You are on page 1of 16

Ocean Engineering 191 (2019) 106474

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Non-linear noise from a ship propeller in open sea condition


M. Cianferra a ,∗, A. Petronio b , V. Armenio a
a
Universitá di Trieste, Dipartimento di Ingegneria e Architettura, Piazzale Europa 1, Trieste, Italy
b
Iefluids S.r.l., Piazzale Europa 1, Trieste, Italy

ARTICLE INFO ABSTRACT

Keywords: In the present paper, we study the hydrodynamic noise generated by a ship propeller in open sea conditions. We
Ship propeller use Large Eddy Simulation for the hydrodynamic field whereas the acoustic field is reconstructed by applying
LES the advective form of the Ffowcs-Williams and Hawkings equation. A dynamic Lagrangian model is adopted
Ffowcs Williams and Hawkings
for the closure of the subgrid-scale stresses and a wall-layer model allows to skip the resolution of the viscous
Hydroacoustics
sub-layer. The acoustic equation is formulated in the integral form and solved through direct integration of the
volume terms. The propeller herein considered is a benchmark case, whose fluid dynamic data are available in
the literature. A grid of about 3 ×106 cells is able to reproduce accurately both integral quantities like thrust
and torque over the propeller, and turbulence propagating downstream in the wake.
Different noise generation mechanisms are investigated separately. The linear terms give rise to a narrow-
band noise spectrum, with a mean peak at the blade frequency and other peaks at frequencies multiple of the
rotational one. The non-linear quadrupole term reveals a broad band noise spectrum; the shaft vortex provides
the largest contribution to the non-linear part of the noise propagated in the far field.

1. Introduction we focus on the analysis of ship propeller in open-sea conditions and


in absence of cavitation.
Generation and propagation of hydrodynamic noise in the marine Over the years, hydrodynamic performance as well as acoustics of
environment is an emerging field of investigation due to its own impact ship propellers have been investigated using experimental as well as
on a number of engineering applications, including, among the others, numerical techniques, in different flow conditions.
the marine propeller design (Slabbekoorn et al., 2010; Rako et al., Several experimental studies collected data and provided important
2013). The International rules are becoming increasingly stringent in observations on the complex propeller wake dynamics. For example,
protected sea regions, allowing the transit only of certified silent ships, fundamental works of Jessup (1989) and Chesnakas and Jessup (1998)
in areas of particular environmental value. The threat of underwater reported an accurate analysis of the three-dimensional complex flow
noise pollution was recognized by the International Maritime Orga- around propellers, obtained through laser Doppler velocimetry mea-
nization (IMO)which adopted, in 2014, the voluntary guidelines for surements. Recent studies provided detailed description of tip vortex
the reduction of underwater noise produced by commercial shipping
evolution and interaction with the hub vortex, under different propeller
to address the impacts on marine life. The development of new and
configurations and work conditions. Felli and Falchi (2018) presented
complete noise prediction tools is hence required for the design of
the comparison of the wake behavior in axis-symmetric and in oblique
advanced environment friendly ships. The underwater noise generated
inflow conditions, for three inclination angles. The authors analyzed
by a ship propeller propagates through two different mechanisms: the
the wake evolution and instability mechanisms through particle image
first is associated with the propeller shape and rotational velocity; it
velocimetry (PIV). In the work of Calcagno et al. (2005) the complex
generates high level of pressure disturbance at the main frequency,
that is the rotation frequency multiplied by the number of blades. interaction between the hull wake and propeller was addressed. In
The second is associated to the flow field around the propeller and Felli et al. (2011) the authors examined the dependence of the wake
the turbulent coherent structures that populate its wake. Moreover, in instability on the spiral-to-spiral distance through the use of three
presence of cavitation, the acoustic signal appears even more loudly propellers with different number of blades. These studies elucidated
due to the presence of the vapor phase, which usually exhibits unsteady mechanisms of interaction and evolution of the vortex structures in the
features and the implosion of vapor bubbles. Hereafter, unless specified, wake. The destabilization of the tip vortices was found to occur before

∗ Corresponding author.
E-mail address: marta.cianferra@dia.units.it (M. Cianferra).

https://doi.org/10.1016/j.oceaneng.2019.106474
Received 30 March 2019; Received in revised form 2 August 2019; Accepted 19 September 2019
Available online 30 September 2019
0029-8018/© 2019 Elsevier Ltd. All rights reserved.
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

that of hub vortices, independent on the number of blades and the coherent over a length longer than the tip vortex. This feature has some
propeller loading conditions. The tip vortex dynamics is characterized relevance in the present study and will be exploited in the next Sections.
by a multi-step grouping mechanism, driven by the mutual inductance To summarize, different numerical techniques, with increasing com-
between adjacent spirals. plexity and computational cost have been applied over the years for the
Numerical simulations of the hydrodynamic field have been carried evaluation of the hydrodynamic field generated by a ship propeller.
out by several authors using different solution techniques. First, the It is now well established that the potential flow theory can provide
potential flow theory has been used (see, among the other Kerwin, useful data for the structural analysis of the propeller; bulk coefficients
1986); it was considered accurate enough to be used in the optimization of the propeller are well evaluated by using low-order models (RANS)
process of a propeller design, either to predict unsteady vibratory both in steady and unsteady conditions; on the opposite side, wall-
forces generated by a propeller or to estimate fluctuating loads on resolving LES provides very accurate pressure and velocity fields, both
an individual blade as input to a structural analysis. However, the in the near propeller region as well as in the wake, although at a
reliability of potential flow theory has been questioned, in particular cost of very expensive and time consuming simulations. Intermediate
regarding the prediction of the three dimensional actual wake, the techniques, like DES or wall-modeled LES (WLES) appear as a good
reproduction of the unsteady trailing vortex wake interaction and the accuracy-computational cost compromise, providing accurate results
inability to reproduce the unsteady leading edge vortex separation. over the propeller and in the wake, using grid size as small as 10 − 102
More recently, due to the continuous increase of the computer power, times that needed using wall-resolving LES. For this reason and with
the Reynolds Averaged Navier Stokes (RANS) equations have been em- the aim of developing a methodology that can be used for engineering
ployed, providing good prediction of the thrust and torque coefficients purposes, WLES is the approach accomplished in the present paper.
in a wide range of advance coefficients (see, among the others, Baek Apart for the prediction of the hydrodynamic characteristics of the
et al., 2015). Note that these coefficients represent the main output propeller, the computational fluid dynamic analysis can be useful for
needed in the design process of the propeller. For this reason the use the prediction of the noise propagated in the far field. In fact, the
of RANS is nowadays a common practice for engineering purposes. acoustic analysis of a ship propeller is a subject that has aroused
Recently, due to the available computation resources, high resolution growing interest in recent years. Usually, in numerical studies the
simulations of the turbulent field generated by the ship propeller acoustic analogy is employed, meaning that the solution of the hydro-
have been carried out by different research groups. Among the others, dynamic field is decoupled from that of the acoustic one, the latter
Di Mascio et al. (2014) performed detached eddy simulations (DES) of being reconstructed solving the Ffowcs-Williams and Hawkings (FWH)
a propeller placed in oblique flow; the authors adopted a grid of about equation (Ffowcs-Williams and Hawkings, 1969). The equation, formu-
11 × 106 cells, and observed a very complex vortical system composed lated into an integral form, shows that the radiated acoustic disturbance
of a strong tip vortex, less intense trailing vortices due to the variation is given by the sum of linear terms, associated to the pressure over the
of the loading, different blade root vortices and an intense hub vortex. solid surface (loading term) and to the rotational velocity of the blades
Di Felice et al. (2009) carried out both experimental and numerical (thickness term), and of non-linear ones (quadrupole terms), associated
(Large Eddy Simulation, LES) tests of a submarine propeller. Their fine to the presence of the turbulent wake.
mesh simulation, composed of about 4.5 millions cells gave results in
The hydroacoustic analysis of ship propellers (or more complex
agreement with data obtained in physical laboratory. Chase and Carrica
multi-body configurations, including hull, rudder and appendages) has
(2013) performed RANS, DES and Delayed DES (DDES) of a submarine
been based, for long time, on the coupling of RANS and FWH linear (or
propeller, adopting four different grids. The authors concluded that the
porous) equation, since RANS was thought to be able to characterize the
grid refinement has weak effect on thrust and torque evaluation, still
acoustic signature of the propeller, in a relatively simple way. How-
being important for the accurate reproduction of the wake dynamics.
ever, the results obtained from coupling RANS models with different
Worth to be mentioned that in the eddy-resolving simulations reported
methodologies based on the FWH equation showed clearly that both
above, Di Mascio et al. (2014) and Chase and Carrica (2013), solved
in the case of isolated propeller (Ianniello et al., 2013) and propeller
the field up to the wall, while in Di Felice et al. (2009) a wall-layer
together with hull, rudder and appendages (Ianniello et al., 2014),
model approach was adopted, allowing for substantial reduction of the
the calculated noise tends to decrease very fast, as soon as the micro-
number of grid cells.
phone moves away from the body. Moreover, the calculated acoustic
The very recent massive wall-resolving1 simulations of Balaras et al.
pressure is narrow-banded and exhibits predominantly tonal features.
(2015), Kumar and Mahesh (2017) and Posa et al. (2019) stand at the
Moreover, the contribution of thickness and loading terms generated
edge of the binomial accuracy/computational cost. The former carried
by a marine propeller can be extremely low (unlike what happens in
out LES of a propeller in open-water conditions and in presence of an
the aeronautical field), to the point of becoming negligible compared to
upstream appendage at zero incidence, to deal with a realistic configu-
the non-linear effects. These non linear effects, coming from the wake
ration. They adopted the immersed-boundary method and analyzed the
developing downstream a propeller become of fundamental acoustic
flow physics and turbulence in the wake. Kumar and Mahesh (2017)
importance, consequently an accurate and reliable evaluation of the
performed LES using more than 180 millions of grid cells, validated
quadrupole non-linear sources is required and cannot be adequately
their results with experimental data and reported a wide and deep
evaluated by a RANS model.
analysis on the fluid dynamics of the ship propeller. An interesting
aspect concerns the hub vortex, which remains coherent over the length Recent literature has established that LES may be the best method
of the computational domain, equal to 8 propeller diameters. In Posa to reproduce the noise source, since it provides dataset containing the
et al. (2019), LES results were validated against PIV measurements; relevant frequencies for the analysis of radiated noise from a marine
the authors adopted the immersed boundary method with a cylindrical propeller. This because of the fact that the large and relevant turbulent
Eulerian grid composed of about 840 million nodes. These studies scales are directly resolved, whereas the small unresolved scales are
have clearly shown the superiority of eddy-resolving simulations with uncorrelated and do not contribute to the radiated noise.
respect to RANSs when the aim is to study the wake dynamics. Most Keller et al. (2018) studied a five blades propeller using LES and
of the studies on this topic (in addition to the above mentioned refer- analyzed in detail the noise emission given by the thickness and loading
ences) have reported on the presence of strong hub vortex, remaining terms, by isolating the contribution of different circular sections on the
propeller surface. Results show that although the individual blades are
strongly tonal in the rotor plane, the propeller is acoustically compact
1
meaning that the near wall turbulent structures are directly resolved and at low frequency and the tonal sound interferes destructively in the far
a no-slip condition is applied at the solid wall. field. Also, the majority of the propeller sound is generated by localized

