Wear 464–465 (2021) 203529

Contents lists available at ScienceDirect

Wear
journal homepage: [Link]

Numerical assessment of cavitation erosion risk using incompressible

simulation of cavitating flows
Mohammad Hossein Arabnejad a, *, Urban Svennberg b, Rickard E. Bensow a
a
Mechanics and Maritime Sciences, Chalmers University of Technology, 412 96, Gothenburg, Sweden
b
Kongsberg Maritime Sweden AB, 681 95, Kristinehamn, Sweden

A R T I C L E I N F O A B S T R A C T

Keywords: In this paper, a numerical method to assess the risk of cavitation erosion is proposed, which can be applied to
Cavitation incompressible simulation approaches. The method is based on the energy description of cavitation erosion,
Cavitation erosion which considers an energy transfer between the collapsing cavities and the eroded surface. The proposed
Risk assessment
framework provides two improvements compared with other published methods. First, it is based on the kinetic
Incompressible simulation
energy in the surrounding liquid during the collapse instead of the potential energy of collapsing cavities, which
avoids the uncertainty regarding the calculation of the collapse driving pressure in the potential energy equation.
Secondly, the approach considers both micro-jets and shock-waves as the mechanisms for cavitation erosion,
while previous methods have taken into account only one of these erosion mechanisms. For validation, the
proposed method is applied to the cavitating axisymmetric nozzle flow of Franc et al. (2011), and the predicted
risk of cavitation erosion is compared with the experimental erosion pattern. This comparison shows that the
areas predicted with high erosion risk agree qualitatively well with the experimental erosion pattern. Further­
more, as the current method can be used to study the relationship between the cavity dynamics and the risk of
cavitation erosion, the hydrodynamic mechanism responsible for the high risk of cavitation erosion at the
inception region of the sheet cavity is investigated in detail. It is shown for the first time that the risk of cavitation
erosion in this region is closely tied to the separation of the flow entering the nozzle.

1. Introduction erosion are an attractive alternative as they can be applied in the early
stage of the design.
Hydrodynamic cavitation is unavoidable in high-performance hy­ Owing to significant progress over the past decade, current numer­
draulic machineries, such as propellers, water turbines, pumps, and ical simulations of cavitating flows are capable of reproducing the large-
diesel injectors. This phenomenon occurs when the pressure drop in an scale cavity dynamics controlling cavitation erosion. With this capa­
accelerating liquid flow leads to the formation of pockets of vapor, bility, it has recently become feasible to develop methods that can
known as cavities. The collapse of these cavities near the surface is predict the cavitation erosion risk based on numerical results. Several of
associated with a high mechanical load in the material, which can such approaches have been proposed in the literature, and they can be
eventually lead to material loss and cavitation erosion. This material loss categorized into two groups: methods based on compressible simula­
significantly increases operating costs of hydraulic machinery; there­ tions or incompressible simulations. In the case of compressible simu­
fore, it is essential to assess the risk of cavitation erosion in the design lation, the strength of collapse-induced shock-waves captured by a
process. Traditionally, the cavitation erosion risk is assessed by applying compressible simulation are analyzed to assess the risk of cavitation
experimental methods on the prototype of a newly designed machine. erosion. These methods have been applied mostly to high-speed cavi­
These experimental methods include visual assessment of collapsing tating flows where the time-scale of the flow is comparable to the time-
cavities using high-speed videos [1] complemented by paint test and/or scale of collapse-induced shock waves. Koukouvinis et al. [5] and Örley
acoustic measurement [2,3,4]. Such experimental methods are, how­ et al. [6] investigated cavitation erosion in high-speed cavitating flows
ever, expensive and mostly used in the late stage of the design process. in diesel injectors using compressible methods and investigated the
Therefore, numerical methods capable of assessing the risk of cavitation hydrodynamic mechanisms leading to cavitation erosion. Mihatsch et al.

* Corresponding author.
E-mail address: [Link]@[Link] (M.H. Arabnejad).

[Link]
Received 27 April 2020; Received in revised form 13 October 2020; Accepted 27 October 2020
Available online 2 November 2020
0043-1648/© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ([Link]
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 1. Energy-based description of cavitation erosion.

[7] studied a cavitating flow in an axisymmetric nozzle with the aim to dependent on the threshold of the method according to Eskilsson and
assess the risk of cavitation erosion. They showed that the areas pre­ Bensow [17]. Alternatively, Koukouvinis et al. [18] proposed an erosion
dicted with high risk of cavitation erosion agree well with the experi­ risk indicator as a function of the total derivative of pressure and vapor
mental erosion pattern by Franc et al. [8]. Blume and Skoda [9] fraction which then was applied to the cavitating flow in the experiment
examined a high-speed cavitating flow over a hydrofoil with the aim to of Franc et al. [8]. Dular et al. [19] developed a method to estimate the
assess risk of cavitation erosion. Using the location of aggressive col­ risk of cavitation erosion based on the micro-jet hypothesis. Peters et al.
lapses and pressure peaks on the foil surface, they were able to predict [20,21] proposed a similar method considering the micro-jet mechanism
the areas with high risk of cavitation erosion. It should be mentioned of cavitation erosion and applied the method to the cavitating flows
that compressible methods have been applied to low-speed cavitating around a ship propeller in model- and full-scale. Eskilsson and Bensow
flows around propellers and foils by Budich et al. [10,11] and Arabnejad [17] compared three of the above mentioned numerical erosion indi­
et al. [12]. However, the simulations in these studies are inviscid due to cator and concluded that further research is necessary to develop reli­
the high computational cost of compressible methods for low-speed able numerical erosion assessment methods.
cavitating flows. A sub-category of incompressible erosion assessment methods,
Alternatives to the above compressible approaches are methods which has been identified by Van et al. [22] to be more suitable for
where incompressible simulations are used to assess the risk of cavita­ numerical erosion assessment, is based on energy description of cavi­
tion erosion. The aim then is to estimate the risk of cavitation erosion tation erosion [1,23,24]. This description considers a balance between
based on the flow properties in the simulation. Ochiai et al. [13] used a the risk of cavitation erosion and the potential energy of collapsing
method where Lagrangian bubbles are injected in the simulation of vapor structures, which is assumed to be proportional to the vapor
cavitating flow, and the risk of cavitation erosion is assessed based on content of collapsing cavity structures and the collapse driving pressure.
acoustic pressure emitted from these bubbles. Similarly, Peters and el Using the energy description of cavitation erosion, a few methods have
Moctar [14] proposed an erosion assessment method based on been proposed [24,25] and applied [26,27,28] in the literature. All of
Lagrangian bubbles present in a multi-scale Euler–Lagrange simulation these methods, however, possess the uncertainty regarding the defini­
of cavitating flows. Krumenacker et al. [15] developed a numerical tion of the collapse driving pressure, which is also noted by Schenke
erosion assessment using the acoustic energy of bubble/cavity implo­ et al. [25]. According to Vogel and Lauterborn [29], the collapse driving
sion. This acoustic energy was obtained from Rayleigh-Plesset software pressure for a single collapsing bubble can be reasonably approximated
which takes input from Eulerian simulation of cavitating flows. Li et al. by the pressure measured far from the collapse center; however, for
[16] developed a method to assess the risk of cavitation erosion based on complex unsteady cavitating flows with several cavities interacting with
the accumulation of time derivative of pressure on the surface. The each other, it remains uncertain how this driving pressure should be
method was successfully applied to a cavitating flow over a foil; how­ obtained.
ever, the predicted cavitation erosion risk by this method is highly In this paper, a new numerical method to assess the risk of cavitation

2
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 2. Volume split of the surrounding liquid of a collapsing cavity.

