Ocean Engineering 211 (2020) 107633

Contents lists available at ScienceDirect

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

Analysis of the blade boundary-layer flow of a marine propeller using a RANS

solver
João Baltazar a ,∗, Daniela Melo a , Douwe Rijpkema b
a
Marine Environment and Technology Center, Instituto Superior Técnico, University of Lisbon, Av. Rovisco Pais 1, Lisbon, 1049-001, Portugal
b
Maritime Research Institute Netherlands, 2 Haagsteeg, 6708 PM Wageningen, The Netherlands

ARTICLE INFO ABSTRACT

Keywords: In this paper a comparison between RANS simulations carried out with the 𝑘 − 𝜔 SST turbulence model and
Marine propeller ̃ 𝜃 transition model, and experimental measurements for marine propeller P4119 is made. The experiments
𝛾 − R𝑒 𝑡
Blade boundary-layer flow were conducted at the David Taylor Model Basin and comprehended three-dimensional velocity components
RANS solver
measurements of the blade boundary-layer using a LDV system in uniform conditions. The present work
Transition modeling
includes an estimation of the numerical errors that occur in the simulations, analysis of the propeller blade flow,
chordwise and radial components of the boundary-layer velocity profiles and boundary-layer characteristics.
From this comparison and depending on the selected inlet turbulence quantities, we conclude that the transition
model is able to predict the extent of laminar and turbulent regions observed in the experiments. Moreover,
the predicted velocity profiles show good agreement with the experimental measurements and verified the
location of transition.

1. Introduction (2012), Hasuike et al. (2017), and the correlation-based transition

RANS (Reynolds-Averaged Navier–Stokes) solvers are becoming a
widely used tool for the analysis of marine propellers both in model-
and full-scale conditions. The first applications of a RANS method to in-
vestigate the scale-effects on the propeller performance characteristics
are attributed to Uto (1993) and Stanier (1998). Similar studies were
carried out by Funeno (2002), Krasilnikov et al. (2009) and Sánchez-
Caja et al. (2014) on the prediction of the viscous flow at different
Reynolds number regimes for open propellers. Furthermore, Abdel-
Maksoud and Heinke (2002), Rijpkema and Vaz (2011) and Bulten and
Nijland (2011), among others, analyzed the scale-effects on ducted pro-
pellers. In the aforementioned literature, the used turbulence models
were developed for the simulation of a wide range of fully turbulent
engineering flows, and they are not able to accurately predict transition
at typical model-scale Reynolds numbers. In Eça and Hoekstra (2008),
several eddy-viscosity turbulence models were used in the prediction of
the flow over a flat plate at different Reynolds numbers. The compari-
son showed the inability of the different turbulent models to simulate
transition from laminar to turbulent flow.
The recent developments in transitional sensitive turbulence models
gave the possibility to solve the viscous flow at different Reynolds num-
bers regimes. The most popular transition sensitive turbulence models
are the model proposed by Walters and Cokljat (2008) based on the
concept of laminar kinetic energy, see for instance Wang and Walters
model proposed by Langtry (2006) based on the concepts of inter-
mittency 𝛾 and local transition onset momentum thickness Reynolds
number Re ̃ 𝜃 . Since the full release to the scientific community (Langtry
𝑡
and Menter, 2009), the 𝛾−Re ̃ 𝜃 is becoming a very popular model for the
𝑡
̃ 𝜃 tran-
simulation of transitional flows. Several applications of the 𝛾 − Re 𝑡
sition model to marine propellers may be found in Bhattacharyya et al.
(2016a), Shin and Andersen (2017), Moran-Guerrero et al. (2018) and
Yao and Zhang (2018). In these studies, a significant improvement is
observed in the comparison between RANS simulations and model tests
performed without forced transition. A better agreement is obtained
for the propeller forces, and also in the occurrence of flow transition
and its location from the comparison with paint-tests. Therefore, the
𝛾−Rẽ 𝜃 transition model opens the opportunity to improve the modeling
𝑡
of transitional flows using RANS solvers.
Traditionally, the prediction of scale-effects is based on simple
extrapolation procedures from model-scale experiments carried out in
a towing tank or cavitation tunnel. At present, the most common
procedure is the 1978 ITTC performance prediction method (ITTC,
1978, 2017). Alternatively, the combination of increased computa-
tional power and the recent developments in turbulence modeling
makes the RANS solver a viable tool for the prediction of the scale-
effects. Scaling approaches based on the RANS simulations have been

∗ Corresponding author.
E-mail address: joao.baltazar@tecnico.ulisboa.pt (J. Baltazar).

