Professional Documents
Culture Documents
LSTC-LS-DYNA-ICFD-THE-1.1-1 i
Document Information
Confidentiality external
Document Identifier LSTC-LS-DYNA-ICFD-THE-1.1-1
Prepared by LSTC
R
Iñaki Çaldichoury, LSTC
R
Rodrigo R. Paz
Approved by LSTC
R
Facundo Del Pin, LSTC
R
Rodrigo R. Paz
Number of pages 61
Date created Wednesday 8th October, 2014
Distribution External
ii LSTC-LS-DYNA-ICFD-THE-1.1-1
Contents
1 Introduction 1
1.1 Purpose of this Document . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Document Information 2
LSTC-LS-DYNA-ICFD-THE-1.1-1 iii
5.2.3 Spatial discretization . . . . . . . . . . . . . . . . . . . . . . . 25
5.2.4 Coupling with the thermal solver . . . . . . . . . . . . . . . . 26
5.2.5 Watching and controlling the analysis for thermal analysis. . . 27
5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8 Turbulence models 37
8.1 The RANS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
8.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
8.1.2 The Reynolds Averaged Equations . . . . . . . . . . . . . . . 37
8.1.3 The k- model . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
8.2 The LES model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
8.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
8.2.2 LES equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
8.2.3 The Smagorinsky model . . . . . . . . . . . . . . . . . . . . . 40
8.3 Wall function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
8.4 Turbulence synthesis method for Large Eddy Simulations . . . . . . . 41
8.5 The Random Flow Generator in Python: How Inlet turbulence gen-
erator works (basically) in ICFD/LS-DYNA . . . . . . . . . . . . . . 42
8.6 Smirnov’s algorithm to generate inhomogeneous anisotropic turbu-
lence at inflow boundaries in LES runs . . . . . . . . . . . . . . . . . 43
iv LSTC-LS-DYNA-ICFD-THE-1.1-1
9.3 Initial Volume mesh building steps . . . . . . . . . . . . . . . . . . . 50
9.4 Local refinement tools . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9.4.1 Imposing the mesh size on a volume domain . . . . . . . . . . 51
9.4.2 The boundary layer mesh . . . . . . . . . . . . . . . . . . . . 53
9.5 Volume mesh remeshing criteria . . . . . . . . . . . . . . . . . . . . . 53
9.5.1 Mesh deformation criteria . . . . . . . . . . . . . . . . . . . . 53
9.5.2 Using adaptivity . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.6 Watching the mesh quality . . . . . . . . . . . . . . . . . . . . . . . . 55
9.6.1 Mesh quality output file . . . . . . . . . . . . . . . . . . . . . 55
9.6.2 Terminal information . . . . . . . . . . . . . . . . . . . . . . . 56
LSTC-LS-DYNA-ICFD-THE-1.1-1 v
1 Introduction
1.1 Purpose of this Document
A detailed description of the analytical equations solved by the incompressible fluid
solver is given as well as the numerical methods used. Additional descriptions of
the equations used for some of the main features of the solver is also provided. The
objective of this document is to offer the reader who wishes to understand and use
the incompressible fluid solver of LS-DYNA a precise and easy way to understand
insight of the formulas, notions and theories employed by the solver.
LSTC-LS-DYNA-ICFD-THE-1.1-1 1
2 Document Information
2 LSTC-LS-DYNA-ICFD-THE-1.1-1
3 Notations and physical variables
Table 2 provides the meaning of each symbol and the SI unit of measure while table
3 provides the conventions and operators used :
LSTC-LS-DYNA-ICFD-THE-1.1-1 3
4 The ICFD solver
4.1 Fluid Mechanics equations
4.1.1 The incompressibility condition
The present solver aims to solve the incompressible Navier-Stokes equations. Tradi-
tionally, a fluid may be considered incompressible when the Mach number is lower
than 0.3, i.e
V
M = ≤ 0.3, (1)
a
where V is the velocity of the flow relative to a fixed object and a is the speed of
sound in the medium (in air, we thus have v ≤ 370 km/h).