erosion is presented. Similar to Fortes-Patella et al. [24] and Schenke jets due to the collapse of some bubbles near the surface (Fig. 1c and d).
et al. [25], the present method uses the energy balance between the When these shock-waves or micro-jets hit the surface, a portion of their
collapsing cavities and cavitation erosion. However, in order to avoid acoustic or kinetic energy is absorbed by the material (Fig. 1e). Ac­
the uncertainty regarding the definition of collapse driving pressure, the cording to Fortes-Patella et al. [30], if the absorbed energy by the ma­
developed method is based on the kinetic energy in the surrounding terial exceeds a certain threshold, which is a function of material
liquid of collapsing cavities instead of the potential energy stored in properties, cavitation erosion can occur. It can be noted from the
these cavities. The method is then applied to a cavitating axisymmetric energy-based description that the kinetic energy in the surrounding
nozzle stagnation flow, and the erosion pattern obtained by the present liquid of a collapsing cloud cavity is transferred to the nearby material,
method is compared with the experimental material removal by Franc normally considered occurring through two mechanisms, shock-waves
et al. [8]. and micro-jets, at the level of bubble-scale. Capturing the detail of this
This paper is divided into five sections. Following this introduction, energy transfer directly in engineering simulations is not computation­
the developed method is presented starting with the theoretical frame­ ally possible, both considering the very high mesh resolutions needed in
work of estimating the energy released by a collapsing cavity followed the Eulerian approach as well as considering compressibility effects and
by some implementation details on how the cavity dynamics is traced its time scale. Here, we instead present the development of a framework
leading to the erosion risk estimate. Then the numerical set-up and test which can model this energy cascade based on the flow quantities at the
case used for validation of the method are described. The results are level of macro-scale cavities in the Eulerian incompressible simulation of
presented including a detailed discussion on the cavitation development cavitating flows. For this modeling, first the dynamics of collapsing
and the hydrodynamic mechanisms leading to erosion in this flow as macro-scale cavities are traced during the simulation and the kinetic
well as the comparison between the predicted risk of cavitation erosion energy in the surrounding liquid of these collapsing cavities is estimated.
using the developed method and the experimental erosion pattern. Then, a subgrid modeling is provided which determines the portion of
this kinetic energy which is transmitted to the material surface either
2. Estimation of erosion risk through an approximation of a pressure wave or a micro-jet. In the
following subsections, first, the theoretical description of estimating this
The energy-based description of cavitation erosion caused by a kinetic energy is presented and then the implementation of tracking the
macro-scale cloud cavity containing a large number of bubbles is shown cavity dynamics and the erosion estimation is described.
in Fig. 1. This description suggests that when a macro-scale cloud cavity
is created in the low-pressure region, the surrounding liquid gains po­
2.1. Theoretical description of estimating the kinetic energy
tential energy (Fig. 1a). As this cloud cavity moves into the pressure
recovery region and the collapse starts, most of the potential energy
The kinetic energy in the surrounding liquid of a collapsing cavity
converts into the kinetic energy stored in the inward motion of the liquid
can be obtained from,
while the rest dissipates away from the cavity (Fig. 1b). The dissipated
∫
energy can be in the form of the internal energy due to viscosity or 1
Ek = ρl Ur2 dV, (1)
acoustic energy when the shock-waves upon the collapse of bubbles in Vl 2
the boundary of the cloud cavity radiate away from the cavity. At the
end of the collapse, depending on the distance between the cloud cavity where Vl is the volume of the surrounding liquid, ρl is the liquid density,
and the nearby surface, the kinetic energy of the liquid is converted into and Ur is the collapse-induced velocity in the surrounding liquid. The
acoustic energy carried by collapse-induced shock-waves as a result of discretized form of equation (1) over a finite volume mesh can be
bubbles collapsing away from the surface and/or focused into the micro- written as,

3
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 3. Volume increments (red transparent sphere or spherical cap shells) used in the integration of the kinetic energy in far-field (the solid gray sphere or spherical
cap represents the interface of the volume split shown in Fig. 2). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web
version of this article.)

∑1 error related to approximating the kinetic energy in the far-field volume.

Ek = 2
ρi Vi Ur,i , (2)
i∈Vl
2 It should be mentioned that the sensitivity of the predicted risk of
cavitation erosion to the chosen value for the radius of the sphere or the
where ρi and Vi are, respectively, the density and the volume of the cell i spherical cap is examined in the results section. Based on the volume
located in the surrounding liquid, and Ur,i is the collapse-induced ve­ split, the kinetic energy in the surrounding liquid is split into the near-
locity in the cell i which can be obtained from, field kinetic energy, Ek,1 , and the far-field kinetic energy, Ek,2 ,
( )
→ → → / ⃒→ ⃒
⃒ ⃒ Ek = Ek,1 + Ek,2 . (4)
Ur,i = U i − U c ⋅ d i ⃒ d i ⃒. (3)
The near-field kinetic energy can be obtained directly from equation
→ → (2) while the far-field kinetic energy, Ek,2 is approximated with the
In the above equation, U i is the velocity in the cell i, U c is the
assumption that the collapse-induced velocity in the far-field volume,
volume-averaged velocity of the cells containing the collapsing cavity,
→ Ur , is only a function of the distance from the center of the cavity, r. It
and d i is the vector connecting the centers of cell i and the center of should be mentioned that this assumption is true for a cavity with
collapsing cavity. arbitrary shape if the distance from the cavity center is significantly
Using equation (2) to obtain the kinetic energy of the surrounding larger than the characteristic length of the cavity. Applying this
liquid requires a loop over the cells in the entire domain at each time assumption and using the volume increments, dV, in Fig. 3a, b, and c,
step for each collapsing cavity which is computationally expensive. As a Ek,2 can be approximated as
remedy for this high computational cost, the surrounding liquid of a

⎧∫ h ( )
∫∞
( ) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎪ 1 1
∫ ⎪
⎪ ρ l Ur2 4πr2 dr + ρl Ur2 2πr2 + 2πrh dr h > 5 3 3Vc /4π
1 ⎨ R 2 h 2
(5)
s
Ek,2 = ρl Ur2 dV ≈ ∫∞ ,
V2 2 ⎪
⎪ 1 ( ) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎪
⎩ ρl Ur2 2πr2 + 2πrh dr h ≤ 5 3 3Vc /4π
Rsc 2

collapsing cavity is split into two volumes, a near-field volume and far-
field one. The near-field volume, Vl1 in Fig. 2b and c, includes the liquid where Rs and Rsc are, respectively, the radius of the sphere and spherical
inside the sphere or spherical cap with radius five times larger than the cap, shown in Fig. 2b and c, and h is the normal distance between the
radius of the sphere with the same volume of the collapsing cavity while center of the cavity and the nearest wall. To be able to compute the
the far-field volume, Vl2 in Fig. 2b and c, contains the liquid outside of above integral, the distribution of the collapse-induced radial velocity,
this sphere or spherical cap. The choice of the radius of the sphere or the Ur , should be estimated. The continuity equation for the mixture
spherical cap is somewhat arbitrary, balancing that choosing a larger enclosed by the volume increments in Fig. 3a, b, and c gives,
near-field volume increases the computational cost of the method,
compromising the applicability of the current method in an industrial
Ur ρl A = V̇ v (ρl − ρv ), (6)
application while selecting a smaller near-field volume may increase the
where A is the surface area of the volume increment and V̇ v is the time

4
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 4. Simulation of collapsing spherical mixture cloud cavities (a) simulation configuration, (b–f) the ratio between the kinetic energy in the surrounding liquid and
the initial potential energy and the ratio between the cavity volume and the initial volume of the cavity as a function of time during the collapse.

derivative of vapor content in the cloud cavity. Note that the left hand In order to validate the modeling presented above, a series of
side of equation (6) is the liquid mass flux across the volume increment collapsing spherical mixture cloud cavities with different distances from
and the right hand side is the mass transfer rate inside the volume the wall is simulated. The vapor fraction in these mixture clouds is
increment. Assuming that ρl ≫ρv and substituting the definition of sur­ initialized using a Gaussian distribution with the maximum vapor
face areas A in Fig. 3a, b, and c, the distribution of induced radial ve­ fraction of 0.8 at the center of the cloud. Using this method of initiali­
locity can be obtained from, zation, the resultant clouds represent cloud cavities in the homogeneous
⎧ mixture approach which are obtained by volume averaging spherical
⎪
⎪ V̇ v V̇ v √ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ clouds of bubbles on a coarse mesh. The series of simulations consists of
⎪
⎪
⎨ 4πr2 for Rs ≤ r < h ; 2πr2 + 2πrh for r ≥ h h > 5 3Vc /4π
3
⎪
variations in different ambient pressure, p∞ = [1, 2, 3, 4, 10] bar; initial
Ur = . (7)
⎪
⎪ V̇ v √ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ radius R0 = [2, 4, 8] mm; and initial vapor content, Vv,0 = [4.23 × 10− 7 ,
⎪
⎪ π 8.42 × 10− 7 ,1.71 × 10− 6 ]m3 . The results are consistent for all conditions
3
⎪
⎩ h ≤ 5 3V c /4
2πr2 + 2πrh
and we here limit the presentation to Fig. 4 and the mixture cloud cavity
Substituting the above equation in equation (5) and evaluating the with p∞ = 10 bar, R0 = 8 mm, and Vv,0 = 1.71 × 10− 6 m3 . For each
integral gives the far-field kinetic energy as, collapsing cavity, the exact instantaneous kinetic energy is obtained by
⎧ ( ) summing the kinetic energy in the cells of the computational domain,
1 1 ln(2) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
and the approximate kinetic energy is calculated by equation (9). Fig. 4
⎪ 2
⎪
⎪ ρl V̇ v − + h > 5 3 3Vc /4π
⎨ 8πRs 8π h 4πh
Ek,2 ≈ . (8) shows the evolution of the ratio between the kinetic energy in the sur­
( ) ( )
⎪
⎪
⎪ 1 h + Rsc √ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ rounding liquid, obtained by both exact and approximate formulations
⎩ ρl V̇ 2v ln h ≤ 5 3Vc /4π
3