https://doi.org/10.1016/j.oceaneng.2020.107633
Received 4 February 2020; Received in revised form 6 May 2020; Accepted 8 June 2020
0029-8018/© 2020 Elsevier Ltd. All rights reserved.
J. Baltazar et al.
Fig. 1. Schematic representation of the boundary-layer regimes on the suction side of
a propeller blade for model-scale Reynolds numbers.
Source: Taken from Kuiper (1981).
proposed by Müller et al. (2009) and Bhattacharyya et al. (2016b) for
open and ducted propellers, respectively.
The flow over the propeller at model-scale Reynolds numbers is
mostly in the critical Reynolds number regime, where laminar, tran-
sition and turbulent flow regimes may occur simultaneously. Visual-
ization of the flow at these Reynolds numbers, for example with the
help of paint tests, typically shows regions of laminar and turbulent
flow on the blade, as for example, discussed in Kuiper (1981). The
character of the boundary-layer on the suction side of a propeller blade
at model-scale Reynolds number is illustrated schematically in Fig. 1.
For the upper radii (A–B), a short laminar separation bubble due to
a strong adverse pressure gradient is present near the leading-edge,
making the boundary-layer turbulent over the rest of the chord. The
separation radius at which the short laminar separation bubble occurs
is found at B–C. The region (C–D) shows a laminar flow region with
transition to turbulent flow downstream. The chordwise position of the
transition region depends on the Reynolds number, but is generally
located at some distance from the leading-edge. At the lower radii (D–
E), a laminar separation region can be observed due to the very low
sectional Reynolds number in combination with thicker root sections.
Therefore, for an accurate prediction of the performance character-
istics at both model- and full-scales with a RANS solver, the selected
model for turbulence closure must have the ability to simulate different
flow regimes including transition, since a different boundary-layer
structure of the flow over the blades arises from the change in Reynolds
number (Fig. 1). The 𝑘 − 𝜔 SST (Shear Stress Transport) turbulence
model (Menter et al., 2003) is currently one of the most widely used
turbulence models for marine propeller applications. However, fully
turbulent boundary-layers are assumed all over the blade surface, and
consequently do not obtain accurate predictions that can be compared
to experimental results. Rijpkema et al. (2015) analyzed model-scale
propellers using two different eddy-viscosity turbulence models. The
comparison with paint-tests showed that the predicted limiting stream-
lines are more circumferentially oriented for all radii, which indicates
an overestimation of the turbulent flow region. Therefore, the 𝛾 − Rẽ 𝜃
𝑡
transition model proposed by Langtry and Menter (2009) may be seen
as an alternative to improve the propeller performance prediction at
model-scale.
In a previous study, the 𝑘 − 𝜔 SST turbulence model and the 𝛾 − Rẽ 𝜃
𝑡
transition model were combined with a RANS solver for two marine
propellers, for which paint-tests have been conducted and experimental
open-water data is available (Baltazar et al., 2018). The analysis of the
2
Ocean Engineering 211 (2020) 107633
numerical simulations focused on a qualitative comparison between the
limiting streamlines and paint-tests observations, and propeller force
prediction in open-water conditions. From this study, an improvement
in the blade boundary-layer flow is obtained with the transition model,
whereas the simulations with the turbulence model predict a too large
turbulent region on the blade.
Alternatively, the experimental study of the laminar/turbulent flow
in the vicinity of a rotating propeller blade conducted by Jessup (1989)
may be used for a better quantitative comparison and understanding of
the blade boundary-layer flow. The experiments were carried out in the
David Taylor Model Basin (DTMB) 24 in. water tunnel about marine
propeller P4119, where two blades were applied with leading-edge
roughness and the remaining was smooth. The experiments compre-
hended three-dimensional velocity component measurements of the
blade boundary-layer and wake using a LDV system in uniform inflow
conditions.
From these experimental measurements, a comparison with the
numerical results obtained from a RANS solver using the 𝑘 − 𝜔 SST
turbulence model and 𝛾 − Re ̃ 𝜃 transition model is presented in this
𝑡
paper. The present work includes an estimation of the numerical er-
rors involved in the simulations, analysis of the propeller blade flow,
chordwise and radial components of the boundary-layer velocities and
boundary-layer characteristics. From this study and depending on the
̃ 𝜃 transition model is able
selected inlet turbulence quantities, the 𝛾 − Re
𝑡
to predict the extent of laminar and turbulent regions observed in the
experiments.
This paper is organized as follows: the numerical method including
the turbulence and transition models is presented in Section 2; the
propeller geometry, grid generation scheme, computational domain
and boundary conditions are described in Section 3; the comparison
between the numerical results and the experimental measurements is
shown in Section 4; in Section 5 the main conclusions are drawn.
2. Numerical method
2.1. RANS equations
The RANS equations are solved using a so-called absolute for-
mulation. This means that the flow velocity, 𝑈𝑖 with 𝑖 = 1, 2, 3 or
(𝑈𝑋 , 𝑈𝑌 , 𝑈𝑍 ), is defined in the absolute or inertial earth-fixed frame of
reference, with the equations being solved in the propeller-fixed frame
of reference 𝑥𝑖 , which is rotating with angular velocity 𝛺. This allows to
perform steady simulations for open-water conditions. For the absolute
velocity formulation, the RANS equations for a steadily rotating frame
and using the Boussinesq eddy-viscosity hypothesis may be written in
the following form:
𝜕𝑈𝑖
= 0,
𝜕𝑥𝑖
[ ( )] (1)
𝜕(𝑉𝑗 𝑈𝑖 ) 1 𝜕𝑃 𝜕 𝜕𝑈𝑖 𝜕𝑈𝑗
+ 𝜀𝑖𝑗𝑘 𝛺𝑗 𝑈𝑘 = − + (𝜈 + 𝜈𝑡 ) + ,
𝜕𝑥𝑗 𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑖
where 𝑉𝑖 represents the 𝑖th Cartesian component of the velocity vec-
tor defined in the propeller-fixed frame of reference, 𝑃 the modified
pressure defined as 𝑃 = 𝑝 + 2∕3𝜌𝑘, 𝑝 the static pressure, 𝑘 the
turbulence kinetic energy, 𝜌 the fluid density, 𝜈 and 𝜈𝑡 the kinematic
and turbulent viscosities, respectively, and 𝜀 the Levi-Civita symbol.
In this formulation, the momentum equations contain only one source
term 𝜀𝑖𝑗𝑘 𝛺𝑗 𝑈𝑘 .
2.2. Turbulence and transition models
̃ 𝜃
In the present work, the 𝑘 − 𝜔 SST turbulence model and the 𝛾 − Re 𝑡
transition model are chosen for turbulence closure.
The 𝑘 − 𝜔 SST turbulence model is a widely used and robust
two-equation turbulence model. The variables of the model are the
turbulence kinetic energy 𝑘 and the turbulence dissipation rate 𝜔. This
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

Fig. 2. Overview of the grid with 37.6 million cells.

model was originally introduced in 1994 by Menter to deal with the Table 1
strong sensitivity to the free-stream conditions of the original 𝑘 − 𝜔 Overview of the grid sizes and number of cell faces on a single blade. Maximum 𝑦+
values refer to the design condition.
turbulence model and to improve the prediction of the boundary-layer
Grid Volume Blade 𝑦+max
flow with adverse pressure gradients (Menter, 1994). In the present
work, the 2003 version of the 𝑘 − 𝜔 SST model is used (Menter et al., G1 37.6M 73.9k 0.20
G2 21.0M 42.3k 0.24
2003). The transport equations to calculate the modeled turbulence
G3 9.9M 25.6k 0.31
kinetic energy and turbulence dissipation rate are: G4 6.1M 18.6k 0.39
[ ]
𝜕(𝑉𝑗 𝑘) 𝜕 ( ) 𝜕𝑘 G5 1.9M 8.6k 0.51
=𝑃𝑘 − 𝐷𝑘 + 𝜈 + 𝜎𝑘 𝜈𝑡 , G6 0.9M 4.9k 0.66
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗
[ ] (2)
𝜕(𝑉𝑗 𝜔) 𝜕 ( ) 𝜕𝜔 ( ) 𝜎 𝜕𝑘 𝜕𝜔
=𝑃𝜔 − 𝐷𝜔 + 𝜈 + 𝜎𝜔 𝜈𝑡 + 2 1 − 𝐹1 𝜔2 ,
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 𝜔 𝜕𝑥𝑗 𝜕𝑥𝑗
which only includes a production term 𝑃𝜃𝑡 . The terms 𝑃𝛾 , 𝐸𝛾 and 𝑃𝜃𝑡
with the turbulent viscosity defined as
and the constants 𝜎𝑓 and 𝜎𝜃𝑡 are given in Langtry and Menter (2009).
𝑎1 𝑘
𝜈𝑡 = ( ), (3) The coupling between the 𝑘 − 𝜔 SST turbulence model and the
max 𝑎1 𝜔, 𝑆𝐹2 ̃ 𝜃 transition model is made by an effective intermittency 𝛾eff
𝛾 − Re 𝑡
where 𝑃𝑘 and 𝐷𝑘 are the production and dissipation terms of the which is used to control the production and dissipation terms in the
turbulence kinetic energy, respectively, 𝑃𝜔 and 𝐷𝜔 the production and 𝑘 transport equation, Eq. (2). The definition of the turbulent viscosity
dissipation terms of the turbulence dissipation rate, respectively, 𝑆 is of the 𝑘 − 𝜔 SST turbulence model remains unchanged. This model is
the strain-rate magnitude, 𝐹1 and 𝐹2 auxiliary functions, and 𝑎1 , 𝜎𝑘 , 𝜎𝜔 ̃ 𝜃 −SST model, because it makes use
also sometimes known as the 𝛾 − Re
and 𝜎𝜔2 are constants of the model. The complete model and constant 𝑡
̃ 𝜃 , in addition to the 𝑘 and 𝜔 equations of
of the equations for 𝛾 and Re
values are given in Menter et al. (2003). 𝑡