In fluid dynamics, the continuity equation states that, in any steady state process,
the rate at which mass enters a system is equal to the rate at which mass leaves the
system. The continuity equation is analogous to Kirchhoff’s current law in electric
circuits.
The differential form of the continuity equation is
∂ρ
+ ∇ · (ρ~u) = 0. (2)
∂t
If ρ is a constant, as in the case of incompressible flow, the mass continuity equation
simplifies to a volume continuity equation
∇ · ~u = 0. (3)
4 LSTC-LS-DYNA-ICFD-THE-1.1-1
For near incompressible flows we have that
∂ui ∂ui
, (7)
∂xi ∂xj
and the last term in the brackets may be neglected from Equation (6). Then,
∂ui ∂uj
σij ≈ −pδij + µ + . (8)
∂xj ∂xi
In the same way, the term ∂σij /∂xj in the momentum equation may be simplified
for near incompressible flows as
∂σij ∂p ∂ ∂ui ∂uj
=− δij + µ +
∂xj ∂xj ∂xj ∂xj ∂xi
∂p ∂ ∂ui ∂ ∂uj
=− δij + µ +µ
∂xj ∂xj ∂xj ∂xj ∂xi
(9)
∂p ∂ ∂ui ∂ ∂uj
=− δij + µ +µ
∂xj ∂xj ∂xj ∂xi ∂xj
∂p ∂ ∂ui
≈− δij + µ .
∂xj ∂xj ∂xj
The above expression simplifies the viscous term so that we end up with the following
system
∂ 2 ui
∂ui ∂ui ∂p
ρ + uj = − +µ + ρfi in Ω (10)
∂t ∂xj ∂xi ∂xj ∂xj
∂ui
= 0 in Ω (11)
∂xi
LSTC-LS-DYNA-ICFD-THE-1.1-1 5
where v i indicates the function that is imposed on the boundary.
The other type of boundary conditions are imposed on the free surface. These are
the most difficult to identify as the shape of the free surface has to be computed
ahead. The principles that form the basis for the free surface boundary conditions
are stated as follow
• stress tangential to the surface must vanish,
• stress normal to the surface must exactly balance any externally applied normal
stress.
In this case these conditions are expressed as
ti = σij nj = ti on Γf , (13)
where nj is the direction cosine of the outward normal on the boundary with respect
to the xj axis. Both surfaces Γv and Γf are two disjoint non-overlapping subsets of
the boundary Γ.
The set of initial conditions for the Lagrangian Navier-Stokes problem will be given
specifying the velocity and pressure at the initial time
vi (xi , 0) = vi0 (xi ), (14)
p(xi , 0) = p0 (xi ), (15)
where the initial velocity vi0 has to satisfy the incompressibility constrain ∂vi /∂xi = 0.
6 LSTC-LS-DYNA-ICFD-THE-1.1-1
4.2 The Fractional Step method
4.2.1 Introduction
For the time integration of the Navier-Stokes equation a projection method was
adopted ([6], [7], [27]).These methods stand out from classical monolithic schemes
as pressure and velocity are uncoupled. Thus four linear systems of equations are
obtained, namely three for the momentum equation and one to solve the incompress-
ibility constrain.
The present scheme consists of three steps until the final velocity is achieved ([25]).
In the first step, a predictor velocity u∗i is computed. This velocity does not satisfy
the incompressibility constrain given by Equation (5). Thus, in the second step the
velocity ui is projected into a space of divergence free vector field and a Poisson
equation of pressure is obtained. This pressure will correct ui to get a final step
divergence free velocity un+1
i . The final step consists of moving the particles to their
n+1
new time step position x . If this is done iteratively then convergence is achieved
when the particles always move to the same final position. This class of schemes are
now widely used in practice and have been rigously analyzed in ([20], [12], [16], [30],
[32], [10], [8]).