4πh Rsc during the collapse, and the initial potential energy, Ep0 = p∞ Vv,0 . The
comparisons between the exact and approximate formulation indicate
Using equations (2), (4) and (8), the final form of the kinetic energy that the approximate kinetic energy agrees well with the exact one. It
in the surrounding liquid can be written as, can also be seen that when the collapse starts, the collapse-induced ki­
∑1
2
netic energy in the surrounding liquid increases progressively. This ki­
Ek ≈ Ek,1 + Ek,2 = ρi Vi Ur,i netic energy reaches its maximum, around 65% of the initial potential
i∈Vl1
2
⎧ ( ) energy, before the end of the collapse and suddenly decreases to almost
1 1 ln(2) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
zero at the end of the collapse. It should be mentioned that the 35% of
⎪ 2
⎪ ρ V̇
⎪ l v 8πR − + h > 5 3 3Vc /4π
⎨ 8πh 4π h
+
s
. (9) the potential energy which is not converted to the kinetic energy is
( ) ( )
⎪
⎪
⎪ 1 h + Rsc √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ dissipated due to viscous effects. As collapse proceeds, the velocity
⎩ ρl V̇ 2v ln h ≤ 5 3 3Vc /4π
4πh Rsc gradient in the surrounding liquid becomes very high. This high gradient
then actives the viscous terms in the momentum equations which are

5
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

2.2.2. Cavity tracking

In order to track the collapsing cavities between consecutive time
steps, the cavity extracted in the previous section are stored in a list
called cavityListNew. The collapsing cavities in the previous time step are
also kept and stored in a list called cavityListOld. The main task in
tracking the cavities is to find a best match for a collapsing cavity in
cavityListNew from cavityListOld. The methodology for finding the best
match is based on overlap detection and it is similar to the one presented
in Silver and Wang [32]. In each time step, a table called overlap table is
created by checking the overlap between cavities in cavityListNew and
cavityListOld. A schematic view of the overlap table is shown in Fig. 5.
The table has m rows and n columns where m and n are, respectively, the
size of cavityListNew and cavityListOld. Initially, the values in the table
are set to zero. After the overlap detection, the value stored in the row i
and column j of this table is set to 1 if the ith cavity in cavityListNew
overlaps with the jth cavity in cavityListOld. Based on the non-zero
values in overlap table, the following events can be detected.

• Creation: If all of the values in row i are zero, the cavity i in cav­
ityListNew does not overlap with any cavity in cavityListOld, therefore
Fig. 5. A schematic view of the overlap table. it is a newly detected collapsing cavity. For this cavity, the maximum
kinetic energy in the surrounding liquid, Ek,max,i , is equal to the ki­
responsible for the dissipation of the potential energy. It is also inter­ netic energy calculated from equation (9).
esting to note that regardless of the distance between the wall and the • Continuation: If the value in row i and column j is one while other
cavity, the ratio between the volume of the cavity at maximum kinetic values in the row and the column are zero, the cavity i in cav­
energy and the initial volume of the cavity is 0.07. This value will be ityListNew and the cavity j in cavityListOld overlap only with each
used later to determine which mechanism of cavitation erosion, micro- other. In this case, the cavity i is the continuation of the cavity j. The
jet or shock-wave, is important. maximum kinetic energy of cavity i, Ek,max,i , is then obtained from,
( )
Ek,max,i = max Ek,max,j , Ek,i , (12)
2.2. Implementation of the method
where Ek,i is the instantaneous kinetic energy of the cavity i and Ek,max,j is
According to the energy approach shown in Fig. 1, the maximum the maximum kinetic energy of the cavity j.
kinetic energy near the end of collapse, Ek,max , is focused to the material
through shock-wave or micro-jet mechanism; therefore Ek,max should be • Merge/Break up: If there are several non-zero values in the row i, it
used to estimate the aggressiveness of collapsing events. However, is assumed that several cavities in cavityListOld have merged together
and formed the cavity i in cavityListNew. Similarly, if there exists
obtaining this maximum kinetic energy for each collapsing cavity re­
quires that the cavity is tracked up to its eventual collapse as Ek,max more than one non-zero values in the column j, it is assumed that the
cavity j in cavityListOld has broken up into several cavities in cav­
occurs before the end of collapse (shown in Fig. 4). In the following
ityListNew. In both cases, the cavities in cavityListNew formed by
sections, the algorithms used for detecting the collapsing cavities in each
break-up or merge are treated as newly detected collapsing cavities,
time step and tracking them between consecutive time steps are
therefore the maximum kinetic energy in the surrounding liquid,
explained.
Ek,max,i , is equal to the kinetic energy calculated from equation (9).
2.2.1. Detecting collapsing cavities • Collapse: If all of the values in column j are zero, the cavity j in
The algorithm used to identify collapsing cavities is similar to the one cavityListOld does not have overlap with any cavities in cavityListNew.
It is then assumed that the cavity j in cavityListOld has collapsed in
used by Vallier [31]. At each time step, a list of cells with vapor fraction,
αv , larger than a threshold (αv > 0.01) and negative total time derivative the new time step.
of vapor content (V̇ v < 0.0) is created. The latter condition simply im­
For each collapsed cavity detected by the above algorithm, the
plies that the collapse has been initiated in these cells. The algorithm
collapse locations, Ci , the maximum kinetic energy, Ek,max,i , and the
goes through this list and extracts the collapsing cavities from the cells in
volume of the cavity at maximum kinetic energy, Vk,max,i , are written out
the list that are neighbours. In order to reduce the computational cost of
as the output of the cavity tracking algorithm.
the implemented tool, the collapsing cavities are detected close to their
eventual collapse, when all of the cells containing the cavity have a
2.2.3. Indicator of cavitation erosion risk
negative total time derivative of vapor content. For each detected
Based on the energy approach, the kinetic energy in the surrounding
collapsing cavity, i, the kinetic energy in the surrounding liquid, Ek,i , is
liquid of collapsing cavities is transferred to the nearby material by
obtained from equation (9). The volume of the cavity, Vi , and the cavity
shock-wave or micro-jet mechanisms. These collapsing cavities are
center of volume, Ci , are also calculated from,
mostly cloud of bubbles which are transported to high-pressure regions
∑
k by the flow. For the cloud cavities collapsing far from the surface, the
Vi = Vcell,j (10) collapse of the bubbles inside the cloud produce shock-waves, therefore
it can be assumed that the kinetic energy in the surrounding liquid is
j=1

∑k converted to acoustic energy carried by the collapse-induced spherical

j=1 Ccell,j Vcell,j
Ci = (11) shock-waves. When these shock-waves hit the surface, a fraction of their
Vi acoustic energy is transferred to the surface. This fraction is estimated by
Leclercq et al. [33] using a discrete solid angle projection on a triangular
where k is the number of cells containing the collapsing cavity and Vcell,j
surface element. Similarly, Schenke and van Terwisga [34] introduced a
and Ccell,j are, respectively, the volume and the center of the cell j.

