Ship Resistance Simulations with OpenFOAM

Kevin Maki

University of Michigan
Ann Arbor, MI USA

6th OpenFOAM Workshop

13 - 16 June 2011
The Pennsylvania State University
State College, PA, USA

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 1 / 22

Introduction

simulate flow around ship moving steadily in calm water

high Reynolds number 106 model scale, 109 full scale
longest wave is 2πF 2 L
in deep water the Kelvin angle is approximately 19 deg
seek primarily the force on the body, also position of body

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 2 / 22

Outline

1 Introduction

2 Governing Equations
Volume of Fluid
Momentum and Continuity Equations

3 Numerics
interFoam solver

4 Solution Settings

5 Wigley Hull Tutorial

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 3 / 22

Interface tracking versus interface capturing

OpenFOAM has both interface

capturing and interface
tracking solvers
most ship hydrodynamics
solvers use interface
capturing, volume-of-fluid,
level set, or a combination, to
solve ship hydrodynamics
problems.
Can a ship move (at model or
full scale) and not generate a
breaking wave?

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 4 / 22

Volume of Fluid
use scalar indicator function to represent
the phase of the fluid in each cell (was γ
in versions ≤ 1.5, is α in versions ≥ 1.6).

α = α(x, t) (1)
µ(x, t) = µwater α + µair (1 − α) (2)
ρ(x, t) = ρwater α + ρair (1 − α) (3)

The density and viscosity are material

properties of the fluids.

Dα
= 0 (4)
Dt
∂α
+ u · ∇α = 0 (5)
∂t
∂α
+ ∇ · uα = 0 (6)
∂t
Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 5 / 22
Volume-of-fluid with compression
the α function transitions from 1 to 0 over an infinitesimal
thickness. This leads to difficulty in approximating the gradient of
α, and results in smearing of the interface.
One remedy, is to use a modified governing equation. The
modification should return solutions of the original equation for the
time evolution of the interface, but help by keeping the interface
crisp.
∂α
+ ∇ · uα + ∇ · wα = 0
∂t
u is the physical velocity field, and w is an artificial velocity field
that is directed normal to and towards the interface.
∂α
+ ∇ · uα + ∇ · w(α(1 − α)) = 0
∂t
the user can specify the relative magnitude of the artificial velocity
(using cAlpha)
Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 6 / 22
Momentum, dynamic pressure
Full Reynolds-averaged momentum equations for the velocity U
and pressure P in a fluid with density ρ and dynamic viscosity µ

∂ρU h i
+ ∇ · UU = −∇P + ρg + ∇ · (µ + µt )(∇U + ∇U> )
∂t
Express the pressure in terms of a hydrostatic component, and
the remainder or that due to dynamic or non-zero velocity p

P= ρg · x + p
| {z } |{z}
hydrostatic dynamic

Governing equation in terms of dynamic pressure

∂ρU h i
+ ∇ · UU = −∇p − g · x∇ρ + ∇ · (µ + µt )(∇U + ∇U> )
∂t

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 7 / 22

Momentum, viscous stress

See Henrik Rusche’s Thesis, pg 156

h i
∇ · µeff (∇U + ∇U> ) = ∇ · (µeff ∇U) + ∇ · (µeff ∇U> )
= ∇ · (µeff ∇U) + ∇U · ∇µeff + µeff ∇(∇ · U)
= ∇ · (µeff ∇U) + ∇U · ∇µeff

Final form of the momentum equation:

∂ρU
+ ∇ · UU = −∇p − g · x∇ρ + ∇ · (µeff ∇U) + ∇U · ∇µeff
∂t
Note, ∇ρ is zero away from the interface, and VERY large along
the interface.

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 8 / 22

Boundary conditions
Body

U = 0
∂p/∂n = 0
∂α/∂n = 0

Inlet

U = U∞
Centerplane: ∂p/∂n = 0
symmetryPlane 
1 if z < 0
Top α =
0 otherwise
∂U/∂n = 0 Outlet
p = 0
∂U/∂n = 0
α = 0
p = 0
∂α/∂n = 0
Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 9 / 22
 simulation.

interFoam
VOF for interface capturing
PISO for pressure velocity coupling
unknowns: z
p_rgh p dynamic pressure
y
p P total pressure (P = p + ρg · x)
Space domain
alpha1 α volume
x fraction
U U velocity vector
phi Sf · Uf velocity flux
rhoPhi Sf · ρf Uf mass flux ∆t t

gh g · xP hydrostaticTime pressure
domain over density at cell center
ghf g · xf Figurehydrostatic
2.1: Discretisationpressure
of the solution over
domain density at face center

P Sf

d N

Maki (UofM) Figure 2.2: Parameters

Training in finite
Session: Ship volume discretisation 6th OpenFOAM Workshop
Resistance 10 / 22
interFoam algorithm

1 solve transport equation for volume fraction

2 generate linear systems for momentum components U, V , W ,
using convection and viscous terms only
3 (optional) solve for momentum components using old values of
pressure gradient and density gradient
4 form the pressure Poisson equation, and solve (may loop over this
for non-orthogonal correction update)
5 update velocity with pressure gradient
6 update face flux with pressure contribution
7 update turbulence quantities

PISO loop over steps 4-6.