2
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

unsteadiness at the blade tip, which is caused by shedding of the tip the kinematic viscosity, 𝜖𝑖𝑗𝑘 is the Levi-Civita tensor, 𝜏𝑖𝑗 is the SGS stress
vortex. tensor, coming from filtering the non-linear terms of the Navier–Stokes
The experimental study of Felli et al. (2014) reported the analysis of equations:
the noise sources over a rudder located behind a marine propeller. The ( )
𝜏𝑖𝑗 = 𝑢𝑖 𝑢𝑗 − 𝑢𝑖 𝑢𝑗 . (3)
investigation concerned near-field pressure fluctuations measurements
on the face and back surfaces of the rudder, at different deflection Simulations are carried out within OpenFoam 2.3.0 framework, an
angles. A novel wavelet-based filtering procedure was presented, able open source software package based on finite volume discretization.
to separate the acoustic and hydrodynamic components of the recorded The set of Eqs. (1) and (2) are solved by means of the Pressure-Implicit
pressure signals. The study highlighted that the acoustic perturbation is with Splitting of Operation (PISO) algorithm adopting a second-order
mainly correlated with the unsteady load variations of the rudder and linear upwind scheme in space, and a backward time integration,
to the shear layer fluctuations of the propeller streamtube. ensuring a second order overall accuracy. The freely available solver
The importance of the evaluation of the quadrupole term in hy- is customized in order to properly include the rotational effects.
drodynamics was highlighted, among the others by Cianferra et al.
(2019), who used the acoustic analogy in conjunction with LES fields 2.1. Dynamic Lagrangian model
to evaluate the noise produced by a finite size cylinder. The authors
showed that the direct evaluation of the non-linear volume terms can The SGS stress tensor is here modeled using a dynamic model in
be carried out in a very efficient way in a class of problems (including conjunction with a Smagorinsky closure. Some more details on the
hydrodynamics) where compressible effects in the fluid dynamic field formulation are reported in Cintolesi et al. (2016, 2015). The deviatoric
are unimportant. part of the SGS stress tensor reads as:
The authors also showed that the quadrupole terms can give contri- 𝛿𝑖𝑗 2
bution in a wide range of frequencies. In a companion paper (Cianferra 𝜏𝑖𝑗 − 𝜏𝑘𝑘 = −2𝜈sgs 𝑆 𝑖𝑗 and 𝜈sgs = 𝑐𝑠2 𝛥 |𝑆 𝑖𝑗 |, (4)
3
et al., 2018a), the authors used the same methodology to study the √
effect of the shape of simple bluff bodies on the emitted noise. They | |
where |𝑆 𝑖𝑗 | = 2𝑆 𝑖𝑗 𝑆 𝑖𝑗 is contraction of strain rate tensor of the large
| |
showed that sharp corner geometries are more loudly than a sphere scales 𝑆 𝑖𝑗 = (1∕2)[(𝜕𝑢𝑖 ∕𝜕𝑥𝑗 ) + (𝜕𝑢𝑗 ∕𝜕𝑥𝑖 )], 𝛥 is the filter width, that is set
and that a prolate spheroid produces a noise one order of magnitude to be the cubic root of the cell volume, 𝑐𝑠 is the Smagorinsky constant.
smaller than a cube or a sphere. Also a direct comparison between a The dynamic Lagrangian model for SGS contribution developed by
cube and a square cylinder showed that the quadrupole terms produced Meneveau et al. (1996) is adopted in order to compute Smagorinsky
by the former is much less relevant than that produced by the elongated constant. It computes the constant 𝑐𝑠 dynamically by averaging over the
body, due to the persistence of the wake in the latter case. fluid–particle Lagrangian trajectories. This model is particularly suited
In the present study, we extend and deepen the work presented when the flow field does not exhibit directions of homogeneity.
in Cianferra et al. (2018b) on the noise radiated from a ship propeller The constant is computed through the formula:
in open sea condition and in absence of cavitation. Our main scope
ℐ𝐿𝑀 (𝐱, 𝑡)
is to evaluate the complete sound pressure level, including the low- 𝑐𝑠2 = . (5)
frequency contribution related to the non-linear quadrupole terms, ℐ𝑀𝑀 (𝐱, 𝑡)
associated to the coherent structures in the wake. In order to obtain The numerator and the denominator are solution of the following
reliable turbulent data we perform LES of the turbulent field around equations:
the rotating body. Further, to avoid the need of using huge grid, also 𝜕ℐ𝐿𝑀 𝜕ℐ 1( )
in view of engineering applications, wall-layer modeling is employed. + 𝑢𝑘 𝐿𝑀 = 𝐿𝑖𝑗 𝑀𝑖𝑗 − ℐ𝐿𝑀 , (6)
𝜕𝑡 𝜕𝑥𝑘 t
The paper is organized as follows: in Section 2 we report the govern- 𝜕ℐ𝑀𝑀 𝜕ℐ 1( )
ing equations for the fluid dynamic field together with the turbulence + 𝑢𝑘 𝑀𝑀 = 𝑀𝑖𝑗 𝑀𝑖𝑗 − ℐ𝑀𝑀 , (7)
𝜕𝑡 𝜕𝑥𝑘 t
SubGrid Scale (SGS) model and the acoustic equation. In Section 3
the propeller benchmark case is described and in Sections 4 and 5 we where t = 1.5𝛥(ℐ𝐿𝑀 ℐ𝑀𝑀 )−1∕8 is the relaxation time-scale, and
discuss respectively the fluid dynamic and the acoustic results. Finally,
𝐿𝑖𝑗 = 𝑢̂ ̂̂
𝑖 𝑢𝑗 − 𝑢𝑖 𝑢𝑗 , (8)
in Section 6 we give concluding remarks. ( )
2 ̂ ̂ ̂
𝑀𝑖𝑗 = 2𝛥 |𝑆|𝑆 𝑖𝑗 − 4|𝑆|𝑆 𝑖𝑗 . (9)
2. Numerical model
The angular hat denotes the additional test-filter operation per-
The problem under investigation is axis-symmetric, with a constant formed at a filter width as large as twice the implicit grid filter. Since
rotation rate. Under these conditions, the governing equations can be the quantity 𝑐𝑠2 is assumed to be always positive, a numerical clipping
solved in the frame of reference rotating with the propeller, introduc- is performed on factors ℐ𝐿𝑀 and ℐ𝑀𝑀 in order to avoid artificial
ing body forces accounting for rotational effects. The incompressible negative values.
Navier–Stokes equations are thus formulated for the absolute velocity
vector, as proposed in Kumar and Mahesh (2017), with the Coriolis 2.2. Wall-layer model
and centrifugal body forces that take into account the rotational effects,
along with the continuity equation: As discussed in the Introduction, LESs of ship propellers have been
𝜕𝑢𝑖 carried out at a model scale using wall-resolving LES. This means that
= 0, (1) the near wall layer together with the turbulent structures are directly
𝜕𝑥𝑖
resolved with a computational cost which increases with 𝑅𝑒2.5 (Piomelli
𝜕𝑢𝑖 𝜕 ( ) 𝜕𝑝 𝜕 2 𝑢𝑖 𝜕𝜏𝑖𝑗
+ 𝑢𝑖 𝑢𝑗 − 𝑢𝑖 𝜖𝑗𝑘𝑙 𝜔𝑘 𝑥𝑙 = − 𝜖𝑖𝑗𝑘 𝜔𝑗 𝑢𝑘 + 𝜈 − , (2) and Balaras, 2002). The literature wall-resolving LESs of ship propeller
𝜕𝑡 𝜕𝑥𝑗 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
in open sea conditions have generally employed a number of grid
where the overbar denotes filtering operation. Filtering is performed cells of the order of 108 -109 , requiring the use of a large number of
implicitly, using the grid cell as top-hat filter in the physical space. In cores on supercomputers. These simulations have exploited flow fea-
Eqs. (1) and (2) 𝑢𝑖 represents the component of the absolute velocity tures not observable in RANS-like simulations as well as in laboratory
in the 𝑖-direction, 𝑝 is the kinematic pressure, 𝑥𝑖 is the 𝑖-component experiments. Also, alternative strategies, intermediate between RANS
of the Cartesian coordinates in the rotating frame of reference, 𝜔𝑖 is and wall-resolving LES, are hybrid techniques, like, among the others,
𝑖-component of the angular velocity of rotating frame of reference, 𝜈 is Detached-Eddy simulation (DES) and wall-layer modeled LES (WLES).