6
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

continuous form of the solid angle projection, which does not require the and the collapse-induced high pressure on the surface is generated by
projection on triangular surface elements. Using this continuous form, spherical shock-waves. Similarly, it is assumed here that for the cavities
the amount of energy absorbed by a surface element j due to the with γ ≥ 3.0, the collapse of these cavities produce only shock-waves
collapsing cavity i can be calculated from, and the kinetic energy in the surrounding liquid of these collapsing
⎛ ⎞ cavities is converted to acoustic energy of the shock wave (Eac,i =
→
1 ⎜ d i,j ⋅→
nj ⎟ Ek,max,i ). The energy absorbed by the material due to collapse of these
Emat.,j = ⎝ ⃒→ ⃒3 ⎠Aj Eac,i , (13) cavities is then obtained from equation (13). For the collapsing bubble
4π ⃒ d i,j ⃒
on the wall (γ = 0.0), Dular et al. [36] showed that cavitation erosion is
solely caused by the micro-jet. Here, we assume that the same is true for
→
where Emat.,j is the energy absorbed by the surface element j, d i,j is the the collapsing cavities on the surface, therefore the kinetic energy in the
vector connecting the center of the collapse and the center of the surface surrounding liquid is assumed to be focused to the kinetic energy of the
element, → nj is the normal unit vector of the surface element, Aj is the micro-jet (Ekm,i = Ek,max,i ) and that equation (15) is used to obtain the
area of the surface element and Eac,i is the acoustic energy in the shock- energy absorbed by the surface. For cloud cavities with 0 < γ < 3.0, it is
wave due to the collapse. Note that in equation (13), collapse-induced assumed that a number of bubbles in these clouds, ns , collapse away
shock-waves are assumed to decay according to linear acoustic theory, from the surface leading to the formation of shock-waves while the rest
which is inline with the work by Johnsen and Colonius [35] who collapse near the surface and produce micro-jets. For the bubbles
investigated the collapse of gas bubbles using a compressible solver. For collapsing away from the surface, a portion of the acoustic energy car­
the collapsing cavities on the surface, the bubbles inside the cavities ried by the collapse-induced shock-waves is transmitted away from the
undergo non-spherical collapse leading to the formation of micro-jets. cloud cavity which can cause erosion. The rest of this acoustic energy is
For these cavities, it can be then assumed that the kinetic energy in transferred back into the cloud cavity due to acoustic interaction leading
the surrounding liquid converts to the kinetic energy carried by to a higher driving pressure for the bubbles collapsing near the surface.
micro-jets, Ekm,i . This kinetic energy is assumed to be uniformly trans­ Using these assumptions, the distribution of the kinetic energy between
ferred to the surface, which is hit by the micro-jets. For the collapsing the shock-wave and micro-jet mechanisms for collapsing cavities with
cavity i, the surface includes the surface elements that are covered by the 0 < γ < 3.0 can be written,
approximate projected area of the cavity on the nearby surface, Aproj,i . ns
(
ns
)
This area can be obtained from, Eac,i = β Ek,max,i , Ekm,i = 1 − β Ek,max,i , (17)
nt nt
( )2/3
3V0,i
Aproj,i = π (14) where β is the portion of acoustic energy transmitted away from the
4π
cloud cavity and nt is the total number of bubbles inside the cloud. To
obtain the exact distribution of the kinetic energy between the two
where V0,i is the initial volume of the cavity i which can be approximated
erosion mechanisms from the above equation, nnst and β should be known
by V0,i = (1.0 /0.07)Vk,max,i according to Fig. 4. Using the above as­
sumptions, the amount of energy absorbed by a surface element j due to as the function of the initial stand-off ratio, γ, which requires a detailed
the cavity i collapsing near the surface can be calculated from, investigation of collapsing clouds of bubbles with different distances
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ from the wall. As mentioned earlier, this detailed investigation is not
⎧ ( )2/3 available in the literature; therefore, we simply assume that β nnst changes
⎪ Ekm,i Aj ⃒⃒ ⃒⃒
⎪
⎪ 3V0,i
⎪
⎪ di,j < + h2i,j linearly with γ with the conditions,
⎨ Aproj,i 4π
Emat.,j = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ , (15) ⎧ n
⎪
⎪ ( )2/3 ⎪ s
⎪
⎪
⎪ 0 ⃒ ⃒ 3V 0,i ⎨ β n = 1.0 if γ = 3.0
⎪
⃒di,j ⃒ ≥ + h2i,j
⎩ (18)
t
4π .
⎪
⎪ ns
⎩ β = 0.0 if γ = 0.0
nt
where hi,j is the normal distance between the collapse center of cavity i
and the surface element j, and Ekm,i is the kinetic energy stored in the To discuss the implication of the above mentioned linear assumption,
micro-jet. we consider a cloud cavity with γ = 1.0 and assume that half of the
To determine which mechanism is dominant for collapsing cloud acoustic energy produced by the collapsing bubbles away from the
cavities, simulations or experimental investigations of collapsing clouds surface in this cloud is absorbed by the bubbles collapsing near the
of bubble with different distances from the wall are needed. These surface (β = 1/2). With this assumption, the assumed linear distribution
simulations or experimental investigations are not available in the indicates that 2/3 of the collapsing bubbles produce shock-waves while
literature and preforming them is out of scope of this paper; therefore in the rest produce micro-jets. We remark that the bubbles forming micro-
the present work, we follow the works by Ochiai et al. [13] and Dular jets are not restricted to bubbles which are near the wall at the beginning
et al. [36] on collapsing single bubbles near the surface. These authors of the collapse. According to the simulation by Ma et al. [37], due to the
concluded that the mechanism for cavitation erosion for collapsing non-uniform distribution of the pressure around the cloud collapsing
bubbles depends on the initial stand-off ratio, γ, of the bubbles which is near a wall, a jet like motion forms toward the wall which pierces the
defined as, cloud. This jet brings bubbles from the location away from the wall to
/√ the regions near the wall leading to a larger number of bubbles pro­
γ=h 33V0 /4π (16) ducing micro-jets. Substituting the linear distribution of nnst based on the
conditions in equation (18) into equation (17), the kinetic energy in the
where h is the distance between the bubble and the wall and V0 is the surrounding liquid is divided between the kinetic energy in the micro-jet
initial volume of the bubble. The simulations by Ochiai et al. [13] have and the acoustic energy of the shock-wave based on the stand-off dis­
shown that for the bubbles with γ ≥ 3.0, the collapse is almost spherical, tance of the cavity as,

7
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 6. Configuration for an axis-symmetric nozzle stagnation flow, a) schematic view of the configuration and the expected cavitation pattern seen in the exper­
iment, b)Computational domain and mesh topology.

( γ ) γ
Ekm,i = 1 − Ek,max,i , Eac,i = Ek,max,i , (19) (
∑
2
)
3.0 3.0
τ= αk μk S, (24)
and the absorbed energy by the surface element, j, is obtained from, k=1

⃒ ⃒ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2/3
̅
⎧
Ekm,i Aj ⃒⃒ ⃒⃒ 3V0,i where S is the mixture strain tensor. To account for the effect of the sub-
⎪ 2
⎛ ⎞ ⎪
⎪
⎪ ⃒di,j ⃒ < 4π
+ hi,j grid scale turbulence, we adopted the wall-adapting local eddy-viscosity
→
1 ⎜ d i,j ⋅→ ⎨ Aproj,i
⎪
nj ⎟ (WALE) model proposed by Nicoud and Ducros [43]. In this model, the
Emat.,j = ⎝ ⃒→ ⃒3 ⎠Aj Eac,i + ̅.
4π ⃒ d i,j ⃒ ⎪
⎪
⎪ ⃒ ⃒ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2/3 sub-grid scale tensor, τsgs , is written as,
⎪
⎪ ⃒ ⃒ 3V0,i
⎩ 0 ⃒di,j ⃒ ≥ + h2i,j
⃒ ⃒ 4π 2
τsgs − ksgs I = − 2νsgs S, (25)
3
(20)
where ksgs is the sub-grid kinetic energy and νsgs is the sub-grid scale
According to the experimental study by Okada et al. [38] and the
turbulent viscosity which can be obtained from,
numerical study by Fortes-Patella et al. [30], there is a linear relation­
√̅̅̅̅̅̅̅
ship between the volume loss due to cavitation erosion after the incu­ νsgs = Ck Δ ksgs . (26)
bation period and the total energy stored in the eroded surface. Using
this linear relationship, an indicator of cavitation erosion risk for the In the above equation, Δ is the cell length scale, Ck , the model con­
surface element, j, can be defined as, stant, is assumed to be 1.6 and ksgs , the sub-grid kinetic energy, can be
calculated from,
1∑ ni
Emat.,j,i
EIj = , (21) ( 2 )2
( d d )3
ts i=1 Aj S̃ S̃
Cw Δ
ksgs = (( )5/2 ( d d )5/4 )2 , (27)
Ck
where ts is the simulation time and ni is the number of collapse events S̃S̃ + S̃ S̃
detected during the simulation. It should be noted that in equation (21),
the total absorbed energy is divided by the simulation time and the area d
where S̃ and S̃ are, respectively, the resolved-scale strain rate tensor and
of the surface element to make the defined erosion indicator indepen­ traceless symmetric part of the square of the velocity gradient tensor,
dent of these two parameters. and Cw , the model constant, is assumed to be 0.325. The cavity dynamics
is captured by Transport Equation Modeling (TEM), where a transport
3. Numerical set-up
equation for the liquid volume fraction, αl , is solved. This equation
reads,
The above method is implemented in a modified version of the
interPhaseChangeFoam solver from the OpenFOAM-2.2. x framework ∂ ( l l) ( )
α ρ + ∇ ⋅ αl ρl ũ = ṁ, (28)
[39]. The solver has been validated and used to study cavitating flows by ∂t
Bensow and Bark [40], and Asnaghi et al. [41]. The governing equations
are the incompressible Navier Stokes equations for two-phase (liquid-­ where ṁ is the mass transfer term which accounts for vaporization and
vapor) isothermal flows. Using the homogeneous mixture assumption condensation. Here, the Schnerr-Sauer model [44] is used for this term.
and applying LES low pass filter [42], the filtered equations for the The mass transfer term is written as the summation of condensation, ṁαlc
mixture of liquid-vapor can be written as, , and vaporization, ṁαlv , terms as,
( )
∂
(ρ) + ∇ ⋅ (ρũ) = 0, (22) ṁ = αl ṁαlv − ṁαlc + ṁαlc , (29)
∂t

∂ ( )
(ρũ) + ∇ ⋅ (ρũ ⊗ ũ) + ∇ ⋅ ([pI − τ]) + ∇ ⋅ τsgs = 0, (23) where ṁαlv and ṁαlc are obtained from,
∂t √̅̅̅̅̅̅̅√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
ρl ρv
l3 2 1
where ρ, ũ, and p are, respectively, the phasic filtered density, the Favre ṁαlc = Cc α max(p − pv , 0), (30)
ρRB 3ρl |p − pv |
phasic filtered velocity vector, and the phasic filtered pressure, I is the
identity tensor, τ is the viscous stress tensor and τsgs is the sub-grid scale √̅̅̅̅̅̅̅√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
tensor in the mixture momentum equations. Adopting the homogeneous (
ṁαlv = Cv 1 + αNuc − α
) 3 ρl ρv
l 2 1
min(p − pv , 0). (31)
mixture assumption and assuming that dynamic viscosity in each phase, ρ RB 3ρl |p − pv |
μk , is constant, the mixture viscous stress tensor, τ, can be obtained from,
In equations (30) and (31), Cc and Cv are set to 1, pv is the vapor