dui 1.5un+1
i − 2uni + 0.5un−1
i
≈ (18)
dt ∆t
However, for clarity purposes, the following demonstrations will use simple forward
differences approximation keeping in mind that the procedure is similar with back-
ward differences approximation
n+θ
un+1 − u?i + u?i − uni
i ∂ n+1 n n ∂ ∂ui ∂ui
ρ = −p + γp − γp δij + µ − ρuj + ρfi (20),
∆t ∂xi ∂xj ∂xj ∂xj
LSTC-LS-DYNA-ICFD-THE-1.1-1 7
where u?i are fictitious variables called fractional velocities and γ is a scalar value
that varies from 0 to 1 and that will be explored later. The value of θ = 0 implies
an explicit forward Euler scheme, θ = 1 is an implicit backward Euler scheme and
θ = 0.5 is the Crank-Nicholson second order semi-implicit scheme. In the present
analysis θ = 1 will be adopted. We can see that the pressure term has been taken out
from the implicit-explicit term [·]n+θ and only the implicit part is taken into account.
Rewriting Equation (20), and reordering some terms, we get
∆t ∂
un+1
i u?i
− + = u?i −p n+1
+ γp δij + uni −
n
ρ ∂xi
n+θ (21)
∆t ∂pn
∆t ∂ ∂ui ∂ui
γ δij + µ + ρfi − ρuj ,
ρ ∂xj ρ ∂xj ∂xj ∂xj
The last three terms of Equation (21) will be related to u?i in such a way that we
may write
8 LSTC-LS-DYNA-ICFD-THE-1.1-1
4.2.3 Equation of Incompressibility
The main objective of this phase is to obtain a pressure field that, when replaced in
Equation (23), satisfies the incompressibility constraint. Once the intermediate ve-
locity u?i has been computed from Equation (24), the end of step velocity is obtained
by adding to u?i the dynamical effect of the still unknown pressure pn+1 , which must
be determined such that the incompressibility constraint from Equation (5) remain
satisfied.
Taking Equation (23) and applying the divergence operator on both sides
∂ n+1 ? ∆t ∂ ∂ n+1 n
(u − ui ) = −p + γp δij . (25)
∂xi i ρ ∂xi ∂xi
By means of Equation (5) the first term on the left hand side vanishes and we may
write
∂ ? ∆t ∂ ∂ n+1 n
(−ui ) = −p + γp δij . (26)
∂xi ρ ∂xi ∂xi
which is a linear elliptic equation of pressure. Solving Equation (26) with the appro-
priate boundary conditions will determine the new time step pressure.
LSTC-LS-DYNA-ICFD-THE-1.1-1 9
4.3 Spatial discretization by the Finite Element Method
4.3.1 The FEM system
Let us first introduce the weak form of the equations of motion that have been already
decoupled in Equations (27), (28) and (29). We will use indistinctly the letter N to
identify either trial functions or interpolating functions. Multiplying the equations
by the trial function N and integrating over the volume Ω:
u?i :
n+1
µ ∂ ∂u?i
Z Z Z
? ? ∂ ūi
Ni ui dΩ − ∆t Ni dΩ + ∆tNi uj dΩ =
Ω Ω ρ ∂xj ∂xj Ω ∂xj
(30)
Z Z n Z
Ni ∂p
Ni uni dΩ − γ∆t δij dΩ + ∆t Ni fi dΩ,
Ω Ω ρ ∂xi Ω
pn+1 : Z Z
∂ N ∂ ∂
N (−u?i )dΩ = ∆t −p n+1 n
+ γp δij dΩ, (31)
Ω ∂xi Ω ρ ∂xi ∂xi
un+1
i :
Z Z Z
Ni ∂
Ni un+1
i dΩ = Ni u?i dΩ + ∆t −p n+1 n
+ γp δij dΩ. (32)
Ω Ω Ω ρ ∂xi
where the parameters ∆t and γ are considered constant during the integration.
Integrating by parts the stress term in Equation (30) and the pressure term in (31),
use of the divergence theorem and the boundary conditions of Equation (13) and
(Equation 12) let us arrive to the following set of equations:
u?i :
µ ∂Ni ∂u?i ∂ ūn+1
Z Z Z
?
