Proceedings of the Nineteenth (2009) International Offshore and Polar Engineering Conference

Osaka, Japan, June 21-26, 2009

Copyright © 2009 by The International Society of Offshore and Polar Engineers (ISOPE)
ISBN 978-1-880653-53-1 (Set); ISSN 1098-618

A Study on the Design Optimization of an AUV by Using Computational Fluid Dynamic Analysis
Taehwan Joung*, Karl Sammut*, Fangpo He*, and Seung-Keon Lee**

*School of Computer Science, Engineering and Mathematics, Faculty of Science & Engineering, Flinders University, Adelaide, Australia
**Department of Naval Architecture and Ocean Engineering, Pusan National University, Busan, Korea

ABSTRACT reliable, and also convenient for users.

Autonomous Underwater Vehicles (AUV's) provide an important
means for collecting detailed scientific information from the ocean
depths. The hull resistance of an AUV is an important factor in
determining the power requirements and range of the vehicle. This
paper describes a design method using Computational Fluid Dynamics
(CFD) for determining the hull resistance of an AUV under
development. The CFD results reveal the distribution of hydrodynamic
values (velocity, pressure, etc.) of the AUV with a ducted propeller.
The optimization of the AUV hull profile for reducing the total
resistance is also discussed in this paper. This paper demonstrates that
shape optimization in conceptual design is possible by using a
commercial CFD package. The optimum designs to minimize the drag
force of the AUV were carried out, for a given object function and
constraints.
KEY WORDS: AUV; CFD (Computational Fluid Dynamics);
Optimum design; Drag force; Drag Coefficient (CD)
In this paper, Computational Fluid Dynamics (CFD) tools are evaluated
with the purpose of obtaining the hydrodynamic parameter (velocity,
pressure, etc.) estimates of an AUV with a ducted propeller. The design
of an AUV is optimized using CFD analysis to minimise drag force.
The methods reported in this paper for optimisation by CFD code are as
follows: (1) CFD results analysis and comparison with theoretical or
empirical equation for validation of reliability, (2) evaluation of an
automatic element meshing method that generates a boundary layer
which allows for appendages such as fins and ducts, and produces a
stable and robust analysis, and (3) searching and identifying optimum
design variables to produce minimum resistance.
INITIAL HULL DESIGN AND DRAG ESTIMATION OF
THE AUV
Hull Design

INTRODUCTION At a conceptual design stage, the hull of an AUV can be divided into
distinct sections, namely the nose, middle section, tail, and propeller
An unmanned AUV (Autonomous Underwater Vehicle) is a versatile duct. The AUV hull has been designed based on the Myring hull profile
research tool for maritime archaeology, oceanographic and marine equations (Prestero, 2001) which is known to produce minimum drag
biology studies, defense applications, and oil and mineral exploration force for a given fineness ratio, that is, ratio of its length to its
and exploitation programs. Rapid progress in AUV development is maximum diameter (l/d). The curve shapes of the nose and tail sections
steadily increasing the reliability and endurance of such vehicles to are determined from equations (1) and (2), respectively.
operate in the harsh marine environment. Much work, however, still
1/ n
needs to be done in terms of optimizing the hull design to minimize
1 ⎡ ⎛ x−a⎞ ⎤
2
drag and increase propulsion efficiency. r ( x) = d ⎢1 − ⎜ ⎟ ⎥ (1)
2 ⎣⎢ ⎝ a ⎠ ⎦⎥
In previous studies, designers have employed empirical formulas or
used experimental derived data to estimate drag force of ships or
submerged bodies such as AUVs. However, the conventional empirical 1 ⎡ 3d tan(θ ) ⎤
r ( x) = d −⎢ 2 − ( x − ( a + b)) 2
formula is not able to accurately compute the drag of complex hull 2 ⎣ 2c c ⎥⎦ (2)
forms with appendages protruding from the hull. Although
experimental testing using a tow tank or cavitation tank can produce ⎡ d tan(θ ) ⎤
+⎢ 3 − 2 ⎥ ( x − ( a + b))3
very accurate predictions of drag, such testing requires considerable ⎣c c ⎦
time and effort, and expensive test facilities to obtain the vehicle’s
hydrodynamic characteristics. Consequently, a new drag estimation The designed shape of the AUV hull based on “Myring equation” and
method is needed for development of a specific AUV, which can be “NACA profile (NACA 6721)” is shown in Fig. 1. As the propeller
applied to a conceptual design. The new method should be efficient, blades rotate through the water, they generate high-pressure areas