8
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Table 1
Description of the mesh resolutions used in this paper.
Grids n1 n2 n3 ntotal

CM 25 41 43 242k
MM 37 61 65 661K
FM 47 80 85 1623k

pressure, αNuc is the initial volume fraction of nuclei, and RB is the radius
of the nuclei which is obtained from
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
3 1 + αNuc − αl
(32)
3
RB = .
4π n0 αl

The initial volume fraction of nuclei is calculated from

3
π n0 dNuc
αNuc = 6
3
π n0 dNuc
, (33)
1+ 6

where the average number of nuclei per cubic meter of liquid volume, Fig. 7. Frequency spectra of the total vapor volume signal for the simulations
n0 , and the initial nuclei diameter, dNuc , are assumed to be 1012 and 10− 5 with different mesh resolutions.
m, respectively. We remark that by selecting these values, the minimum
pressure in the simulations becomes very close to the vapor pressure
Table 2
which mimics the equilibrium assumption made in barotropic cavitation
Mean and standard deviation of the total vapor volume in the simulations with
models.
different mesh resolutions.
In order to discretize the convective and diffusion terms in the mo­
Simulation
mentum equations, a TVD limited linear interpolation scheme, and a 〈Vc 〉 σVv

standard linear interpolation are, respectively, used. A first-order up­ CM 1.5 × 10 7

−
4.4 × 10− 8
wind scheme is used to discretize the convective term in the liquid MM 1.9 × 10− 7 3.7 × 10− 8
fraction, and the temporal terms are discretized using a second-order FM 1.4 × 10− 7 2.2 × 10− 8
implicit scheme.

3.1. Test case non-dimensional distance in tangential direction, z+ , at radial location

of r = 0.015 mm in these mesh resolutions changes between 670 and
In order to validate the method presented in this paper, the cavi­ 335. It should be mentioned that this mesh resolution is not enough to
tating flow in an axisymmetric nozzle is simulated. This configuration capture the flow detail in the boundary layer. However, according to
reproduces the experiments by Franc et al. [8] in which cavitation Franc and Michel [47], the cavity dynamics presented in this paper, i.e.
erosion is investigated. Fig. 6a shows a schematic view of the flow unsteady sheet cavity due to re-entrant jet, is mostly inertia driven and
configuration. The flow enters a converging nozzle, which is connected viscous effects, such as boundary layer, do not play a role in this cavi­
to a pipe with a diameter of 16 mm. The flow then is deflected by the tation dynamics. For all simulations, a fixed time step is used, so that the
plate, placed 2.5 mm away from the pipe exit, and discharges through a maximum Courant number is around 1.
disk. At the edge of the pipe exit where the flow experiences a sharp
turn, the pressure drops, and a sheet cavity, attached to the upper wall of 4. Results
the disk, forms. Fig. 6b shows the computational domain and mesh to­
pology. The computational domain includes only 1/8 of the geometry Here, first the comparison of the cavity dynamics captured in the
with a symmetry boundary condition used for the side planes. The same simulations with different mesh resolutions and the experiment by
approach is used in the numerical study by Gavaises et al. [45]. Similar Gavaises et al. [45] is made. The hydrodynamic mechanisms of cavita­
to the experiments by Franc et al. [8], the flow rate at the inlet is set to tion erosion are identified and discussed in some detail, including the
6.25 l/s and the pressure at the outlet, pout is adjusted so that the cavi­ flow features responsible for the erosion near the sheet inception; this
tation number, σ , is 0.9. This cavitation number is defined as, has not previously been presented in the literature. The section then
continues by comparing the predicted risk of cavitation erosion using
pout
σ= , (34) the developed method and the experimental erosion pattern by Franc
pin − pout et al. [8], followed by discussion on the effect of mesh resolution on the
predicted risk of cavitation erosion and the relation between the cavi­
where pin is the pressure at the inlet.
tation dynamics and the predicted risk. Lastly, the effect of simulation
To investigate the effect of mesh resolution on the predicted erosion
time and adding a threshold to the erosion risk indicator to mimic the
pattern, three grids have been created. Table 1 represents the descrip­
response of material to aggressive collapse events is discussed.
tion of these grids in the regions where the cavitation is expected to
The numerical results show that the cavitating flow studied in this
occur. In order to check whether these mesh resolutions are adequate for
paper exhibits an unsteady sheet cavity with periodic behaviour gov­
LES, we used the method proposed by Pope [46] where the ratio be­
erned by multiple dominant frequencies. Similar observation has been
tween the sub-grid scale kinetic energy and total kinetic energy is
made in the numerical studies by Peters et al. [20] and Mihatsch et al.
examined. For all three simulations, it was found that this ratio is smaller
[7]. To be able to compare our numerical results with similar numerical
than 0.2 in the cavitating regions, indicating that the resolutions are
and experimental studies in literature, the dominant frequencies are
enough for LES as more than 80% of total kinetic energy is resolved. In
expressed in term of Strouhal number which is defined as,
all three mesh resolutions, the averaged non-dimensional wall distance,
y+ , is around 1 while the averaged non-dimensional stream-wise dis­
tance in the cavity region, x+ , varies from 700 to 350. The

9
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 8. Cavitation pattern in one cycle corresponding to the high dominant frequency in the numerical simulation and the experiment by Gavaises et al. [45] (The
solid red lines in the simulation and dashed white lines in the experiment represent r = 25mm, Ts and t are, respectively, the high-frequency shedding period and the
reference time, and the cavitation pattern in the simulation is shown by iso-surfaces of αl = 0.9). (For interpretation of the references to colour in this figure legend,
the reader is referred to the Web version of this article.)

fLc dominant frequencies can be seen. The high frequency (Sr,h ) is related to
Sr = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅, (35)
2(pin − pv )/ρ the periodic shedding of the cloud cavity due to the re-entrant jet. This
frequency corresponds to Sr,h = 0.27 which agrees well with the nu­
where f is the frequency, Lc is the maximum length of the sheet cavity, merical study by Mihatsch et al. [7] and the reported Strouhal number
and pin is the inlet pressure. Fig. 7 shows the frequency spectra in the corresponding to unstable sheet cavity (0.25–0.35 according to Franc
term of the Strouhal number for the simulations with three mesh reso­ and Michel [47]). The harmonic of this frequency can also be seen in the
lutions, obtained by taking Fast Fourier Transform (FFT) of the history of frequency spectra which corresponds to 2Sr,h = 0.54. The frequency
the total vapor content in the entire domain. In all three simulations, two spectra also shows that there exists a low dominant frequency (Sr,l ) in all

Fig. 9. Radial velocity on the tangential planes where the disturbances on the sheet cavity occur, (a) instantaneous radial velocity, (b) averaged radial velocity (The
instantaneous and averaged interface of cavitating regions, shown by αl = 0.9 and αl = 0.9, are marked by while lines).

10
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 10. Numerical and the experimental erosion pattern, (a) the predicted areas with high risk of cavitation erosion in the simulation with different mesh resolution
(white lines represents the location of eroded areas in the experiment) and (b) the erosion pattern in the experiment by Franc et al. [8]).