Ni ui dΩ + ∆t dΩ + ∆tNi u?j i dΩ =
Ω Ω ρ ∂xj ∂xj Ω ∂xj
(33)
n
Z Z Z Z ?
N i ∂p ∂u i
Ni uni dΩ − γ∆t δij dΩ + ∆t Ni fi dΩ + Ni nj dΓ,
Ω Ω ρ ∂x i Ω Γ ∂x j Γf
pn+1 :
Z Z
∂N ? 1 ∂N ∂ n+1 n
ui dΩ = ∆t p dΩ − γp δij +
Ω ∂xi Ω ρ ∂xi ∂xi
Z (34)
? ∆t ∂ n+1 n
Ni ui + (−p + γp )δij ni dΓ,
Γ ρ ∂xi Γv
10 LSTC-LS-DYNA-ICFD-THE-1.1-1
un+1
i :
Z Z Z
Ni ∂
Ni un+1
i dΩ = Ni u?i dΩ + ∆t −p n+1 n
+ γp δij dΩ, (35)
Ω Ω Ω ρ ∂xi
where ni is the outward normal vector to the Γv boundary. Comparing the boundary
term in Equation (34) and the end of step velocity (Equation (35)) we may write:
Z Z
∆t ∂
?
N ui + (−p n+1
+ γp )δij ni dΓ = N [ũn+1
n
i ]Γv ni dΓ. (36)
Γ ρ ∂x i Γv Γ
To complete this set of equations we need to specify the essential and natural bound-
ary conditions for Equation (33) and Equation (34). In the first case the derivative
of the velocity normal to the free surface will vanish resulting in the following natural
boundary condition
?
∂ui
nj = 0. (37)
∂xj Γf
Also free-slip or no-slip velocity conditions over the walls may be imposed. In the
second case there will be two sets of boundary conditions namely essentials over Γf
and Γv , we have
p=0 on Γf , (38)
ũn+1
i
n+1
= ui on Γv , (39)
which means that the value of pressure will be imposed over the free surface and the
normal velocity of the fluid over particles with a given known velocity will take the
known normal velocity of the particles.
The full version of the discrete problem takes place when the unknown fields are
approximated over the finite elements. Then within an element, the velocity and
pressure will be represented by
ui = Nl uli , (40)
p = Nl pl , (41)
where l represents a node of the element. Replacing Equation (40) and Equation
(41) in Equation (33) and (35) a final set of equations in a more compact form is
obtained
LSTC-LS-DYNA-ICFD-THE-1.1-1 11
ρMu?i + ∆tK(µ)u?i + S(ρūn+1
i )u?j = ρMuni − γ∆tGpn + ∆tF, (42)
1 n+1 ? 1 n
∆tL p = Dui + γ∆tL p − Ũ, (43)
ρ ρ
ρM̄un+1
i = ρM̄u?i − ∆tG(pn+1 − γpn ), (44)
Z
M= Nqn+1 Nln+1 dΩ, (45)
Ω
∂Nqn+1 ∂Nln+1
Z
K(µ) = µ dΩ, (46)
Ω ∂xi ∂xi
∂Nqn+1 n+1
Z
D= Nl dΩ, (47)
Ω ∂xi
Z n+1
n+1 ∂Nl
G= Nq dΩ, (48)
Ω ∂xi
1 ∂Nqn+1 ∂Nln+1
Z
1
L= dΩ, (49)
ρ Ω ρ ∂xi ∂xi
Z
F= Nqn+1 fi dΩ, (50)
Ω
Z
n+1 ∂Nl
S(ρui ) = ρNq ūi dΩ, (51)
Ω ∂xi
Z
Ũ = Nqn+1 un+1
i ni dΓ. (52)
Γ
The indices (q, l) denote the nodes of an element and n + 1 implies that the interpo-
lating functions have been computed at time step tn+1 . The matrix M̄ in Equation
(44) is actually the lumped mass matrix M of Equation (45) obtained by adding the
elements of M over a row and placing the result on the diagonal of the matrix.
Equations (42),(43) and (44) form the system that needs to be solved.