3
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

These hybrid techniques were already used for simulation of open 𝑢𝑏 is the value of the velocity at the wall. In our case 𝑢𝑏 corresponds to
water ship propellers (see the Introduction) and have the advantage the tangential velocity given by rotation.
to allow for the reduction of grid points in the region close to the Thus, the information from the flow (first cell off the wall) is used
blades and the use of high resolution in the wake, responsible of the to determine local wall stress, which is then fed back to the LES in the
quadrupole component of the noise propagated in the far field. WLES form of the proper momentum flux at the wall due to normal diffusion.
is commonly in use in the scientific community, since it greatly reduces Further, since the numerical evaluation of the leading terms of 𝑆 𝑖𝑗 in
the computational cost of the simulation (which increases with Re), the computation of 𝜈𝑆𝐺𝑆 , becomes increasingly wrong with decreasing
since the inertial part of the boundary layer is resolved through the LES grid resolution, they are replaced by the analytical values based on the
whereas the viscous near wall region is modeled through a wall layer location of 𝑦𝑝 , as proposed in Stocca et al. (2011):
model which is required to reproduce correctly the wall shear stress.
⎧ 𝑢 𝑢
Previous studies (Piomelli and Balaras, 2002; Radhakrishnan and ⎪ 𝑆 𝑥𝑧 = 𝜏 𝑢 , 𝑆 𝑦𝑧 = 𝜏 𝑣 𝑖𝑓 𝑦+
𝑝 > 11
Piomelli, 2008; Wang, 1999, 2000) addressed the problem of the wall- ⎪ 𝜅𝑦𝑝 𝑢𝑝 𝜅𝑦𝑝 𝑢𝑝
⎨ 𝑢2 𝑢2 (11)
layer models based on the logarithmic law. They compared the per- ⎪ 𝑆 𝑥𝑧 = 𝜏 𝑢 , 𝑆 𝑦𝑧 = 𝜏 𝑣 𝑖𝑓 𝑦+
𝑝 ≤ 11
formance of the equilibrium wall-layer model with improved versions ⎪ 𝜈 𝑢𝑝 𝜈 𝑢𝑝

of it, namely considering also both pressure gradient and advective or
unsteady terms. For example, in Wang (2000) the friction coefficient of Consequently the value of eddy viscosity near the wall adjusts consis-
an airfoil was computed using different wall-layer model LESs and com- tently with the imposed stress, as described and validated in Fakhari
pared with wall resolved LES. The best results were obtained through et al. (2018).
the Two-Layer Model (TLM), which involves the use of two separate
grids. The results obtained with other wall-layer models, including the 2.3. Acoustic model
one adopted in our work, were all somewhat inaccurate in particular
in the reverse pressure-gradient region, however the equilibrium wall- To reconstruct the acoustic field we adopt the advective form of
layer model reproduced the peak value of the skin friction coefficient the Ffowcs Williams and Hawkings (FWH) equation, given in Cianferra
accurately. Further, the results presented in Radhakrishnan and Pi- et al. (2019, 2018a) and reported here:
omelli (2008) where an oscillating boundary layer was studied (a very 𝜕
□2𝐶 𝐻(𝑓 )𝑝̂ = {[𝜌 𝑣 + 𝜌(𝑢𝑛 − 𝑣𝑛 )] 𝛿(𝑓 )} (12)
complex problem subject to periodic acceleration–deceleration during 𝜕𝑡 0 𝑛
the oscillatory cycle), exhibited acceptable agreement between the 𝜕 𝜕2
− {[𝑝̃𝑛̂ 𝑖 + 𝜌𝑢𝑖 (𝑢𝑛 − 𝑣𝑛 )] 𝛿(𝑓 )} + [𝑇 𝐻(𝑓 )]
wall-layer model LES and resolved LES. Finally, Wang (1999) computed 𝜕𝑥𝑖 𝜕𝑥𝑖 𝜕𝑥𝑗 𝑖𝑗
the far-field noise generated by the sharp trailing edge and found
where
that the predictions of the calculations with approximate boundary [ ]
conditions with logarithmic model did not match the resolved ones in 𝜕2 𝜕2 𝜕2 𝜕2
□2𝐶 = − 𝑐02 + 2𝑈0𝑗 + 𝑈0𝑖 𝑈0𝑗
2 𝜕𝑥 𝜕𝑥 𝜕𝑡𝜕𝑥 𝜕𝑥
the frequency range associated with scales generated in the separation 𝜕𝑡 𝑖 𝑗 𝑗 𝑖 𝜕𝑥𝑗
region, whereas at other frequencies the agreement was acceptable. To represents the convective form of the D’Alembert operator, being 𝑈0𝑖
summarize, although it is well known in literature that the equilibrium the mean flow velocity along 𝑖-direction.
wall-layer model is a basic model, it has also been recognized that In Eq. (12) the function 𝑓 describes the surface:
it is robust in a number of complex-flow cases, including curvature,
local acceleration and stratification. The main drawback of this model 𝑆 = {𝐱 ∶ 𝑓 (𝐱, 𝑡) = 0}. (13)
is in the reproduction of massive separation which is not the case The function 𝑓 is such that 𝑓 (𝐱, 𝑡) < 0 within S and 𝑓 (𝐱, 𝑡) > 0 outside
investigated in our manuscript. 𝑆. 𝐻 and 𝛿 are the Heaviside and Dirac-Delta functions respectively
In the case of ship propellers in open sea conditions, the equilibrium and indicate where the integration is defined. Specifically, if 𝑉 is the
stress model is expected to provide reasonable accurate results, as volume outside the closed surface 𝑆, 𝐻(𝑓 ) = 1 for 𝐱 ∈ 𝑉 and 𝐻(𝑓 ) = 0
already shown in Di Felice et al. (2009). for 𝐱 within 𝑆, while 𝛿(𝑓 ) ≠ 0 only for 𝐱 ∈ 𝑆 and ∇𝐻 = ∇𝑓 𝛿(𝑓 ) = 𝐧𝛿(𝑓 ).
To account for the desired stress at the wall, the boundary condi- The common assumption is that |∇𝑓 | = 1 and 𝑓 is a smooth function,
tions for the scalar field 𝜈𝑆𝐺𝑆 need to be updated at every time step of defined everywhere. Note that the use of Dirac and Heaviside functions
the simulation, the procedure is as follows: is a mathematical formalism that allows to limit the acoustic sources
(i) The wall stress 𝜏𝑤 is obtained from instantaneous horizontal distribution to specific surfaces or fluid regions.
velocity at the first off-the-wall centroid based on law of the wall: 𝑣𝑛 is the velocity of the surface 𝑆 projected along the outward
⎧ 1 normal to 𝑆 (represented by the unit vector 𝐧), ̂ 𝜌 is the constant density
⎪ 𝑙𝑛(𝑦+ +
𝑝 ) + 𝐵 𝑖𝑓 𝑦𝑝 > 11 of the flow field, 𝑝̃ is the pressure distribution on the surface 𝑆 which
𝑢+
𝑝 =⎨ 𝑘 (10)
⎪ 𝑦+ 𝑖𝑓 𝑦+ may coincide with the body surface. In this case, Eq. (12) is referred
𝑝 𝑝 ≤ 11
⎩ to as FWH direct formulation and it verifies 𝑢𝑛 = 𝑣𝑛 . Alternatively, the
√ surface 𝑆 may describe a not-physical surface, external to the body,
where 𝑢+ 𝑝 = 𝑢𝑝 ∕𝑢𝜏 = 𝑢2 + 𝑣2 ∕𝑢𝜏 is the modulus of instantaneous
often referred to as porous domain or control surface. In this case, the
non dimensional velocity at the first off-the-wall computational node 𝑃
net flow across 𝑆 is given by the difference 𝑢𝑛 − 𝑣𝑛 ≠ 0. In the present
placed at distance 𝑦𝑝 from the wall, 𝑘 = 0.41 is the von Kármán constant
work the FWH direct formulation is adopted, since it allows a separate
and 𝐵 = 5.1. Note that 𝑢 and 𝑣 represent the two components of the
analysis of the linear and non-linear terms contribution.
tangential velocity to the wall. More precisely, in this section, 𝑥, 𝑦, 𝑧 are,
The differential form (12) may be turned into the following integral
respectively, the streamwise, spanwise and the wall-normal directions
form by considering the free-space three-dimensional advective Green’s
and 𝑢, 𝑣, 𝑤 the corresponding velocity components. The modulus of the
function:
friction velocity 𝑢𝜏 is calculated from the velocity 𝑢𝑝 at each 𝑃 point
and depends on the distance from the wall 𝑦+ 𝑝 , either from linear or 𝑝̂ = [thickness + loading]𝜏 𝑑𝑆 + [quadrupole]𝜏 𝑑𝑉 (14)
logarithmic law. Then, the wall shear stress 𝜏𝑤 is calculated from the ∫ ∫
friction velocity as 𝜏𝑤 = 𝜌𝑢2𝜏 . The first two source terms on the right-hand-side of Eq. (12) give
(ii) The wall stress 𝜏𝑤 is used to determine the boundary condition rise to surface integrals (thickness and loading terms in Eq. (14)) and
for the turbulent viscosity 𝜈𝑆𝐺𝑆 considering the relation 𝜏𝑤 = (𝜈 + the procedure is described in detail in Najafi-Yazdi et al. (2011). The
𝜈𝑆𝐺𝑆 )𝛥𝑢∕𝛥𝑦, being the gradient of velocity 𝛥𝑢∕𝛥𝑦 = (𝑢𝑝 − 𝑢𝑏 )∕𝑦𝑝 , where transformation of the third (quadrupole) term of Eq. (12) requires the