696
behind each blade and low pressure areas in front of each blade. It is
this pressure differential that provides the force to drive the vessel.
However, losses occur at the tip of each blade as water escapes from
the high pressure side of the blade to the low pressure side, resulting in
reduced efficiency in terms of pushing the vessel forward. To obtain the
most thrust, a propeller must move as much water as possible in a given
time. A nozzle will reduce these propeller losses, especially when a
high thrust is needed at a low vehicle speed. A Rice speed nozzle
profile has been employed for our AUV design concept, since its
coefficient of drag is over 17 times less than that of a conventional Kort
nozzle. A section of a designed nozzle has been developed based on the
‘NACA profile (NACA 6721)’ as showed in Table 1.
Fig. 1 A conceptual design of the AUV
Table 1. Section of the Nozzle - NACA profile (NACA 6721)
Thickness 21 % of wing chord (Length of chord = 1)
Position 0.7 Position of max camber (0 - 1)
Camber 0.06 % of wing chord (Length of chord = 1)
Estimation of the Resistance Factor by an Empirical Formula
The drag caused by an axis-symmetric AUV moving forward under the
water is a direct result of the viscosity of the water. The effect of
viscosity can be considered as two separate factors. That is, one is the
skin friction drag which is caused by viscosity shear force of a fluid
flowing along the hull, and the other is the form drag caused by
development of a boundary layer and the resulting difference of
pressure distribution between front and stern of the vehicle (Phillips A.
et al., 2007). To estimate the coefficient of friction drag (CF) caused by
viscosity, the ITTC (International Towing Tank Conference ) 1957
correlation line’ which is a commonly used method for estimation of
the skin friction component, is considered here. The correlation line as
a function of Reynolds number can be considered as follows (Sang-
Jeon Lim, 1971).
⎛ 0.075 ⎞
C F 1957 = ⎜⎜ ⎟
2 ⎟
(3)
⎝ log( RN − 2) ⎠
The coefficient of the friction drag (CF), however, should be multiplied
by form factor (1+k) in order to complete estimation. The form factor is
a function of the hull shape. Hoerner (Hoerner 1965) proposed the form
factor as a function of the hull length (l) and hull diameter (d) as
follows.
(1 + k ) = 1 + 1.5(d / l )3 / 2 + 7(d / l )3 (4)
697
Estimation of the Drag Force
The axial drag force acting on a body moving at a constant velocity in a
fluid medium is approximated by the expression (5). The drag force of
an AUV, therefore, can be estimated by using the following formula
(Michael, 2003).
1
Rf = ρC f A f u u = X u u u u (5)
2
where, Cf is the coefficient of friction drag obtained from equation (3)
and (4), ρ is the density of the fluid, Af is a submerged area of a hull
and u is the fluid velocity or the advanced speed of the AUV. This
estimation for the drag force can provide useful information for the
powering requirements at the early stage of design. However, the
estimation of the drag obtained only by considering form factor (1+k)
could be uncertain, if there is complex shape equipment attached on the
AUV such as an antenna, DVL (Doppler Velocity Log), ADCP
(Acoustic Doppler Current Profilers), camera or other protruding
equipment.
CFD ANAYSIS METHOD FOR ESTIMATING THE AUV DRAG
Governing Equation of the CFD Analysis
The fluid flow around the AUV has been modelled using the
commercial CFD code ANSYS CFX 11.0. For these calculations, the
governing equations are the Navier-Stokes equations and continuity
equation under the assumption of incompressible fluid. The Navier-
Stokes equations considered in ANSYS CFX 11.0 is the isothermal
Reynolds Averaged Navier-Stokes (RANS) shown in equations (6) and
(7) (Wilcox, 1998; Seo et al, 2005)
∂v j
=0 (6)
∂x j
∂ρvi ∂ρvi ∂p ∂τ ij
+ ui =− + + Fi (7)
∂t ∂x j ∂xi ∂xi
where, Einstein’s summation convention is adopted, that is, xj (j=1, 2
and 3) denote the axes of the orthogonal coordinate system. t is time; vi
is the i-th component of the fluid velocity; ρ is a density of the fluid; τij
is a viscosity stress tensor; and Fi is external force of i-th element due
to the gravity, thrust power, torque and so on.
The Pre-processing for the CFD Analysis
Modelling for CFD Analysis
The AUV hull shape is computed in MS-EXCEL and exported to
ANSYS Workbench Design Modeller 11.0. The 3D model was
exported to ANSYS-CFX-MESH or ANSYS-ICEM-CFD and meshed
to generate the nodes and elements. To produce high quality structured
grids, the elements are produced by careful parameterization of the
AUV hull using script files for driving the meshing package ANSYS-
ICEM-CFD. ANSYS-CFX-MESH, however, was employed at the final
stage for optimization, because of its auto-meshing function and its
link-up with ANSYS-DesigneXplorer, which is the optimization
program in ANSYS-WORKBENCH. and allows solutions to be quickly generated by control of the iteration
times. What is required is a tool that is not only easy to use, but also
Mesh Generation able to provide a satisfactory reliability analysis result with the desired
degree of convergence.
‘Tetrahedral’ and ‘Pyramid’ elements are normally employed for
generating nodes and elements in the fluid domain. These elements are
suitable for representation of a complex geometry such as a nozzle.
However, such elements are not suitable to resolve the boundary layer
adjacent to the solid body (Nishi, 2007). Therefore, ‘Prism’ elements
are selected for generating meshes around the body surface as these are
the most appropriate for a boundary layer mesh (ANSYS-CFX, 2007).