12 LSTC-LS-DYNA-ICFD-THE-1.1-1
4.3.2 Predictor corrector scheme
In the previously defined system of equations it can be observed that in order to be
able to solve Equation (42), an estimation of ūn+1
i needs to be given. The choice of
n+1
ūi can be the object of discussions as it needs to be the closest possible to the
actual value of un+1
i . An estimated average value of : ūn+1
i = 2uni − uin−1 is being
used by default for the first solve of the system. However, the solver also provides an
option to trigger an iterative procedure on the system. Once u∗ and pn+1 have been
determined using Equation (42) and (43), they can be re-injected in the system in
order to form an iterative loop
where k is the number of iterations (the vector indices do not show for clarity pur-
poses).
This procedure aims to bring more precision to the final pressure and velocity values
but is often very time consuming. It must be therefore used with caution in specific
cases. For stability purposes, this method is authomatically used for the first ICFD
time step where iterations are conducted. Otherwise, it can be accessed through the
ICFD CONTROL SPLIT card (see Keyword manual).
LSTC-LS-DYNA-ICFD-THE-1.1-1 13
4.4 Integration scheme stabilization
4.4.1 Pressure stabilization
The system formed by Equations (42), (43) and (44) can be solved as such but it
can be shown that stability issues arise [8] depending on the time step size. It can be
anticipated that if ∆t is very small, stability problems may occur, especially for the
second order scheme γ = 1 used by the solver. The orthogonal subscale stabilization
presently used has been studied by [9].
Let’s start by observing that the DM−1 G represents the Laplacian operator at the
continuous level. When using discrete elements, this leads to the approximation
where τ is a stabilization parameter which depends on the local element sizes and
behaves as
h2
τ ≤C , (61)
ν
where C is a constant.
This stabilization term is consistent with the approximation introduced by Equation
(55) which means that the stabilization term will get smaller when using a finer mesh
i.e when getting closer to the hypothetical continuous level.
14 LSTC-LS-DYNA-ICFD-THE-1.1-1
4.4.2 Convection stabilization
The idea behind the stabilization of the convection term is similar to the pressure
stabilization and has been developed in [9]. It consists of adding a low order stabi-
lization term and recover the higher order of the advection operator by subtracting
a higher order term. This term is built by projecting the velocity orthogonal to the
finite element space. We will only give the final system here that will consist of
with
Z
∂Nq n ∂Nl n
Su = u u dΩ, (67)
Ω ∂xi i ∂xi i
Z
∂Nq n
Sy = u Nl dΩ, (68)
Ω ∂xi i
Z
n ∂Nl
C= Nq ui dΩ, (69)
Ω ∂xi
LSTC-LS-DYNA-ICFD-THE-1.1-1 15
4.5 Watching and interpreting the analysis
We have seen in the previous sections that the fractional step method consists in
segregating the velocity and pressure terms thus resulting in Equation (42) and (43)
that must be solved implicitly before finally updating the velocity using Equation
(44). When running a CFD only analysis, it is possible for the user to track the
solving of these systems through the terminal window. Assuming that the maximum
level of output information defined in the ICFD CONTROL OUTPUT card has been
triggered, this kind of information extracted from a 2D case, will appear:
After each ICFD time step, the solver will give the number of iteration that were
needed in order to solve Equation (42) in the three directions of velocity (two in
2D cases) as well as the associated residual. The solver will then proceed to solve
Equation (43) and will also output the number of iterations as well as the associated
residual. Finally, the solver will give the number of iterations needed in order to
solve Equation (44) in the three directions (two in 2D cases) as well as the associated
residual. The solver will then proceed to the next ICFD time step. This way, the
user can make sure that the residuals stay low at each time step and that the analysis
is not diverging.
16 LSTC-LS-DYNA-ICFD-THE-1.1-1
4.6 Non-Inertial Reference Frame
The Navier-Stokes equation for incompressible flows may be modified to allow the
simulation in a moving reference frame. The reference frame considered here is
rotating at a constant speed and thus it is undergoing acceleration with respect
to the stationary frame. The rotating frame is then called a non-inertial reference
frame. This feature is useful in application such as wind turbines, turbomachinery
and every other rotating problem that admits a steady state solution.