4
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

manipulation of the double spatial derivative of the Green’s function The near wall resolution is such to obtain a distance of the first
and the procedure is described in detail in Cianferra et al. (2019, grid point off the wall of about 𝑦+ = 40, as required by the use of
2018a). wall layer models. A grid of about 3 millions of grid cells was built.
Then, they have formulated a criterion for selecting cases where the The grid refinement in the wake region was aimed at avoiding the
computation of the time-delays can be omitted and, finally assessed artificial smearing of the helical tip vortex in the wake, since it may
different methodologies for the evaluation of the quadrupole terms. The give a significant contribution to the non-linear part of the noise. The
authors concluded that, when the source is compact, quantified by a present level of wake refinement was already tested in Cianferra et al.
value of the Maximum Frequency Parameter 𝑀𝐹 𝑃 = 1∕𝛥𝑑𝑒𝑙 𝑓𝑚𝑎𝑥 < 1, (2018b).
where 𝛥𝑑𝑒𝑙 = (max |𝐲 − 𝐱𝑚𝑖𝑐 |−min |𝐲 − 𝐱𝑚𝑖𝑐 |)∕𝑐0 with 𝐲 ∈ 𝑉 , and 𝑓𝑚𝑎𝑥 the The WLES was started from a solution obtained with a RANS model
highest frequency at which the fluid dynamic phenomena is observed, on the same grid, in order to avoid the initial development stage. Also,
the direct integration of the volume terms can be performed neglecting after about 60 revolutions obtained by the WLES, the fluid dynamic
the time delays and the solution is more accurate when compared with data were collected for about 5 revolutions.
other solution techniques (i.e. porous method). Since this is the case The time step of the LES is 𝑑𝑡 = 10−4 s giving a Courant number
in the present study, we calculate the FWH non-linear terms by direct ∼ 0.4. The flow fields were printed out every 𝑑𝑡𝑓 𝑙𝑢𝑖𝑑 = 0.01333 s (which
integration over volume regions including the wake. corresponds to 72 degrees) to carry out the phase-average analysis of
Finally, differently than in Cianferra et al. (2019, 2018a), here we the fluid dynamics field, and every 𝑑𝑡𝑎𝑐𝑜 = 0.0027 s (which corresponds
consider a rotating frame of reference, since the CFD simulation is car- to about 15 degrees) for the acoustic analysis. Adopting the sampling
ried out likewise. Thus, when computing the distance 𝑟 = |𝐱(𝑡) − 𝐲(𝜏)| time 𝑑𝑡𝑎𝑐𝑜 , an entire revolution (360 degrees) is discretized with just 22
inside the integrals, where 𝐱 is the measurement point (also referred snapshots. This implies a loss of information on the high frequencies,
to as probe or microphone) and 𝐲 is the source point (which varies but allows to collect data for several revolutions, in order to study the
along the solid surface or along the volume), one needs to consider the appearance of lower frequency modes.
measurement point 𝐱 as it were rotating with angular velocity 𝛺 = |𝜔|. We use the standard PISO algorithm, implemented in OpenFoam
(version 2.3), but modified by including the non-inertial forces in the
3. Hydrodynamics of the propeller momentum equation, as described in Eq. (2). As boundary conditions,
at the inlet section we give a uniform flow field with the advance
In this study we consider a benchmark propeller, the SVA VP1304, velocity 𝑈0𝑥 = −4 m∕s, at lateral boundaries we set a slip condition and
whose complete documentation, including geometry, experimental data at the outlet we impose a zero gradient condition. At the solid walls of
and numerical results, is available online.2 The data for both the open ̄ 𝑟,
the shaft, hub and blades, the tangential velocity is given as 𝑢𝑡 (𝑟) = 𝛺×̄
water and the installed configurations were presented and discussed in with 𝑟̄ radial distance from the rotation axis, and 𝛺̄ = (2𝜋𝑛, 0, 0), with
the International Symposiums on Marine Propulsors in 2011 and 2015 𝑛 = 15 𝑟𝑝𝑠. For the pressure, we set 𝜕𝑝∕𝜕𝑛 = 0 to all boundaries of the
respectively. computational domain, 𝑛 being the direction normal to the surface. The
In the present study, we carry out the analysis of the uniform flow constant flow density and the speed of sound are set to 𝜌 = 998 Kg∕m3
case of an isolated propeller, at design advance ratio and Reynolds and 𝑐0 = 1400 m∕s respectively, these quantities being necessary for the
number: acoustic analysis.
𝑈 𝑈 𝐷
𝐽𝑣 = 0 = 1.0683, 𝑅𝑒 = 0 = 889 680, (15)
𝑛𝐷 𝜈 4. Fluid-dynamic analysis
where 𝑈0 = −4 m∕s is the advance velocity, 𝑛 is the rotational velocity
in rps and 𝐷 = 0.25 m the diameter of the propeller. The simulation results are compared with available experimental
For the value 𝐽𝑣 = 1.0683, the thrust (𝐾𝑇 ) and torque (𝐾𝑄 ) coeffi- data. For the value of the advance coefficient 𝐽 = 1.068 we obtain the
cients are: following values of the force and torque coefficients:
𝑆 𝑄
𝐾𝑇 = = 0.3538, 𝐾𝑄 = = 0.09096, (16) 𝐾𝑇 = 0.3414, 𝐾𝑄 = 0.09051 (17)
𝜌𝑛2 𝐷4 𝜌𝑛2 𝐷5
where 𝑆 and 𝑄 are the thrust and the torque provided by the propeller. corresponding to errors of
In the experimental set-up the rotation direction is right-handed. The 𝑒𝐾𝑇 = 3.5%, 𝑒𝐾𝑄 = 0.5% (18)
solid body is composed of the five-blade propeller and the shaft, placed
downstream, according to the typical arrangement for isolated (or with respect to experimental data.
equivalently open sea) propeller performance assessment. The coefficients are reproduced satisfactorily; in previous studies
We use a cylindrical numerical domain, as sketched in Fig. 1(a); the carried out using WLES (see for example (Di Felice et al., 2009), where
diameter is equal to 7𝐷, and the length is 10𝐷 as suggested in Kumar an equilibrium wall-layer model was adopted), thrust and torque were
and Mahesh (2017). obtained with error of about 5% with respect to the experimental
The reference coordinate system is chosen such that the blades of values.
the propeller are located at the plane 𝑥 = 0 and the flow is in the We also observed a dependency of the coefficients 𝐾𝑇 and 𝐾𝑄 on
direction of negative 𝑥, so that 𝑈0,𝑥 = −4 m∕s. The axis of rotation the grid size, in agreement with previous studies (see for example Chase
of the frame of reference is 𝑟 = 0. and Carrica, 2013; Di Felice et al., 2009). In the previous test discussed
The plane containing the blades is located 3𝐷 downstream of the in Cianferra et al. (2018b), we used a grid of about 6 million cells,
inlet and 7𝐷 upstream of the outlet section. The grid is obtained in the same as the one described here in the wake region, but refined close
following way: we generate a cylindrical O-grid non uniform along the to the propeller boundaries (mean distance of the first grid point off-
streamwise direction. The cells are clustered in the regions occupied by the-wall ∼ 15𝑦+ ). The results concerning the mean values of thrust
the shaft and the blades and coarsened moving towards the cylindrical and torque coefficients were slightly worse than those obtained in the
lateral boundaries; Then we adopt the OpenFoam tool SnappyHexMesh present study, but the time spectra of 𝐾𝑇 and 𝐾𝑄 appeared qualitatively
to build the mesh, merging geometry of the propeller within the cylin- comparable with that reported in Kumar and Mahesh (2017), namely
drical grid. The mesh quality parameters fit the OpenFoam criteria of broadband and with peaks on both revolution and blade frequency.
aspect ratio, skewness and non-orthogonality. With the present resolution near the wall, the values of 𝐾𝑇 and 𝐾𝑄
are closer to the experimental data and the time signal exhibits smaller
fluctuations, similarly to what reported in Chase and Carrica (2013),
2
https://www.sva-potsdam.de/en/potsdam-propeller-test-case-pptc. for a resolved DES. Fig. 2(a) shows the signal in time, expressed as

5
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 1. Numerical fluid dynamic domain and acoustic control domains: two cylinders, parallel to the 𝑥-axis, of radius 0.14 m (domain A) and 0.08 m (domain B).