Fig. 2 shows the various meshed sections which are merged, and
embodied for the CFD analysis by the “ANSYS-CFX-MESH” mesh
generator. The size of the fluid domain around the tank is big enough so
as to not cause any error due to blocking effects if the walls of the tank
significantly restrict flow around the hull. The water tank size should be
decided after thorough preliminary analysis, and experimental or
theoretical prediction so as not to make the fluid domain too small thus
adversely affecting the CFD analysis, nor unduly large which would
unnecessarily increase the computation time.

In order to determine the velocity of the AUV, the inlet velocity from
the front of the water tank was set to be equivalent to the advanced
AUV velocity. The constraint of the outlet (opposite side of the inlet) is
that there is no relative pressure, and the ‘Free Surface Condition: no
slip condition’ was allocated to the remaining sides of the water tank.
Given that both the AUV geometry and the boundary conditions are
symmetrical about the centre plane, the symmetric condition was used
for modelling the AUV instead of the total model as shown Fig. 3.

Turbulence Model

Various turbulence models, such as Menter’s Shear Stress Transport

(SST) model which is based on a combined (κ-ω) model at the wall and
(κ-ε) model in the bulk flow, and Speziale-Sarkar-Gatski model (SSG),
have been proposed to provide solutions to the Reynolds stresses in
terms of known quantities to allow closure of the RANS by ANSYS Fig. 2 Hybrid (Tetrahedral and Layer) meshes generated adjacent to the
CFX. From the various turbulence models, κ-ε model and Shear Stress
Transport (SST) were considered in this study. The reason for this main body and nozzle of the AUV
selection is that the κ-ε model is a commonly used turbulence model for
engineering simulations due to its robustness and application to a wide
range of flows, while the SST model is better at predicting separation
(ANSYS-CFX, 2007) likely to be found at the aft of the AUV. For
SST, the wall boundary treatment available in CFX switches
automatically from a low-Reynolds number formulation to a wall-
function κ-ω treatment based on grid density. The Y+ close to the walls
is kept below 100 in-line with the CFX recommendations. From CFD
analysis of the AUV, the drag predictions from SST and κ-ε model
demonstrate a high degree of correspondence.