To account for the rotating velocity the fluid velocity is transformed
va = vr + vΩ ,
with
vΩ = Ω × r
where va is the absolute fluid velocity (the velocity viewed from the stationary frame),
vr is the relative velocity (the velocity viewed from the rotating frame) and vΩ is the
velocity of the rotating frame, Ω is the angular velocity (considered constant in this
analysis) and r is the vector from the axis of rotation.
Apart from the velocity transformations a rotating frame induce forces due to the
Coriolis and centrifugal accelerations which are given by −2Ω × v and −Ω × Ω × r
respectively.
Depending where the observer is placed the Navier-Stokes equations may be written
in a relative form or in absolute form.
For the absolute form:
∂v ∂v ∂p ∂v
ρ( + (v − vΩ ) )=− + µ 2 − ρ(Ωi × v) + Fi
∂t ∂xj ∂xi ∂xj
In the relative reference frame the equations are re-written in terms of vr
∂vr ∂vr ∂p ∂v
ρ( + vr )=− + µ 2 − ρ(2Ωi × vr + Ωi × Ωi × ri ) + Fi
∂t ∂xj ∂xi ∂xj
The equations above are valid for steady state solutions within the rotating frame.
Nevertheless these equations can also be applied to unsteady problems with rea-
sonable results. The equation expressed in absolute reference frame is of particular
interest since it allows the definition of rotating sub-domains.
LSTC-LS-DYNA-ICFD-THE-1.1-1 17
5 Structure and Thermal coupling
5.1 Fluid Structure Interaction (FSI)
5.1.1 Types of FSI coupling
Fluid-structure interaction problems involving an incompressible viscous flow and
elastic non linear-structure have been solved in the past using different methods.
The monolithic approach considers the fluid and the solid as a single domain with
the fluid and solid equations solved together in a coupled way. However, solving the
pressure together with the rest of the unknowns (typically velocities or displacements)
is too expensive from the computational point of view : the non-linear system to be
solved is large and ill-conditioned with always non-defined positive matrices.
A second approach would be to segregate the pressure from the velocity in the mono-
lithic scheme. This would still imply that the fluid and solid equations are solved in
a coupled way in the same system and would therefore require an implementation
of the solid equations directly in the ICFD solver. This would not be a practical
solution and would contradict the objective of the present solver which is to make
fully use of LS-DYNA’s mechanical solver capabilities in order to solve complex fluid
structure interaction problems.
A third approach would be by using a partitioned (or staggered) method ([15], [13],
[26], [22]) where the fluid and solid equations are uncoupled and that therefore allows
using specifically designed codes on the different domains and offer significant benefits
in terms of efficiency: smaller and better conditioned subsystems are solved instead
of a single problem. It is the method adopted by the present solver in order to solve
FSI problems.
In the partitioned approach, two schemes are distinguished : loosely (or weakly) ([21])
coupled scheme or strongly coupled scheme ([28], [29], [31], [33], [14], [35], [11]). Both
are available when using the ICFD solver in LS-DYNA. Loosely coupled schemes
require only one solution of either field per time step in a sequentially staggered
manner and are thus particularly appealing in terms of efficiency. However, they
tend to become unstable when the “added-mass effect” is significant([5], [3]).
In fluid mechanics, added mass or virtual mass is the inertia added to a system
because an accelerating or decelerating body must move some volume of surrounding
fluid as it moves through it, since the object and fluid cannot occupy the same
physical space simultaneously. However the name “added-mass effect” has been
used in the literature to indicate the numerical instabilities that typically occur in
the internal flow of an incompressible fluid whose density is close to the structure
density. Other causes for instabilities include the elasticity coefficients or a time
18 LSTC-LS-DYNA-ICFD-THE-1.1-1
step too small. The added mass effect therefore does usually not occur in aero-
elasticity problems as the solid’s density is a lot higher than the fluid’s density, but
it becomes very important in several other applications such as bio-mechanics where
the materials are normally muscles and arteries and the fluid is blood.