The 𝛾 − Rẽ 𝜃 transition model proposed by Langtry and Menter the SST turbulence model (Menter et al., 2003). Therefore, this model
𝑡
(2009) is coupled with the 𝑘 − 𝜔 SST turbulence model of Menter et al. corresponds to a four-equation transition SST turbulence model.
(2003) for the simulation of the flow around the marine propeller at
model-scale including transition. This transition model contains two
transport equations and accounts for transition due to free-stream tur- 2.3. Flow solver
bulence intensity, pressure gradients and separation. One is a transport
equation for intermittency 𝛾:
[( ) ] The RANS equations are solved using the ReFRESCO code, devel-
𝜕(𝑉𝑗 𝛾) 𝜕 𝜈 𝜕𝛾 oped within a cooperation led by MARIN for hydrodynamic applica-
= 𝑃𝛾 − 𝐸𝛾 + 𝜈+ 𝑡 , (4)
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜎𝑓 𝜕𝑥𝑗 tions (ReFRESCO, 2019). A finite-volume technique with cell-centered
where 𝑃𝛾 and 𝐸𝛾 are the production and relaminarisation terms, respec- collocated variables is used for solving the equations. A pressure-
tively. The blending between the laminar and turbulent flow regimes correction equation based on the SIMPLE algorithm is used to ensure
is made by the intermittency, which changes between zero and one, mass conservation. A segregated approach is adopted here for the
respectively. In the present formulation, the intermittency is also set solution of all transport equations. The implementation is face-based,
equal to one in the freestream instead of a small value as in the original which permits grids with elements consisting of an arbitrary number of
model (Menter et al., 2002). The second one is a transport equation for
̃ 𝜃: faces. For the convective flux terms, a second-order scheme (QUICK) is
the local transition onset momentum thickness Reynolds number Re 𝑡
[ ] applied for the momentum equations, and a first-order upwind scheme
̃ 𝜃)
𝜕(𝑉𝑗 Re ( ) 𝜕 ̃
Re is applied for the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re ̃ 𝜃 transition
𝑡 𝜕 𝜃𝑡
𝑡
= 𝑃𝜃𝑡 + 𝜎𝜃𝑡 𝜈 + 𝜈𝑡 , (5)
𝜕𝑥𝑗 𝜕𝑥𝑗 𝜕𝑥𝑗 model. In this study, ReFRESCO version 2.3 is used.

3
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

3. Propeller geometry and numerical set-up

The marine propeller P4119 is considered in the present study. The

propeller has three blades, a diameter of 0.305 m and an expanded-
area ratio of 0.5. The pitch-diameter ratio and chord-length at 0.7 of
the propeller radius are 1.084 and 0.141 m, respectively. The propeller
has no skew or rake.
For this propeller six nearly-geometrically similar multi-block struc-
tured grids are generated using the commercial grid generation soft-
ware GridPro (GridPro, 2019). The grids for the entire propeller range
from 900 thousand to 37.6 million cells. In Table 1 the number of cells
in the volume and on a single blade, and the maximum dimensionless
wall distance of the first cell height 𝑦+ are listed for each grid. A fine
boundary-layer resolution is obtained, since the maximum 𝑦+ is lower
than one for all grids. Therefore, the boundary-layer is fully resolved
and no wall functions are used. An overview of the grid with 37.6
million cells is presented in Fig. 2.
A cylindrical domain is considered, where the inlet, the outlet and
the outer boundary are located ten propeller radii from the domain
origin. A schematic overview of the computational domain and its
boundaries is presented in Fig. 3, where 𝑅 denotes the propeller ra-
dius. The prescribed boundary conditions in the present work are the
Fig. 3. Schematic overview of the computational domain.
following:

• At the inlet, the uniform velocity and the turbulence level depend-
ing on the turbulence model are prescribed, while the pressure is

̃ 𝜃 transition model (bottom).

Fig. 4. 𝐿∞ (left) and 𝐿2 (right) iterative convergence for grid G1 using 𝑘 − 𝜔 SST turbulence model (top) and 𝛾 − Re 𝑡

4
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

̃ 𝜃 transition model (right).

Fig. 5. Iterative convergence of 𝐾𝑇 and 10𝐾𝑄 for grid G1 using 𝑘 − 𝜔 SST turbulence model (left) and 𝛾 − Re 𝑡

Fig. 6. Convergence of 𝐾𝑇 (left) and 10𝐾𝑄 (right) with grid refinement ratio ℎ𝑖 ∕ℎ1 .

Table 2 𝜈𝑡 ∕𝜈 = 500 at the inlet boundary are specified. The choice of these inlet
Decay of turbulence quantities from inlet (𝑥∕𝑅 = 10) to the vicinity of the propeller
turbulence quantities is based on a previous study (Baltazar et al.,
(𝑥∕𝑅 = 1). Turbulence quantities at 𝑥∕𝑅 = 1 refers to section 𝑟∕𝑅 = 0.7.
2018), where the 𝛾 − Re ̃ 𝜃 transition model has been used for the flow
Model 𝑥∕𝑅 = 10 𝑥∕𝑅 = 1 𝑡
simulation on two marine propellers at model-scale and the results
𝑇𝑢 𝜈𝑡 ∕𝜈 𝑇𝑢 𝜈𝑡 ∕𝜈
have been compared with paint-tests. Table 2 presents the turbulence
̃ 𝜃
𝛾 − Re 1.20% 500.0 1.14% 151.5
̃ 𝜃
𝑡 intensity and eddy-viscosity ratio at the inlet boundary (𝑥∕𝑅 = 10) and
𝛾 − Re 1.50% 500.0 1.37% 222.5
𝑡
𝑘 − 𝜔 SST 1.00% 1.0 0.13% 0.74 plane 𝑥∕𝑅 = 1 upstream of the propeller at the radial section 𝑟∕𝑅 = 0.7.
The results show that the large value selected for the eddy-viscosity
ratio (𝜈𝑡 ∕𝜈 = 500) is crucial in the control of the turbulence decay from
the inlet boundary to the propeller plane (Spalart and Rumsey, 2007),
extrapolated from the interior assuming zero normal derivative and therefore in the prediction of transition occurrence and location.
(Neumann boundary condition). For the 𝑘 − 𝜔 SST turbulence model, standard values, i.e. 𝑇 𝑢 = 1.0%
• At the outlet, an outflow condition of zero normal derivative is and 𝜈𝑡 ∕𝜈 = 1, are assumed at the inlet boundary. In this case, and as
prescribed for all flow variables.
shown in Baltazar et al. (2018), the influence of the inlet turbulence
• At the outer boundary, a constant pressure is set, while a Neu-
quantities on the propeller simulations is negligible.
mann boundary condition is used for all other variables.
• For the propeller blades and hub, no-slip and impermeability
conditions are applied at the surface and Neumann boundary 4. Results
conditions are used for the pressure 𝑃 , intermittency 𝛾 and local
transition onset momentum thickness Reynolds number Re ̃ 𝜃 . In 4.1. General
𝑡
addition, the turbulence kinetic energy 𝑘 is set equal to zero and
𝜔 is specified at the near-wall cell center (Menter, 1994).
Results are presented for the marine propeller P4119 in open-water
In this study, the turbulence conditions. The propeller open-water characteristics are expressed by
√ level is specified at the inlet from the
turbulence intensity 𝑇 𝑢 = 100 2𝑘∕(3𝑉𝐴2 ) and eddy-viscosity ratio 𝜈𝑡 ∕𝜈, the advance coefficient 𝐽 , thrust and torque coefficients 𝐾𝑇 , 𝐾𝑄 , and
where 𝑉𝐴 is the incoming flow velocity and equal to the propeller the open-water efficiency 𝜂0 as follows:
advance speed. Due to the strong dependence of the 𝛾 − Re ̃ 𝜃 transition 𝑉𝐴 𝑇 𝑄 𝐽 𝐾𝑇
𝑡
model on the inlet turbulence quantities, 𝑇 𝑢 = 1.2% and 1.5% and 𝐽= , 𝐾𝑇 = , 𝐾𝑄 = , 𝜂0 = , (6)
𝑛𝐷 𝜌𝑛2 𝐷4 𝜌𝑛2 𝐷5 2𝜋 𝐾𝑄

5
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

Fig. 7. Convergence of 𝐶𝑝 (left) and 𝐶𝑓 (right) at 𝑠∕𝑐 = 0.2 (top) and 0.7 (bottom) on the blade suction side with grid refinement ratio ℎ𝑖 ∕ℎ1 .

