c = 0.3 KL-SST : c = 0.3
0.008 KL-SST : c = 0.2 0.008 KL-SST : c = 0.2
KL-SST : c = 0.1 KL-SST : c = 0.1
KL-Realizable : c = 0.2 KL-Realizable : c = 0.2
0.006 0.006
y (m)
y (m)
0.004 0.004
0.002 0.002
0 0
0 0.2 0.4 0.6 0.8 1 -4 -2 0 2 4 6 8 10 12 14
alpha u (m/s)
Fig. 10 Time-averaged void ratio (left) and velocity (right) proles, KL-SST model
206 E. Goncalves
0.05 0.05
0.04 0.04
0.03 0.03
y (m)
y (m)
0.02 0.02
0.01 0.01
0 0
0 0.025 0.05 0.075 0.1 0 0.025 0.05 0.075 0.1
x (m) x (m)
0.05 0.05
0.04 0.04
0.03 0.03
y (m)
y (m)
0.02 0.02
0.01 0.01
0 0
0 0.025 0.05 0.075 0.1 0 0.025 0.05 0.075 0.1
x (m) x (m)
0.08 0.05
0.04
0.06
0.03
y (m)
0.04
y (m)
0.02
0.02
0.01
0 0
0 0.05 0.1 0.15 0 0.025 0.05 0.075 0.1
x (m) x (m)
Fig. 13 Contours of the density gradient, k + , 2 =0.15, 3 = 0.001 (left) and 3 = 0.025
(right)
set to 0.15. A weak value of 3 leads to the large shedding of two-phase structures. By
increasing 3 , it is possible to simulate a sheet cavity in relatively correct agreement
with the experiments. Unfortunately, the simulation of the second Venturi case using
this value 3 = 0.025 did not provide a correct result. A similar conclusion was done
for the calibration of the extra-dissipation term. As a consequence, it was impossible
to calibrate these compressibility corrections for both Venturi cases (see Decaix and
Goncalves 2013a).
This numerical work investigates the mesh inuence in the near-wall area when a
partial cavity develops and interacts with the turbulent boundary layer. All meshes
contain 251 nodes in the ow direction and vary by the rst cell size and the dis-
tribution in the orthogonal direction. The y+ values obtained from a non cavitating
simulation vary between 9 and 60 near the throat (Fig. 14). All meshes are computed
using a two-layer wall model.
Comparisons are focused on the velocity proles at station 3 and the develop-
ment of the re-entrant jet. Four meshes and four turbulence models (coupled with
the Reboud correction except the SST model) are compared in Fig. 15. According to
Fig. 14 Evolution of y+ 80
251x77
values near the throat, 4 251x62
Venturi 60 251x61
251x59
40
y+
20
0
0 0.2 0.4 0.6 0.8 1
x (m)
208 E. Goncalves
y(m)
0.003 0.003
0.002 0.002
0.001 0.001
0 0
0 5 10 0 5 10
u (m/s) u (m/s)
SA KE
y(m)
0.003 0.003
0.002 0.002
0.001 0.001
0 0
0 5 10 0 5 10
u (m/s) u (m/s)
KL KW-SST
Fig. 15 Time-averaged velocity proles at station 3, mesh and turbulence model comparisons
the turbulence model, dierent behaviour can be highlighted. For the k model,
results provided by all meshes are similar and in very close agreement with the exper-
imental values. For the k model, the re-entrant jet is not predicted using the coars-
est mesh 251 59. With the other meshes, results are relatively similar and in close
agreement with the experiments. For the SpalartAllmaras model, large discrepan-
cies between meshes appear. Using the coarsest and nest meshes, the re-entrant jet
is not predicted. For the intermediate mesh, the recirculating area obtained with the
mesh 251 62 is in better agreement in comparison with the experimental data. The
sensitivity to the near-wall mesh renement seems important with this turbulence
model. Using the Menter SST model, the best result is also obtained with the mesh
251 62. The ner mesh provides a less intense recirculating area and the solution
is degraded when a coarser mesh is used. With the coarsest mesh, the re-entrant jet
is not predicted. To conclude, the near-wall mesh can be determinant for the good
simulation of the re-entrant jet.
Computations were performed to compare the two-layer wall functions and the
TBLE model, associated with the Smith k turbulence model and the mesh
251 62 nodes. Figure 16 shows the velocity and void ratio proles at station 3.
Results obtained with both wall treatments are similar and in very good agreement
with the experimental data. A similar conclusion has been done for the computa-
tion of the second Venturi case (see Goncalves and Decaix (2012)). The two-layer
single-phase wall functions seem to be a good approximation.
For the second Venturi case, the experimental frequency of the periodic shedding
is around 45 Hz and the inlet cavitation parameter inlet 2.15. Similar to the pre-
Numerical Simulation of Cavitating Flows . . . 209
y(m)
y(m)
0.004 0.003
0.002
0.002
0.001
0 0
0 0.2 0.4 0.6 0.8 1 0 5 10
alpha u (m/s)
Fig. 16 Time-averaged velocity and void ratio proles at station 3, two-layer model versus TBLE
vious study, dierent meshes varying by the rst cell size are compared using the
k Reboud turbulence model coupled with a two-layer wall model. The simula-
tion time is around 2 s. The numerical frequency and the cavitation parameter are
given in Table 3.
The ratio between CPU time for a given case and the ne mesh computation is also
given. With the coarsest mesh, the gain is around 25%. About the frequency predic-
tion, with the nest mesh 174 77 computed without wall functions, it was not pos-
sible to capture a periodic self-oscillating cavity with a frequency close to 45 Hz. For
values of the inlet cavitation parameter near 2.15, the obtained frequency is around
30 Hz. For higher values of inlet , no frequency appears. For all other meshes com-
puted with wall functions, a correct frequency can be obtained. According to the
mesh renement, the inlet increase allows to reach the experimental value. For the
coarsest meshes, both the frequency and the inlet value are close to the experimental
data. Moreover, the local velocity proles obtained with all meshes are similar (not
presented here).
210 E. Goncalves
0.04 0.04
0.035 0.035
0.03 0.03
0.025 0.025
y (m)
y (m)
0.02 0.02
0.015 0.015
0.01 0.01
0.005 0.005
0 0
0 0.025 0.05 0.075 0.1 0 0.025 0.05 0.075 0.1
x (m) x (m)
EXP
KL EXP
0.8 KL
KL-Reboud
16 KL-Reboud
KL-SAS 0.7
14
KL-SAS
0.6
12
0.5
P rms / Pav
10
(P-Pv)/Pv
8 0.4
6
0.3
4
0.2
2
0.1
0
0
-2
0.1 0.15 0.2 0.25 0.3
0.15 0.2 0.25 0.3
x-xi
x-xi
Fig. 18 Time-averaged wall pressure evolution (left) and RMS wall pressure uctuations (right)
Numerical Simulation of Cavitating Flows . . . 211
Fig. 19 Iso-surface of the void fraction for the value of 60% at two dierent instants, k SAS
model
0.01
0.01 EXPE
EXPE
0.009
0.009 SA-SAS 3D SA-SAS 3D
SA-SAS 2D 0.008 SA-SAS 2D
0.008
KL-SAS 3D KL-SAS 3D
0.007
0.007 KL-SAS 2D KL-SAS 2D
0.006
0.006
y (m)
y (m)
0.005
0.005
0.004
0.004
0.003
0.003
0.002
0.002
0.001
0.001
0 0
-4 -2 0 2 4 6 8 10 12 14 0 0.2 0.4 0.6 0.8 1
u (m/s) alpha
Fig. 20 Time-averaged velocity (right) and void ratio (left) proles at station 3, mid-span plane
212 E. Goncalves
0.8 EXP
16 EXP
SA-SAS 3D
SA-SAS 3D
SA-SAS 2D 0.7 SA-SAS 2D
14 KL-SAS 3D
KL-SAS 3D
KL-SAS 2D 0.6 KL-SAS 2D
12
10 0.5
P rms / P mean
(P-Pv)/Pv
8
0.4
6
0.3
4
2 0.2
0 0.1
-2
0.1 0.15 0.2 0.25 0.3 0
x-xi (m) 0.15 0.2 0.25 0.3
x-xinlet (m)
Fig. 21 Time-averaged wall pressure (left) and RMS uctuations (right), mid-span plane
The time-averaged wall pressure, plotted in Fig. 21, does not evolve in the same
way for the 3D computations. For the SpalartAllmaras model, the 3D computation
leads to a stronger re-compression area in comparison with the 2D computation. It
is the inverse situation for the k model. The RMS wall pressure uctuations (on
the right) put in evidence large discrepancies between 2D and 3D computations. For
the 2D computations, the peak of uctuations is correctly located with an amplitude
in correct agreement with the experimental measurements. Moreover, the decrease
of the uctuations level in the wake of the cavitation sheet is well captured. This
is not the case for the 3D computations since the level of pressure uctuations is
largely over-estimated in the wake. More details concerning the investigation of 3D
computations can be found in Decaix and Goncalves (2013b).
Dierent simulations were performed to compare the 4-equation sinus model and
the 4-equation Hosangadi model by varying the initial discontinuity velocity uinit
Numerical Simulation of Cavitating Flows . . . 213
0
10 4-eqt sinus 1
4-eqt Hosangadi 4-eqt sinus
Zein 4-eqt Hosangadi
Zein
0.8
P (bar)
alpha
-1
10
0.6
10-2 0.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
2
2 4-eqt sinus 10
4-eqt Hosangadi
Zein
1
u (m/s)
c (m/s)
1
0 10
4-eqt sinus
4-eqt Hosangadi
-1
-2 100
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
Fig. 22 Watergas double rarefaction with cavitation u = 2 m/s, init = 102 , mesh 5000 cells, t =
3.2 ms. Void ratio, pressure, velocity, and speed of sound proles
and the initial value of the void fraction init . For these cases, only the evaporation
process is present and only the Cprod parameter is studied. For the sinus model, the
parameter cbaro is set to 1.31 m/s for all computations.
First, we computed the case uinit = 2 m/s and init = 102 . Solutions are com-
pared in Fig. 22 at time t = 3.2 ms with the two-uid solution of Zein et al. (2010).
The Cprod parameter was calibrated using the two-uid solution and set to 1. Results
obtained with both models are in close agreement with the 2-uid solution for the
void ratio and velocity proles. For the pressure evolution, a small discrepancy
appears for the location of the rarefaction fronts, certainly due to a small dierence
on the mixture speed of sound. An important dierence between 4-equation mod-
els concerns the evaporation fronts, which are not simulated using the Hosangadi
model. A part of the physics is missing with this model. Finally, large discrepancies
are noticed about the mixture speed of sound proles (plotted in log scale). It is due to
the fact that the speed of sound associated to the Hosangadi model is the Wallis one.
Second, we computed the case uinit = 100 m/s and init = 102 . Solutions are
compared in Fig. 23 at time t = 1.5 ms with the two-uid solution of Zein et al.
(2010). The Cprod parameter was calibrated using the two-uid solution and set to
100. Results obtained with both models are in good agreement with the 2-uid solu-
tion for the void ratio and the velocity. In this case, evaporation is much more intense
resulting in a large cavitation pocket where the gas volume fraction is close to 1.
However, this pocket does not contain pure gas but a mixture at thermodynamic equi-
librium. Moreover, thermal eects are not negligible and a large temperature drop
214 E. Goncalves
100 1
4-eqt sinus 4-eqt sinus
4-eqt Hosangad 4-eqt Hosangadi
Zein 0.8 Zein
P (bar)
alpha
10
-1 0.6
0.4
0.2
10-2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
c (m/s)
0
101
-50
0
-100 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
Fig. 23 Watergas double rarefaction with cavitation u = 100 m/s, init = 102 , mesh 5000 cells,
t = 1.5 ms. Void ratio, pressure, velocity and speed of sound proles
appears leading to a pressure drop in the cavitation area. This phenomenon is not
taken into account by both 4-equation models, which explains the large dierence
with the 2-uid solution (see the next section for the use of non isothermal mod-
els). As previously observed, evaporation fronts are not simulated by the Hosangadi
model and large dierences are put in evidence for the mixture speed of sound.
Finally, we computed the case uinit = 100 m/s with a small initial value of gas
init = 1010 . For this case, no reference solutions are available. Various Cprod values
were tested. From the value 106 , the numerical solution did not move any more and
we set the parameter to this value. Solutions are compared in Fig. 24 at time t =
0.2 ms. The rarefaction fronts are stier in comparison with the previous case. The
cavitation area and the evaporation fronts can be clearly observed with the 4-equation
sinus solution, that is not the case for the other model. The mixture speed of sound
is plotted with a logarithmic scale for both 4-equation models. Large variations are
highlighted, from the pure liquid high value to small values in the cavitation pocket.
At the middle of the tube where the void ratio is close to 1 the speed of sound is close
to the vapour value. For the sinus speed of sound, the four waves can be observed
and the ratio between extremal sound velocities is higher than 1000. These variations
make the numerical solving of the system very sti.
For the considered test cases, the domain of evolution of Cprod is [1, 106 ] that is
really questionable. Other comparisons can be found in Goncalves (2013).
Numerical Simulation of Cavitating Flows . . . 215
1
4-eqt sinus 103
4-eqt Hosangadi
0.8
102
P (bar)
c (m/s)
0.6 101 4-eqt sinus
4-eqt Hosangad
0.4 100
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
Fig. 24 Watergas double rarefaction with cavitation u = 100 m/s, init = 1010 , mesh 5000 cells,
t = 0.2 ms. Pressure and speed of sound proles
0.04
0.03
0.02
y
0.01
0
0 0.05 0.1
x
Fig. 27 Contours of the density gradient modulus: 3-equation model (left) and 4-equation model
(right)
y(m)
0.004
0.002
0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 5 10
alpha u (m/s)
transport-equation model did not improve the numerical results in comparison with
the 3-equation sinus model (see Goncalves and Charriere 2014).
4-equation models Two formulations of 4-equation models have been previously
presented in the theoretical part, based on either a mixture of stiened gas EOS
(called 4-equation SG model) or a sinus law (called 4-equation sinus model). The
sheet cavities obtained with these models are compared in Fig. 29 with the plot of
the instantaneous density gradient modulus. Both simulations put in evidence a quite
similar dynamics involving an attached cavity (up to the abscissa x = 0.04 m) and a
recirculating area with the generation of small vapour cloud shedding.