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 11 / 22

Momentum prediction
Total momentum equation:

∂ρU
+ ∇ · UU = −∇p − g · x∇ρ + ∇ · (µeff ∇U) + ∇U · ∇µeff
∂t
in prediction, form linear systems using convection and viscous terms
only:

∂ρU
+ ∇ · UU − ∇ · (µeff ∇U) − ∇U · ∇µeff = 0
∂t

[A]U {U} = {bU }
[A]V {V } = {bV }
[A]W {W } = {bW }

if you “solve” for momentum prediction:

−1
{U} = [A]U · [{bU } − ∇p · i − g · x∇ρ · i]

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 12 / 22

Pressure correction
start with semi-discrete momentum equation

[A]U {U} = [{bU } − ∇p · i − g · x∇ρ · i]

look at equation for a single cell
X
a P UP + aN UN = bP − ∇p − g · x∇ρ

calculate the velocity without ∇ρ and ∇p

X
U?P = aP−1 (bP − aN UN )

interpolation of gradients is bad! (Rhie-Chow). Face flux using starred

velocity
φ? = U?f · Sf
now the flux with the density gradient:

∂ρ −1
φ0 = φ? − g · xf a |Sf |
∂n P,f

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 13 / 22

Pressure correction, cont.
use the continuity equation to find pressure that makes the velocity
discretely divergence free.
X X
∇·U= Uf · Sf = φ=0

φ0 will not satisfy continuity because it is a numerical approximation, and

it does not contain the pressure gradient term. Return to the momentum
equation for a single cell, and note the use of the starred velocity.

UP = U? − aP−1 ∇p − aP−1 g · x∇ρ

insert into continuity X
∇ · aP−1 ∇p = φ0
after solving for p, then update the face flux and velocity
−1
φ = φ? − ∇ · aP,f ∇pf
U = U? − g · xf − ∇p

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 14 / 22

Time-step size

Courant number in simple terms:

∆t
Co = U
∆x
For arbitrary polyhedral finite volume:

U · Sf
Co = ∆t
d · Sf

d is the vector from pole center to neighbor center

if we solve implicit equations, what is an acceptable time step size
based on the Courant number?

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 15 / 22

PISO-settings
momentumPredictor: relatively small additional expense →
recommended
nCorrectors: this is to loop over pressure system, also known as
PISO loops. For strict time accuracy, minimum of 2. Calm-water
resistance, 1 should do.
nNonOrthogonalCorrectors: due to small time step, and use of
nCorrectors, this may be set to 0 in most cases. Perhaps for initial
time steps on bad grids a few may help.
nAlphaCorr: loop over α equation. For time-dependent flows 1-2. For
steady flow like calm-water resistance, 0.
nAlphaSubCycles: this reduces the time-step size for the explicit
integration of the α transport equation. As you increase the time-step
size for the total system of equations, and if you need time accuracy
(maybe retain stability), increase the number of sub-cycles according.
cAlpha: the compression term in the advection of α is scaled by this
parameter. Set to zero to deactivate compression. Set to 1 as a nominal
value.

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 16 / 22

Discretization Settings

time: Euler, Courant number restriction leads to small time steps, first
order accuracy is fine for calm-water resistance.
gradient: linear
divergence: upwind to aid in convergence. vanLeer is second-order
away from extrema. limitedLinearV may be less diffusive than
vanLeer
Laplacian: Gauss linear corrected. Second-order, with correction
for non-orthogonal part.

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 17 / 22

Wigley Hull Experiments

Well used test data. Body fixed and free to sink and trim. SRI
0.08 < F < 0.40
2 × 106 < R < 1 × 107

Item Symbol Value Unit

Length L 4.0 m
Beam B 0.4 m
Draft T 0.25 m
Wetted Surface S 2.3796 m2

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 18 / 22

 +08$ *0++45+2$ 10-97$ :$ 7066*$

Wigley Hull Computations:

Table 5: Resistance coefficients Time Integration
results for experimental values

As it can be seen from Table 4 and 5, the numerical approximations using the CFD solver are in agreement with the
experimental results. Especially the Fine grid computations for Fr# = 0.316 caught 100% convergence with experimental
value.
144K cell coarse grid (Pointwise)

Figure 2: Comparison of Total Resistancecourtesy

Coefficient Results
of Mert with Experimental Value for Fr# = 0.316
Türkol

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 19 / 22

Wigley Hull Computations: Convergence

courtesy of Ensign William Garland

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 20 / 22

Wigley Hull Computations: Full Scale
400 m Ship

courtesy of Ensign William Garland

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 21 / 22
Hull Force Library

control over quantity calculated and the write syntax

Z
Fp = pndS
ZS
Mp = (xf − xo ) × pndS
ZS
Fv = τ̄¯ · ndS
S
Z
Mv = (xf − xo ) × τ̄¯ · ndS
S

column 1:time, 2-4: Fp , 5-7: Fp , 8-10: Mv , 11-13: Mv

Maki (UofM) Training Session: Ship Resistance 6th OpenFOAM Workshop 22 / 22