̃ 𝜃 transition model (right).

Fig. 8. Influence of grid refinement on 𝑉𝑠 on the suction side at 𝑠∕𝑐 = 0.5 and 𝑟∕𝑅 = 0.7. 𝑘 − 𝜔 SST turbulence model (left) and 𝛾 − Re 𝑡

where 𝑛 = 𝛺∕(2𝜋) is the rotation rate in rps, 𝐷 the propeller diameter, 𝑇 is a reference velocity defined as the undisturbed onset velocity at
the propeller thrust and 𝑄 the propeller torque. Other useful quantities the radial position 𝑟. The use of the propeller advance speed on the
are the pressure coefficient 𝐶𝑝 and the skin friction coefficient 𝐶𝑓 , definition of the skin friction coefficient allows a better identification
defined as: of the transition location over the entire propeller blade, whereas for
𝑝 − 𝑝∞ 𝜏𝑤 the pressure coefficient the analysis is carried out at different radial
𝐶𝑝 = , 𝐶𝑓 = , (7) sections and the standard definition based on the undisturbed onset
2
1∕2𝜌𝑉ref 1∕2𝜌𝑉𝐴2
velocity to those sections considered.
where 𝑝∞ is the undisturbed static pressure here defined√at the inlet The RANS simulations are compared with open-water tests and
boundary, 𝜏𝑤 is the local wall shear stress and 𝑉ref = 𝑉𝐴2 + (𝛺𝑟)2 LDV measurements of the blade boundary-layer that were performed

6
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

̃ 𝜃 transition model with 𝑇 𝑢 = 1.2% (top) and 𝑇 𝑢 = 1.5% (middle) for 𝜈𝑡 ∕𝜈 = 500,
Fig. 9. Limiting streamlines and skin friction distribution over the propeller blade using the 𝛾 − Re 𝑡
and 𝑘 − 𝜔 SST turbulence model with 𝑇 𝑢 = 1.0% and 𝜈𝑡 ∕𝜈 = 1 (bottom).

in the DTMB 24 in. water tunnel. The propeller was tested at 10 rps First, the iterative error is analyzed for the RANS simulations with
at an advance coefficient of 𝐽 = 0.833, corresponding to a Reynolds
̃ 𝜃 transition model. For the
the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re
number√ of 9.5 × 105 , based on chord-length 𝑐0.7𝑅 and inflow velocity 𝑡

𝑉0.7𝑅 = 𝑉𝐴2 + (0.7𝑅𝛺)2 at 0.7 of the propeller radius 𝑅. ̃ 𝜃 transition model, the inflow turbulence quantities are set to
𝛾 − Re 𝑡

1.5% and 500 for the 𝑇 𝑢 and 𝜈𝑡 ∕𝜈, respectively. The iterative error is
4.2. Numerical errors
estimated from the norms of the residuals of the momentum equations,
In CFD (Computational Fluid Dynamics) methods the three main
pressure correction, and transport equations of the turbulence and
contributions to the numerical error are: round-off error, iterative error
and discretization error (Oberkampf and Roy, 2010). Since double- transition models. In this work, we have adopted the 𝐿∞ norm:
precision is used in the present calculations, the round-off error is
neglected. Therefore, an analysis of the iterative and discretization
errors involved in the present computations is shown in this section. 𝐿∞ = max |res(𝜙𝑖 )|, 1 ≤ 𝑖 ≤ 𝑁cells , (8)

7
J. Baltazar et al.
̃ 𝜃
Fig. 10. Blade pressure distribution at sections 𝑟∕𝑅 = 0.3, 0.7 and 0.9 using the 𝛾 − Re
transition model with 𝑇 𝑢 = 1.2% and 1.5% for 𝜈𝑡 ∕𝜈 = 500, and 𝑘 − 𝜔 SST turbulence
𝑡
model with 𝑇 𝑢 = 1.0% and 𝜈𝑡 ∕𝜈 = 1. Comparison with experiments (Jessup, 1989).
where 𝑁cells is the total number of grid cells and res(𝜙) refers to the
residuals of the equations for any local flow quantity 𝜙. Complemen-
tary, the 𝐿2 norm of the residuals is also studied:
√
√ ∑𝑁cells
√ 2
√ 𝑖=1 res (𝜙𝑖 )
𝐿2 = . (9)
𝑁cells
The residuals are made non-dimensional using the reference vari-
ables 𝑐0.7𝑅 , 𝑉0.7𝑅 and 𝑝∞ . The 𝐿∞ and 𝐿2 norms of the residuals for
the Cartesian components of the flow velocity 𝑈𝑋,𝑌 ,𝑍 , modified static
pressure 𝑃 , turbulence kinetic energy 𝑘, specific turbulence dissipation
rate 𝜔, intermittency 𝛾 and local transition onset momentum thickness
Reynolds number Re ̃ 𝜃 for the grid with 37.6 million cells (G1) as a
𝑡 𝑡
8
Ocean Engineering 211 (2020) 107633
function of the number of iterations are plotted in Fig. 4. In this study
a convergence criterion of 10−6 has been adopted for the 𝐿∞ and 𝐿2
norms of the flow quantities. The analysis of the results obtained with
the 𝑘 − 𝜔 SST turbulence model shows that convergence of the 𝐿∞ and
𝐿2 norms is obtained for all quantities. However, iterative convergence
is not achieved for the 𝛾 − Re ̃ 𝜃 transition model. In the transition
𝑡
region and depending on the local refinement level, the intermittency
may have abrupt changes and since is limited between zero and one,
this had led to maximum residuals of the order of 10−1 , due to the
jump between the laminar and turbulent regimes. Nonetheless, this
corresponds to a 𝐿2 norm of the 𝛾 residual below 10−4 . For the re-
maining quantities, a 𝐿2 error norm below 10−6 is achieved. In the fully
turbulent computations, a large number of iterations is needed to obtain
normalized residuals below than 10−6 . This results in 80 000 iterations.
Since in the transition model computations, the normalized residuals
stagnate after 20 000 iterations, the computational effort is lower. The
extra computational cost of the turbulence model calculations is around
70% compared to the transition model solution. In order to understand
the influence of the iterative convergence on the propeller forces, the
variation of the thrust and torque coefficients with the number of
iterations is presented in Fig. 5. For the propeller forces using the
̃ 𝜃 transition model, iterative convergence is achieved after 7000
𝛾 − Re 𝑡
iterations, when the 𝐿∞ and 𝐿2 norms of the flow quantities have
not yet stagnated. In this sense, a minor effect of the iterative error
is expected on the prediction of the propeller coefficients.
The discretization error is estimated following the procedure de-
scribed in Eça and Hoekstra (2014). In this procedure, the discretization
error is estimated based on the truncated power series expansion
𝜙𝑖 = 𝜙0 + 𝛼ℎ𝑝𝑖 , (10)
where 𝜙𝑖 is a quantity of interest obtained on grid 𝑖, 𝜙0 is an estimate of
the exact solution, 𝛼 is a constant, ℎ𝑖 is the typical cell size of grid 𝑖 and
𝑝 is the observed order of grid convergence. Since the grids are block-
structured and the same coarsening factor is applied on each direction,
the typical cell size of grid 𝑖 is determined from the total number of grid
cells 𝑁cells by ℎ𝑖 = (1∕𝑁cells )1∕3 . The unknown coefficients in Eq. (10)
are determined from a least-square fit of the numerical solutions for
the six grids. The error estimate is then converted into a numerical
uncertainty 𝑈num , which is meant to provide a 95% confidence interval.
The estimated uncertainty depends on the estimated discretization
error 𝜙𝑖 − 𝜙0 , the standard deviation of the fit and the difference
between the real data point and the value obtained from the fit for
the same grid density. The complete procedure is described in Eça and
Hoekstra (2014).
The results of the numerical uncertainty estimation for the propeller
forces are plotted in Fig. 6. The plots include also the fits and the
apparent orders of convergence. The numerical uncertainties are evi-
denced by the error bars. For the two finest grids (21 and 37.6 million
cells), calculations were run for two days in parallel using 128 CPUs
at the MARIN cluster. For the remaining grid sizes, the calculations
were performed overnight in parallel on a 16 CPU Linux computer.
With the exception of the 𝐾𝑇 predicted by the 𝛾 − Re ̃ 𝜃 transition
𝑡
model, numerical uncertainties lower than 1% are obtained. In these
cases, near first-order convergence is obtained for the thrust coefficient,
predicted with the 𝑘 − 𝜔 SST turbulence model, and near second-order
convergence is obtained for the torque coefficient. For the 𝐾𝑇 predicted
by the 𝛾−Rẽ 𝜃 transition model, the three coarsest grids deviate from the
𝑡
trend and are omitted from the least-square fit. In this case, first-order
convergence is obtained and the uncertainty in the thrust coefficient is
1%. Since this study comprehends the analysis of local flow quantities,
the convergence with grid refinement of the pressure and skin friction
coefficients at non-dimensional chordwise locations 𝑠∕𝑐 = 0.2 and 0.7
on the blade suction side are analyzed in Fig. 7. Some cases do not show
monotonic convergence. In general, apparent orders of convergence
between 1 and 2 are obtained from the least-square fits. One exception
is obtained for the 𝐶𝑓 at 𝑠∕𝑐 = 0.7 using the 𝛾 − Re ̃ 𝜃 transition model,
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