A local analysis is led by studying void ratio and velocity proles, which are pre-
sented in Fig. 30 at station 3. Results obtained with both models are in good agree-
ment with the experiments. The 4-equation SG cavitation model slightly provides a
better prediction of the void ratio wall value. Moreover, this model provides a better
Numerical Simulation of Cavitating Flows . . . 217
Fig. 29 Contours of the density gradient modulus: 4-equation SG model (left) and 4-equation
sinus model (right)
0.04 0.04
0.03 0.03
0.02
y (m)
y(m)
0.02
0.01 0.01
0 0
-0.01 -0.01
0.02 0.04 0.06 0.08 0.02 0.04 0.06 0.08
x(m) x(m)
Fig. 31 Instantaneous contour of void fraction, incompressible code (left) and compressible code
(right)
0.01
EXPERIMENT 0.01 EXPERIMENT
0.008
Compressible station 3 Compressible
Incompressible Incompressible
0.008
0.006
station 3
Y (m)
0.006
Y (m)
0.004
0.004
0.002 0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 5 10
alpha u (m/s)
shedding is well illustrated. On the other hand, the incompressible code captured a
low frequency periodic sheet dynamics without re-entrant jet (around 6 Hz).
Figure 32 shows the longitudinal velocity and void ratio proles for experiments
and numerical results at station 3. For the void ratio proles, both codes provided
a similar solution in good agreement with the experimental data. The recirculating
behaviour with a re-entrant jet is well simulated by the compressible code. On the
contrary, using the incompressible code the re-entrant jet is not captured. At stations
4 and 5, the re-entrant is largely under-estimated (see Goncalves et al. 2010a).
Another comparison is proposed with the incompressible code OpenFOAM. Sim-
ulations were performed on the 4 Venturi case using a similar mesh and the k
SST turbulence model. The considered cavitation models are the 4-equation SG
model and the Kunz model for the CaviFlow and OpenFOAM codes, respectively
(see Charriere et al. 2015).
Figure 33 shows the instantaneous void fraction eld downstream the Venturi
throat for both computations. Using the openFOAM solver, the attached cavity is
extended with void ratio values close to 1 and a small liquid lm can be observed
along the wall from abscissa x = 0.05 m. On the contrary, the attached sheet cav-
ity computed with the compressible CaviFlow solver is in good agreement with the
experiments. Downstream, we can observe the recirculating area with two-phase
structure shedding.
Numerical Simulation of Cavitating Flows . . . 219
Fig. 33 Instantaneous contour of void fraction, incompressible code (left) and compressible code
(right)
0.008 0.01
CaviFlow CaviFlow
OpenFOAM OpenFOAM
Experiment 0.008 Experiment
0.006
0.006
y (m)
y (m)
0.004
0.004
0.002
0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 5 10 15
alpha U (m/s)
Fig. 34 Time-averaged void ratio (left) and velocity (right) proles at station 3
In Fig. 34 are presented the time-averaged velocity and void ratio proles at sta-
tion 3. The compressible solver provides a better prediction of the void ratio prole
due to the presence of a mixture in this area. The OpenFOAM computation over-
predicts the vapour quantity inside the cavity, which can reach 95%. Moreover, the
thickness of the cavity is under-estimated and a pure liquid phase is simulated close to
the wall. As regard to the re-entrant jet, the OpenFOAM simulation badly reproduces
the recirculation area. On the contrary, the recirculating behaviour with a re-entrant
jet is correctly simulated by the compressible code.
To conclude, the topology of the cavitation pocket marks large discrepancies
between compressible and incompressible codes, especially concerning the re-entrant
jet development. Comparisons with measurements suggest that the compressible for-
mulation provides an improved description of the cavitation dynamics compared
with incompressible computations.
Unsteady simulations were performed on the 8 Venturi case, which involves a self-
sustained oscillating sheet cavity. The same mesh 174 56 is used for all simula-
220 E. Goncalves
tions. The simulation time is around 2 s. The considered turbulence model is the
SpalartAllmaras one with the Reboud limiter.
3-equation versus 4-equation We compare the 3-equation and 4-equation models
coupled with the sinus EOS. The obtained frequency is close to the experimental for
both codes (around 45 Hz). A qualitative description of the sheet cavities is illustrated
in Fig. 35 with the instantaneous density gradient contours. Using the 4-equation
model, results provide a shedding phenomenon in better agreement with the experi-
mental data. With the 3-equation model, we can observe an extended attached cavity,
and downstream the uctuating recirculation with two-phase structures shedding.
These shedding are rapidly eliminated and under-estimated in comparison with the
experimental visualizations and the 4-equation simulations.
Figure 36 presents the time-averaged void ratio and longitudinal velocity pro-
les at station 1. Near the throat, the void ratio is largely over-estimated using the
3-equation model. The attached cavity is too extended. Results are similar for the
velocity proles in good agreement with the experiments.
In Fig. 37 are plotted the time-averaged void ratio proles at station 3 and the
time-averaged dimensionless wall pressure distribution. Using the 3-equation model,
the averaged value of the void ratio is null. As commented previously, the cloud
shedding is not enough developed. On the contrary, it is more intense using the 4-
equation model. For the average pressure, both models provide a result in correct
agreement with the experimental data. More comparisons can be found in Goncalves
and Charriere (2014).
To conclude between the 3 and 4-equation models and their ability to compute
both Venturi cases:
0.02 0.02
y (m)
y (m)
0.01 0.01
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
x (m) x (m)
Fig. 35 Contours of the density gradient modulus: 3-equation model (left) and 4-equation model
(right)
y (m)
0.002
0.002
0.001
0 0
0 0.1 0.2 0.3 0.4 0 5 10
alpha u (m/s)
EXPE
10 4-eqt
0.005
3-eqt
5
0 0
0 0.02 0.04 0.06 0.08 0.1 0.15 0.2 0.25 0.3 0.35
alpha x-xi (m)
Fig. 37 Time-averaged void ratio prole at station 3 (left) and wall pressure evolution (right)
Fig. 38 Contours of the density gradient modulus: 4-equation SG model (left) and 4-equation
sinus model (right)
For the quasi-stable cavity, the void ratio transport-equation does not clearly
improve the previous simulations obtained with a 3-equation model.
For the periodic self-oscillating cavity, the two-phase structures shedding is more
intense and in better agreement with the experimental data using the 4-equation
model. The transport of void ratio seems to be a key point to correctly predict the
sheet cavity dynamics.
4-equation models We compare the two formulations of 4-equation models based
on a mixture of stiened gas EOS and a sinus EOS. First, the obtained frequency
using the 4-equation SG model is around 20 Hz instead of 45 Hz. Except the fre-
quency, the overall agreement with the experimental visualizations is correct. The
instantaneous density gradient contours are plotted in Fig. 38 for both models. The
two-phase shedding is well captured by both models.
Figure 39 presents the time-averaged void ratio and longitudinal velocity proles
at station 1. For the velocity prole, both models provide a result in correct agreement
with the experimental data. The void ratio prole is better predicted using the 4-
equation SG model, especially the near-wall values.
Void ratio proles at stations 2 and 3 are plotted in Fig. 40. As previously, a bet-
ter agreement is observed using the 4-equation SG model, but the frequency of the
phenomenon is not correct.
Comparison with an incompressible code Unsteady simulations obtained using
the incompressible IZ and compressible CaviFlow codes are compared. The same
mesh is used (251 61 nodes) and the turbulence model is a k with the Reboud
222 E. Goncalves
0.05
0.04 0.03
0.03
y (m)
y(m)
0.02
0.02
0.01 0.01
0
0 0.02 0.04 0.06 0.08 0 0.02 0.04
x(m) x(m)
Fig. 41 Instantaneous void ratio eld: incompressible code (left) and compressible code (right)
limiter. The 3-equation sinus model is considered for the compressible simulations
and a barotropic sinus model for the incompressible ones. Both codes provided an
oscillating phenomenon with the correct frequency in comparison with the experi-
mental value.
An instantaneous view of the cavity (void ratio contours) obtained with the incom-
pressible code is illustrated in the left part of Fig. 41. Large-scale two-phase structure
shedding is clearly observed. The re-entrant jet is very intense and ows upstream to
the throat of the Venturi, leading to the disappearance of the attached cavity. The
solution obtained with the 3-equation compressible model presented in the right
part of Fig. 41 is dierent. As previously commented, an extended attached cavity is
predicted and two-phase structures convected in the ow are smaller and condense
rapidly. Results are improved with the use of the 4-equation sinus model.
Numerical Simulation of Cavitating Flows . . . 223
0.005
EXPERIMENT 0.005 EXPERIMENT
Compressible Compressible
0.004 Incompressible 0.004 Incompressible
Y (m)
0.002 0.002
0.001 0.001
0 0
0 0.2 0.4 0.6 0.8 1 -5 0 5 10
alpha u (m/s)
0.01
station 2 0.01
0.008 station 2
0.008
0.006
0.006
Y (m)
Y (m)
0.004 0.004
0.002 0.002
0 0
0 0.1 0.2 0.3 0.4 -5 0 5 10
alpha u (m/s)
Fig. 42 Time-averaged void ratio (left) and velocity (right) proles at stations 1 and 2
Figure 42 illustrates the time-averaged void ratio and velocity proles at stations
1 and 2. At station 1, with the compressible code, the recirculating behaviour is very
well captured, yet the void ratio is over-predicted. For the incompressible code, the
void ratio is slightly under-estimated due to the disappearance of the attached cavity.
At station 2, both codes under-evaluated the recirculating area thickness. For the void
ratio prole, values obtained with the incompressible code are largely over-predicted,
whereas values obtained by the compressible code are in better agreement with the
experimental data.
To conclude these studies between compressible and incompressible codes:
The incompressible IZ code seems to be unable to correctly simulate the quasi-
stable Venturi case using a barotropic model and a Reboud limiter. Similarly, the
openFOAM code using the Kunzs void ratio transport-equation model and the
Menter SST model is unable to reproduce the re-entrant jet phenomenon. Is it due
to a problem of waves propagation or compressibility eects?
For the second Venturi case, using the incompressible IZ code a very intense re-
entrant jet is captured, which ows upstream to the throat and completely elimi-
nates the attached cavity. For the compressible CaviFlow code, large-scale cavita-
tion structure shedding are under-predicted using the 3-equation model, whereas
the phenomenon is correctly simulated using the void ratio transport-equation.
224 E. Goncalves
1 4-eqt sinus 1
4-eqt SG
Zein 4-eqt sinus
0.8 0.8 4-eqt SG
Zein
0.6 0.6
P (bar)
alpha
0.4 0.4
0.2 0.2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
100 360
4-eqt sinus
75 4-eqt SG
Zein 340
50
25 320
u (m/s)
T (K)
0
-25 300
-50
280 4-eqt sinus
-75 4-eqt SG
-100 260
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m)
Fig. 43 Watergas double rarefaction with cavitation u = 100 m/s, init = 102 , mesh 5000 cells,
t = 1.5 ms. Void ratio, pressure, velocity and temperature proles
Numerical Simulation of Cavitating Flows . . . 225
100 1
4-eqt sinus
0.8 4-eqt SG
Zein
0.6
P (bar)
alpha
-2
10 00
0 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
x (m) x (m)
Fig. 44 Watergas double rarefaction with cavitation u = 500 m/s, init = 102 , mesh 5000 cells,
t = 0.58 ms. Void ratio and pressure proles
pressure drop is near 0.5 bar under the vapour pressure at the reference temperature.
The value of dPvap dT is set to 130 Pa/K. In Fig. 44 are presented the void ratio and
pressure proles at time t = 0.58 ms. Results obtained with both 4-equation models
are in good agreement with the two-uid solution.
Steady simulations were performed on the freon R-114 Venturi case to investigate
cavitation models. The considered mesh contained 251 77 nodes and the Spalart-
Allmaras turbulence model was used for all simulations.
Isothermal and non isothermal 3-equation models Comparisons between the
isothermal and non isothermal formulations of the 3-equation sinus model are pro-
posed. The value of the constant dPvap dT is set to 5900 K/Pa and the parameter cbaro
is set to 0.74 m/s. A study of the inuence of this parameter is given in Goncalves
and Patella (2010).
For both models, a steady pocket for witch the length is around 80 mm was simu-
lated. Void ratio proles at stations 1 to 3 are plotted in Fig. 45 with the experimental
data. The cavity thickness is well predicted using the isothermal model but the maxi-
mal values of void ratio are too high (85% instead of 45% at station 1). Using the non
isothermal model, the thickness is increased and the maximum values are decreased.
The cooling eect during the phase transition is studied by plotting the temperature
dierence Tref T at station 1 (Fig. 45). The measured wall temperature is 2.1 K.
The wall value is correctly predicted by the non isothermal model. With the isother-
mal model, temperature variations are important inside the cavity and the peak of
temperature decit can reach 12 K.
Simulations were performed with other ow conditions: Tref = 303 K and Uinlet =
22 m/s. Two inlet cavitation parameters were considered: inlet = 0.55 and inlet =
0.59. The goal was to obtain a stable cavitation sheet with a length close to the experi-
mental one. Results provided by the 3-equation non isothermal model are compared
with the experimental data. The wall temperature depression is plotted in Fig. 46
226 E. Goncalves
0.01
0.008
Station 2
0.006
y (m)
0.004
0.002
0
0 0.2 0.4 0.6 0.8 1
alpha
0.012
0.01
Station 3
0.008
y (m)
0.006
0.004
0.002
0
0 0.2 0.4 0.6 0.8 1
alpha
4 Experiment 4 Experiment
Computation Computation
Tref-T
2 2
1 1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(x-xt)/Lcav (x-xt)/Lcav
versus the distance (x xthroat )Lcav for the two inlet cavitation numbers. The tem-
perature decit is well predicted by computations and results are in good agreement
with the experimental data. Yet, for higher inlet velocities, the model fails to predict
the temperature depression (in comparison with the experimental data, the error is
around 50%, see Goncalves and Patella 2010).
3-equation versus 4-equation models Both non isothermal sinus formulations have
been compared on the Venturi case. The void ratio follows an analytical relation
for the 3-equation model, whereas it follows a transport-equation for the 4-equation
model.
Numerical Simulation of Cavitating Flows . . . 227
0.06 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0.06 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
0.04 0.04
y (m)
y (m)
0.02 0.02
0 0
-0.02 -0.02
Fig. 47 Temperature dierence T Tref (K) downstream the Venturi throat: 3-equation model
(left) versus 4-equation model
0.04 0.04
y (m)
y (m)
0.02 0.02
0 0
Fig. 48 Gradient of temperature (K/m) downstream the Venturi throat: 3-equation model (left)
versus 4-equation model
Figure 47 shows the temperature decit T Tref inside the Venturi divergent. The
cooling eect due to the vaporisation process is clearly observed for all simulations
(negative values). A large discrepancy appears between the 3- and 4-equation mod-
els downstream the cavity in the re-compression area. Using the 3-equation model,
the temperature decit is close to zero downstream the cavitation pocket, that is the
temperature goes back to the freestream temperature. On the contrary, we observe
a warming eect downstream the cavity using the 4-equations model. Locally, the
temperature exceeds the freestream temperature and values reach more than 4 K.