of the simulations. In contrast to the high dominant frequency, the low occurs due to a sharp turn at the inception region as it will be shown in
frequency, which corresponds to Sr,l = 0.05 − 0.07, depends on the the paper.
mesh resolution. Similar mesh dependent low dominant frequency has The comparison between cavity dynamics in the present simulations
been observed by Mihatsch et al. [7], although they identified slightly and the experiment in Fig. 8 shows that the large-scale dynamics of the
different range for this frequency (Sr,l = 0.07 − 0.1). It should be noted cavitating structures is qualitatively captured in comparison to the ex­
that the simulations by Mihatsch et al. [7] are obtained using a periments on all tested meshes, although it can be noted that the CM and
compressible inviscid solver, while in the present study, the simulations MM simulations are not sufficiently resolved to correctly represent all
are viscous and based on incompressible approach. Both of these dif­ physics in the flow. This figure shows that the maximum length of the
ferences can explain the discrepancy between the range of the low sheet cavity in CM and MM simulations is larger than in FM simulation.
dominant frequency in our study and the study by Mihatsch et al. [7]. As this sheet cavity transforms to the cloud cavity in step t1 →t2 , the
Table 2 presents the mean and standard deviation of the total vapor resultant cloud cavity becomes larger in the CM and MM simulations.
content which are denoted, respectively, by 〈Vv 〉 and σ Vv . The mean This larger cloud cavity can then travel further downstream, leading to
value of total vapor content changes non-monotonically with the change collapse events at larger radial distances from the pipe exit. These dif­
in mesh resolution, while the standard deviation decreases by increasing ferences between the size of the sheet and cloud cavities and the location
mesh resolutions. This decrease is due to less cycle-to-cycle variation in of the cloud cavity collapse explain the slight mesh dependency of the
the simulations with higher resolutions which will affect the sensitivity predicted risk of cavitation erosion obtained on the simulations with
of the predicted risk to the simulation time as it will be shown later. different mesh resolutions, as detailed below.
Fig. 8 shows the cavity dynamics in one cycle of high-frequency As it can be seen in the zoom-in views of Fig. 8, the interface of the
shedding in the present numerical simulation and the experiment by sheet cavity is disturbed at the inception point near the pipe exit. An
Gavaises et al. [45]. This cavity dynamics can be characterized by the example of these disturbances is marked in Fig. 8d. The comparison
following five steps. 1) t1 →t2 : A large-scale cloud cavity is formed as the between the simulations also shows that these disturbances become
sheet cavity is pinched off from the upper wall (Fig. 8a). While this cloud more significant as the mesh resolution decreases. In order to explain the
cavity is transported downstream by the bulk flow, a new growing sheet cause of these disturbances and their mesh dependency, Fig. 9 presents
cavity is formed on the upper wall (Fig. 8b). 2) t2 →t3 : While the new the instantaneous and averaged radial velocity on tangential planes
sheet cavity is growing, cavitating structures in the shed cloud cavity where the disturbances occur. In this figure, the instantaneous and the
collapse as they are transported downstream. Due to these collapses, the averaged interface of the cavitating regions are also shown by white
cloud cavity has become smaller in Fig. 8c. 3) t3 →t4 : The growing sheet lines. The distribution of the instantaneous radial velocity (Fig. 9a)
cavity reaches its maximum length (Fig. 8d) while all of the vapor shows that when the flow exits the pipe, it separates from the upper-wall
content in the cloud cavity transforms into liquid (between Fig. 8c and due to the sharp turn. This separation zone has a high value of negative
d). 4) t4 →t5 : An upstream moving liquid flow is formed at the end of the velocity (marked by A) and can be seen in all simulations; note though
sheet cavity (between Fig. 8d and e). This liquid flow, often called that some physics responsible for this separation in the CM and MM
re-entrant flow, interacts with the sheet cavity interface as it travels simulations should be considered only qualitatively as the mesh reso­
upstream. This interaction disturbs the interface of the sheet cavity lution in these simulations is coarse. Fig. 9a shows that this reverse flow
(Fig. 8e). 5) t5 →t6 : The re-entrant flow pinches off a large scale cloud is connected to a liquid reverse flow originating from the closure line of
cavity from the sharp turn and a new growing sheet cavity forms on the the sheet cavity (marked by B). The reverse flow marked by B supplies
upper wall (between Fig. 8e and f). packets of liquid at the downstream end of the separation zone which
Note that the shedding mechanism described above have been can travel even further upstream due to the reverse flow in the separa­
extensively observed and studied in the cavitating flow over hydrofoils tion zone. If the upstream moving liquid packets have enough mo­
[48–50]. However, in most of these studies, the foil angle of attack is not mentum to reach the pipe exit, it hits the interface of the sheet cavity at
high enough to create a separation zone at the inception region of sheet the pipe exit. This collision is responsible for the disturbance on the
cavity while in the present flow configuration, this separation zone interface of the sheet cavity seen in the zoom view in Fig. 8.

11
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 11. Contribution of different mechanisms to the predicted risk of cavitation erosion in FM simulation, a) contribution of collapsing cavities with different stand-
off distances, b) contribution of micro-jets and shock-waves to the predicted risk of cavitation erosion.

Fig. 9b shows the averaged radial velocity on the same tangential radial locations. These differences are due to the larger sheet and cloud
plane as Fig. 9a. A region with a negative averaged radial velocity can be cavities in CM and MM simulations which is shown in Fig. 8. Slightly
seen near the wall, which indicates the presence of reverse flow in this higher mesh dependency can be seen in position 3 where the predicted
region. The reverse flow is stronger and also thicker near the pipe exit area with high erosion risk reduces progressively by increasing mesh
due to the separation zone. The comparison between the zoom-in views resolution.
for different simulations in Fig. 9b shows that the negative velocity of Fig. 11a shows the risk of cavitation erosion associated with the
reverse flow in the separation zone decreases as the mesh resolution collapse of cavities with different stand-off ratios. It can be seen that the
increases (lighter blue in zoom-in views as the mesh resolution in­ contribution of collapsing cavities with the stand-off ratio equal to or
creases). As mentioned above, the combined effect of this reverse flow in larger than 3 to the predicted risk of cavitation erosion is insignificant.
the separation zone and the reverse flow at the closure line of the sheet As mentioned in section 2, the kinetic energy in the surrounding liquid
cavity leads to the disturbance on the sheet cavity near the pipe. of these collapsing cavities is assumed to be converted to acoustic energy
Therefore it is expected that the disturbances are more significant in the in shock-waves. As these collapse events are far away from the surface
CM simulation compared to the MM and FM simulations. and the acoustic energy of the shock-wave decays with the distance, the
Fig. 10 compares the experimental erosion pattern by Franc et al. [8] absorbed energy by surface due to the impacts of these shock-waves is
and the areas with high risk of cavitation erosion identified by the expected to be small. Fig. 11b shows the contribution of different
developed method. In the experiment, erosion can be seen in three main mechanisms of cavitation erosion, micro-jets and shock-waves, to the
regions, a region on the lower wall with the radial extension 19 mm < predicted risk of cavitation erosion. It can be seen that although the
r < 32mm, a region on the upper wall with the radial extension contribution of shock-waves are smaller compared to the contribution of
17 mm < r < 27mm, and a region on the upper wall between the pipe micro-jets, the contribution of these two mechanisms are in the same
exit and r = 11 mm. These regions are shown in Fig. 10b, respectively, order, which highlights the importance of considering both mechanisms
by position 1 to 3, and their radial extension are marked by white lines in in the numerical methods predicting the risk of cavitation erosion. We
the numerical results. The comparison between the numerical results in remark that our findings presented here are inline with the description
Fig. 10 shows that regardless of the mesh resolution, the presented of cavitation erosion by Dular and Coutier-Delgosha [51]. In this
method predicts the areas with high risk of cavitation erosion, which are description, it is assumed that the contribution of shock-waves due to
qualitatively comparable with the experimental erosion pattern. It is the collapse of cavities away from the surface is small, which corre­
also seen that the change in the mesh resolution slightly affects the sponds well with the results presented in Fig. 11a. Further, the
radial extension of 1 and 2 as well as the location of position 2. By description of cavitation erosion by these authors assumes that these
increasing the mesh resolution, the radial extension of positions 1 and 2 shock-waves can trigger the collapse of bubbles near the surface which
decreases, and the location of position 2 slightly shifts toward lower can cause erosion through micro-jets. In present study, this acoustic

Fig. 12. Effect of the choice of the radius Rs,sc on the predicted risk of cavitation erosion, a) radial distribution of erosion indicator on the lower wall, b) predicted
areas with high risk of cavitation erosion on the lower and upper wall.

12
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 13. Hydrodynamic mechanism of cavitation erosion risk, a-f) steps in the cavity dynamics in one cycle, g-k) estimated risk of erosion during the steps in the
cavity dynamics (The solid red lines represent r = 25mm and the white lines represent the eroded region in the experiment). (For interpretation of the references to
colour in this figure legend, the reader is referred to the Web version of this article.)

interaction is considered in the cavities with 0 < γ < 3.0. It is assumed radius of the sphere with the same volume of the collapsing cavity. In
that a number of bubbles in these cloud cavities which are away from the order to investigate the effect of this selection on the predicted risk of
surface produce shock-waves. A portion of the acoustic energy carried cavitation erosion, the results from two extra simulations are presented
by these shock-waves is then assumed to be transferred to bubbles near here. In these simulations which are performed on the fine mesh, the
the surface which in turn can cause erosion through the micro-jet radius Rs,sc is assumed to be 3 and 7 times larger than the cloud volume-
mechanism. equivalent radius. Fig. 12a shows the radial distribution of the predicted
As described in section 2.1, the proposed numerical method requires erosion risk on the lower wall using the three values of Rs,sc . It can be
the splitting of the liquid volume around collapsing cavities into near- seen that the predicted risk of cavitation erosion using Rs,sc =
field and far-field volumes. The near-field volume is the liquid inside a √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
3 3 (3Vc /4π ) is slightly shifted toward larger radial positions compared
sphere or spherical cap with radius of Rs,sc . For the results presented so
to the predicted risk using the other two values of Rs,sc . Further, the
far in this section, this radius is selected to be five times larger than the

13
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

Fig. 14. Distribution of the erosion indicator on the lower wall obtained using different simulation times and mesh resolution. mesh resolutions and the distribution
of the erosion depth in the experiment by Franc et al. [8].