The degree of the nozzle (Ψ) defined as the angle of attack of the
nozzle profile was considered as one design variable satisfying the
given constraints (0° ≤ Ψ ≤ 20°) in order to minimize the overall drag
resistance of the AUV body including the nozzle. The above
summarized specifications for the “pre-processing” stage including
mesh generation are showed in Table 2.

CFD ANAYSIS and RESULTS

Procedure for the CFD Analysis

Fig. 3 Overview of the pre-processing for the CFD analysis
The CFD analysis module provided with ANSYS-CFX is easy to use

698
Table 2. Principal conditions employed in the numerical computation
Water tank size 7,000×3,000×3,000 mm3

κ-ε model, Shear Stress

Turbulence model
Transport (SST)

Reynolds number 8.73×105 ∼ 6.11×106

Radius, Length of the AUV 0.25 m, 2 m

Angle of attack of the nozzle 0° ∼ 20° (Design Variable)

132,308 ∼ 158,611
Total no. of elements (nodes)
(30,685 ∼ 45,303)

No. of Tetrahedral Max. 153,878

No. of Prisms; Pyramids Max. 1,313; Max. 3,420

If the results are satisfied at a certain value, a user can stop the run and
move to “post-processing” to see the results. Fig. 4 shows the “solve-
process” being monitored while ANSYS-CFX solver is running. Note
that the value of the X-direction drag force of the AUV hull converged  
after nearly 30 iterations. The user, therefore, can decide the number of (a) Default (Momentum etc.)
iteration for convergence and determine when to stop in order to save
CPU time.

Verification

In order to ensure the verification of the CFD analysis, the bare hull of
the AUV was employed, not considering nozzle part. The drag force
predictions from the CFD results and the ‘ITTC 1957 correlation line’
have a high degree of correspondence as shown in Fig. 5. The results
show that the form factor predicted by equation (2) is useful for the
estimation. There are drag differences between ‘ANSYS-CFX’ and
‘ITTC 1957 correlation line’, because ANSYS-CFX considers the total
drag while the ‘ITTC 1957 correlation line’ only considers the skin
friction drag. The pressure and skin friction distribution along the AUV
are shown in Fig. 6.

The CFD results can thus be validated by the ‘ITTC 1957 correlation
line’, and demonstrated to be reliable and useful for further research
such as optimizing the nozzle shape.

The CFD Results

The empirical result, however, do not include the effects of separation

of vortices at the stern. The detailed information about the velocity and
pressure distribution of the AUV with nozzle can be extracted from the
CFD code, ANSYS-CFX.

The pressure distribution around the AUV (seen in Fig. 7 - 8) shows an

even distribution except for the stagnation point at the bow of the hull.  
The maximum pressure (-1.822e-4Pa) occurs within part of nozzle, and
is a negative pressure and higher compared to the pressure along the (b) Defined by user (Drag)
main body of the AUV. As the fluid passes through the nozzle its Fig. 4 Monitoring during the CFD analysis process
velocity increases as shown in Fig. 9. Along the parallel mid-body, the
boundary layer grows and the flow is accelerated as it reaches the stern.
Large vertical structures, which represent the wake region, form behind
the stern as shown in Fig. 9. The flow is accelerated around the nozzle
up to a maximum velocity of 2.167U0; the maximum velocity of the
fluid in the nozzle is 6.5m/s, when the AUV velocity is 3m/s.