The temperature gradient is illustrated in Fig. 48 for both models. The grey scale
is the same for all computations. We clearly observe dierences on the behaviour
of models in the closure part of the cavitation pocket. Using the 3-equation model,
there is no temperature gradient at the end of the pocket. We just see a small gradient
at the interface in the vicinity of the throat. With the 4-equation model, strong gra-
dients are put in evidence in the closure part of the pocket and downstream in the re-
compression area. Such a phenomenon was depicted in Petkovsek and Dular (2013)
on a hot water Venturi ow due to the collapse of bubbles. Using high-speed infra-red
thermography, authors measured a temperature depression of approximately 0.4 K
in the vicinity of the throat and a temperature rise up to 1.4 K was recorded in the
region of pressure recuperation.
Figure 49 illustrates the numerical void ratio obtained with three models (3-
equation sinus, 4-equation sinus and 4-equation SG models) in comparison with the
measurements at stations 2 and 3. The cavity thickness is clearly better predicted
by the 4-equation sinus model. Using the 3-equation model, the thickness is over-
estimated with a factor 2. An intermediate result is provided by the 4-equation SG
model. Other comparisons can be found in Goncalves (2014).
228 E. Goncalves
The simulation of cavitating ows using 1-uid RANS tools is confronted to open
questions as regard to the turbulence modelling, the thermodynamic path during the
phase transition, the mass transfer modelling and the turbulence-cavitation interac-
tion. In this paper, various turbulence and cavitation models have been compared
and tested in their ability to correctly simulate cavitation pockets which can occur in
a variety of practical cases. Models has been implemented in a compressible RANS
solver including a low Mach number preconditioning algorithm and has been applied
for the simulation of two dierent partial cavities developing along Venturi geome-
tries. Furthermore, a comparison with incompressible pressure-based solvers were
proposed. Results lead to dierent concluding remarks:
(i) For the quasi-stable cavity, it was shown that both tested incompressible codes
were unable to predict such sheet dynamics. The presence of wave propagation
involved in the sheet dynamics is a possible explanation. Moreover, it seems that
a void ratio transport-equation is not necessary to correctly simulate this case using
the compressible code.
(ii) For the periodic self-oscillating cavity, the two-phase structures shedding was
more intense and in better agreement with the experimental data using the 4-equation
model in comparison with the 3-equation model. Using the incompressible code, the
re-entrant jet phenomenon was over-estimated leading to the disappearance of the
attached cavity near the Venturi throat.
On the other hand, the turbulence models comparison put in evidence the impor-
tance of the eddy viscosity limitation (Reboud limiter, SST correction, realizability
constrains, scale-adaptive term) and the near-wall boundary-layer treatment for both
Venturi cases. Moreover, it was impossible to calibrate the compressible or dilata-
tional terms appearing in the mixture TKE equation for both cases using aerody-
namic closures. The modelling of theses terms in cavitating ows is still an open
question. Given that it is extremely dicult to quantify this term, it seems obvi-
ous to report the modelling assumptions to the dissipative scales by developing a
Numerical Simulation of Cavitating Flows . . . 229
large-eddy simulation formalism adapted to cavitating ows. Yet, the 1-uid ltered
equations have not been clearly expressed and investigated.
Finally, a study of cavitation models in the case of thermo-sensitive uid was pre-
sented. A 4-equation model was developed to study thermal eects based on a void
ratio transport equation and assuming a linear variation of the vapour pressure with
the mixture temperature. First validations on inviscid one-dimensional cases showed
the ability of 4-equation non isothermal models to simulate the cavitation develop-
ment in which the running uid is hot water. Secondly, RANS simulations were per-
formed to study a stable cavitation pocket developing along a Venturi geometry in
which the working uid is freon R-114. In comparison with 3-equation simulations,
the cavity thickness was clearly improved. Moreover, a warming eect was exhib-
ited downstream the cavity in the re-compression area. The intensity of this warming
eect is higher than the cooling eect observed near the Venturi throat. This rise of
temperature is due to the collapse of bubbles in the closure part of the pocket. This
new 4-equation model is therefore very attractive to study thermodynamic eects
and cryogenic cavitation.
Acknowledgements The author is grateful to Prof. Salvetti and Prof. dAgostino for their invitation
to the Udine summer school. The author expresses his gratitude to the French Space Agency CNES
and to the SNECMA Space Engines Division to support these studies. The author thanks the IDRIS
CNRS supercomputing centre for providing us their computing resources.
References
Aeschlimann, V., & Barre, S. (2009). PIV-LIF determination of mean velocity eld and Reynolds
stress tensor in a cavitatiting mixing layer. In 7th International Symposium on Cavitation
CAV2009, Ann Arbor, USA.
Aeschlimann, V., Barre, S., & Djeridi, H. (2011). Velocity eld analysis in an experimental cavi-
tating mixing layer. Physics of Fluids, 23, 055105.
Aeschlimann, V., Barre, S., & Djeridi, H. (2013). Unsteady cavitation analysis using phase averag-
ing and conditional approaches in a 2D Venturi. Open Journal of Fluid Dynamics, 3, 171183.
Ahn, S. J., & Kwon, O. J. (2013). Numerical investigation of cavitating ows for marine propulsors
using an untructured mesh technique. International Journal of Heat and Fluid Flow, 43, 259
267.
Barre, S., Rolland, J., Boitel, G., Goncalves, E., & Patella, R. F. (2009). Experiments and mod-
elling of cavitating ows in Venturi: Attached sheet cavitation. European Journal of Mechanics
B/Fluids, 28, 444464.
Barret, M., Faucher, E., & Herard, J. M. (2002). Schemes to compute unsteady ashing ows. AIAA
Journal, 40(5), 905913.
Besnard, D. C., & Harlow, F. H. (1988). Turbulence in multiphase ow. International Journal of
Multiphase Flow, 14(6), 679699.
Bilicki, Z., Kwidzinski, R., & Mohammadein, S. A. (1996). Evaluation of the relaxation time of
heat and mass exchange in the liquidvapour bubble ow. International Journal of Heat and
Mass Transfer, 39(4), 753759.
Bois, G., Jamet, D., & Lebaigue, O. (2010). Towards large eddy simulation of two-phase ow with
phase change: Direct numerical simulation of a pseudo-turbulent two-phase condensing ows.
In 7th International Conference on Multiphase Flow, Tampa, USA.
230 E. Goncalves
Charriere, B., Decaix, J., & Goncalves, E. (2015). A comparative study of cavitation models in a
venturi ow. European Journal of Mechanics B/Fluids, 49, 287297.
Chen, Y., Lu, C. J., & Wu, L. (2006). Modelling and computation of unsteady turbulent cavitation
ows. Journal of Hydrodynamics, 18(5), 559566.
Choi, Y. H., & Merkle, C. L. (1993). The application of preconditioning to viscous ows. Journal
of Computational Physics, 105(2), 207223.
Clerc, S. (2000). Numerical simulation of the homogeneous equilibrium model for two-phase ows.
Journal of Computational Physics, 161(1), 354375.
Coutier-Delgosha, O., Fortes-Patella, R., & Reboud, J. L. (2002). Simulation of unsteady cavitation
with a two-equation turbulence model including compressibility eects. Journal of Turbulence,
3(58).
Davidson, L. (2006). Evaluation of the SST-SAS model: Channel ow, asymmetric diuser and axi-
symmetric hill. In Proceedings of ECCOMAS CFD 2006, Egmond aan Zee, The Netherlands.
Decaix, J., & Goncalves, E. (2012). Time-dependent simulation of cavitating ow with k tur-
bulence models. International Journal for Numerical Methods in Fluids, 68, 10531072.
Decaix, J., & Goncalves, E. (2013a). Compressible eects modelling in turbulent cavitating ows.
European Journal of Mechanics B/Fluids, 39, 1131.
Decaix, J., & Goncalves, E. (2013b). Investigation of three-dimensional eects on a cavitating ven-
turi ow. International Journal of Heat and Fluid Flow, 44, 576595.
Delannoy, Y., & Kueny, J. L. (1990). Two phase ow approach in unsteady cavitation modelling.
Cavitation and Multiphase Flow Forum, ASME-FED, 98, 153158.
Downar-Zapolski, P., Bilicki, Z., Bolle, L., & Franco, J. (1996). The non-equilibrium relaxation
model for one-dimensional ashing liquid ow. International Journal of Multiphase Flow, 22(3),
473483.
Durbin, P. A. (2009). Limiters and wall treatments in applied turbulence modeling. Fluid Dynamics
Research, 41(1), 012203.
Edwards, J. R., & Franklin, R. K. (2000). Low-diusion ux splitting methods for real uid ows
with phase transition. AIAA Journal, 38(9), 16241633.
Elghobashi, S. E., & Abou-Arab, T. W. (1983). A two-equation turbulence model for two-phase
ows. Physics of Fluids, 26(4), 931938.
Fruman, D. H., Reboud, J. L., & Stutz, B. (1999). Estimation of thermal eects in cavitation of
thermosensible liquids. International Journal of Heat and Mass Transfer, 42, 31953204.
Goncalves, E. (2011). Numerical study of unsteady turbulent cavitating ows. European Journal
of Mechanics B/Fluids, 30(1), 2640.
Goncalves, E. (2013). Numerical study of expansion tube problems: Toward the simulation of cav-
itation. Computers & Fluids, 72, 119.
Goncalves, E. (2014). Modeling for non isothermal cavitation using 4-equation model. Interna-
tional Journal of Heat and Mass Transfer, 76, 247262.
Goncalves, E., & Charriere, B. (2014). Modelling for isothermal cavitation with a four-equation
model. International Journal of Multiphase Flow, 59, 5472.
Goncalves, E., & Decaix, J. (2012). Wall model and mesh inuence study for partial cavities. Euro-
pean Journal of Mechanics B/Fluids, 31(1), 1229.
Goncalves, E., & Fortes, R. (2009). Patella. Numerical simulation of cavitating ows with homo-
geneous models. Computers & Fluids, 38(9), 16821696.
Goncalves, E., & Patella, R. F. (2010). Numerical study of cavitating ows with thermodynamic
eect. Computers & Fluids, 39(1), 99113.
Goncalves, E., & Patella, R. F. (2011). Constraints on equation of state for cavitating ows with
thermodynamic eects. Applied Mathematics and Computation, 217, 50955102.
Goncalves, E., Champagnac, M., & Fortes, R. (2010a). Patella. Comparison of numerical solvers
for cavitating ows. International Journal of Computational Fluid Dynamics, 24(6), 201216.
Goncalves, E., & Fortes, R. (2010b). Patella, J. Rolland, B. Pouary, and G. Challier. Thermody-
namic eect on a cavitating inducer in liquid hydrogen. Journal of Fluids Engineering, 132,
111305.
Numerical Simulation of Cavitating Flows . . . 231
Guillard, H., & Viozat, C. (1999). On the behaviour of upwind schemes in the low Mach number
limit. Computers & Fluids, 28(1), 6386.
Hosangadi, A., & Ahuja, V. (2005). Numerical study of cavitation in cryogenic uids. Journal of
Fluids Engineering, 127(2), 267281.
Hu, Z. M., Dou, H. S., & Khoo, B. C. (2011). On the modied dispersion-controlled dissipative
(DCD) scheme for computation of ow supercavitation. Computers & Fluids, 40(1), 315323.
Ihm, S. W., & Kim, C. (2008). Computations of homogeneous-equilibrium two-phase ows with
accurate and ecient shock-stable schemes. AIAA Journal, 46(12), 30123037.
Ishii, C. M., & Hibiki, T. (2006). Thermo-uid dynamics of two-phase ow. Springer.
Jameson, A., Schmidt, W., & Turkel, E. (1981). Numerical solution of the Euler equations by nite
volume methods using Runge-Kutta time stepping schemes. In AIAA Paper, 811259.
Ji, B., Luo, X., Wu, Y., Peng, X., & Duan, Y. (2013). Numerical analysis of unsteady cavitating
turbulent ow and sheding horse-shoe vortex structure around a twisted hydrofoil. Interational
Journal of Multiphase Flow, 51, 3343.
Jones, W. P. (1979). Prediction methods for turbulent ows. In Von Karman Institute. Lecture Series
19792.
Jones, W. P., & Launder, B. E. (1972). The prediction of laminarization with a two-equation model
of turbulence. International Journal of Heat and Mass Transfer, 15, 301314.
Kapila, A. K., Meniko, R., Bdzil, J. B., Son, S. F., & Stewart, D. S. (2001). Two-phase modeling of
deagration-to-detonation transition in granular materials: Reduced equations. Physics of Fluids,
13(10), 30023024.
Kataoka, I., & Serizawa, A. (1989). Basic equations of turbulence in gas-liquid two-phase ow.
International Journal of Multiphase Flow, 15(5), 843855.
Kunz, R. F. Boger, D. A., Stinebring, D. R., Chyczewski, T. S., Lindau, J. W., Gibeling, H. J.,
Venkateswaran, S., & Govindan, T. R. (2000). A preconditioned Navier-Stokes method for two-
phase ows with application to cavitation prediction. Computers & Fluids, 29(8), 849875.
Labourasse, E., Lacanette, D., Toutant, A., Vincent, S., Lubin, P., Lebaigue, O., Caltagirone, J.-P.,
& Sagaut, P. (2007). Towards large eddy simulation of isothermal two-phase ows: Governing
equations and a priori test. The International Journal of Multiphase Flow, 33(1), 139.
Lance, M., Marie, J. L., & Bataille, J. (1984). Construction of a model for liquid phase turbulence
in bubble ow. La Houille Blanche, 34, 255260.
Larocque, J., Vincent, S., Lacanette, D., Lubin, P., & Caltagirone, J.-P. (2010). Parametric study
of LES subgrid terms in a turbulent phase separation. International Journal of Heat and Fluid
Flow, 31, 536544.
Liovic, P., & Lakehal, D. (2012). Subgrid-scale modelling of surface tension within interface
tracking-based Large Eddy and Interface Simulation of 3D interfacial ows. Computers & Flu-
ids, 63, 2746.
Liu, T. G., Khoo, B. C., & Xie, W. F. (2004). Isentropic one-uid modelling of unsteady cavitating
ow. Journal of Computational Physics, 201(1), 80108.