Fig. 2. Fluid dynamic data of ship propeller.

rotational angle, of 𝐾𝑇 − ⟨𝐾𝑇 ⟩∕⟨𝐾𝑇 ⟩, in the range 150 − 550◦ degrees To the best of our knowledge, there are not quantitative literature
of rotation. The two signals refer to two different simulations: the one data on the fluid dynamic field generated by the propeller investigated
reported in a previous work (Cianferra et al., 2018b) and the one in this paper, thus we proceed with qualitative comparison with results
carried out in the present study. The coefficient appears in statistically reported in the high resolution LES of Kumar and Mahesh (2017)
steady conditions, with fluctuations around the mean value. As dis- and Posa et al. (2019), where wall-resolving LES were performed in
cussed above, the fluctuations of the signal relative to the grid more conjunction with huge grids. The authors studied propellers different
refined at the wall are more energetic compared to the signal obtained from the present one, with a different advance ratio 𝐽 and Reynolds
with the present grid. These results are reasonable, in that the wall- number. In Kumar and Mahesh (2017) 𝐽 = 0.889 and 𝑅𝑒 = 8.9 × 105 ,
layer model is designed to work better when the first grid point is while in Posa et al. (2019) three different working conditions were
placed in the log region of the boundary layer, and, thus, in this case considered 𝐽 = 0.74, 0.65, 0.56. Also, both in Kumar and Mahesh (2017)
the thrust and the drag are reproduced with good accuracy. On the and Posa et al. (2019) the shaft is located upstream. Although qualita-
tive, the comparison of non dimensional quantities may be useful to
other hand, the increase of the distance of the first point from the wall
place our data within the context of eddy resolving simulations of ship
reduces the level of fluctuations at the wall. We also carried out a
propellers, also in view of the relevant acoustic analysis.
simulation after switching-off the wall-layer model, thus performing an
Coherent structures are observed up to 4 𝐷 downstream of the
under-resolved LES. The coefficients were strongly underestimated; for
propeller: in Fig. 2(b) the isosurface of the non-dimensional vorticity
example, the value of the thrust coefficient 𝐾𝑇 decreased after about
magnitude |𝜔|𝐷∕𝑈0 = 9.4 is reported. After that length, because of the
7000 iterations, which is approximately one revolution, up to an error
grid coarsening along the 𝑥-direction, the grid resolution is not suffi-
of about 30%. This demonstrates the need to use a wall-layer model in
cient to capture accurately the tip vortex, which, furthermore, tends to
the present case, to reproduce the near-wall dynamics and obtain the lose energy and to dissipate. The tip and hub vortices are structures
correct stress at the wall. commonly observed in the wake of marine propellers (Di Mascio et al.,
The fact that our simulation gives accurate values of the coefficients 2014; Kumar and Mahesh, 2017). In the present case, we reproduce
is not surprising, since the equilibrium wall-layer model is very robust a pulling-type propeller configuration, typical of scale model tests in
and accurate in simulations of non separated boundary layer, as for water basins. This configuration differs from the real scale application,
the case of the blades of the propeller. On the other hand, the fact that where the propeller is in a pushing-type configuration. The main dif-
reasonable results can be obtained making use of wall-layer modeling ference stands in the fact that, being the shaft located downstream
approaches makes the analysis of ship propeller affordable even for of the propeller, we observe a shaft vortex instead of hub coherent
engineering purposes. structures. In fact, from the base of the propeller blades a vortex is

6
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 3. Visualization of the tip-vortex and shaft-vortex coherent structures.

Fig. 4. Phase-average non-dimensional axial velocity (first column), vorticity magnitude (second column), tke (third column). Section perpendicular to the mean flow (𝑥-axis), at
locations (from top panel to bottom panel) 𝑥∕𝐷 = 0.2, −0.2, −0.6.

created, which then twists around the shaft along a helical shape. This In Figs. 4 and 5 we select different planes perpendicular to the
is clearly depicted in Fig. 3 through the Q-criterion: Figs. 3(a) and 3(b) mean flow direction, at locations 𝑥∕𝐷 = 0.2, −0.2, −0.6, −1, −3, −5
depict the isosurface 𝑄 𝐷2 ∕𝑈02 = 40 and 80 respectively. We recall that (respectively from top panel to bottom panel). There, we show the
positive values of the scalar Q indicate flow regions characterized by contour plot of phase average non-dimensional axial velocity, vorticity
vorticity. The case of 𝑄 𝐷2 ∕𝑈02 = 40 is comparable with that reported magnitude and turbulent kinetic energy (TKE) (from left panel to right
in Posa et al. (2019), keeping in mind all the differences between the panel). Phase-average is computed as
two cases. In both figures, the contour of the magnitude of vorticity and ∑𝑚
𝑓 (𝐱, 𝑡0 + (𝑖 − 1) ∗ 𝑇 )
of the scalar 𝑄 reveals the shaft vortex to be more energetic than the tip 𝑓̄(𝐱) = 𝑖=1
vortex. This has important effects in the acoustic analysis, reported in 𝑚
the next Section 5. Our results are consistent with those of Kumar and where 𝑓̄ is the phase-average value of any field 𝑓 and 𝑇 = 1∕𝑛𝑁 is
Mahesh (2017), where the hub vortex was observed to remain coherent the period associated to blade passage. Consistently with Kumar and
over the entire length of the computational domain. Mahesh (2017), the upstream region is characterized by the suction

7
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 5. Phase-average non-dimensional axial velocity (first column), vorticity magnitude (second column), tke (third column). Section perpendicular to the mean flow (𝑥-axis), at
locations (from top panel to bottom panel) 𝑥∕𝐷 = −1, −3, −5.

effect of the propeller. This clearly appears looking at the axial velocity, As a further consequence of the grid coarseness, with respect to the
vorticity and TKE (Fig. 4) which also show that the propeller-induced reference wall-resolving LESs, we observe the weakening of tip vortex
perturbations do not produce significant turbulence. The level of TKE signatures in the downstream region at about 2 diameters, (see Fig. 5)
in the upstream region is comparable with that of Kumar and Ma- whereas it occurs more downstream (at about 3 diameters) in Kumar
hesh (2017) although a bit lower, which is probably due to our grid and Mahesh (2017).
resolution compared to the reference ones (as large as 100 time our Fig. 6 shows the axial component of vorticity fields at 𝑥∕𝐷 =
one). However, the TKE contour plot near the propeller (second row −0.4, −0.6 downstream of the propeller. The spiral roll-up of the blade
panel) shows significant traces of turbulent fluctuations. We observe a wake vortex sheet is not as well-defined as in Kumar and Mahesh
phenomenon similar to that described in Kumar and Mahesh (2017), (2017). However, it is clear that clockwise vorticity is released by the
namely, as soon as the tip vortex becomes unstable, the TKE starts propeller tip, while a counterclockwise spiral mode characterizes the
increasing. At about three diameters downstream of the propeller and motion around the shaft. The opposite sign between tip and shaft vortex
beyond, both the tip and the shaft vortex break down completely, may be observed also in Fig. 7 and it is reported in Kumar and Mahesh
producing TKE (see last two rows of Fig. 5). Note that, the highest (2017) (they observe opposite sign between tip and hub vortex). In
values of TKE and vorticity are observed around the shaft. Comparing Fig. 7, non-dimensional 𝑧-component of vorticity contour is depicted on
the vorticity magnitude with those of Posa et al. (2019), we note the plane 𝑧 = 0. We can anticipate that, from an acoustic point of view,
a similar behavior and values, except for the high values we have shaft and tip vortex are relevant, both having considerable amount of
around the shaft, which in their case decrease faster downstream. This vorticity. Note that the shaft vortex is commonly observed and referred
may be due to the different blade geometry and work conditions, but to as hub vortex, being the shaft located upstream. In previous studies,
also because of the presence of the shaft downstream, which in the (see for example Okulov and Sørensen, 2007), the hub vortex is found
present case may help to keep the vortex structure consistent in space to be stronger than the tip vortex, being generated by the coalescence
(moreover, we remind that we set the slip condition on the shaft), and of the blade root vortices and thus having a strength that is N times
prevents from the interaction and coalescence of the hub vortex with larger than that of the tip vortex. In the present study, the flipped shaft
the tip vortex. configuration influences the flow dynamics and the interaction among

8
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 6. Phase-average axial component of non-dimensional vorticity. Section perpendicular to the mean flow (𝑥-axis), at locations 𝑥∕𝐷 = −0.2 (left panel) and 𝑥∕𝐷 = −0.6 (right
panel).