699
Fig. 5 Comparisons of drag predictions for the AUV (w/o duct)
Fig. 6 Cp and Cf distribution around the AUV at 3 m/s (w/o duct)
Fig. 7 Pressure contour around the AUV hull at 3 m/s
700
Fig. 8 Pressure distribution around the AUV at 3 m/s
Fig. 9 Velocity contour around the AUV hull at 3 m/s
OPTIMIZATION BY THE CFD ANAYSIS
Optimum Design Method
A design optimization was carried out to find the optimum value of the
nozzle angle, since it has a high sensitivity value with respect to the
total resistance and drag, as well as AUV manoeuvrability. The object
function for optimum design is the X-direction (+ is the advanced
direction) drag force, and the constraint is the nozzle angle (0°～20°).
For the purpose of finding the optimum value of the object function, the
three tools, that is, 3D-CAD program, mesh generation program and
ANSYS-DesignXplorer (DX), are linked together. ANSYS-
DesignXplorer is the optimization program, which sets up the relations
between input parameter from the 3D CAD program (DesignModeler
(DM), AUTOCAD etc.) and output parameters. The typical application
workflows of ANSYS-DesignXplorer comprise two different methods
as shown in Fig. 10. The ANSYS-DesignXplorer provides three
optimization methods; Design of Experiments (DOE), Variational
Technology (VT) and Design for Six Sigma (DFSS) (ANSYS-CFX,
2007). From the three methods, Design of Experiments (DOE) was
employed for the optimization due to its simplicity and reliability. That
is, DOE creates a matrix and builds an approximation model, and
optimization is performed by sampling against this approximation
method.
(Ψ) by DOE has a high correspondence with the Direct Searching
Method, therefore, the results of the DOE method can be accepted to be
reliable.
From the results of the CFD analysis, shown in Fig. 13(a), it can be
seen that if the nozzle angle is lower than the optimum angle (-11°),
then the value of the object function (drag force) becomes much higher,
even though mass flow of the water is a bit larger, when compared with
the drag force at the optimum angle. On the other hand, if the nozzle
angle is higher than optimum angle (-11°), not only does the value of
the object function (drag force) become much higher, but also vortices
can occur behind of the nozzle which reduce the nozzle efficiency, as
shown in Fig. 13(c).

Fig. 10 The typical application workflow of ANSYS-DesigneXplorer

The methods of ‘Design of Experiment (DOE)’ and ‘Direct Searching

Method (What-if)’ were employed for the nozzle optimization as the
optimum design methods. ‘Direct searching method’ is a crucial
method but has a drawback, in that excessive computing effort is spent
repeating the analysis process as many times as the number of samples.
An alternative to ‘Direct Searching Method’ is the DOE method.
Where possible the analysis for design variable(Ψ) in the constraints is
carried out by the ‘Direct Searching Method’, whereas DOE is the
method for carrying out a CFD analysis for an abstracted sample point Fig. 11 X-directional drag force acting on the AUV (DOE)
only. In this way, it is, therefore, possible to reduce considerable CPU
running time for the analysis.
In this study, the sampling points are selected by Central Composite
Design (CCD) to carry out the DOE method, before finding optimum
value of the design variable. That is, sample points were abstracted by
CCD, and the optimal design variable which has the optimum value of
the object function was then identified.

The DOE method by CCD sampling was verified by using the Direct
Searching Method to ensure its reliability.

Results of the Optimum Design

The results of the CFD analysis conducted by DOE and Direct

Searching Method are shown in Fig. 11 and Fig. 12, respectively. The
results show that they have a similar trend and a high degree of
correspondence.