Menter, F. R. (2010a). The scale-adaptive simulation mathod for unsteady turbulent ow predic-
tions. Part 1: Theory and model description. Flow, Turbulence and Combustion, 85(1), 113138.
Menter, F. R. (2010b). The scale-adaptive simulation method for unsteady turbulent ow predic-
tions. Part 2: Application to complex ows. Flow, Turbulence and Combustion, 85(1), 139165.
Menter, F. R. (1994). Two-equation eddy-viscosity turbulence models for engineering applications.
AIAA Journal, 32(8), 15981605.
Menter, F. R. (1997). Eddy viscosity transport equations and their relation to the k model.
Journal of Fluids Engineering, 119, 876884.
Menter, F. R., & Egorov, Y. (2004). Re-visiting the turbulent scale equation. In Proceeding IUTAM
Symposium; One Hundred Years of Boundary Layer ResearchGottingen.
Merci, B., Steelant, J., Vierendeels, J., Riemslagh, K., & Dick, E. (2000). Computational treatment
of source terms in two-equation turbulence models. AIAA Journal, 38(11), 20852093.
Merkle, C. L., Feng, J., & Buelow, P. E. (1998). Computation modeling of the dynamics of sheet
cavitation. In 3rd International Symposium on Cavitation CAV1998, Grenoble, France.
232 E. Goncalves
Le Metayer, O., Massoni, J., & Saurel, R. (2004). Elaborating equations of state of a liquid and its
vapor for two-phase ow models. International Journal of Thermal Sciences, 43, 265276.
Petitpas, F., Massoni, J., Saurel, R., Lapedie, E., & Munier, L. (2009). Diuse interface model for
high speed cavitating underwater systems. International Journal of Multiphase Flow, 35, 747
759.
Petkovsek, M., & Dular, M. (2013). IR measurements of the thermodynamic eects in cavitating
ow. International Journal of Heat and Fluid Flow, 44, 756763.
Reboud, J.-L., Stutz, B., & Coutier, O. (1998). Two-phase ow structure of cavitation: Experiment
and modelling of unsteady eects. In 3rd International Symposium on Cavitation CAV1998,
Grenoble, France.
Roe, P. L. (1981). Approximate Riemann solvers, parameters vectors, and dierence schemes. Jour-
nal of Computational Physics, 43, 357372.
Sarkar, S. (1992). The pressure-dilatation correlation in compressible ows. Physics of Fluids,
4(12), 26742682.
Saurel, R., Cocchi, J. P., & Barry, P. (1999). Butler. Numerical study of cavitation in the wake of a
hypervelocity underwater projectile. Journal of Propulsion and Power, 15(3), 513522.
Saurel, R., Petitpas, F., & Abgrall, R. (2008). Modelling phase transition in metastable liquids:
application to cavitating and ashing ows. Journal of Fluid Mechanics, 607, 313350.
Saurel, R., Petitpas, F., & Berry, A. (2009). Simple and ecient relaxation methods for interfaces
separating compressible uids, cavitating ows and shocks in multiphase mixtures. Journal of
Computational Physics, 228(5), 16781712.
Schmidt, D., Rutland, C., & Corradini, M. L. (1999). A fully compressible, two-dimensional model
of small, high-speed, cavitating nozzles. Atomization and Sprays, 9, 255276.
Schmidt, S. J., Sezal, I. H., & Schnerr, G. H. (2006). Compressible simulation of high-speed
hydrodynamics with phase change. In European Conference on Computational Fluid Dynamics
ECCOMAS 2006, Delft, The Netherlands.
Senocak, I., & Shyy, W. (2002). A pressure-based method for turbulent cavitating ow computa-
tions. Journal of Computational Physics, 176(2), 363383.
Shin, B. R., Iwata, Y., & Ikohagi, T. (2003). Numerical simulation of unsteady cavitating ows
using a homogeneous equilibrium model. Computational Mechanics, 30, 388395.
Singhal, A. K., Athavale, M. M., Li, H., & Jiang, Y. (2002). Mathematical basis and validation of
the full cavitation model. Journal of Fluids Engineering, 124(3), 617624.
Sinibaldi, E., Beux, F., & Salvetti, M. V. (2006). A numerical method for 3D barotropic ows in
turbomachinery. Flow Turbulence Combustion, 76, 371381.
Smith, B. R. (1994). A near wall model for the k l two equation turbulence model. In AIAA 94
2386, 25th Fluid Dynamics ConferenceColorado Springs, Colorado.
Spalart, P. R., & Allmaras, S. R. (1994). A one-equation turbulence model for aerodynamic ows.
La Recherche Arospatiale, 1, 521.
Srinivasan, V., Salazar, A. J., & Saito, K. (2009). Numerical simulation of cavitation dynamics
using a cavitation-induced-momentum-defect (CIMD) correction approach. Applied Mathemat-
ical Modelling, 33, 15291559.
Stutz, B. (1996). Analyse de la structure diphasique et instationnaire de poches de cavitation. Ph.D.
thesis, Institut National Polytechnique de Grenoble.
Toutant, A., Chandesris, M., Jamet, D., & Lebaigue, O. (2009). Jump conditions for ltered quan-
tities at under-resolved discontinuous interface. Part 1: Theoretical development. International
Journal of Multiphase Flow, 35, 11001118.
Turkel, E. (1987). Preconditioned methods for solving the incompressible and low speed compress-
ible equations. Journal of Computational Physics, 172(2), 277298.
Utturkar, Y., Wu, J., Wang, G., & Shyy, W. (2005). Recent progress in modelling of cryogenic
cavitation for liquid rocket propulsion. Progress in Aerospace Sciences, 41, 558608.
Venkateswaran, S., Lindau, J. W., Kunz, R. F., & Merkle, C. L. (2002). Computation of multiphase
mixture ows with compressibility eects. Journal of Computational Physics, 180(1), 5477.
Numerical Simulation of Cavitating Flows . . . 233
Ventikos, Y., & Tzabiras, G. (2000). A numerical method for the simulation of steady and unsteady
cavitating ows. Computers & Fluids, 29(1), 6388.
Viegas, J. R., & Rubesin, M. W. (1983). Wallfunction boundary conditions in the solu tion of
the NavierStokes equations for complex compressible ows. In AIAA 831694, 16th Fluid and
Plasma Dynamics Conference, Danver, Massachussetts, 1214 July 1983.
Vincent, S., Larocque, J., Lacanette, D., Toutant, A., Lubin, P., & Sagaut, P. (2008). Computers &
Fluids, 37, 898906.
Vortmann, C., Schnerr, G. H., & Seelecke, S. (2003). Thermodynamic modeling and simulation of
cavitating nozzle ow. International Journal of Heat and Fluid Flow, 24, 774783.
De Vuyst, F., Ghidaglia, J. M., & Le Coq, G. (2005). On the numerical simulation of multiphase
water ows with changes of phase and strong gradients using the Homogeneous Equilibrium
Model. International Journal on Finite Volumes, 2(1), 136.
Wallis, G. (1967). One-dimensional two-phase ow. New York: McGraw-Hill.
Weller, H., Tabor, G., Jasak, H., & Fureby, C. (1998). A tensorial approach to computational con-
tinuum mechanics using objected-oriented techniques. Computers in Physics, 12, 620631.
Wilcox, D. C. (1988). Reassement of the scale-determining equation for advanced turbulence mod-
els. AIAA Journal, 26(11), 12991310.
Wilcox, D. C. (1998). Turbulence modeling for CFD. DCW Industries Inc.
Wu, J., Wang, G., & Shyy, W. (2005). Time-dependent turbulent cavitating ow computations with
interfacial transport and lter-based models. International Journal for Numerical Methods in
Fluids, 49(7), 739761.
Zein, A., Hantke, M., & Warnecke, G. (2010). Modeling phase transition for compressible two-
phase ows applied to metastable liquids. Journal of Computational Physics, 229(8), 2964
2998.
Zhang, X. B., Qiu, L. M., Gao, Y., & Zhang, X. J. (2008). Computational uid dynamic study on
cavitation in liquid nitrogen. Cryogenics, 48, 432438.
Zhou, L., & Wang, Z. (2008). Numerical simulation of cavitation around a hydrofoil and evaluation
of a RNG k model. Journal of Fluid Engineering, 130(1), 011302.
Numerical Simulation of Cavitating Flows
in Complex Geometries
1 Introduction
Cavitating ows occur in many engineering devices, as, for instance, rocket turbop-
umps, turbomachinery, hydrofoils, marine propellers, or nozzles. It is well known
that cavitation has strong eects, usually negative, on performance and life of such
devices. These kinds of engineering problems have some common features, which
impact on physical modeling and numerical methodology adopted in numerical sim-
ulations of these ows. First, the uid ows over solid walls having complex geome-
try and in many cases moving. Moreover, the characteristics of the inlet ow stream
are only partially known; for instance, the size and distribution of active cavitation
nuclei are usually not known. In general, for engineering applications, the aim is to
obtain predictions of the global performance of the device and of the eects of cavita-
tion rather than a detailed description of the ow or of cavitation phenomena. On the
other hand, the various eects of cavitation may have dierent importance depend-
ing on the engineering application; for instance, the analysis of liquid-propellant fed
turbopumps for space application put especial prize on the suction and dynamic per-
formance of the machine rather than on its resistance to erosion and other long-term
eects of cavitation, which are typically a major concern in other applications.
From a physical viewpoint, cavitating ows are characterized by complex phe-
nomena interacting each other, such as change of phase, evolving vapor to liquid
M.V. Salvetti ()
DICI, University of Pisa, Pisa, Italy
e-mail: [email protected]
interfaces and turbulence. Therefore, the modeling of cavitating ows is a very com-
plex task. Several models exist in the literature of dierent levels of complexity. We
focus here on the so-called one-uid or homogeneous-ow models, in which the cavi-
tating ow is described in terms of a single uid or mixture, whose properties depend
on the void fraction. We refer to the contributions by Goncalves and by Saurel et al.
in this volume for more complex two-phase cavitation models. Homogeneous-ow
cavitation models are based on the following assumptions: (i) local kinematic equi-
librium, i.e., same local velocity for both liquid and vapor phases; (ii) local thermo-
dynamic equilibrium, i.e., local temperature and pressure equality between phases;
and (iii) instantaneous vaporization or condensation processes. Clearly, this kind of
modeling is not suitable if strong non-equilibrium eects between the two phases are
present. On the other hand, these models are attracting because of their simplicity
and because they a-priori have the capability of describing the global eects of cav-
itation, which are the most interesting for practical applications. In the framework
of the previous simplifying assumptions, the thermodynamic properties for the pure
liquid, the pure vapor, and the mixture must be specied through an equation of state
(EOS). Dierent ways have been followed in the literature, e.g., tabulated EOS and
thermodynamic properties (e.g., Clerc 2000 or Ventikos and Tzabiras 2000), entropy
maximization procedure (e.g., Barberon and Helluy 2005), partial mass and partial
density properties (e.g., Meng and Yang 2003), and saturation equations and ther-
modynamics tables (e.g., Saurel et al. 1999). Homogeneous-ow cavitation models
using the assumption of a barotropic behavior for pure liquid and vapor and for the
liquidvapor mixture (see e.g., Delannoy and Kueny 1990; dAgostino et al. 2011;
Qin et al. 2003; Liu et al. 2004) are considered herein.
In the homogenous-ow description, the physical properties of the ow change
dramatically between the zones of pure liquid and the cavitating regions. In partic-
ular, in non-cavitating zones the ow is almost incompressible, while the cavitating
mixture is described as a highly compressible uid, characterized by speed of sound
values of several orders of magnitude lower than those of the pure liquid. More-
over, an abrupt transition from the wetted (incompressible) to the cavitating (highly
supersonic) regimes occurs. These features yield strong diculties for numerical
simulation, in spite of the simplifying assumptions made and the apparent simplic-
ity of barotropic homogeneous-ow cavitation models. It is clear, on the basis of the
previous considerations, that the numerical schemes must be designed in order to
cope with nearly incompressible and highly compressible regions coexisting in the
ow. Two opposite approaches can be found in the literature: adaptation to compress-
ible ows of methods developed for incompressible ows, usually through pressure
correction (e.g., Srinivasan et al. 2009; Senocak and Shyy 2002; Coutier-Delgosha
et al. 2003) or adaptation to the incompressible limit, usually through ad-hoc pre-
conditioning, of compressible ow solvers (e.g., Sinibaldi et al. 2006; Bilanceri et al.
2010; Goncalves and Patella 2009; Coutier-Delgosha et al. 2005; Kunz et al. 2000).
Each one of these approaches has advantages and drawbacks. Incompressible solvers
may lead to erroneous acoustic speeds in the cavitating regions, while compressible
solvers have well-known accuracy and eciency/robustness problems in the low
Mach number regions (Sinibaldi et al. 2006; Guillard and Viozat 1999).
Numerical Simulation of Cavitating Flows in Complex Geometries 237
+ () + () j (, )
t xj j xj j xj
(, ) = () + (, , ) . (1)
xj t j
238 M.V. Salvetti
In Eq. (1) the Einstein notation is used and the vector of unknowns is dened as
follows:
= (, u, v, w, k, )T , (2)
where is the density; u, v, and w are the velocity components in the x, y, and z direc-
tions; k is the turbulent kinetic energy; and is the turbulent dissipation. The other
terms appearing in Eq. (1) are the convective uxes Fj , the turbulence contribution
to convective uxes F j , the viscous laminar uxes Vj , the viscous turbulent uxes V
j ,
and the source term related to the k- model. Their expressions are the following:
( )
1 = 0, 2 k, 0, 0, 0, 0 T
( 3 )
2 = 0, 0, 2 k, 0, 0, 0 T (4)
( 3 )T
3 = 0, 0, 0, 3 k, 0, 0
2
( )T
k
1 = 0, 11 , 12 , 13 , ,
x1 x1
( )T
k
2 = 0, 21 , 22 , 23 , , (5)
x2 x2
( )T
k
3 = 0, 31 , 32 , 33 , ,
x3 x3
( )T
1 k 1
1 = 0, 11 , 12 , 13 , ,
k x1 x1
( )T
1 k 1
2 = 0, 21 , 22 , 23 , , (6)
k x2 x2
( )T
1 k 1
3 = 0, 31 , 32 , 33 , , ,
k x3 x3
in which p is the pressure, ij are the components of the viscous stress tensor, while
and k are empirical parameters.