̃ 𝜃 transition model with different

Fig. 11. Predicted chordwise component 𝑉𝑠 of the flow velocity on the suction side at 𝑟∕𝑅 = 0.7 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re 𝑡
inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).

where 𝑝 = 0.6. Lower numerical uncertainties are obtained with the Table 3
𝑘−𝜔 SST turbulence model. We note that large numerical uncertainties Estimated discretization error |𝜙1 − 𝜙0 |, apparent order of convergence 𝑝 and numerical
uncertainty 𝑈num for 𝐾𝑇 , 10𝐾𝑄 , 𝐶𝑝 and 𝐶𝑓 at 𝑠∕𝑐 = 0.2 and 0.7 on the blade suction
̃ 𝜃 transition
are estimated for the skin friction coefficient with the 𝛾 − Re 𝑡 side.
model. For sake of completeness, the apparent orders of convergence Model 𝑘 − 𝜔 SST ̃ 𝜃
𝛾 − Re 𝑡
and estimated numerical uncertainties are summarized in Table 3. The
𝜙 |𝜙1 − 𝜙0 | 𝑝 𝑈num |𝜙1 − 𝜙0 | 𝑝 𝑈num
discretization errors estimated from the least-square fits |𝜙1 − 𝜙0 |, are
𝐾𝑇 6.0 × 10−4 1.0 0.6% 1.2 × 10−3 1.8 1.0%
also presented in the table. In general, larger discretization errors are
10𝐾𝑄 1.5 × 10−4 1.8 0.2% 1.1 × 10−3 1.6 0.5%
estimated for the 𝛾 − Rẽ 𝜃 transition model.
𝑡 𝐶𝑝(0.2) 4.3 × 10−4 2.0 0.8% 3.1 × 10−4 1.0 3.7%
Since not only global quantities, but also local flow quantities are 𝐶𝑓(0.2) 4.3 × 10−5 1.0 3.6% 1.5 × 10−3 2.0 27.8%
analyzed and compared with the experimental results, the influence 𝐶𝑝(0.7) 5.3 × 10−5 2.0 0.6% 1.0 × 10−3 1.0 2.8%
of the grid density on the propeller forces and blade boundary-layer 𝐶𝑓(0.7) 8.8 × 10−4 – 3.6% 2.4 × 10−3 0.6 9.8%
velocity profiles are discussed. The variation of the thrust and torque
coefficients, and open-water efficiency for each grid compared to the
finest grid is listed in Table 4. Differences lower than 1% are obtained

9
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

Table 4
Variation of the force coefficients with grid density compared to the finest grid.
Model 𝑘 − 𝜔 SST ̃ 𝜃
𝛾 − Re 𝑡

Grid 𝛥𝐾𝑇 𝛥𝐾𝑄 𝛥𝜂0 𝛥𝐾𝑇 𝛥𝐾𝑄 𝛥𝜂0

G6 1.0% 1.4% −0.4% −0.6% 2.4% −2.8%
G5 0.6% 0.8% −0.3% −0.7% 1.4% −2.1%
G4 0.4% 0.3% 0.0% −0.6% 0.6% −1.1%
G3 0.2% 0.2% 0.0% −0.6% 0.3% −0.8%
G2 0.1% 0.1% 0.0% −0.2% 0.1% −0.3%

Table 5
Pressure (𝑝 ) and friction (𝑓 ) contributions to the propeller thrust and torque and
open-water efficiency. Comparison with experimental data.
Model 𝑇𝑢 𝜈𝑡 ∕𝜈 𝐾𝑇𝑝 𝐾𝑇𝑓 𝐾𝑇
̃ 𝜃
𝛾 − Re 1.2% 500 0.150 −0.00227 0.148
𝑡
̃ 𝜃
𝛾 − Re 1.5% 500 0.148 −0.00313 0.145
𝑡
𝑘 − 𝜔 SST 1.0% 1 0.146 −0.00448 0.142
Exp. – – – – 0.146
Model 𝑇𝑢 𝜈𝑡 ∕𝜈 10𝐾𝑄𝑝 10𝐾𝑄𝑓 10𝐾𝑄
̃ 𝜃
𝛾 − Re 1.2% 500 0.254 0.0161 0.270
𝑡
̃ 𝜃
𝛾 − Re 1.5% 500 0.250 0.0226 0.272
𝑡
𝑘 − 𝜔 SST 1.0% 1 0.247 0.0312 0.278
Exp. – – – – 0.280
Model 𝑇𝑢 𝜈𝑡 ∕𝜈 – – 𝜂0
̃ 𝜃
𝛾 − Re 1.2% 500 – – 0.727
𝑡
̃ 𝜃
𝛾 − Re 1.5% 500 – – 0.707
𝑡
𝑘 − 𝜔 SST 1.0% 1 – – 0.676
Exp. – – – – 0.692

for the grid with 9.9 million cells (G3). Fig. 8 presents the chordwise
component 𝑉𝑠 of the blade boundary-layer velocity profile on the
suction side at 0.5 of the chord-length and 0.7 of the propeller radius.
Results of all grid sizes are compared for both models. The velocity
profile is presented along the normal direction, where 𝑦 is the wall
distance to blade surface and 𝑐 the chord-length. A minor influence
of the grid refinement level on the blade velocity profiles predicted
by the 𝑘 − 𝜔 SST turbulence model is observed. However, convergence
̃ 𝜃 transition model velocity profiles is difficult to obtain.
of the 𝛾 − Re 𝑡
The maximum differences in the velocity profile between the grids with
9.9 million cells and 37.6 million cells are of the order of 4%. The
results show that fine grids are required to capture the flow details.
Therefore, the results obtained with the finest grid (37.6 million cells)
are considered for the analysis of the blade flow and comparison with
experimental data.