the vortical structures. However, a very energetic vortex is created


which is rolled up around the shaft and remains stable downstream.
This is an important issue which deserves discussion. The configuration
herein studied is that typically employed in open-water laboratory scale
tests, which substantially differ from the real scale application. This
is typically done under the assumption that the presence of the shaft,
which has to sustain the propeller during the test, does not affect the
loads (thrust and torque) over the propeller. Although this is true for
the loads, it has to be noticed that the downward shaft substantially
modifies the structure of the wake, since its own interaction with
the tip vortex differs from the mutual influence between tip and hub
vortex. In other words, if one is interested in the analysis of the wake
developing downward a propeller, the pulling type test is not the
best suited for that. In Figs. 8 and 9 we report the contour plot of Fig. 7. 𝑧−component of non-dimensional vorticity. 𝑥𝑦-plane at 𝑧 = 0.
phase average non-dimensional pressure, root mean square of pressure
and turbulent viscosity 𝜈𝑆𝐺𝑆 (from left panel to right panel). The
different planes, perpendicular to the mean flow direction, are located
as well as for the fluid dynamics of the wake, which is relevant for the
at 𝑥∕𝐷 = 0.2, −0.2, −0.6, −1, −3, −5 (respectively from top panel to
non-linear quadrupole contribution to the noise.
bottom panel). The low pressure cores of tip vortices are evident up
to 𝑥∕𝐷 = −1. Downstream, at 𝑥∕𝐷 = −3, −5, the region of low pressure
5. Acoustic analysis
around the shaft emphasizes the presence of the vortex. The pressure
fluctuation associated to the tip vortex are visible near the propeller,
up to a distance of 0.6𝐷, while the region inside the shaft vortex As mentioned, the computation of FWH volume integrals is feasible
contains large pressure fluctuations, up to 5 diameters downstream. The only in case of compact sources, thus when time delays are unimpor-
contour plot of eddy viscosity normalized with the molecular viscosity tant. Otherwise the storing process of all the fluid dynamic data is too
shows satisfactory values in the tip vortex regions and around the shaft. expensive in terms of RAM and CPU time. In literature, it is common to
A comparison with previous Figs. 4 and 5 shows that the turbulent find the assumption of compact noise source. In Cianferra et al. (2019)
transfer of energy occurs in regions characterized by high vorticity. a rigorous discussion was proposed, based on a dimensional analysis
The fact that vortex cores are not evident and the contour plots show involving source length, maximum frequency and speed of sound, and
instead continuous strips is due to the phase average procedure. We giving a criterion based on the value of the non-dimensional parameter
may conclude that the contour plot of the three quantities are in good MFP.
qualitative agreement with the results reported in Kumar and Mahesh In presence of a moving body, such as a rotating propeller, a
(2017), although the working condition is different and considering consideration on the importance of time delays can be addressed by
that our grid in much coarse. calculating the retarded surface 𝛴 (see Ianniello et al., 2013). 𝛴 is
Finally, the flow field in the wake of the blade is depicted in Fig. 10: defined for each time 𝑡 of signal reception and for each observation
contours of radial (top left panel) and axial (top right panel) non- point 𝐱 (therefore it would be 𝛴 = 𝛴(𝑡, 𝐱) ), and is the surface
dimensional velocity fields are extracted on a 𝑟𝑥-plane, with 𝑟 being the constituted by the points 𝑦(𝜏), such that 𝑡 − 𝜏 − |𝐱 − 𝐲(𝜏)|∕𝑐0 = 0. In
radial direction. The plane is located at the end of the trailing edge, as this case the observer 𝐱 is at rest and the source 𝐲 rotates and changes
sketched in Fig. 10 (bottom panel). The vortex sheet in the trailing edge in time. The procedure of calculating 𝛴 consists of: (𝑖) choose a time 𝑡∗
wake of the blade can be observed in both axial and radial velocity of acoustic signal reception and an observer location 𝐱; (𝑖𝑖) calculate 𝜏
fields. The jump in radial velocity is quite sharp, as also evidenced with the bisection method on the function 𝑓 (𝜏) = 𝑡∗ − 𝜏 − |𝐱 − 𝐲(𝜏)|∕𝑐0
in Kumar and Mahesh (2017). where 𝐲(𝜏) are the points on the propeller that are rotating and 𝐱 is
Overall, our results exhibit a good accuracy, when compared to a microphone location; (𝑖𝑖𝑖) once 𝜏 ∗ is found that satisfies 𝑓 (𝜏 ∗ ) = 0,
high-resolution reference simulations which take advantage of grids as we calculate 𝐲(𝜏 ∗ ) that corresponds to the retarded surface. For high
large as 100 times our one. This is true for the propeller coefficients rotational values of the Mach number (as for helicopter blades) the

9
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 8. Phase-average non-dimensional pressure (first column), root mean square of pressure (second column), turbulent viscosity (third column). Section perpendicular to the
mean flow (𝑥-axis), at locations (from top panel to bottom panel) 𝑥∕𝐷 = 0.2, −0.2, −0.6.

retarded surface is substantially different from the solid surfaces of red line in both Figs. 12 and 13), with a frequency equal to 𝑛𝑁,
the blade, meaning that the time delays play a relevant role in the being 𝑛 the number of revolutions per second and 𝑁 the number
propagation of the acoustic signal. Conversely, at very low rotational of blades. Very close to the propeller (microphone A) the amplitude
Mach numbers, as those typical of ship propellers, the retarded surface peak reaches about 150 dB, whereas, moving away from the blades,
coincides with the propeller surface. This is clearly shown in Fig. 11 for it decreases to about 130 dB (microphone B). This result is consistent
the case herein investigated. Thus, source pressure signals emitted by with literature experimental data (Felli et al., 2011). To be noted
the propeller reach the observer at the same time, meaning that, they that the quadrupole, non-linear, term significantly contributes at both
do not overlap over time. In this case, direct integration of kernels is locations. It exhibits an important presence of lower frequencies. Also,
allowed without the need to consider the time delays. For these reasons, comparatively, it contributes more to the overall noise, with respect
hereafter, we evaluate the acoustic pressure using the direct integration to linear terms, at microphone B which is farther than A from the
of the volume terms as described in Cianferra et al. (2019). We first
propeller. In other words, the rate at which the signal associated to
discuss the propagation of the noise in the near field and, successively,
the non-linear terms decreases with distance is slower than that of
in the far field.
linear terms. This important issue has been also discussed in Ianniello
(2016). There, the author investigated how the various terms of the
5.1. Near field results
FWH equation decay, for the archetypal problem of a rotating point
Two measurement points are selected on the propeller plane 𝑥 = 0, source and found that a simple rule based on the 1∕𝑟𝑛 decay law cannot
at distances 0.6 𝐷 (mic A) and 0.8 𝐷 (mic B) from the axis of rotation be applied.
respectively, thus just out of the tip of the blades. When computing the The spectrum of the signal, expressed in a logarithmic scale, is
FWH integral volume terms, we adopt the acoustic control domain A reported in Fig. 13 for microphones A and B respectively. We recall
of Fig. 1(b), in order to include the entire wake contribution and the that the Sound Spectrum Level is evaluated as 𝑆𝑃 𝐿 = 20 log10 (𝐴∕𝑝𝑟𝑒𝑓 )
tip vortex. dB, being A the amplitude of the signal and 𝑝𝑟𝑒𝑓 = 1 𝜇 Pa the reference
Fig. 12 shows the time signal of the acoustic pressure. The signature pressure for underwater measurements. As already observed in the time
of the linear terms is periodic and practically monochromatic (see the signals, the frequency analysis clearly shows the contribution of the

10
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 9. Phase-average non-dimensional pressure (first column), root mean square of pressure (second column), turbulent viscosity (third column). Section perpendicular to the
mean flow (𝑥-axis), at locations (from top panel to bottom panel) 𝑥∕𝐷 = −1, −3, −5.

quadrupole terms in the range of low frequencies. A main peak is term. To this aim, we show in Fig. 15 an instantaneous snapshot (at the
at about 15–20 Hz, in correspondence of the revolution frequency n plane 𝑧 = 0) of the scalar 𝜕 2 𝑇𝑖𝑗 ∕𝜕𝑥𝑖 𝜕𝑥𝑗 , related to the Lighthill tensor
and other peaks are in between n and nN. To be noted that in the of Eq. (12), which constitutes the instantaneous non-linear (in terms of
farther microphone B the maximum pressure amplitude of about 140 velocity field) source of the acoustic wave equation. The figure shows
dB is associated to the low frequency peak and, thus, to the non linear that the most energetic contribution to the Lighthill tensor and thus
contribution to the noise. to the non-linear source of noise may be the energetic helical vortex
We evaluate the noise at a third microphone C, located downstream that forms around the propeller shaft. This vortex structure appears
at (−1𝐷, 0.8𝐷, 0) in the wake region. In Fig. 14 the SPL is shown for even more significant than the tip vortex. Also, its spatial distribution,
this latter microphone. As for the previous cases, and even more, the specifically the distance between two successive vortices along the
non linear contribution to the noise is here predominant, in particular
mean flow direction, is consistent with the low frequency observed in
in the range of lower frequencies. At the revolution frequency n, the
the SPL. Our results are consistent with those of Di Mascio et al. (2014),
linear part contribution amounts to about 70 dB, whereas the non-
Felli et al. (2011) who reported that the wave number associated to the
linear contribution is a bit less than 140 dB. Also, the shape of the
shaft vortex is 1∕𝑁 times that associated to the tip vortex.
overall spectrum is completely different from that of the linear part;
The fluid dynamic pressure induced on the shaft generates noise as
specifically, the former exhibits larger contributions in the range of low
frequencies whereas the opposite is true for the latter. The peak on the well, however its amplitude is four orders of magnitude lower than the
main frequency 𝑛𝑁 reaches about 100 dB, while the peak at the lower pressure noise induced by the shaft vortex. To show it, we evaluated
frequency 15–20 Hz is a bit below 140 dB. The behavior depicted above linear terms by integrating on the shaft surface alone: the resulting
indicates that the noise source persists downstream mainly due to the pressure radiated from the shaft is given in Fig. 16; it is compared with
contribution of the wake, whereas the noise generated by the pressure the FWH pressure given by the quadrupole volume terms, evaluated by
radiated from the propeller, as known, decreases with distance. The integrating over the control domain B of Fig. 1(b). By considering a
comparison with previous Fig. 13(b) highlights this remark. narrow control domain B, it was possible to consider the shaft vortex
An interesting issue is to understand which turbulent coherent contribution alone. Note, in Fig. 16 the different reference system along
structure mainly contributes to the composition of the non-linear source the 𝑦-axis. The frequency captured by both the FWH integrals (volume