The formulation of the source term () is:
Numerical Simulation of Cavitating Flows in Complex Geometries 239
( ( ))
2 ui uj 2 uk ui
= kij t + ij
3 xj xi 3 xk xj
k = +
2 (7)
= c1 k c2
( k )
() = 0 0 0 0 k T ,
being the production term of the turbulent kinetic energy and c1 and c2 are empir-
ical parameters.
The molecular and turbulent viscosities are, respectively, , dened in the follow-
ing (Eq. (12)) and t dened as follows:
k2
t = C , (8)
C being an additional empirical parameter. For all the empirical parameters con-
tained in the k- model, the standard values proposed in Launder and Spalding (1974)
are used herein.
Finally, is the source term due to reference frame rotation:
= (2 + ( ))
( )T
T (9)
(, , ) = 0, , 0, 0 ,
where g , , V , and pc are the liquid parameters, aLsat is the liquid sound speed at
saturation, = 1 Lsat , and L = L (, ) is a given function (see dAgostino
et al. 2011 for its physical interpretation and for more details). The resulting uni-
ed barotropic state law for the liquid and for the cavitating mixture only depends
on the two parameters TL and . For instance, for water at TL = 293.16 K, the
other parameters involved in (10) and (11) are psat = 2806.82 Pa, Lsat = 997.29
kg/m3 , sL = 5 1010 Pa1 , g = 1.67, = 0.73, V = 1.28, pc = 2.21 107 Pa and
aLsat = 1415 m/s (Torre et al. 2010). Note that, despite the model simplications
leading to a unied barotropic state law, the transition between wetted and cavitat-
ing regimes is extremely abrupt. Indeed, the sound speed falls from values of order
103 m/s in the pure liquid down to values of order 0.1 or 1 m/s in the mixture. The
corresponding Mach number variation renders this state law very sti from a numer-
ical viewpoint.
As for the denition of the molecular viscosity, a simple model, which is linear
in the cavitating regime, is considered:
if Lsat
L
() = v if v (12)
v + (1 )L otherwise,
in which v and L are the molecular viscosity of the vapor and liquid, respectively,
which, consistently with the assumptions made in the adopted cavitation model, are
considered constant and computed at T = TL .
3 Numerical Methodology
where Ci is the boundary of Ci and nk are the components of the unit vector normal
to Ci . It is possible to reformulate (13) in the following semi-discrete form:
Numerical Simulation of Cavitating Flows in Complex Geometries 241
di
i + ij + i = i + i , (14)
dt jN(i)
1
i = d, (15)
i Ci
where i is the cell volume, and N(i) represents the set of neighbors of the ith cell.
The numerical discretization of the convective ux crossing the boundary Cij shared
by Ci and Cj (positive towards Cj ) is denoted ij , while i , i , and i are the numer-
ical discretizations for, respectively, the viscous uxes and the source terms.
Since the key ingredients of the proposed numerical discretization are the numerical
approximation of the convective uxes and time advancing, the 1D inviscid case is
used as a rst step for the denition of the numerical convective uxes:
()
+ =0 (16)
t x
( )T
where = (, u, )T and () = u, u2 + p, u , denoting a passive scalar.
The associated semi-discrete problem is the following:
di
xi + i,i+1 i1,i = 0, (17)
dt
xi being the width of the nite-volume cell centered around node i.
The numerical scheme herein considered belongs to the so-called approximate Rie-
mann solvers, and, more particularly, to the HLL schemes (Harten et al. 1983), which
assume one intermediate state between two acoustic waves in the approximate Rie-
mann problem. The simplest HLL scheme is the so-called Rusanov scheme (Rusanov
1961; Park and Kwon 2003):
(i ) + (j ) ij
i,j = (j i ), (18)
2 2
242 M.V. Salvetti
where the rst term of the right-hand side is a centered approximation of the uxes
and the second one the upwind part, introducing numerical viscosity, with ij = sR
being an upper bound for the fastest wave speed. A classical choice for ij is the
largest absolute value of the Roe matrix eigenvalues (Roe 1981). A suitable Roe
matrix for a generic barotropic EOS was dened in Sinibaldi et al. (2006); its eigen-
values are the following:
1 = u ij + a ij , 2 = u ij a ij and 3 = u ij ,
whereas a ij , which can be considered as a Roe average for the sound speed, is dened
as:
p(j ) p(i ) if
j i j i
a ij = (20)
( )
a , p( ) if i = j = .
Therefore, we obtain
ij = max(|p |) = |uij | + a ij .
p
The Rusanov scheme, as all the HLL schemes, is known to introduce excessive
numerical diusion in the presence of contact discontinuities. Indeed, due to the
assumption of a two-wave conguration with only one intermediate state, the two
intermediate states of the exact Riemann solution are averaged, smearing, in this
way, the contact discontinuity. In the barotropic case, since density, velocity and
pressure are continuous across the contact discontinuity, the presence of two dier-
ent intermediate states in the Riemann problem is only due to the passive scalar (see
Sinibaldi 2006). Moreover, only one eigenvalue is associated with the contact dis-
continuity. Thus, a possible way to counteract the excessive diusion of the Rusanov
scheme is to introduce an anti-diusive term in the discretization of the third equa-
tion of system (16). Based on this idea, the following LD-HLL (Low-Diusive HLL)
was proposed in Bilanceri et al. (2010):
(i ) + (j ) 0 0
1 ij
i,j = 0 ij 0 (j i ). (21)
2 2
0 0 |uij |
Qij
Numerical Simulation of Cavitating Flows in Complex Geometries 243
Compared with the original Rusanov scheme, the rst two equations, which are
related to the acoustic waves, are unchanged, while for the third equation, i.e., the
one directly related to the contact discontinuity, the diusive part of the scheme has
been reduced (see Bilanceri et al. 2010 for more details).
As mentioned in the Introduction, compressible ow solvers suer of accuracy
and eciency problems for nearly incompressible ows, as encountered in the non-
cavitating ow regions. Following the approach proposed by Guillard and Viozat
(1999), it was shown in Bilanceri (2011) that, as expected, also the LD-HLL scheme
(21) suers of accuracy problems in the low Mach number limit. Therefore, suitable
preconditioning is needed. It was also shown in Bilanceri (2011) that the correct
asymptotic behavior can be recovered by acting on the acoustic terms of the matrix
Qij . This can be done by multiplying Qij by the following preconditioning matrix:
1 0 0
Pij = 0 0 , (22)
0 0 1
in which {
106 if M 106 |uij |
= (M) = . M= . (23)
min(M, 1) otherwise a ij
(i ) + (j ) 1 0 0
1 ij
i,j = 0 ij 0 (j i ). (24)
2 2
0 0 |uij |
Qij
Note that since preconditioning acts only on the upwind part of the scheme, con-
sistency in time is preserved. Moreover, preconditioning does not aect the low-
diusion correction introduced for the passive scalar equation.
Let us consider an implicit backward Euler method applied to the semi-discrete prob-
lem (17):
xi n ( )
i + n i,i+1 n i1,i = ni,i+1 ni1,i , (25)
t
where n () = ()n+1 ()n . To avoid the solution of a nonlinear system at each time
step, a linearization of n ij is usually adopted. A way to obtain such a linearization
is to nd two matrices D1 and D2 such that
244 M.V. Salvetti
In this case (25) is reduced to the following block tridiagonal linear system:
( )
Bi,n
1
n
i1 + B
i,n n
0
i + B
i,n n
1
i+1 = n
i,i+1 i1,i ,
n
(27)
where
i,n
B1 = D1 (i1 , i )
n n
i,n xi
B0 = I + D1 (ni , ni+1 ) D2 (ni1 , ni ) (28)
t
i,n
B1 = D2 (ni , ni+1 ).
Qnij
n ij,u = (n j n i ) n,n+1 , (30)
2 ij
n Qij
in which n,n+1 (n+1 n+1 ).
ij 2 j i
Linearization (29) is obtained just by neglecting the term n,n+1
ij
. This term can be
neglected as long as that the solution is regular enough to satisfy
n+1
j
n+1
i
O(x) and n Qij O(t). (31)
Numerical Simulation of Cavitating Flows in Complex Geometries 245
Even if the assumption (31) is in general a reasonable one, there are situations of
practical interest in which it is not satised. Indeed, if a discontinuity is present the
magnitude of the term n+1 j
n+1
i
can be large independently of the size of x.
Moreover, the term Qij can also be large. This can happen again in the presence
n
of large variations of the ow velocity but also when the speed of sound has a sti
change in magnitude. The latter is a typical situation occurring in the presence of
cavitation. A more complete linearization was proposed by Bilanceri et al. (2010)
for cavitating barotropic ows, which takes into account, at least in an approximate
way, the term n,n+1
ij
. This linearization exploits the fact that Qij is a diagonal matrix
such that any of its diagonal coecients, qk , can be written as a composite function
of two variables, a and u . Then, through dierentiation, by neglecting terms of higher
order and after some mathematical developments (see Bilanceri et al. 2010 for the
detailed procedure), the following approximation of the previously neglected term
is obtained:
1 1
n,n+1 Kij n i Kji n j , (32)
ij 2 2
in which the generic element of Kij is dened by the following expression:
( )( )
( ) qk u q a
Kij km = + k n
Wj,k Wi,k
n
, (33)
u Wi,m a Wi,m
where Wi,m denotes the m-th element of the vector i . Finally, considering ijn,n+1
from (32) and (33) in the evaluation of the upwind part of the numerical ux time
variation, the following more complete approximation is obtained instead of (29):
( )
1
ij
n
A(i ) + Qij Kij n i
n n
2
( ) (34)
1
+ A(j ) Qij + Kji n j .
n n
2
These partial derivatives are numerically computed through centered nite dier-
ences (see Bilanceri 2011 for more details).
( ) (36)
+ hi+1 i+1 ni ni+2 i+1
i+ 1 = i+1 (1 ) + 1 + 3 ,
2 2 i+ 2 hi+1 i+ 2 hi+2
( )
3n+1
i
4ni + n1
i
xi + n i+ 1 n i 1 = n n , (37)
2t 2 2 i+ 12 i 12
0 = n
i,s s
1 i1 + 0 s i + 1 s i+1 = i s = 0, , m 1
i,s i,s s
n+1 = m ,
in which
i,s
1 = D1 ( i1 , i )
s s
i,s 3xi
0 = I + D1 ( si , si+1 ) D2 ( si1 , si )
2t
i,s
1 = D2 ( si , si+1 )
( )
s = xi (3 s 4n + n1 ) + s s ,
i 2t i i i i+ 12 i 12
D1 and D2 being the generic matrices of the approximation (26). Only a few iter-
ations are needed to obtain second-order accuracy in time (Bilanceri 2011; Martin
and Guillard 1996).
The preconditioned LD-HLL scheme (24), previously dened for the 1D case, can
be extended to 3D turbulent ows by dening ij as the integral over Cij of the
248 M.V. Salvetti
p 0
01 ( 32 p )0 2 ( 32 p )0 ( 32 p )0 0
0
p
n + 3 n n n n 0
( 32 p ) ij,1 ( 32 p ) 2ij,1 ij,2p ( 32 p ) ij,1 ij,3
0 n n n + 3 n n 0 0
Sij = ( 32 p ) ij,2 ij,1 ( 32 p ) ij,2 ( 32 p ) 2ij,2 ij,3p
0
(39)
0 nij,3 nij,1 nij,3 nij,2 nij,3 + 3 0
0 0 0 0
p
3 0
0 0 0 0 0 3
p
where
1 1
ij = i + i ij , +ij = j j ij (41)
2 2
where ij is the vector joining the i-th node with the j-th one and i is an approx-
imation of the gradient in the i-th cell, possibly taking into account slope limiters.
The denition of j ij in this work is based on the same approach considered in
Camarri et al. (2004), Farhat et al. (1991) when limiters have to be included. First, a
linear approximation |Th of the gradient in each tetrahedron Th is considered:
|Th = k Tk , (42)
h
ki(Th )
where i(Th ) is the set of nodes belonging to Th , Tk are the barycentric coordinates
h
associated with the kth-vertex of Th , while t(i) is the set of the tetrahedrons which
Numerical Simulation of Cavitating Flows in Complex Geometries 249
share the ith vertex. Once |Th is available, a centered approximation of the gra-
dient in the i-th cell is derived as follows:
Th |Th
ht(i)
ci = , (43)
Th
ht(i)
Th being the volume of the h-tetrahedron. Following Farhat et al. (1991), it is now
possible to obtain a second-order MUSCL reconstruction operator with slope lim-
iters by dening i ij and j ij by:
{ ( )
i ij = minmod{j i , 2ci ij (j i )}
(44)
j ij = minmod{j i , 2cj ij j i },
where (i,Th ) is the P1 nite-element test function associated to the vertex i restricted
to the tetrahedron Th and hT is given by
L
Th = .
4
Li(Th )
Th ( Th Th Th
)T
1 = 0, 11 , 12 , 13
Th ( T T T
)T
2 = 0, 21h , 22h , 23h (47)
( )T
Th = 0, Th , Th , Th ,
3 31 32 33
where
( )
T 2 (L,Th )
ij h = uL,k ij
3 xk
Li(Th )
)
(L,Th ) (L,Th )
+ uL,i + uL,j , (48)
xj xi
The discretization of the turbulence source term appearing in (7), and, in particular,
of the term , is carried out as previously described for the viscous terms.
The source term due to the frame rotation is discretized as follows:
( )
0 ( )
i = i = i , (49)
2 i i + i i
The time discretization for the 3D numerical method can be readily obtained from its
1D counterpart. Let us consider a second-order accurate in space and time approach
(the modications for the rst-order case are straightforward). Then the implicit for-
mulation of (14) is
3
i n i + n ij + n i n (i + i ) =
2 t
n
jN(i)
1 ( ) n
i n n1 + ni ij ni + ni + ni . (50)
2 t i
jN(i)
( )
2 (L,Th )
n ij h =
T
n uL,k ij
3 xk
( h)
Li T
( ))
(L,Th ) (L,Th )
+ n uL,i + n uL,j . (51)
xj xi
By approximating n uL,k as
unL,k 1 n ( n 2)
n uL,k = n L + n uL,k + O ( t) , (52)
nL L
a second-order time linearization for the viscous uxes is achieved. With a similar
approach a second-order time linearization can be obtained also for the source terms.
Once n ij , n i , and n (i + i ) are given, the second-order defect correction
approach described in Sect. 3.1 can be applied also for the 3D case.