4.3. Propeller blade flow

In this section the propeller blade flow predictions using the tur-
bulence and transition models are examined. For the analysis of the
different flow solutions, the limiting streamlines and skin friction co-
efficient 𝐶𝑓 on the blade surface are presented in Fig. 9. The location
where flow transition occurs is identified by the sudden rise of the skin
friction distribution and the change in the orientation of the limiting
streamlines towards the chordwise direction. The results obtained with
the 𝛾 − Rẽ 𝜃 transition model present a laminar flow region with
𝑡 experimental results in Table 5. Due to the growth in the turbulent
transition to turbulent flow at some distance from the leading-edge.
A strong sensitivity to the inlet turbulence quantities in the prediction
of laminar-to-turbulent transition is observed. The domain of laminar
flow reduces with the increase of the turbulence intensity at the inlet.
As expected, the 𝑘 − 𝜔 SST turbulence model leads to turbulent flow on
the propeller blades. In this case, a short laminar flow region is present
along the leading-edge. This comparison shows that detailed experi-
mental information is crucial for the selection of the inlet turbulence
quantities when using the 𝛾 − Rẽ 𝜃 transition model.
𝑡
Fig. 12. Predicted chordwise component 𝑉𝑠 of the flow velocity on the pressure side
at 𝑟∕𝑅 = 0.7 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re ̃ 𝜃 transition model with
𝑡
different inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).
The predicted open-water characteristics, including the pressure and
friction contributions to the force coefficients, are compared with the
flow region, an increase in the magnitude of the friction contribution
is observed. At the same time, a decrease in the pressure contribution
is observed, which may be explained by a decambering effect due to
a thicker boundary-layer. A better agreement with the experimental
torque coefficient is obtained with the 𝑘 − 𝜔 SST turbulence model.
For the thrust coefficient and open-water efficiency, a better agreement
̃ 𝜃 transition model
with the experimental data is obtained for the 𝛾 − Re 𝑡
with 𝑇 𝑢 = 1.5% and 𝜈𝑡 ∕𝜈 = 500 at the inlet boundary.

10
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

̃ 𝜃 transition model with different

Fig. 13. Predicted radial component 𝑉𝑟 of the flow velocity on the suction side at 𝑟∕𝑅 = 0.7 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re 𝑡
inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).

In addition to the propeller forces, the predicted blade pressure propeller frame of reference and is calculated as follows:
distribution is compared with experimental measurements. Since the [ ] [ ]
pressure distributions presented in Jessup (1989) were derived from the 𝛥𝑝𝑡 = 𝑃 + 1∕2𝜌 𝑉𝑋2 + 𝑉𝑌2 + 𝑉𝑍2 − 𝑃inlet − 1∕2𝜌 𝑉𝐴2 + (𝛺𝑟)2 , (12)
measured chordwise and radial velocity components at the boundary- with 𝑃inlet = 𝑝∞ + 2∕3𝜌𝑘inlet , which results in a negative variation of
layer edge, a similar approach is considered in this study for the the total pressure inside the boundary-layer. In the present work, the
RANS simulations. Following Jessup (1989) the pressure coefficient is boundary-layer thickness is defined by the distance normal to the wall
calculated by: 2 )
to a point where the total pressure loss coefficient 𝐶𝛥𝑝𝑡 = 𝛥𝑝𝑡 ∕(1∕2𝜌𝑉ref
( ) ( )
𝑉𝑠 2 𝑉𝑟 2 is equal to −0.01, corresponding approximately to the definition of
𝐶𝑝 = 1 − − , (11) 99.5% of the free-stream velocity at that point.
𝑉ref 𝑉ref
√ The comparison between the calculated pressure distributions with
with 𝑉ref = 𝑉𝐴2 + (𝛺𝑟)2 , where 𝑉𝑠 and 𝑉𝑟 are the chordwise and radial the turbulence and transition models and the experimental data is
velocity components in the propeller frame of reference calculated at presented at the radial sections 𝑟∕𝑅 = 0.3, 0.7 and 0.9 in Fig. 10. The
the edge of the boundary-layer 𝛿, respectively. The estimated boundary- non-dimensional chordwise location is defined as 𝑠∕𝑐. Similar results
layer thickness 𝛿 is obtained from the total pressure loss 𝛥𝑝𝑡 in the are obtained between the turbulence model and transition model. A

11
J. Baltazar et al.
Fig. 14. Predicted chordwise component 𝑉𝑠 of the flow velocity on the suction side at 𝑟∕𝑅 = 0.9 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re 𝑡
inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).
reasonable to good agreement between the numerical results and the
experimental data is obtained at the radial sections 𝑟∕𝑅 = 0.7 and
0.9. At the radial section 𝑟∕𝑅 = 0.3, some discrepancies between the
calculations and experiments are visible in the pressure distribution.
The larger disagreement observed at 𝑟∕𝑅 = 0.3 in comparison with
the other radial sections, may be due to the boundary-layer thick-
ness corrections applied by Jessup (1989) to the chordwise location
of the measurements. This correction is especially important towards
the blade root, due to the thicker blade sections. The values larger
than 1.0 obtained at the trailing edge indicate lack of accuracy in
the determination of the measured chordwise positions. In addition,
some differences are observed near the blade leading-edge for all radial
sections. Jessup (1989) suggested that these differences may be due to
errors in the geometry model.
12
Ocean Engineering 211 (2020) 107633
̃ 𝜃 transition model with different
4.4. Comparison of velocity profiles
In this section a comparison of the velocity profiles in the blade
boundary-layer between the RANS simulations and the experiments
is made. The chordwise 𝑉𝑠 and radial 𝑉𝑟 components of the flow
velocity were measured by a LDV system comprising a one compo-
nent lens system with a fiber optic probe. The measurement volume,
formed by the interception of the two transmission beams, is placed
tangential to the blade surface at each measurement point. As the
propeller rotates, velocity measurements along the boundary layer are
acquired. The chordwise and radial velocity components are measured
by setting the plane that contains the transmission beams tangentially
or perpendicularly to the blade surface, respectively. In order to fully
capture the large flow gradients near the blades, a measurement volume
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