11
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 10. non-dimensional radial (top left panel) and axial (top left panel) velocity at blade wake. The plane is located at the blade wake, as shown in the bottom panel.

depend on the frequency 𝑓 since 𝜆 = 𝑐0 ∕𝑓 . Therefore, the most distant


microphone 𝐴2 can be considered as located in the acoustic far field,
at least in the range of high frequencies.
The energetic SPL observed in correspondence of microphone A
appears smoothed out once moving far from the propeller. In particular,
high frequencies associated to turbulence and small scale fluid motion,
are not persistent over long distances and characterize the near field
only. We remind that microphone A is located in proximity of the
blades, where pressure impulses, related to blade passage, are intense.
This is the reason of the high pressure amplitudes (about 150 dB)
recorded at microphone A. Comparing the two far-field probes A1 and
A2, a mean decay of about 40 dB is observed. At probe A2 the main
peak at 𝑛𝑁 Hz drops down to values below 0 dB and the signal exhibits
a low frequency band.
In order to place our results within the context of the general theory
of the far field noise propagation, we recall the law of attenuation of
spherical propagation, which rules the rate of decay of an impulsive
source. It is given by the formula
( )
𝑑2
𝐿𝑑2 = 𝐿𝑑1 − 20 log10 (19)
𝑑1
Fig. 11. Retarded surface 𝛴 (red) and propeller surface (black). Note that they are where 𝐿𝑑1 (𝐿𝑑2 ) is the pressure level (in dB) at distance 𝑑1 (𝑑2 ). Eq. (19)
practically overlapped. is valid for a point source, and if 𝑑2 = 10 𝑑1 the pressure level drop is
𝐿𝑑1 − 𝐿𝑑2 = 20 dB. To validate it, we carried out a test on a monopole
field, used as a simple source of noise.
and surface) corresponds to 15–20 Hz. Interestingly, this noisy and The result of the acoustic analogy applied to the monopole field for
energetic vortex structure is not present in full scale applications and, in two microphones placed at distances 10 m and 100 m from the source
some sense, it can be considered as an artifact generated by the ’pulling gives an amplitude decay, from 0.0055 to 0.05 Pascal, corresponding to
type’ architecture of the laboratory scale test. a drop of 20 dB, which is consistent with Eq. (19). However, according
to our results of Fig. 17, where we observe a mean decay of 40 dB (in
5.2. Far field decay the range of frequency 10–100 Hz) between two microphones A1 and
A2 located at distance 𝑑2 = 10 𝑑1 , it is unclear whether this rule can
In this subsection we analyze the propagation of noise in the far be applied to the case of complex noise sources, such as a flow field
field. Two microphones, 𝐴1 and 𝐴2, are located over the line 𝑥 = 𝑧 = 0, given by a propeller. Moreover, we observe that the decay does not
at distances 𝑦 = 8 𝐷 and 𝑦 = 80 𝐷, respectively from the axis of occur homogeneously with respect to the frequencies. In Fig. 18 the
symmetry. In Fig. 17, their signals are compared with that of previous difference between the sound pressure level of microphones A1 and A2
microphone A. To be precise, as acoustic far field it is considered is reported. We have separated the linear part (red dashed line) from
a microphone placed at distance more than 2 wavelengths 𝜆, which the nonlinear one (black solid line). As regards the linear part, a decay

12
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 12. Time signal of the acoustic pressure of FWH non-linear terms (black dashed line) and FWH linear terms (red solid line): (a) microphone A; (b) microphone B.

Fig. 13. Sound spectrum level of FWH non-linear terms (black solid line), FWH linear terms (red dashed line) and the entire FWH resulting signal (blue dash–dot line): (a)
microphone A; (b) microphone B. Control domain A of Fig. 1(b).

of about 70 dB from the microphone A1 to A2 occurs at the fundamental


frequency of 75 Hz; Other peaks are relevant, which indicate a strong
decay of the linear part, at frequencies 2𝑛 = 30 Hz and 3𝑛 = 45 Hz. The
decay of the non-linear part is also different from the analytical ideal
value, being of the order of 70 Hz at the frequency 𝑛 and maintaining
high values in a broad part of the spectrum. This behavior has to
be attributed to the complexity and non linearity of the noise source
which deserves additional studies. It must be said that the analysis of
acoustic decay between microphone A1 and A2 does not correspond to
a true analysis between two microphones in the far field. Therefore,
interaction phenomena of acoustic waves are still present, instead of
pure propagation of a signal. However, we note that to consider the far
field for low frequencies it would be necessary to measure the noise at
very great distances. For example, an acoustic wave of frequency 5 Hz
would reach the far field, in our case, at 2400 diameters away. An other
aspect closely related to the way the signal decays with distance is the
directivity.
The directivity is here evaluated as the pressure root mean square
given by solving the FWH equation at different measurement points
equidistant from the source. We consider 100 measurement points, Fig. 14. Sound spectrum level of FWH volume terms (black solid line), FWH linear
terms (red dashed line) and the entire FWH resulting signal (blue dash–dot line) for
arranged on circumferences of radius 10 𝐷 over four different planes
microphone C.
𝑥 = 0, 𝑥 = −2, 𝑦 = 0 and 𝑦 = −2 (sketched in Fig. 19). The results
are depicted in Figs. 20 and 21. The distribution of pressure level
over the planes 𝑥 = 0 and 𝑥 = −2 (Fig. 20 top panels) is consistent
with theory and literature results, since the radial symmetry of the directivity of linear terms on the planes 𝑦 = 0 and 𝑦 = −2 (Fig. 20
problem is recovered in the computation of the directivity. On the other bottom panels). This behavior has been already observed in previous
hand, the dipole-type acoustic field is observable when considering the research (see, among the others, (Lidtke, 2017)).

13
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 15. Snapshot of the scalar quantity 𝜕 2 𝑇𝑖𝑗 ∕𝜕𝑥𝑖 𝜕𝑥𝑗 at the symmetry plane 𝑧 = 0.

Fig. 18. Decibel decay in the far field. Difference between microphone A1 and A2.
FWH volume terms (black solid line) and FWH linear terms (red dashed line). Mean
values of FWH volume terms (dots) and FWH linear terms (crosses).

Fig. 16. Microphone C, acoustic pressure over time of FWH volume terms (solid line)
and FWH linear terms (dashed line). Control domain B of Fig. 1(b).

Fig. 19. Sketch of the points selected for the computation of the directivity, lying on
the four planes 𝑥 = 0, 𝑥 = −2, 𝑦 = 0 and 𝑦 = −2.

that the nonlinear contribution is about 3 orders of magnitude higher


than linear one (left-bottom panels of Figs. 20 and 21).

6. Conclusions

In the present paper, we study the noise radiated from a marine


propeller in open sea condition using the acoustic analogy. The fluid
dynamic field is simulated using large eddy simulation with wall-layer
models whereas the acoustic field is obtained using the Ffowcs Williams
Fig. 17. Sound spectrum level of FWH volume terms (black solid line), FWH linear and Hawkings equation.
terms (solid-crossed red line) and the entire FWH resulting signal (blue dash-dotted
line): Top lines, mic. A; Middle lines, mic. A1; bottom lines, mic. A2.
As regards the fluid dynamic field, WLES gives accurate results, both
for the integral quantities (thrust and torque) and for the turbulent
field developing downward in the wake. The accuracy of the results
was assessed through comparison with available experimental data and
The results of the FWH non linear part appear even more interesting. high-resolution LES carried out by other authors.
Although the axial symmetry exhibited over the planes 𝑥 = 0 and The integral formulation of the FWH equation is first formulated in
𝑥 = −2 still holds (see Fig. 21 top panels), very interesting patterns the advective form and then solved using a direct volume integration
are evidenced over the planes 𝑦 = 0 and 𝑦 = −2, depicted in Fig. 21, for the non-linear quadrupole terms. In the present case, this is possible
bottom panels. because the assumption of compact source has been demonstrated to be
valid through the retarded surface method.
Considering the fact that the negative cos(𝑎) is in correspondence
The solution of the complete FWH equation allows to evaluate sepa-
of the wake, one may conclude that the propeller wake gives rise rately the linear and non linear contributions to the noise propagation.
to consistent pressure fluctuations, propagating in the downstream In agreement with literature studies, the linear part produces a
direction and contributing to the far field noise. Finally, to be noted narrow-banded spectrum with a main peak at the blade frequency. The

14
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

Fig. 20. Directivity of FWH linear terms. Radius 𝑟 = 10 𝐷.