In order to reduce the memory and computational requirements, we adopted a
weakly coupling between the mass and momentum balance equations and the equa-
tions for k and , i.e., in the implicit time advancing of the mass and momentum
equations, we consider k and frozen at time step n, while when advancing the
equations for k and , , u, v, and w are frozen. This permits to successively solve
two linear systems of smaller dimensions: one composed by 4 4 blocks, which
updates the ow variables, , u, v, w, and another one for the turbulent variables,
k, , composed by 2 2 blocks. Note that, even if this introduces an additional
approximation to the discretization technique described in the previous sections, a
linear system composed by modied blocks as previously described can be consid-
ered as a particular instance of the DeC approach (see also Bilanceri 2011). As a
consequence, even if for this case we do not have theoretical results of a nite termi-
nation property, the second-order approximation should be recovered asymptotically
when a full convergence of the DeC iteration is obtained.
In this section the modeling and numerical tools previously described are applied
to the simulations of the ow in a real three-blade axial inducer designed for liquid-
propelled rockets (Torre et al. 2010). It is a three-blade inducer with a tip blade radius
252 M.V. Salvetti
of 81-mm and 2-mm radial clearance between the blade tip and the external case (see
Fig. 1).
Experimental data are available for all the numerical simulations described in
the following. The pressure jump between two dierent stations has been measured
for a wide range of working conditions: from small to large mass ow rates, non-
cavitating and cavitating conditions, and dierent values of the rotational speed z .
The results are presented in terms of the mean adimensionalized pressure jump as
a function of the adimensionalized discharge :
P Q
= = , (53)
L 2z R2T R2T z RT
where Q is the discharge, RT is the radius of the tip of blade, L is the density of
the liquid, and z is the angular velocity. Note that the numerical pressure jump is
averaged over one complete revolution of the inducer.
Numerical Simulation of Cavitating Flows in Complex Geometries 253
Note that, except when dierently stated, the simulations do not include turbulence
eects.
As shown in Table 1, all the simulations in non-cavitating conditions use the same
rotational velocity of 1500 rpm. In the - plane the experimental curves of the
performances of the inducer are roughly independent from the rotational velocity z
(Torre et al. 2010). As a consequence, validating the numerical tool for a specic
rotational velocity and dierent ow rates should validate the proposed numerical
tool for a generic rotational velocity.
Table 2 shows the results for the non-cavitating simulations. It clearly appears
that the lower is the discharge , the worse are the results. Already with the coarsest
grid G1, rather satisfactory results are obtained for intermediate and high discharge
values, Ind2 and Ind1, respectively. The quantitative agreement is further improved
considering the more rened grid G2 for the case Ind2. Conversely, for the low dis-
charge case, Ind3, the simulations with the grid G1 and G2 greatly overestimate the
pressure jump by, respectively, 41% and 30%. The magnitude of this error could be
ascribed to the backward ow between the inducer blades and the external case. The
correct resolution of this ow is of crucial importance for the determination of the
performance of an inducer. Since the smaller is the mass ow rate the greater is the
254 M.V. Salvetti
backow (see Fig. 2), we investigated two possible explanations of this behavior.
The rst one is the distance of the inlet from the inducer nose, which could be not
large enough to avoid spurious eects on the solution; the second one is turbulence
eect, which for this case could be important. The results of the rst simulations
for the longer computational domain, G1L-Ind3, show that even if there is a small
eect, i.e., a decrease of the error from 41% to 39%, this is not the source of the error.
Instead the results of the simulation G1L-Ind3-T, i.e., the one done considering the
RANS model, show that in this case turbulence is a key issue. Indeed, in this case
the error falls down to 12%, less than the error obtained with the rened grid G2 in
laminar conditions. As expected the eects of turbulence are particularly important
near the gap between the blades and the external case, as it is shown in Fig. 3 by
considering the isocontours of k.
This strongly aects the backow and, thus, the pressure jump. This also explains
why for larger ow rates, for which the backow is less important, the eects of
turbulence are not so strong and a good agreement with experimental data can be
obtained also in laminar simulations.
The mass ow rate for the cavitating cases is large enough to prevent the issues
related to the backow previously described, and thus only laminar simulations are
considered. The results for the cavitating conditions, reported in Table 3, show that
the rst grid G1 is not enough rened to correctly describe cavitation for this case.
The pressure jump is greatly overestimated: for these conditions the error is related
to the underestimation of the cavitating region: the experimental data for = 0.056
show a large cavitating zone and consequently the performance of the inducer is sig-
nicantly deteriorated. Instead, in the simulation with grid G1, the extension of the
cavitating region is greatly underestimated and, as a consequence, the numerical
Numerical Simulation of Cavitating Flows in Complex Geometries 255
Fig. 2 Comparison of the averaged axial velocity for the three non-cavitating benchmarks, grid
G1
Fig. 3 Cross section of the averaged k eld, simulation G1L-Ind3 (view of the shorter domain)
5 Concluding Remarks
A numerical methodology has been presented for the simulation of cavitating ows
in realistic engineering problems through a barotropic homogeneous-ow cavita-
tion model. The key ingredients of this methodology are (i) the use of unstructured
Numerical Simulation of Cavitating Flows in Complex Geometries 257
grids to deal with complex geometries and local renement, (ii) a low-diusive HLL
scheme for convective uxes, and (iii) implicit time advancing together with a lin-
earization in time accounting for speed of sound variations and a defect correction
approach to reduce the computational costs.
As an example of a real engineering problem, the application to the ow in a real
turbopump inducer designed for liquid-propelled rockets in both cavitating and non-
cavitating conditions. The obtained results showed a global good agreement with the
experimental data and the proposed methodology appeared to be robust and ecient,
leading to simulations requiring reasonable computational costs also in cavitating
conditions.
Acknowledgements Marco Bilanceri and Franois Beux are gratefully acknowledged for their
precious contribution to the work presented in this contribution.
References
Barberon, T., & Helluy, P. (2005). Finite volume simulation of cavitating ows. Computers and
Fluids, 34, 832858.
Bilanceri, M. (2011). Numerical simulations of barotropic ows in complex geometries. Ph.D.
thesis, Aerospace Engineering, University of Pisa. https://etd.adm.unipi.it/theses/available/etd-
03232011-094909/.
Bilanceri, M., Beux, F., & Salvetti, M. V. (2010). An implicit low-diusive hll scheme with com-
plete time linearization: Application to cavitating barotropic ows. Computers and Fluids, 39,
19902006.
Camarri, S., Salvetti, M. V., Koobus, B., & Dervieux, A. (2004). A low diusion MUSCL scheme
for LES on unstructured grids. Computer and Fluids, 33, 11011129.
Clerc, S. (2000). Numerical simulation of the homogeneous equilibrium model for two-phase ows.
Journal of Computational Physics, 161, 354375.
Coutier-Delgosha, O., Fortes-Patella, R., & Delannoy, Y. (2003). Numerical simulation of the
unsteady behavior of cavitating ow. International Journal of Numerica Methids in Fluids, 42,
527548.
Coutier-Delgosha, O., Fortes-Patella, R., Reboud, J. L., Hakimi, N., & Hirsch, C. (2005). Numer-
ical simulation of cavitating ow in 2D and 3D inducer geometries. International Journal for
Numerical Methods in Fluids, 48, 135167.
dAgostino, L., Rapposelli, E., Pascarella, S., & Ciucci. A. (2011). A modied bubbly isen-
thalpic model for numerical simulation of cavitating ows. In 37th AIAA/ASME/SAE/ASEE Joint
Propulsion Conference, Salt Lake City, UT, USA.
Delannoy, Y., & Kueny, J. L. (1990). Cavity ow predictions based on the Euler equations (pp.
153158). ASME Cavitation and Multiphase Flow Forum.
Delis, A. I., Skeels, C. P., & Ryrie, S. C. (2000). Implicit high-resolution methods for modelling
one-dimensional open channel ow. Journal of Hydraulic Research, 38(5), 369382.
Farhat, S., Fezoui, L., & Lanteri, S. (1991). Computational uid dynamics with irregular grids on
the connection machine. INRIA Rapport de Recherche 1411, INRIA. http://hal.inria.fr/inria-
00075149/PDF/RR-1411.pdf.
Goncalves, E., & Fortes, R. (2009). Numerical simulation of cavitating ows with homogeneous
models. Computers and Fluids, 38, 16821696.
Guillard, H., & Viozat, C. (1999). On the behaviour of upwind schemes in the low Mach number
limit. Computers and Fluids, 28, 6386.
258 M.V. Salvetti
Harten, A., Lax, P. D., & van Leer, B. (1983). On upstreaming dierencing and Godunov-type
schemes for hyperbolic conservation laws. SIAM Review, 25(1), 3561.
Kunz, R. F., Boger, D. A., Stinebring, D. R., Chyczewski, T. S., Lindau, J. W., Gibeling, H. J., et al.
(2000). A preconditioned Navier-Stokes method for two-phase ows application to cavitation
prediction. Computers and Fluids, 29, 849875.
Launder, B., & Spalding, D. (1974). The numerical computation of turbulent ows. Computer Meth-
ods in Applied Mechanics and Engineering, 3(2), 269289.
Leveque, R. J. (1994). Numerical methods for conservation laws. Basel: Birkhuser-Verlag.
Liu, T. G., Khoo, B. C., & Xie, W. F. (2004). Isentropic one-uid modelling of unsteady cavitating
ow. Journal of Computational Physics, 201, 80108.
Luo, H., Baum, J. D., & Lhner, R. (2001). An accurate, fast, matrix-free implicit method for com-
puting unsteady ows on unstructured grids. Computers and Fluids, 30, 137159.
Martin, R., & Guillard, H. (1996). A second order defect correction scheme for unsteady problems.
Computers and Fluids, 25(1), 927.
Meng, H., & Yang, V. (2003). A unied treatment of general uid thermodynamics and its appli-
cation to a preconditioning scheme. Journal of Computational Physics, 189, 277304.
Park, S. H., & Kwon, J. H. (2003). On the dissipation mechanism of Godunov-type schemes. Jour-
nal of Computational Physics, 188(2), 524542. doi:10.1016/S0021-9991(03)00191-8.
Qin, Q., Song, C. C. S., & Arndt, R. E. A. (2003). A virtual single-phase natural cavitation model
and its application to CAV2003 hydrofoil. In Proceedings of CAV2003Fifth International Sym-
posium on Cavitation, Osaka (Japan), November 2003.
Roe, P. L. (1981). Approximate Riemann solvers, parameter vectors, and dierence schemes. Jour-
nal of Computational Physics, 43, 357372. doi:10.1016/0021-9991(81)90128-5.
Rusanov, V. V. (1961). The calculation of the interaction of non-stationary shock waves with bar-
riers. Journal of Computational Mathematics and Physics USSR, 1, 267279.
Saurel, R., Cocchi, J. P., & Butler, P. B. (1999). Numerical study of cavitation in the wake of a
hypervelocity underwater projectile. Journal of Propulsion and Power, 15(4), 513522.
Senocak, I., & Shyy, W. (2002). A pressure-based method for turbulent cavitating ow computa-
tions. Journal of Computational Physics, 176, 363383.
Sinibaldi, E. (2006). Implicit preconditioned numerical schemes for the simulation of three-
dimensional barotropic ows. Ph.D. thesis, Scuola Normale Superiore di Pisa.
Sinibaldi, E., Beux, F., & Salvetti, M. V. (2006). A numerical method for 3D barotropic ows in
turbomachinery. Flow, Turbulence and Combustion, 76(4), 371381.
Srinivasan, V., Salazar, A. J., & Saito, K. (2009). Numerical simulation of cavitation dynamics
using a cavitation-induced-momentum-defect (cimd) correction approach. Applied Mathemati-
cal Modelling, 33, 15291559.
Toro, E. F. (1997). Riemann solvers and numerical methods for uid dynamics. Springer.
Torre, L., Pace, G., Miloro, P., Pasini, A., Cervone, A., & dAgostino, L. (2010). Flow instabilities
on a three bladed axial inducer at variable tip clearance. In: 13th International Symposium on
Transport Phenomena and Dynamics of Rotating Machinery, Honolulu, Hawaii, USA.
van Leer, B. (1979). Towards the ultimate conservative dierence scheme V: A second-order sequel
to Godunovs method. Journal of Computational Physics, 32(1), 101136.
Ventikos, Y., & Tzabiras, G. (2000). A numerical method for the simulation of steady and unsteady
cavitating ows. Computers and Fluids, 29, 6388.
Yee, H. C. (1987). Construction of explicit and implicit symmetric TVD schemes and their appli-
cations. Journal of Computational Physics, 68(1), 151179.
From Cavitating to Boiling Flows
1 Introduction
Cavitating, boiling and evaporating are three phenomena that involve phase transi-
tion in multi-phase ows. They appear in countless engineering applications: steam
generators, marine propellers, liquid fuel combustion, etc. Yet, modelling these
phenomena remains an unsettled problem.
R. Saurel ()
University Institute of France, Paris, France
e-mail: [email protected]
O. Le Mtayer
Aix-Marseille University, CNRS IUSTI, Marseille, France
e-mail: [email protected]
R. Saurel P. Boivin
Aix-Marseille University, CNRS M2P2, Marseille, France
e-mail: [email protected]
R. Saurel
RS2N, Marseille, France
In the present work a unied approach is provided and its ability to model and
compute cavitating and boiling ows is shown with computational examples. Most
of the scientic material results from former investigations by Saurel (2008), Le
Martelot et al. (2013, 2014). The ow model is a hyperbolic system of partial dif-
ferential equations with Gibbs free energy relaxation. The thermodynamic closure is
built without ambiguity and results in a mixture EOS valid in pure liquid, pure vapour
and two-phase mixture. The presence of non-condensable phase may be considered
as well. The sound speed is dened in all space of variables. Nonlinear waves, such
as shocks have a clear denition too. Last, the formulation is valid for any ow speed
and any uid density ratio at interfaces.
mechanical and thermal relaxation are assumed to be sti, it results in the following
system of partial dierential equations:
()1 ( )
+ div ()1 u = 0, or alternatively + div(u) =0
t t
()2 ( ) Y1
+ div ()2 u =0 + div(Y1 u) =0
t t
u ( )
+ div u u + pI =0
t
E
+ div ((E + p)u) =0 (1)
t
T = T = T,
1 2
e = Y1 e1 (T, p) + Y2 e2 (T, p),
(2)
p1 = p2 = p,
1 + 2 = 1 or alternatively, Y1 v1 (T, p) + Y2 v2 (T, p) = v
In this algebraic system, the phases are in pressure equilibrium and each one occu-
pies its own sub-volume or volume fraction k . System (2) corresponds to a nonlinear
system of two equations with the two unknowns T and p. To determine its explicit
solution, EOS for the phases have to be provided. This is addressed in the forthcom-
ing section. This thermodynamic closure diers signicantly from that of ideal gas
mixtures. In ideal mixtures each uid occupies the entire volume, this assumption
replacing the last equality of System (2). Also, the pressure is dened by the Daltons
law p = k pk instead of the pressure equilibrium condition. In formulation (1) with
thermodynamic closure (2) each phase occupies its own volume (and not the entire
one) and evolves in temperature and pressure equilibrium with the other phase.