characterized by a reference length of approximately 0.075 mm was

used. Seeding particles were added to the flow to improve signal-
to-noise ratio. In the propeller model, two blades had leading-edges
roughened with 60 μm distributed roughness and the other blade was
smooth (Jessup, 1989).
Figs. 11 to 13 show the chordwise and radial boundary-layer ve-
locity profiles at 0.7 of the propeller radius developed on the suction
and pressure sides at different chordwise locations, respectively. In
addition, Figs. 14 and 15 show the chordwise and radial boundary-layer
velocity profiles developed at the 0.9 radius, respectively. These pro-
files are shown in the propeller-fixed frame of reference. The numerical
predictions are compared with the experimental boundary-layer pro-
files measured on the smooth and tripped blades. The legends present
the estimated boundary-layer thicknesses of the numerical results. The
grid resolution inside the blade boundary-layer consists of more than
30 layers with an expansion ratio of 1.2 along the direction normal to
the blade surface.
From the streamwise measurements at 0.7 of the propeller radius,
the smooth blade appears to have transitioned on the suction side by
0.7 of the chord. In this case, the best agreement with the smooth
blade measurements is obtained for the 𝛾 − Re ̃ 𝜃 transition model with
𝑡
𝑇 𝑢 = 1.2% before transition. After transition, a better match is obtained
̃ 𝜃 transition model with 𝑇 𝑢 = 1.5%. On the pressure side,
for the 𝛾 − Re 𝑡
the best agreement is achieved for the 𝛾 − Re ̃ 𝜃 transition model with
𝑡
𝑇 𝑢 = 1.5%, since the results with 𝑇 𝑢 = 1.2% predict a too large laminar
flow region. The suction side radial profiles at 0.7 radius on the smooth
blade predicted by the 𝛾 − Re ̃ 𝜃 transition model with 𝑇 𝑢 = 1.2% are
𝑡
also laminar showing a very large outward radial flow near the wall.
At 𝑠∕𝑐 = 0.77 the measured outward radial flow reduces significantly
and a better agreement is obtained with 𝛾 − Re ̃ 𝜃 transition model using
𝑡
𝑇 𝑢 = 1.5%.
From the measurements on the smooth blade at 0.9 of the propeller
radius, transition seems to occur between 0.3 and 0.4 of the section
chord. The streamwise velocity profiles obtained from the 𝛾 − Re ̃ 𝜃
𝑡
transition model with both 𝑇 𝑢 = 1.2% and 1.5% show that transition
is predicted at more aft-chord locations. The measured radial profiles
on the smooth blade show small outward flow near the wall. Similar
profile shapes are obtained with the 𝛾 − Rẽ 𝜃 transition model using
𝑡
𝑇 𝑢 = 1.5%.
The velocity profiles predicted by the 𝑘 − 𝜔 SST turbulence model
approach the measurements on the tripped blade. However, inflected
profiles are obtained at the chordwise positions close to the leading
edge from the tripped blade measurements at 0.7 of the propeller
radius. According to Jessup (1989), these discrepancies are attributed
to the influence of the leading-edge roughness on the blade flow. For
the radial section 0.9, the tripped blade shows a turbulent profile at
0.3 chord without the inflection. These variations may be due to the
amount of distributed roughness along the radial direction (Jessup,
1989).
Fig. 15. Predicted radial component 𝑉𝑟 of the flow velocity on the suction side at
4.5. Boundary-layer characteristics 𝑟∕𝑅 = 0.9 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re ̃ 𝜃 transition model with
𝑡
different inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).

From the predicted velocity profiles, the boundary-layer thickness 𝛿,

displacement thickness 𝛿 ∗ and shape factor 𝐻 are calculated and repre-
sented in Figs. 16 and 17. In this work, the boundary-layer parameters
are predicted from the streamwise velocity profiles and the integrals
estimated by numerical integration from the wall to the boundary-
layer thickness. The predicted displacement thickness and shape factor
are also compared with the calculations of Jessup (1989) from the
measured velocity profiles.
The evolution of the boundary-layer thickness along the chord-
wise direction shows the increase in thickness due to transition from
laminar-to-turbulent flow regime. For the displacement thickness and
shape factor, agreement between the numerical simulations and the
experimentally determined boundary-layer parameters is difficult to
obtain. In general, the results obtained for the 𝑘 − 𝜔 SST turbulence
model approach the experimental values of the tripped blade. The
same trend is observed between the 𝛾 − Re ̃ 𝜃 transition model and the
𝑡
smooth blade experiments. In general, a better agreement is obtained
with 𝑇 𝑢 = 1.5%. However, as previously discussed in the analysis of
the velocity profiles, the agreement at 0.7 of the propeller radius on
the suction side improves with 𝑇 𝑢 = 1.2%. The experimental and
numerical shape factors show large values in the laminar flow region,
which reduce to turbulent profile values around 1.4 typical of two-
dimensional flows. The drops observed in the shape factor with the
development of the boundary-layer correlate well with the estimated
transition locations based on the velocity profiles.

13
J. Baltazar et al. Ocean Engineering 211 (2020) 107633

̃ 𝜃
Fig. 16. Boundary-layer thickness 𝛿 (top), displacement thickness 𝛿 ∗ (middle), and shape factor 𝐻 (bottom) at 𝑟∕𝑅 = 0.7 using the 𝑘 − 𝜔 SST turbulence model and 𝛾 − Re 𝑡
transition model with different inlet turbulence quantities. Comparison with experimental data (Jessup, 1989).

5. Conclusions
In this paper, a comparison between RANS calculations and ex-
perimental data has been presented. The analysis is carried out for
marine propeller P4119, for which three-dimensional velocity com-
ponent measurements on the blade boundary-layer are available at
model-scale. In the model propeller, two blades had rough leading-
edges and the other blade was smooth. For the RANS calculations two
models are considered: the commonly-used 𝑘 − 𝜔 SST eddy-viscosity
turbulence model, and the 𝛾 − Re ̃ 𝜃 transition model. The influence
𝑡
of the iterative errors, discretization errors and boundary conditions
on the propeller flow predictions has been analyzed. The 𝛾 − Re ̃ 𝜃
𝑡 𝑡
transition model does not satisfy the iterative convergence criterion.
Still, its influence on the propeller forces is assumed to be small
due to the fast convergence of the thrust and torque coefficients.
The numerical uncertainty of the propeller forces (integral quantities)
and pressure and skin friction coefficients (local quantities) has been
determined from an estimation of the discretization error based on
systematically refined grids. Low numerical uncertainties (less than
2%) are obtained for the propeller force coefficients. For the local
quantities, small numerical uncertainties (less than 4%) are predicted
with the 𝑘 − 𝜔 SST turbulence model. For the 𝛾 − Rẽ 𝜃 transition model,
𝑡
larger uncertainties are obtained (more than 9%), especially for the skin
friction coefficient. The numerical predictions of the velocity profiles
have been compared with blade boundary-layer measurements. An
agreement is found between the turbulent velocity profile predicted by
the 𝑘 − 𝜔 SST turbulence model and the measured velocities from the
̃ 𝜃 transition model, the agreement with the
tripped blade. For the 𝛾 − Re
smooth blade velocity measurements is highly dependent on the inlet
turbulence quantities. In this work, from the selected inlet turbulence
quantities, a qualitative agreement is obtained for the blade boundary-
layer velocity profiles. However, this information may not be available,