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
radial extension of the predicted risk using Rs,sc = 3 3 (3Vc /4π ) is the downstream end of the sheet cavity. According to Arabnejad et al.
slightly smaller. However, the comparison between the erosion assess­ [42], the interaction between this re-entrant jet and the downstream end
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
ment using Rs,sc = 5 3 (3Vc /4π) and Rs,sc = 7 3 (3Vc /4π ) shows that this of the sheet cavity leads to the detachment of cavity structures. Similar
assessment is the same in these two simulations. The maximum risk of detachment of cavity structures can be seen in Fig. 13e. These structures
cavitation erosion in these simulations occurs in the same radial location can collapse due to high-pressure around the closure line and produce a
and the radial extension of the predicted erosion risk is almost the same. high risk of cavitation erosion as it can be seen in Fig. 13j. In step t5 →t6
Fig. 12b presents the areas with high risk of cavitation erosion on the (Fig. 13e → Fig. 13f), the re-entrant jet reaches the region near the pipe
lower and upper wall using the three values of Rs,sc . The comparison in exit and pinches off a cloud cavity from the upper wall. Fig. 13k shows
that in this step, collapse events occur underneath and on the top of the
this figure shows that the predicted high-risk areas are qualitatively the
detached cavity which can cause a high risk of cavitation erosion on
same in the simulations with different values Rs,sc . From this compari­
both walls.
son, it can be concluded that the areas with high risk of cavitation
The steps in cavitation dynamics and its associated risk of cavitation
erosion predicted by the developed method are not sensitive to the
erosion in the position 3, shown in Fig. 10, are presented in the zoom-in
choice of Rs,sc .
views in Fig. 13. Comparison between the locations of high erosion risk
Fig. 13 shows the steps in cavitation dynamics as well as their
and cavity dynamics indicates that the high risk of cavitation erosion
associated risk of cavitation erosion for one shedding cycle in FM
occurs mostly in the region where there is a disturbance in the interface
simulation. It can be seen in this figure that the risk of cavitation erosion
of the sheet cavity near the pipe exit. At the location of these distur­
in position 1 and 2, shown in Fig. 10, is not restricted to the cavity dy­
bances as shown in Fig. 9, there is a liquid reverse flow augmented by
namics in a specific step. In the step t1 →t2 (Fig. 13a → Fig. 13b), as the
the separation zone. When this liquid reverse flow reaches the pipe exit,
detached cloud rolls downstream, aggressive collapse events occur in its
it hits the flow exiting the pipe. This collision can increase the pressure
upstream and downstream ends which induce a high risk of cavitation
locally leading to aggressive collapse events in the pipe exit which are
erosion on both walls (Fig. 13g). More aggressive collapse events can be
responsible for the high risk of cavitation erosion in position 3.
seen in the step t2 →t3 (Fig. 13b → Fig. 13c) when the cloud cavity travels
Fig. 14 shows the tangentially averaged distribution of the erosion
further downstream and starts to shrink. In the step t3 → t4 (Fig. 13c →
risk indicator, EI, on the lower wall in the simulations with different
Fig. 13d), the traveling cloud suddenly collapses leading to a high risk of
mesh resolutions. These distributions are obtained using different
cavitation erosion on both walls. The comparison between the erosion
simulation times in order to investigate the effect of this parameter. It
risk in this step and in other steps indicates that this large-scale collapse
can be seen that the distribution of EI in all three simulations does not
of the cloud cavity is associated with the highest risk of cavitation
change significantly if the simulation time is larger than 20 shedding
erosion compared to the other collapse events in the cycle. During the
periods, Ts . However, the sensitivity of this distribution to the simula­
steps t1 →t4 , a new sheet cavity appears and grows to its maximum length
tion times smaller than 20Ts is not the same in the simulations with
which can be seen in Fig. 13d. In the step t4 →t5 , a re-entrant jet forms at

14
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

different mesh resolutions. This sensitivity is higher in CM simulation collapse events with high risk of cavitation erosion near the inception
compared to MM and FM simulations which can be due to higher cycle- region of the sheet cavity.
to-cycle variation in CM simulation as shown in Table 2. Comparison The results presented in this paper show that the proposed method is
between converged distributions of EI (t ≥ 20Ts ) in all simulations also able to identify areas with high risk of cavitation erosion in a simple
shows that the maximum value of EI is not mesh dependent while the geometry such as an axisymmetric nozzle. In order to examine this
radial location of this maximum value and the extent of erosion risk capability in geometries relevant to marine applications, the proposed
slightly depend on the mesh resolution. As discussed earlier, this mesh method will be applied to the cavitating flows in a commercial water jet
dependency is due to a different dynamics of sheet and cloud cavities in pump and the results will be compared with the experimental erosion
the simulations with different mesh resolutions. Fig. 14d compares the assessment as the future work.
converged distribution of EI in FM simulation (green line) with the
erosion depth profile in the experiment by Franc et al. [8]. It can be seen Author statement
that the radial extension of EI distribution is quite larger than the
extension of the erosion depth profile. This discrepancy is due to the Mohammad Hossein Arabnejad: Conceptualization, Methodology,
definition of the erosion indicator in equation (21) which do not Software, Validation, Investigation, Writing-Original Draft, Visualiza­
consider the response of material to the absorbed energy. Due to this tion, Urban Svennberg: Supervision, Writing- Reviewing and Editing,
deficiency, the energy transferred to the surface elements by all of the Rickard E. Bensow: Supervision, Writing- Reviewing and Editing,
collapse events contributes to the risk of cavitation erosion while the Funding acquisition.
amount of this transferred energy for some collapse events might not be
high enough to cause erosion. To consider only the effect of highly
Declaration of competing interest
aggressive events, one can modify the definition of the erosion indicator
as,
The authors declare that they have no known competing financial
⎧
⎪ Emat.,j,i Emat.,j,i interests or personal relationships that could have appeared to influence
⎪
⎪ > th.
1 ∑ni ⎨ Aj Aj the work reported in this paper.
EIj = , (36)
ts i=1 ⎪
⎪ Emat.,j,i
⎪ 0
⎩
Aj
≤ th. Acknowledgments

where th. is the threshold above which the absorbed energy per area is Financial support for this work has been provided by the EU H2020
high enough to cause erosion. Obtaining this threshold as a function of Project CaFE, a Marie Skłodowska-Curie Action Innovative Training
the material properties is a subject of future work. However, to show Network Project, grant number 642536, and Kongsberg Maritime
that adding a threshold to the definition of the erosion indicator can through the University Technology Center in Computational Hydrody­
improve the results, Fig. 14d presents the distribution of the modified namics hosted at the Division of Marine Technology, Department of
erosion indicator for different values of threshold. The distributions are Mechanics and Maritime Sciences at Chalmers. The authors would like
normalized by their maximum values to able to show them in one figure. also thank Dr. Abolfazl Asnaghi for his insightful comments on the
It can be seen that by increasing the threshold, the extension of the manuscript. The simulations were performed on resources provided by
estimated risk of erosion becomes closer to the experimental erosion Chalmers Center for Computational Science and Engineering (C3SE).
depth profile.
References
5. Conclusions
[1] G. Bark, N. Berchiche, M. Grekula, Application of Principles for Observation and
Analysis of Eroding Cavitation-The Erocav Observation Handbook, EROCAV
This paper presents a new method to assess the risk of cavitation Report, Dept. of Naval Architecture, Chalmers University of Technology, Göteborg,
erosion using incompressible simulations of cavitating flows. The Sweden, 2004.
method is based on the energy balance between the cavitating structures [2] M. Van Rijsbergen, E. Foeth, P. Fitzsimmons, A. Boorsma, High-speed video
observations and acoustic-Impact measurements on a NACA0015 foil, in:
and cavitation erosion suggested by Hammitt [23]. In contrast to pre­ Proceedings of the 8th International Symposium on Cavitation, 2012. Singapore.
vious methodologies [24,27,34] in which the potential energy of [3] W. Pfitsch, S. Gowing, D. Fry, M. Donnelly, S. Jessup, Development of
collapsing cavities has been used for erosion assessment, the presented measurement techniques for studying propeller erosion damage in severe wake
fields, in: 7th International Symposium on Cavitation, 2009. Michigan, USA.
method uses the kinetic energy in the surrounding liquid to estimate the [4] Y. Cao, X. Peng, K. Yan, L. Xu, L. Shu, A qualitative study on the relationship
risk of cavitation erosion. The developed method then estimates how to between cavitation structure and erosion region around a 3d twisted hydrofoil by
this kinetic energy is transfer to the surface through two well-known painting method, in: Fifth International Symposium on Marine Propulsors, Finland,
2017.
mechanisms of cavitation erosion, shock-waves and micro-jets. [5] P. Koukouvinis, N. Mitroglou, M. Gavaises, M. Lorenzi, M. Santini, Quantitative
In order to validate the method, the cavitating flow in a stagnation predictions of cavitation presence and erosion-prone locations in a high-pressure
nozzle flow is simulated using three mesh resolutions, and the areas with cavitation test rig, J. Fluid Mech. 819 (2017) 21–57.
[6] F. Örley, S. Hickel, S.J. Schmidt, N.A. Adams, Large-eddy simulation of turbulent,
predicted high risk of cavitation erosion are compared with the erosion cavitating fuel flow inside a 9-hole diesel injector including needle movement, Int.
pattern in the experiment by Franc et al. [8]. It is shown that regardless J. Engine Res. 18 (3) (2017) 195–211.
of mesh resolution, the predicted areas with high erosion risk are [7] M. Mihatsch, S. Schmidt, N. Adams, Cavitation erosion prediction based on analysis
of flow dynamics and impact load spectra, Phys. Fluids 27 (10) (2015) 103302.
qualitatively in good agreement with the experiment. The agreement
[8] J. Franc, M. Riondet, A. Karimi, G.L. Chahine, Impact load measurements in an
with the experimental results improves with mesh resolution due to an erosive cavitating flow, J. Fluid Eng. 133 (12) (2011), 121301.
improved prediction of the cavity extent and dynamics on the finer [9] M. Blume, R. Skoda, 3d flow simulation of a circular leading edge hydrofoil and
mesh. assessment of cavitation erosion by the statistical evaluation of void collapses and
cavitation structures, Wear 428 (2019) 457–469.
Using the proposed method, the relationship between the cavity [10] B. Budich, S. Schmidt, N. Adams, Numerical investigation of a cavitating model
dynamics and the risk of cavitation erosion at the inception region of the propeller including compressible shock wave dynamics, in: Fourth International
sheet cavity is investigated. It is shown that the high risk of cavitation Symposium on Marine Propulsors,, 2015. Texas, USA.
[11] B. Budich, S. Schmidt, N. Adams, Numerical simulation of cavitating ship propeller
erosion in this region is closely related to the separation zone in this flow and assessment of erosion aggressiveness, in: Fourth International Symposium
region. Due to this separation zone, the reverse liquid flow underneath on Marine Propulsors,, 2015. Texas, USA.
the sheet cavity gains momentum and hits the flow exiting the pipe [12] M.H. Arabnejad, A. Amini, M. Farhat, R.E. Bensow, Hydrodynamic mechanisms of
aggressive collapse events in leading edge cavitation, J. Hydrodyn. 32 (1) (2020)
which increases the pressure locally. This high pressure can trigger 6–19.