At interfaces separating pure liquid and pure gas the assumption of single tem-
perature seems unrealistic as interface conditions, in the absence of mass transfer
and heat diusion, reduce to normal equal velocities and equal pressures, implying
arbitrary temperature jumps. However, when heat diusion is present an additional
interface condition appears, corresponding to temperatures equality. Thus System
(2) is valid for the computation of interfacial ows when:
From Cavitating to Boiling Flows 263
Heat diusion is present and when it is possible to resolve the heat diusion layer,
as for example in the boiling ow congurations that will be considered latter.
This is similar to laminar ames computations.
A mushy zone is present at the interface. This is the case for example with ash-
ing and super-cavitating ows where the interface is not a clear discontinuity but a
sharp mixture layer in which thermodynamic relaxation and heat exchanges occur
intensively. This is similar to turbulent ames computations, where it is not possi-
ble to resolve all heat diusion and chemical relaxation layers at subscale, but for
which turbulent heat conduction imply global turbulent ame propagation.
System (1) with closure (2) is thus valid for the computation of:
local interface dynamics when heat transfer is considered (this eect will be
inserted later),
global (or macro-scale) interface dynamics when a micro-scale or subscale struc-
ture such as a mushy zone is present. Indeed, at subscale, heat transfer is neces-
sarily present, imposing temperature equality at the interface.
When phase transition is addressed Gibbs free energy relaxation terms have to be
considered in one of the mass equations:
Y1 uY1
+ = (g2 g1 ) (3)
t x
where gk = hk Tsk denotes the phase k Gibbs free energy. hk and sk represent the
specic enthalpy and specic entropy. represents a relaxation parameter that con-
trols the rate at which thermodynamic equilibrium is reached. The way this relax-
ation parameter is estimated will be addressed later. The entropy equation associated
to System (1)(3) reads,
s su (g2 g1 )2
+ = , (4)
t x T
with the following denition for the mixture entropy: s = Y1 s1 + Y2 s2 .
The same type of remark as before with temperature equality at interfaces is
needed to explain the validity of a single velocity model to compute interfacial ows
with phase transition. When heat diusion and Gibbs energy relaxation are addressed
and solved at the interface, i.e., when the interface structure is solved, three velocities
appear at the global scale even if a single local velocity is present in the ow model.
The three velocities that appear at the global scale are the liquid one, the vapour one
and the phase transition front one.
This is similar to ames computation in which the reactive NavierStokes equa-
tions are appropriate to compute the reacting and burnt gas dynamics as well as ame
front dynamics.
The model has been shown to converge to exact sharp interface solutions in
Le Martelot et al. (2014). When the interface has more complex structure with tur-
bulent mixing at subscale, the same three velocities appear with dierent dynamics.
264 R. Saurel et al.
RT
p= c2 (5)
1 b
r
where R denote the specic gas constant (R = ) with r = 8.314 J/mol/K and
W
W (kg/mol) the molar mass. b represents the specic covolume, i.e. the volume occu-
pied by the molecules (m3 /kg) and c is a constant associated to attractive eects.
This EOS contains a fundamental drawback schematized in Fig. 1 where an isen-
trope for the VdW EOS is shown.
EOS (5) thus presents unphysical behaviour during the phase change process. The
following interpretations follow:
Equation (5) is aimed to close the balance equations of mass, momentum and
energy of the mixture. Unlike System (1), the volume or mass concentrations are
not considered in such formulations. A possible reason for choosing models with
From Cavitating to Boiling Flows 265
Fig. 1 Thermodynamic path along an isentrope showing an expansion process starting from a pure
( p )
liquid to the pure vapour region. The square sound speed c2 = v2 is well dened in the pure
v s
liquid and pure gas but is undened in the two-phase region, between the metastable liquid and the
metastable gas
In the following we adopt an approach where each phase has its own EOS, each
EOS being thermodynamically consistent (convex) with well dened sound speed.
The connexion between the two phases is done through a kinetic path instead of a
thermodynamic one. Determination of the kinetic relaxation rate will be addressed
later. Schematic representation of the thermo-kinetic approach of phase transition is
shown in Fig. 2. This idea was promoted by Saurel et al. (2008).
In this frame, liquid and gas require their own EOS, these ones being linked by
the phase diagram. The building of such EOS has been addressed in Le Mtayer et al.
(2004) on the basis of the stiened gas (SG) EOS. The main formulas for this EOS
read for a given phase k = 1, 2:
266 R. Saurel et al.
pk (k , ek ) = k (k 1)(ek qk ) k pk ,
pk + pk
Tk (pk , k ) = , (6)
k Cvk (k 1)
Tk k
gk (pk , Tk ) = (k Cvk qk )Tk Cvk Tk ln + qk .
(pk + pk )k 1
For a given phase, the following parameters are needed: k , pk , Cvk , qk , and
qk . As shown by Le Mtayer et al. (2004) there is no diculty to determine these
parameters once the saturation curves for the liquid and gas are known. For liquid
water and steam, the uid parameters, optimized in the [300500 K] temperature
range, are the following:
Liquid water:
Water vapour:
These parameters are used in the computational examples of boiling and cavitating
ows.
Expressing the thermal EOS of System (6) dierently the following relation
appears:
pk (k , Tk ) = k (k 1)Cvk Tk pk . (7)
From Cavitating to Boiling Flows 267
The agitation part is k (k 1)Cvk Tk and is similar to the RT part of (5). The attrac-
tive part (pk ) is here a constant while it is density varying with the VdW EOS.
The repulsive part (1 b)1 is neglected in the SG formulation. Its consideration
is addressed in Le Mtayer and Saurel (2016).
The sound speed of a given phase reads,
pk + pk
ck = k . (8)
k
With this formulation the phases sound speed are always dened as pk > pk . Hav-
ing in hands EOS (6) for each phase we now address building of the EOS for the
mixture.
Mixture EOS
The mixture EOS is the one that closes System (1) with the help of mixture rules (2)
on the basis of formulation (6) for each phase. The algebraic system to solve is thus,
where e and v are obtained from the resolution of (1) and combination of the mass,
momentum and energy equations. The unknowns in (9) are consequently the mixture
temperature T and pressure P. Combining the caloric and thermal EOSs of (6) the
energy for a given phase reads,
pk
ek (Tk , k ) = Cvk Tk + + qk .
k
Inserting this expression in the rst equation of (6) the specic volume as a function
of pressure and temperature is obtained:
(k 1)Cvk Tk
vk (Tk , pk ) = . (10)
pk + pk
Inserting this expression in one of the EOS (6) the internal energy as a function of
pressure and temperature is obtained:
(pk + k pk )Cvk Tk
ek (Tk , pk ) = + qk .
pk + pk
( )1
Yi Cvi (p + i pi )
T(e, p, Y1 ) = (e q ) , (11)
p + pi
i
( )
(i 1)Yi Cvi 1
T(v, p, Y1 ) = v , (12)
i
p + pi
with q = Y1 q1 + Y2 q2 .
Eliminating the temperature from these two equations the EOS for the pressure
is obtained as solution of the following quadratic equation:
a 2 p2 + a 1 p + a 0 = 0
a2 = Y1 Cv1 + Y2 Cv2
a1 = Y1 Cv1 (p2 + 1 p1 (1 1)Q) + Y2 Cv2 (p1 + 2 p2 (2 1)Q)
a0 = Q((1 1)Y1 Cv1 p2 + (2 1)Y2 Cv2 p1 ) + p1 p2 (1 Y1 Cv1 + 2 Y2 Cv2 )
where
Q = (e q )
Fig. 3 Representation of the mechanical equilibrium mixture sound speed (in lines) and the
pressure-temperature equilibrium mixture sound speed (in dashed lines). Both sound speeds present
a non-monotonic behaviour versus volume fraction. The mechanical equilibrium sound speed is
always slightly higher than the pressure-temperature equilibrium one
From Cavitating to Boiling Flows 269
From the results shown in Fig. 3 it appears that there is no practical need to com-
pute the mixture sound speed that is quite complex and computationally expensive
(its explicit expression is given in Le Martelot et al. (2014)). As it is always lower
than the mechanical equilibrium one, this one is preferred in the numerical computa-
tions, as better stability is guaranteed with this estimate. The mechanical equilibrium
sound speed obeys the well known Wood (1930) formula:
1
= 12 + 22 (14)
cw
2
1 c1 2 c2
The volume fractions are determined from the resolution of System (1) and the spe-
cic volumes computed by (10) with the pressure given by (13) and temperature
given by (12). The phase sound speeds are given by (8).
When dealing with phase transition the relaxation rate present in the concentration
equation (3) has to be specied. Following Saurel et al. (2008), the guess is
appropriate in most situtations. In other words, pure liquid and pure gas are allowed
to have any temperature during the hyperbolic step and instantly reach thermody-
namic equilibrium at the saturation conditions where both liquid and gas are present
(interface and mixture zones). This assertion is justied as follows:
When dealing with the direct numerical simulation of boiling ows, the uids
become metastable as a consequence of heat diusion. Heat conduction is slow
and controls the global rate of phase change. Thus, one can assume that phase
transition occurs at any rate, given it is greater than heat conduction. For the sake
of simplicity, innitely fast relaxation is adopted.
When dealing with cavitating ows, it is assumed that the uids contain enough
impurities to have many nucleation sites. Around evaporating interfaces, phase
transition happens and their collective eects result in macroscopic cavitation
fronts surrounded by mushy zones. In these mushy zones the interfacial area is
so large that heat and mass exchange are intense enough so that a ow model with
a unique temperature and sti Gibbs energy relaxation is appropriate.
When none of the uids is metastable (pure liquid state and pure gas state) no
thermo-chemical relaxation occurs and the ow model reduces to single phase
equations, with appropriate thermodynamics.
With this approach the thermodynamic path that the uid follows during an isen-
tropic expansion is shown in Fig. 4.
As the thermo-chemical solver is used as soon as metastable states appear, the
eective thermodynamic path that the uid follows is that of Fig. 4. Such a path could
be reproduced by a reduced version of the ow model (1) with three equations only
270 R. Saurel et al.
yy
liq
gas
xx
Fig. 4 As metastable states are transformed stiy to equilibrium mixture states the eective ther-
modynamic path corresponds in the two-phase region to a line whose slope has been exaggerated
in the gure. The slope is non zero but is weak, as sound propagates slowly in two-phase mixtures
(mixture mass, mixture momentum and mixture energy). However, the sound speed
of such a model is not convenient, as it is costly to compute, and most importantly,
discontinuous at the phase diagram boundaries.
Here, the ow model has a sound speed always dened and continuous. As men-
tioned earlier, Eq. (14) is a fair approximation. Also, extending System (1) to the
presence of non-condensable gas is straightforward while it is non-trivial with a more
reduced model.
The presence of a sti relaxation term in (3) does not result in computational
diculties as integration of such term is never addressed: in the following, only the
equilibrium state is required.
The relaxed solution corresponding to thermodynamic equilibrium is obtained
considering the mixture mass and mixture energy denition:
1
v= = Y1 v1 + Y2 v2 = cst = v0 (15)
e = Y1 e1 + Y2 e2 = cst = e0 (16)
1 1
where Y1 = and Y2 = 2 2 = 1 Y1 are the varying mass fractions of both
phases. In the following the liquid and its vapour are denoted respectively by the
subscripts 1 and 2.
The specic volumes and internal energies are given by the SG EOS:
(k 1)Cvk Tk
vk = , (17)
pk + p,k
( )
(k 1)p,k
ek = Cvk Tk 1 + + qk . (18)
pk + p,k
All parameters appearing in relations (17), (18) and (19) are computed in order
to satisfy the experimental liquid/vapour saturation curves (Le Mtayer et al. 2004).
From Cavitating to Boiling Flows 271
The nal relaxed state, denoted by the superscript corresponds to the thermo-
dynamic equilibrium state. The liquid and vapour phases have a common pressure,
temperature and Gibbs free energy. Equality of Gibbs free energy of both phases,
Tk k
gk = hk Tk sk = (k Cvk qk )Tk Cvk Tk ln + bk Pk + qk , (19)
(p + p,k )k 1
T (p ) = Tsat (p ) (20)
v0 = Y1 v1 (p ) + Y2 v2 (p ) = Y1 v1 (p ) + (1 Y1 )v2 (p ), (21)
with,
(k 1)Cvk T (p )
vk (p ) = . (22)
p + p,k
Variables v1 and v2 thus correspond to the saturated specic volumes of the liquid
and vapour phases respectively.
A rst relation linking the liquid mass fraction and the pressure is obtained from
(21):
v (p ) v0
Y1 = 2 . (23)
v2 (p ) v1 (p )
In relation (23) the existence of physical solution is fullled by the following condi-
tion:
0 < Y1 < 1 v1 (p ) < v0 < v2 (p ) (24)
By using once more the saturation relation (20), the total energy equation (16)
becomes,
e0 = Y1 e1 (p ) + Y2 e2 (p ) = Y1 e1 (p ) + (1 Y1 )e2 (p ), (25)
where, ( )
(k 1)p,k
ek (p ) = Cvk T (p ) 1 + + qk . (26)
p + p,k
A second relation linking the liquid mass fraction and the nal pressure is obtained,
e0 e2 (p )
Y1 = . (27)
e1 (p ) e2 (p )
272 R. Saurel et al.
Lv (p ) = h2 (p ) h1 (p ). (29)
h2 (p ) (e0 + p v0 )
Y1 = (30)
h2 (p ) h1 (p ).
Equating relations (23) and (30) leads to an equation where the nal pressure p is
the only unknown,
h2 (p ) (e0 + p v0 ) v2 (p ) v0
= 0. (32)
h2 (p ) h1 (p ) v2 (p ) v1 (p )
4 Hyperbolic Solver
For the sake of simplicity, the presentation of the solver is one-dimensional, and
limited to rst order. Details about higher order extensions may be found in Toro
(2009).
System (1) is written in compact form as,
From Cavitating to Boiling Flows 273
U F(U)
+ = 0, (33)
t x
[ ]t
where U = u E Y is the conservative variables vector, and F =
[ ]t
u u2 + p (E + p)u uY represents the associated ux vector.