Fig. 21. Directivity FWH non linear terms. Radius 𝑟 = 10 𝐷.

non-linear contribution appears of great interest. It is mainly associated except for the peak on the blade frequency generated by the linear
with the shaft vortex, that persists for a long distance at the rear of terms.
the propeller; it is characterized by high level of vorticity and strongly The analysis of the propagation of the noise in the far field, shows
characterizes the acoustic field. Therefore, the acoustic signature given that ’thumb-like’ rules based on simple analytical formulations may be
by the FWH equation is found to consist mainly of quadrupole noise, very inaccurate. Specifically, in presence of a complex noise source

15
M. Cianferra et al. Ocean Engineering 191 (2019) 106474

like a ship propeller, the decay of the noise in the far field is not Di Felice, F., Felli, M., Liefvendahl, M., Svennberg, U., 2009. Numerical and experi-
constant over the frequency spectrum, but appears much stronger at mental analysis of the wake behavior of a generic submarine propeller. In: First
International Symposium on Marine Propulsors, Trondheim, Norway.
fundamental frequencies. The linear part exhibits a strong decay at
Di Mascio, A., Muscari, R., Dubbioso, G., 2014. On the wake dynamics of a propeller
the fundamental frequency 𝑛𝑁 as well as at frequencies multiple of operating in drift. J. Fluid Mech. 754, 263–307.
the rotational one 𝑛. The non-linear part decays faster in the range of Fakhari, A., Cintolesi, C., Petronio, A., Roman, F., Armenio, V., 2018. Numerical
low frequencies. This has to be attributed to the fact that the main simulation of hot smoke plumes from funnels, technology and science for the ships
turbulent structures contributing to the far field noise are the well- of the future. In: 19th International Conference on Ship & Maritime Research. IOS
Press, pp. 238–245.
known tip vortex and the shaft vortex (or hub vortex, depending on Felli, M., Camussi, R., Di Felice, F., 2011. Mechanisms of evolution of the propeller
the geometric configuration investigated). The latter appears more wake in the transition and far fields. J. Fluid Mech. 1–49.
energetic and noisy than the former. Felli, M., Falchi, M., 2018. Propeller wake evolution mechanisms in oblique flow
The directivity points out the dipole-type acoustic field, gener- conditions. J. Fluid Mech. 845, 520–559.
Felli, M., Grizzi, S., Falchi, M., 2014. A novel approach for the isolation of the sound
ated by the FWH linear contribution, which was already observed in
and pseudo-sound contributions from near-field pressure fluctuation measurements:
previous studies. Also, it exhibits a strong asymmetry due to the non- Analysis of the hydroacoustic and hydrodynamic perturbation in a propeller-rudder
linear contribution; this is expected although it is noteworthy that the system. Exp. Fluids 55 (1).
contribution of the linear part is much smaller than the non-linear one, Ffowcs-Williams, J.E., Hawkings, D.L., 1969. Sound generation by turbulence and
especially in the far field and in the direction of the wake. surfaces in arbitrary motion. Philos. Trans. Roy. Soc. 264 (A1151), 321–342.
Ianniello, S., 2016. The Ffowcs Williams–Hawkings equation for hydroacoustic analysis
Finally, it is advisable to dwell on the particular configuration of of rotating blades. Part 1. The rotpole. J. Fluid Mech. 797, 345–388.
this benchmark case, which has the shaft flipped downstream. The Ianniello, S., Muscari, R., Di Mascio, A., 2013. Ship underwater noise assessment by
presence of the shaft in the middle of the wake affect the dynamics of the acoustic analogy. Part I: Nonlinear analysis of a marine propeller in a uniform
the (otherwise called) hub vortex, observed in previous experimental flow. J. Mar. Sci. Tech. 18 (4), 547–570.
Ianniello, S., Muscari, R., Di Mascio, A., 2014. Ship underwater noise assessment by
and numerical studies. The different evolution of the downstream vor-
the acoustic analogy. Part II: Hydroacoustic analysis of a ship scaled model. J. Mar.
tex also considerably changes the acoustic signature. Thus, we would Sci. Tech. 19 (1), 52–74.
conclude that a configuration of this type can be misleading for realistic Jessup, S.D., 1989. An Experimental Investigation of Viscous Aspects. of Propeller Blade
noise measurements. Flow (Ph.D. thesis). School of Engineering and Architecture, Catholic University of
America.
Keller, J., Kumar, P., Mahesh, K., 2018. Examination of propeller sound production
Acknowledgment
using large eddy simulation. Phys. Rev. Fluids 3 (6), 064601.
Kerwin, J.E., 1986. Marine propellers. Ann. Rev. Fluid Mech. 18, 367–403.
This work was supported by Region FVG-POR FESR 2014–2020, Kumar, P., Mahesh, K., 2017. Large eddy simulation of propeller wake instabilities. J.
Fondo Europeo di Sviluppo Regionale, Project PRELICA ’’Metodologie Fluid Mech. 814, 361–396.
Lidtke, A.K., 2017. Predicting Radiated Noise of Marine Propellers using Acoustic Analo-
avanzate per la progettazione idroacustica dell’elica navale‘‘.
gies and Hybrid Eulerian-Lagrangian Cavitation Models (Ph.D. thesis). University
Southampton.
References Meneveau, C., Lund, T.S., Cabot, W.H., 1996. A Lagrangian dynamic subgrid-scale
model of turbulence. J. Fluid Mech. 319, 353–385.
Baek, D.G., Yoon, H.S., Jung, J.H., Kim, K.S., Paik, B.G., 2015. Effects of the advance Najafi-Yazdi, A., Bres, G.A., Mongeau, L., 2011. An acoustic analogy formulation for
ratio on the evolution of a propeller wake. Comput. Fluids 118, 32–43. moving sources in uniformly moving media. Proc. R. Soc. Lond. Ser. A Math. Phys.
Balaras, E., Schroeder, S., Posa, A., 2015. Large-eddy simulations of submarine Eng. Sci. 467, 144–165.
propellers. J. Ship Res. 59 (4), 227–237. Okulov, V.L., Sørensen, J.N., 2007. Stability of helical tip vortices in rotor far wake.
Calcagno, G., Di Felice, F., Felli, M., Pereira, F., 2005. A stereo-PIV investigation of a J. Fluid Mech. 576, 1–25.
propellers wake behind a ship model in a large free-surface tunnel. Mar. Technol. Piomelli, U., Balaras, E., 2002. Wall layer models for large eddy simulations. Ann. Rev.
Soc. J. 39 (2), 94–102. Fluid Mech. 34, 349–374.
Chase, N., Carrica, P.M., 2013. Submarine propeller computations and application to Posa, A., Broglia, R., Felli, M., Falchi, M., Balaras, E., 2019. Characterization of the
self-propulsion of DARPA suboff. Ocean Eng. 60, 68–80. wake of a submarine propeller via large-Eddy simulation. Comput. Fluids 184,
Chesnakas, C., Jessup, S., 1998. Experimental characterization of propeller tip flow. In: 138–152.
Proc. 22nd Symposium on Naval Hydrodynamics, Washington. Radhakrishnan, S., Piomelli, U., 2008. Large-eddy simulation of oscillating boundary
Cianferra, M., Armenio, V., Ianniello, S., 2018a. Hydroacoustic noise from different layers: Model comparison and validation. J. Geophys. Res. 113, C02022.
geometries. Int. J. Heat Fluid Flow 70, 348–362. Rako, N., Fortuna, C.M., Holcer, D., Mackelworth, P., Nimak-Wood, M., Pleslic’, G.,
Cianferra, M., Ianniello, S., Armenio, V., 2019. Assessment of methodologies for Sebastianutto, L., Vilibic’, I., Wiemann, A., Picciulin, M., 2013. Leisure boating
the solution of the Ffowcs Williams and Hawkings equation using large-eddy noise as a trigger for the displacement of the bottlenose dolphins of the CresLovsinj
simulations of incompressible single-phase flow around a finite-size square cylinder. archipelago (northern Adriatic Sea, Croatia). Mar. Pollut. Bull. 68, 77–84.
J. Sound Vib. (in press). Slabbekoorn, H., Bouton, N., van Opzeeland, I., Coers, A., ten Cate, C., Popper, A.N.,
Cianferra, M., Petronio, A., Armenio, V., 2018b. Hydrodynamic noise from a propeller 2010. A noisy spring: the impact of globally rising underwater sound levels on fish.
in open sea condition. In: 19th International Conference on Ship & Maritime Trends Ecol. Evol. 25 (7), 419–427.
Research, Trieste, Italy. Stocca, V., Armenio, V., Sreenivasan, K.R., 2011. Improved wall-layer model for
Cintolesi, C., Petronio, A., Armenio, V., 2015. Large eddy simulation of turbulent forced-convection environmental LES. In: ERCOFTAC Series, vol. 15.
buoyant flow in a confined cavity with conjugate heat transfer. Phys. Fluids 27 Wang, M., 1999. LES With wall models for trailing-edge aeroacoustics. In: Center
(9), 95–109. for Turbulence Research, Annual Research Briefs. Res. Stanford Univ. Calif., pp.
Cintolesi, C., Petronio, A., Armenio, V., 2016. Large-eddy simulation of thin film 355–364.
evaporation and condensation from a hot plate in enclosure: first order statistics. Wang, M., 2000. Dynamic wall modeling for LES of complex turbulent flows. Cent.
Int. J. Heat Mass Transf. 101, 1123–1137. Turbul. Res. Annu. Res. Briefs 24, 1–250.

16

You might also like