15
M.H. Arabnejad et al. Wear 464–465 (2021) 203529

[13] N. Ochiai, Y. Iga, M. Nohmi, T. Ikohagi, Study of quantitative numerical prediction [32] D. Silver, X. Wang, Tracking and visualizing turbulent 3D features, IEEE Trans.
of cavitation erosion in cavitating flow, J. Fluid Eng. 135 (1) (2013), 011302. Visual. Comput. Graph. 3 (2) (1997) 129–141.
[14] A. Peters, O. el Moctar, Numerical assessment of cavitation-induced erosion using a [33] C. Leclercq, A. Archer, R. Fortes-Patella, F. Cerru, Numerical cavitation intensity on
multi-scale euler–lagrange method, J. Fluid Mech. 894 (2020). a hydrofoil for 3d homogeneous unsteady viscous flows, International Journal of
[15] L. Krumenacker, R. Fortes-Patella, A. Archer, Numerical estimation of cavitation Fluid Machinery and Systems 10 (3) (2017) 254–263.
intensity, in: IOP Conference Series: Earth and Environmental Science, vol. 22, [34] S. Schenke, T. van Terwisga, An energy conservative method to predict the erosive
2014, 052014. aggressiveness of collapsing cavitating structures and cavitating flows from
[16] Z. Li, M. Pourquie, T. van Terwisga, Assessment of cavitation erosion with a urans numerical simulations, Int. J. Multiphas. Flow 111 (2019) 200–218.
method, J. Fluid Eng. 136 (4) (2014), 041101. [35] E. Johnsen, T. Colonius, Numerical simulations of non-spherical bubble collapse,
[17] C. Eskilsson, R.E. Bensow, Estimation of cavitation erosion intensity using CFD: J. Fluid Mech. 629 (2009) 231–262.
numerical comparison of three different methods, in: Fourth International [36] M. Dular, J. Požar, T. Zevnik, et al., High speed observation of damage created by a
Symposium on Marine Propulsors, 2015. Texas, USA. collapse of a single cavitation bubble, Wear 418 (2019) 13–23.
[18] P. Koukouvinis, G. Bergeles, A cavitation aggressiveness index within the Reynolds [37] J. Ma, C. Hsiao, G.L. Chahine, Euler-Lagrange simulations of bubble cloud
averaged Navier Stokes methodology for cavitating flows, J. Hydrodyn. 27 (4) dynamics near a wall, J. Fluid Eng. 137 (4) (2015).
(2015) 579–586. [38] T. Okada, Y. Iwai, S. Hattori, N. Tanimura, Relation between impact load and the
[19] M. Dular, B. Stoffel, B. Širok, Development of a cavitation erosion model, Wear 261 damage produced by cavitation bubble collapse, Wear 184 (2) (1995) 231–239.
(5–6) (2006) 642–655. [39] H.G. Weller, G. Tabor, H. Jasak, C. Fureby, A tensorial approach to computational
[20] A. Peters, H. Sagar, U. Lantermann, O. el Moctar, Numerical modelling and continuum mechanics using object-oriented techniques, Comput. Phys. 12 (6)
prediction of cavitation erosion, Wear 338 (2015) 189–201. (1998) 620–631.
[21] A. Peters, U. Lantermann, O. el Moctar, Numerical prediction of cavitation erosion [40] R.E. Bensow, G. Bark, Implicit LES predictions of the cavitating flow on a propeller,
on a ship propeller in model-and full-scale, Wear 408 (2018) 1–12. J. Fluid Eng. 132 (4) (2010), 041302.
[22] T. Van, P. Fitzsimmons, E. Foeth, Z. Li, Cavitation erosion: a critical review of [41] A. Asnaghi, A. Feymark, R.E. Bensow, Improvement of cavitation mass transfer
physical mechanisms and erosion risk models. 7th International Symposium on modeling based on local flow properties, Int. J. Multiphas. Flow 93 (2017)
Cavitation, 2009. Michigan, USA. 142–157.
[23] F.G. Hammitt, Observations on cavitation damage in a flowing system, J. Fluid [42] M.H. Arabnejad, A. Amini, R.E. Bensow, M. Farhat, Numerical and experimental
Eng. 85 (3) (09 1963) 347–356, 0098–2202. investigation of shedding mechanisms from leading-edge cavitation, Int. J.
[24] R. Fortes-Patella, J. Reboud, L. Briancon-Marjollet, A phenomenological and Multiphas. Flow 119 (2019) 123–143.
numerical model for scaling the flow aggressiveness in cavitation erosion, in: [43] F. Nicoud, F. Ducros, Subgrid-scale stress modelling based on the square of the
EROCAV Workshop, Val de Reuil, France, vol. 11, 2004, pp. 283–290. velocity gradient tensor, Flow, Turbul. Combust. 62 (3) (1999) 183–200.
[25] S. Schenke, T. Melissaris, T. van Terwisga, On the relevance of kinematics for [44] J. Sauer, Instationär kavitierende strömungen-ein neues modell, basierend auf
cavitation implosion loads, Phys. Fluids 31 (5) (2019), 052102. front capturing (vof) und blasendynamik, Diss., Uni Karlsruhe, 2000.
[26] R. Fortes-Patella, A. Archer, C. Flageul, Numerical and experimental investigations [45] M. Gavaises, F. Villa, P. Koukouvinis, M. Marengo, J. Franc, Visualisation and LES
on cavitation erosion, in: IOP Conference Series: Earth and Environmental Science, simulation of cavitation cloud formation and collapse in an axisymmetric
vol. 15, IOP Publishing, 2012, 022013. geometry, Int. J. Multiphas. Flow 68 (2015) 14–26.
[27] T. Melissaris, N. Bulten, T. van Terwisga, On the applicability of cavitation erosion [46] S.B. Pope, Turbulent Flows, Cambridge University Press, 2000, [Link]
risk models with a urans solver, J. Fluid Eng. 141 (10) (2019), 101104. 10.1017/CBO9780511840531.
[28] T. Melissaris, S. Schenke, N. Bulten, T.J.C. van Terwisga, On the Accuracy of [47] J. Franc, J. Michel, Fundamentals of Cavitation, vol. 76, Springer Science &
Predicting Cavitation Impact Loads on Marine Propellers, Wear, 2020, 203393. Business Media, 2006.
[29] A. Vogel, W. Lauterborn, Acoustic transient generation by laser-produced [48] E. Foeth, T. van Terwisga, C. van Doorne, On the collapse structure of an attached
cavitation bubbles near solid boundaries, J. Acoust. Soc. Am. 84 (2) (1988) cavity on a three-dimensional hydrofoil, J. Fluid Eng. 130 (7) (2008), 071303.
719–731. [49] M. Callenaere, J. Franc, J. Michel, M. Riondet, The cavitation instability induced
[30] R. Fortes-Patella, G. Challier, J. Reboud, A. Archer, Energy balance in cavitation by the development of a re-entrant jet, J. Fluid Mech. 444 (2001) 223–256.
erosion: from bubble collapse to indentation of material surface, J. Fluid Eng. 135 [50] K. Laberteaux, S. Ceccio, Partial cavity flows. part 1. cavities forming on models
(1) (2013), 011303. without spanwise variation, J. Fluid Mech. 431 (2001) 1–41.
[31] A. Vallier, Simulations of Cavitation - from the Large Vapour Structures to the [51] M. Dular, O. Coutier-Delgosha, Numerical modelling of cavitation erosion, Int. J.
Small Bubble Dynamics, PhD thesis, Lund University, 2013. Numer. Methods Fluid. 61 (12) (2009) 1388–1410.

16