Godunov scheme
System (33) is a hyperbolic system of conservation laws, with wave speeds u, u ceq
and u + ceq .
The rst-order Godunov method reads,
t
Uin+1 = Uin (F Fi12 ), (34)
x i+12
under the CFL stability condition,
x
t . (35)
max(u ceq )
The cell boundary uxes Fi12 are computed with the following approximate
solver.
HLLC solver
The HLLC solver (Toro et al. 1994) is an approximate Riemann solver for the Euler
equations that is easy to adapt to the present real gas context. In the HLLC context,
each wave is considered as a discontinuity:
a contact discontinuity wave, with characteristic speed SM , through which pressure
and velocity is constant
two discontinuities, with characteristic speed SL and SR (left facing and right facing
waves), through which mass fractions Yi are constant.
As shown in Fig. 5, each of these waves separate the states: (U L , U L , U R , U R ), from
which it is possible to compute the appropriate ux between the two cells during the
time step.
In the example depicted in Fig. 5, UL is the appropriate state to compute the inter-
cell ux F, since it is the state which persists at the physical position of the cell
discontinuity during the time step. In a more general context, there are four possible
cases depending on the signs of the three wave velocities:
case a SL > 0 F = F(UL )
case b SR < 0 F = F(FR )
case c SM > 0, SL < 0 F = F(UL )
case d SM < 0, SR > 0 F = F(UR )
274 R. Saurel et al.
SL SM
UL
UR
SR
UL
UR
x
Fig. 5 Schematic representation of the Riemann problem emerging at each cell boundary. Three
nonlinear waves are emitted, each one being considered as a discontinuity
Following Davis (1988), the right and left velocities are approximated as,
SR = max((u + cw )R , (u + cw )L ), (36)
SL = min((u cw )R , (u cw )L ). (37)
where Sk denotes the speed of the kth discontinuity and [...] denotes the jump of a
given variable across the discontinuity.
Under HLL approximation, the intermediate wave speed is given by
pR pL + R uR (uR SR ) L uL (uL SL )
SM = u = (39)
R (uR SR ) L (uL SL )
Using again the RankineHugoniot jump conditions, one can derive the following
relations for the intermediate state:
= uR SR
R R S S
M R
u = u = SM
R
pR = pR + R (uR SR )(uR SM ) (40)
E = E + pR (uR SM ) S (u S )
R R R (uR SR ) M R M
Y = YR,i
R,i
From Cavitating to Boiling Flows 275
f
left state right state
Fig. 6 3D Godunov method applied to computational control volumes made of tetrahedrons. The
code is cell-centered, and the Riemann problem is solved at each face. tetrahedron center. face
center
These relations fully dene the intercell set of conservative variables UR , and con-
sequently the ux F(UR ) given by,
FR = FR SR (UR UR ). (41)
Substituting the L index to the R index in the above formula leads to a similar expres-
sion for the remaining ux F(UL ).
The computational examples that follow were achieved with the DALPHADT
code based on tetrahedron meshes. In this frame, the Godunov method reads,
4
t
n+1
Uk,i = Uk,i
n+1
(F )f Sf , (42)
Vi f =1 k
5 Cavitating Flows
F G
H
A B C
D
E
(a) (b)
(c) (d)
Fig. 8 Experimental photographs of the break o cycle observed in the 8 Venturi nozzle with
the boundary conditions aforementioned. Flow direction is from right to left. A cavitation pocket
appears, extends and separates in two sub-pockets, one transported with the mean ow and another
one collapsing close to the nozzle. Courtesy of S. Barre (LEGI)
prescribed pressure at the outlet. The imposed conditions at the inlet are the fol-
lowing,
Fig. 9 Computed volume fraction of water vapour. This example shows the same four dierent
parts as those observed during the experimental studies and shown in Fig. 8. The mean attached
cavity length is about 45 mm, in perfect agreement with the experiments
stream and starts to collapse while the attached part starts another cycle (d). The
mean attached cavity length value is 45 5 mm while the quasi-periodic vapour
cloud shedding frequency is about 45 Hz.
To compute this unstable ow a 2D unstructured mesh containing 52450 cells is
used. The grid is rened at the throat in order to capture the cavitation pocket. The
average cell size is 0.13 mm at throat and 0.8 mm elsewhere.
The explicit scheme summarized in Sect. 4 is extended to time implicit integration
and the Riemann problem is preconditioned with the method detailed in Le Martelot
et al. (2013). The ow is computed during 1.8 s of physical time, this time being long
enough to obtain a quasi-stationary ow with quasi-periodic vapour clouds shedding.
An example of the obtained cloud shedding is shown in the volume fraction contours
of Fig. 9. Examining the water vapour volume fraction contours oscillations, we were
able to determine a vapour pocket shedding frequency of about 43 Hz, in excellent
agreement with the measured frequency for the pressure uctuations (45 Hz.).
By performing measurements during every cycle, an average attached cavity
length of about 45 mm has been measured from the computations. These results
show very good agreement with the experiments. Indeed, experimental measure-
ments gave a mean attached cavity length equals to 45 5 mm.
These results show that it is possible to reproduce the large structures of such
cavitating ows with a ow model free of parameters. In particular, no turbulence
model is used.
6 Boiling Flows
To deal with DNS-type of boiling ows, System (1) has to be enhanced by introduc-
ing additional physical eects:
buoyancy,
surface tension,
heat conduction.
278 R. Saurel et al.
Surface tension eects are considered through the Continuum Surface Force (CSF)
method of Brackbill et al. (1992). The capillary force is modelled as,
1
F = Y (43)
where represents the surface tension coecient (N.m1 ), represents the local
curvature (m1 ) and Y1 is the mass fraction of phase 1. The local curvature reads,
( )
Y
= div . (44)
|Y|
Fg = g,
Heat conduction is inserted in the total energy equation of the model through
where the mixture thermal conductivity is given by
the Fourier law q = c T
c = 1 1 + 2 2 and k represents the thermal conductivity of phase k. Details are
given in Le Martelot et al. (2014).
Inserting these extra eects, the ow model now becomes:
Y1 ( )
+ div Y1 u = (g2 g1 )
t
+ div (u) = 0
t
u ( ) (45)
+ div u u + PI = Y1 + g
t
E ( )
+ Y
+ div ((E + P)u) = div c T 1 u + g u
t
A conservative form is available as well,
Y1 ( )
+ div Y1 u = (g2 g1 )
t
+ div (u) = 0
t
u ( ) (46)
+ div u u + PI m = g
t
E + |Y1 | ( )
1 |)u m u c T
+ div (E + P + |Y = g u
t
From Cavitating to Boiling Flows 279
( )
1 Y
Y 1
where m = 1 |I
|Y .
1 |
|Y
This system is closed by the EOS (13). The entropy equation associated to System
(45) reads,
( )
S
c T (g2 g1 )2 (T)2
+ div Su = + c (47)
t T T T2
A closed and adiabatic rectangular domain (12 cm 7 cm) in which the lower half
is lled with saturated liquid water and the upper half is lled with saturated vapour
is considered.
The surface tension coecient is set to = 73.103 N.m1 , the contact angle
is taken constant and equal to = 45 and the gravity acceleration is set to |g| =
9.81 m.s2 .
At start, the initial volume fraction of vapour is vap = 0.0001 in the lower half
domain and vap = 0.9999 in the upper part. Moreover, the initial pressure and tem-
perature are initialised with the hydrostatic gravity prole with the
constraint T = TSAT (P) in each cell. The bottom wall temperature is set constant at
TSAT (Patm ) + 15K.
The numerical scheme presented in the former section is rendered time implicit
and the Riemann solver is low Mach pre-conditioned (Le Martelot et al. 2014). A
mesh made of 960 560 cells is used in the computations that follow.
Three vapour bubbles (radius = 3 mm) are set initially at the bottom wall as shown
in the rst image of Fig. 10. The volume fraction of vapour inside these bubbles is
vap = 0.9999. The computed vapour mass function is shown in the same gure.
The rst instants show the three rst bubbles moving toward the surface due to
buoyancy eects while, as the bottom wall of the box is heated, liquid water begins
to boil, creating a vapour lm. The boiling phenomenon appears as consequence
of wall heating eect that renders the liquid locally slightly overheated. Indeed, the
assumption made is that the liquid contains enough impurities to not accept over-
heating, as pure liquids are able to become metastable while real liquids are not.
Therefore, using the phase transition solver of Sect. 4 after checking that the liquid
is not in stable state, the equilibrium state is computed on the basis of Eq. (32) and a
mixed cell appears. From this nucleation cell site and merging eects due to sur-
face tension, convection and inertia, new bubbles appear. Once created, theses new
bubbles begin to rise and, as there is now again liquid in contact with the bottom
wall, new bubbles appear behind them and begin to grow.
It is worth to mention that the bottom wall of the box is a perfect surface, where
uniform temperature and constant contact angle are set. According to the velocity
280 R. Saurel et al.
Fig. 10 Vapour mass fraction contours at times t = 0 s, t = 50 ms, t = 100 ms, t = 200 ms, t =
300 ms and t = 400 ms. A vapour lm appears at the rst instants. As a result of wall eects com-
bined with capillary ones a big elongated bubble is created at the centre and breaks up in several
bubbles. New bubbles appear at the bottom wall as a consequence of wall heating
proles, the rst bubbles seem to be created between the convective rolls, where the
velocity is the lowest, as shown in Fig. 10.
7 Conclusions
Boiling, evaporation and cavitation are essentially the same phenomenon, driven by
phase transition, but in dierent pressure and temperature conditions. Consequently,
From Cavitating to Boiling Flows 281
they ought to be described by the same model, and deriving such a model is the object
of the present work. The ow model is essentially a hyperbolic system with relax-
ation terms. Coping with boiling, evaporation and cavitation, it isin the authors
knowledgethe rst approach showing such a wide range of applicability. It is also
a very robust computational approach of phase transition compared to existing alter-
natives. Although not shown here, our preliminary investigations show that the ow
model can be extended to cope not only with a liquidvapour couple, but also with
multi-component gas mixtures.
Acknowledgements Part of this work has been carried out in the framework of the Labex MEC
(ANR-10-LABX-0092) and of the A*MIDEX project (ANR-11-IDEX-0001-02), funded by the
Investissements dAvenir French Government program managed by the French National Research
Agency (ANR). We also acknowledge funding from ANR through project ANR-14-CE22-0014. Pr.
Stphane Barre (LEGI) is also gratefully acknowledged for providing the photographs of Fig. 8.
References
Barre, S., Rolland, J., Boitel, G., Goncalves, E., & Fortes Patella, R. (2009). Experiments and mod-
eling of cavitating ows in Venturi: Attached sheet cavitation. European Journal of Mechanics-
B/Fluids, 280(3), 444464.
Brackbill, J. U., Kothe, D. B., & Zemach, C. I. (1992). A continuum method for modeling surface
tension. Journal of Computational Physics, 1000(2), 335354.
Cahn, J. W., & Hilliard, J. E. (1958). Free energy of a nonuniform system. I. Interfacial free energy.
The Journal of Chemical Physics, 280(2), 258267.
Chaves, H. (1984). Changes of phase and waves on depressurization of liquids with high specic
heat. NASA STI/Recon Technical Report N, 84, 25003.
Coutier-Delgosha, O., Fortes-Patella, R., & Reboud., J.-C. (2003). Evaluation of the turbulence
model inuence on the numerical simulations of unsteady cavitation. Journal of Fluids Engi-
neering, 1250(1), 3845.
dAgostino, L., & Salvetti, M. V. (2008). Fluid dynamics of cavitation and cavitating turbopumps
(Vol. 496). Springer Science & Business Media.
Davis, S. F. (1988) Simplied second-order Godunov-type methods. SIAM Journal on Scientic
and Statistical Computing, 90(3), 445473.
Goncalves, E., & Patella, R. F. (2009). Numerical simulation of cavitating ows with homogeneous
models. Computers & Fluids, 380(9), 16821696.
Jamet, D., Lebaigue, O., Coutris, N., & Delhaye, J. M. (2001). The second gradient method for the
direct numerical simulation of liquidvapor ows with phase change. Journal of Computational
Physics, 1690(2), 624651.
Kapila, A. K., Meniko, R., Bdzil, J. B., Son, S. F., & Stewart, D. S. (2001). Two-phase modeling of
deagration-to-detonation transition in granular materials: Reduced equations. Physics of Fluids
(1994present), 130(10), 30023024.
Le Martelot, S., Nkonga, B., & Saurel, R. (2013). Liquid and liquid-gas ows at all speeds. Journal
of Computational Physics, 255, 5382.
Le Martelot, S., Saurel, R., & Nkonga, B. (2014). Towards the direct numerical simulation of nucle-
ate boiling ows. International Journal of Multiphase Flow, 66, 6278.
Le Mtayer, O., & Saurel, R. (2016). The Noble-AbelStiened-Gas equation of state. Physics of
Fluids 28(4), 046102.
282 R. Saurel et al.
Le Mtayer, O., Massoni, J., & Saurel, R. (2004). Elaboration des lois dtat dun liquide et de sa
vapeur pour les modles dcoulements diphasiques. International Journal of Thermal Sciences,
430(3), 265276.
Lund, H. (2012). A hierarchy of relaxation models for two-phase ow. SIAM Journal on Applied
Mathematics, 720(6), 17131741.
Meniko, R., & Plohr, B.J. (1989). The Riemann problem for uid ow of real materials. Reviews
of Modern Physics, 610(1), 75.
Murrone, A., & Guillard, H. (2005). A ve equation reduced model for compressible two phase
ow problems. Journal of Computational Physics, 2020(2), 664698.
Saurel, R., Petitpas, F., Abgrall, R., et al. (2008). Modelling phase transition in metastable liquids:
Application to cavitating and ashing ows. Journal of Fluid Mechanics, 6070(1), 313350.
Simoes-Moreira, J. R., & Shepherd, J. E. (1999). Evaporation waves in superheated dodecane. Jour-
nal of Fluid Mechanics, 382, 6386.
Sinibaldi, E., Beux, F., & Salvetti, M. V. (2006). A numerical method for 3D barotropic ows in
turbomachinery. Flow, Turbulence and Combustion, 760(4), 371381.
Toro, E. F. (2009). Riemann solvers and numerical methods for uid dynamics: A practical intro-
duction. Springer Science & Business Media.
Toro, E. F., Spruce, M., & Speares, W. (1994). Restoration of the contact surface in the HLL-
Riemann solver. Shock Waves, 40(1), 2534.
Wood, A. B. (1930). A textbook of sound. London: G. Bell and Sons Ltd.