14
J. Baltazar et al.
Fig. 17. Boundary-layer thickness 𝛿 (top), displacement thickness 𝛿 ∗ (middle), and
shape factor 𝐻 (bottom) on the suction side at 𝑟∕𝑅 = 0.9 using the 𝑘 − 𝜔 SST
̃ 𝜃 transition model with different inlet turbulence quantities.
turbulence model and 𝛾 −Re
𝑡
Comparison with experimental data (Jessup, 1989).
̃ 𝜃 transition model
which limits the predictive capabilities of the 𝛾 − Re
𝑡
at model-scale, specially because the selected inlet values are not real-
istic from the physical point of view. Finally, boundary-layer integral
quantities obtained from the measured and computed velocity profiles
have been analyzed. The evolution of these quantities along the chord-
wise direction shows typical laminar and turbulent flow behaviors.
The comparison between the RANS calculations and the experimental
measurements shows that the considered models can be used, not only
for the prediction of integral quantities, but also for the analysis of the
local features of the propeller blade flow. Nevertheless, the selection
of the inlet turbulence quantities is crucial in the prediction of the
transition location, and in the accuracy of the model-scale propeller
performance prediction. Therefore, the assessment of the predictive
15
Ocean Engineering 211 (2020) 107633
̃ 𝜃 transition model requires the analyses of
capabilities of the 𝛾 − Re 𝑡
not only the propeller forces, but also of the blade boundary-layer flow
for the identification of transition location.
CRediT authorship contribution statement
João Baltazar: Conceptualization, Methodology, Investigation,
Writing - original draft, Writing - review & editing. Daniela Melo:
Formal analysis, Visualization. Douwe Rijpkema: Software, Resources,
Supervision.
Declaration of competing interest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
influence the work reported in this paper.
Acknowledgments
The authors are indebted to José Falcão de Campos, Luís Eça and
Rui Lopes for many fruitful discussions on this topic.
References
Abdel-Maksoud, M., Heinke, H., 2002. Scale effects on ducted propellers. In:
Proceedings of the 24th Symposium on Naval Hydrodynamics. Fukuoka, Japan.
Baltazar, J., Rijpkema, D., Falcão de Campos, J.A.C., 2018. On the use of the 𝛾 − Re ̃ 𝜃
𝑡
transition model for the prediction of the propeller performance at model-scale.
Ocean Eng. 170, 6–19.
Bhattacharyya, A., Krasilnikov, V., Steen, S., 2016a. Scale effects on open water
characteristics of a controllable pitch propeller working within different duct
designs. Ocean Eng. 112, 226–242.
Bhattacharyya, A., Krasilnikov, V., Steen, S., 2016b. A CFD-based scaling approach for
ducted propellers. Ocean Eng. 123, 116–130.
Bulten, N., Nijland, M., 2011. On the development of a full-scale numerical towing
tank Reynolds scaling effects on ducted propellers and wakefields. In: Proceedings
of the Second International Symposium on Marine Propulsors. Hamburg, Germany.
Eça, L., Hoekstra, M., 2008. The numerical friction line. J. Mar. Sci. Technol. 13 (4),
328–345.
Eça, L., Hoekstra, M., 2014. A procedure for the estimation of the numerical uncertainty
of CFD calculations based on grid refinement studies. J. Comput. Phys. 262,
104–130.
Funeno, I., 2002. On viscous flow around marine propellers - hub vortex and scale
effect. J. Kansai Soc. N. A 238, 17–27.
GridPro, 2019. http://www.gridpro.com.
Hasuike, N., Okazaki, M., Okazaki, A., Fujiyama, K., 2017. Scale effects of marine
propellers in POT and self propulsion test conditions. In: Proceedings of the Fifth
International Symposium on Marine Propulsors. Espoo, Finland.
International Towing Tank Conference, 1978. Final report and recommendations of
the propulsion committee to the 15th ITTC. In: Proceedings of the 15th ITTC. the
Hague, the Netherlands.
International Towing Tank Conference, 2017. 1978 ITTC Performance Prediction
Method. Technical Report ITTC - Recommended Procedures and Guidelines
7.5-02-03-01.4, Revision 07.
Jessup, S.D., 1989. An Experimental Investigation of Viscous Aspects of Propeller Blade
Flow (Ph.D. Thesis). The Catholic University of America.
Krasilnikov, V., Sun, J., Halse, K., 2009. CFD investigation in scale effects on propellers
with different magnitude of skew in turbulent flow. In: Proceedings of the First
International Symposium on Marine Propulsors. Trondheim, Norway.
Kuiper, G., 1981. Cavitation Inception on Ship Propeller Models (Ph.D. Thesis). Delft
University of Technology.
Langtry, R., 2006. A Correlation-Based Transition Model using Local Variables for
Unstructured Parallelized CFD Codes (Ph.D. Thesis). University of Stuttgart.
Langtry, R., Menter, F., 2009. Correlation-based transition modeling for unstructured
parallelized computational fluid dynamics codes. AIAA J. 47, 2894–2906.
Menter, F., 1994. Two-equation eddy viscosity turbulence models for engineering
applications. AIAA J. 32, 1598–1605.
Menter, F., Esch, T., Kubacki, S., 2002. Transition modelling based on local variables.
In: Proceedings of the 5th International Symposium on Engineering Turbulence
Modelling and Measurements. Mallorca, Spain.
Menter, F., Kuntz, M., Langtry, R., 2003. Ten years of industrial experience with the
SST turbulence model. In: Proceedings of the Fourth International Symposium on
Turbulence, Heat and Mass Transfer, Vol. 4. pp. 625–632.
Moran-Guerrero, A., Gonzalez-Gutierrez, L.M., Oliva-Remola, A., Diaz-Ojeda, H.R.,
2018. On the influence of transition modeling and crossflow effects on open water
propeller simulations. Ocean Eng. 156, 101–119.
J. Baltazar et al.
Müller, S.-B., Abdel-Maksoud, M., Hilbert, G., 2009. Scale effects on propellers for large
container vessels. In: Proceedings of the First International Symposium on Marine
Propulsors. Trondheim, Norway.
Oberkampf, W.L., Roy, C.J., 2010. Verification and Validation in Scientific Computing.
Cambridge University Press..
ReFRESCO, 2019. http://www.refresco.org.
Rijpkema, D., Baltazar, J., Falcão de Campos, J.A.C., 2015. Viscous flow simulations
of propellers in different Reynolds number regimes. In: Proceedings of the Fourth
International Symposium on Marine Propulsors. Austin, Texas, USA.
Rijpkema, D., Vaz, G., 2011. Viscous flow computations on propulsors: verification,
validation and scale effects. In: Proceedings of the International Conference on
Developments in Marine CFD. London, UK.
Sánchez-Caja, A., González-Adalid, J., Pérez-Sobrino, M., Sipilä, T., 2014. Scale effects
on tip loaded propeller performance using a RANSE solver. Ocean Eng. 88,
607–617.
Shin, K.W., Andersen, P., 2017. CFD analysis of scale effects on conventional and
tip-modified propellers. In: Proceedings of the Fifth International Symposium on
Marine Propulsors. Espoo, Finland.
16
Ocean Engineering 211 (2020) 107633
Spalart, P., Rumsey, C., 2007. Effective inflow conditions for turbulence models in
aerodynamic calculations. AIAA J. 45, 2544–2553.
Stanier, M., 1998. The application of RANS code to investigate propeller scale effects.
In: Proceedings of the 22th ONR Symposium on Naval Hydrodynamics. Washington,
DC, USA.
Uto, S., 1993. Computation of incompressible viscous flow around a marine propeller.
J. Soc. Nav. Archit. Japan 173, 67–75.
Walters, D.K., Cokljat, D., 2008. A three-equation eddy-viscosity model for reynolds-
averaged Navier–Stokes simulations of transitional flow. ASME J. Fluids Eng. 130
(12), 121401.1–121401.14.
Wang, X., Walters, K., 2012. Computational analysis of marine-propeller performance
using transition-sensitive turbulence modeling. ASME J. Fluids Eng. 134 (7),
071107.1–071107.10.
Yao, H., Zhang, H., 2018. Numerical simulation of boundary-layer transition flow of a
model propeller and the full-scale propeller for studying scale effects. J. Mar. Sci.
Technol. 23, 1004–1018.