As shown in Fig. 11, the optimum value of the design variable (Ψ) was
obtained as -11.11° from the CFD analysis by DOE, and the values of
the object function (drag force) were 3.00 N (@ 1 m/s), 11.12 N(@ 2
m/s), and 23.98 N (@ 3 m/s). The result of the Direct Searching Fig. 12 X-directional drag force acting on the AUV (Direct Searching
Method, which used for verifying the DOE-CCD result, showed that Method)
the optimum value of the design variable (Ψ) was obtained as -11°, and
the values of object function were 2.98 N (@ 1 m/s), 11.05 N (@ 2
m/s), and 23.76 N (@ 3 m/s). The optimum value of the design variable

701

(a) 0° of the Nozzle (Initial value)
(b) -11° of the Nozzle (Optimum value)
(c) -20° of the Nozzle
Fig. 13 Pressure and Velocity contour around the AUV at 3m/s
CONCLUSIONS
In this paper, 3D modelling of an AUV shape, auto-meshing for
element generation, and CFD analysis have been conducted using a
commercial CFD program. As a result of the CFD analysis, pressure
and velocity distribution around the AUV, and drag force were
obtained. A CFD analysis was also carried out for finding the optimum
value of a design value, i.e., the nozzle angle, based on given
constraints. In summarizing the results of this study, we conclude the
following,
(1) In comparison with conventional methods, precise and reliable CFD
results were obtained and verified by the ‘ITTC 1957 correlation line’.
The results showed that the CFD method can be employed for
estimation of the total resistance, even though the shape of the vehicle
is complex.
(2) Auto mesh generation with boundary layer inclusion can be used for
faster convergence. The convergence time can be highly reduced by
monitoring the object function and controlling the iteration times.
702
(3) Two possible optimum design methods, ‘Design of Experiment
(DOE)’ and ‘Direct Searching Method (What-if)’, have been
researched and employed for finding the optimum design value (nozzle
angle) with the minimum value of the object function (drag force of the
AUV).
Future Work
The velocity and pressure distribution around the AUV and drag
estimation, which are difficult to obtain from an experimental tow tank
test, were obtained by CFD analysis. Similarly, the optimum design
value of the nozzle angle was also obtained using CFD analysis. The
effect of the propulsion force have not however been considered in this
paper.
Relative rotation efficiency (ηR), thrust deduction fraction (t), and wake
factor (w) of the nozzle should be carried out for more reliable
optimization of the shape. A CFD analysis including a rotating
propeller is now being studied, and will be verified by an experimental
test in a tow tank.
ACKNOWLEDGEMENTS
The authors would like thank to Professor Yoshiki Nishi of Yokohama
National University, Professor Wataru Koterayama, and Professor
Masahiko Nakamura of Kyushu University in Japan for their practical,
technical support, and productive international cooperation.
REFERENCES
ANSYS Inc. (2007). "ANSYS-CFX Ver. 11.0 Manual,” ANSYS Ltd,
2007
Hoerner, S.F. (1965). "Fluid-Dynamic Drag", Published by the Author
Lim, SJ et al. (1971). "Principles of Naval Architecture - Korean
version", Dae-hwan text Co., pp 462-472.
Michael, V. Jakuba (2003). “Modeling and Control of an Autonomous
Underwater Vehicle with Combined Foil/Thruster Actuators,”
Massachusetts Institute of Technology, pp. 33-62.
Nishi, Y., Kashiwagi, M., Koterayama, W., Nakamura, M., Samuel
S.Z.H., Yamamoto, I., Hyakudome, T. (2007). "Resistance and
Propulsion Performance of an Underwater Vehicle Estimated by a
CFD Method and Experiment," Proc 17th Annual Int Ocean and
Polar Eng Conf, Lisbon, Portugal, ISOPE, Vol 2.
Phillips, A., Furlong, M., and Turnock, S.R. (2007). "The Use of
Computational Fluid Dynamics to Access the Hull Resistance of
Concept Autonomous Underwater Vehicles." OCEAN '07 IEEE
Aberdeen.
Prestero, T. (2001). "Verification of a Six-Degree of Freedom
Simulation Model for the REMUS Autonomous Underwater
Vehicle", M. Sc. thesis at M. I. T., pp 14-19.
Seo, YK et al. (2005). "Computational Fluid Dynamics," Dong-A
University, pp 187-201.
Wilcox, D.C. (1998), “Turbulence Modeling for CFD”, DCW Industries.