Strongly coupled schemes require the convergence of the fluid and solid variables at
the interface and give, after an iterative process, the same results as non-partitioned
schemes. However, they are also subject to the added mass effect resulting in a
non convergence of the solution. Special stabilization techniques must therefore be
developed in order to diminish its influence and aim for a wider range of applications
for fluid structure interactions. This stabilization method will be developed in the
next section. Figure (1) gives a summary of the different FSI couplings possible,
those implemented in the ICFD solver and the influence of the added mass effect.
Figure 1: Summary of the different FSI couplings and the numerical troubles induced
by the added-mass effect.
LSTC-LS-DYNA-ICFD-THE-1.1-1 19
Laplace equation is built. It can be shown that this procedure greatly improves the
convergence of the FSI coupling. It should be noted that this is still a partitioned
approach and convergence may still be challenging for some kind of problems.
−1
ρsolid ∆tλ ∆tG
τs = + + , (74)
∆t Jh2 Jh2
Z
e ∂Nsf ∂Nsf
L̄ = dΩ (75)
Ω ∂xj ∂xj s
e
where λ and G are Lame parameters, J the Jacobian matrix and L̄ is the Laplace
matrix of the solid elements evaluated only with the shape function Nsf that are
different from zero on the fluid-structure interface.
Equation (70) may also be written as
20 LSTC-LS-DYNA-ICFD-THE-1.1-1
τs
β= , (77)
τf
and
e
X
L̄τf = (τf L̄ ), (78)
This means that the Laplace interface matrix L̄τf may be neglected for small values
of the β parameter. This is for instance the case when ρs >> ρf and the added-mass
effect is not present. However, for other physical properties, the β parameter may
not be negligible and the Laplace interface matrix must be evaluated in order to
obtain good results. It has been shown ([17]) that the β L̄tauf term improves the
convergence when ρf ≈ ρs , small time steps and when using soft materials.
LSTC-LS-DYNA-ICFD-THE-1.1-1 21
Loose FSI coupling
22 LSTC-LS-DYNA-ICFD-THE-1.1-1
Figure 3: Strong FSI interaction resolution scheme.
LSTC-LS-DYNA-ICFD-THE-1.1-1 23
5.1.5 Watching and controlling the analysis for FSI cases
We have seen in the previous sections that two FSI couplings were available for
the user. In cases of strong FSI coupling and assuming that the maximum level of
output information defined in the ICFD CONTROL OUTPUT card has been trig-
gered, some information about the Newton loop and the convergence of the coupled
structural and fluid systems can be extracted
As described in Figure (3), after the first time step, the solver will first solve the
fluid equations. The forces computed will then be communicated to the solid me-
chanics solver. The three averaged components appear next to T Force along with a
residual value. The solid mechanics solver will then return the node displacements
as boundary conditions for the fluid solver and the procedure will be repeated until
convergence has been reached ([17]).
24 LSTC-LS-DYNA-ICFD-THE-1.1-1
5.2 Thermal coupling
5.2.1 Introduction
Heat transfer is a discipline of thermal engineering that concerns the generation, use,
conversion, and exchange of thermal energy and heat between physical systems. The
ICFD solver offers the possibility to solve and study the behavior of temperature
flow in fluids. Potential applications are numerous and include refrigeration, air
conditioning, building heating, motor coolants, defrost or even heat transfer in the
human body. Furthermore, the ICFD thermal solver is fully coupled with the thermal
solver using a monolithic approach which allows to solve complex problems where
both heated structures and flows are present and interact together.
∂T ∂T ∂ 2T
+ uj −α =f in Ω (79)
∂t ∂xj ∂xj xj
where α is a positive constant called thermal diffusivity and f is a potential source
of heat.
This formulation is incomplete if the appropriate set of boundary conditions and
initial conditions is not specified. The user can specify the temperature on the
boundaries resulting in Dirichlet conditions
~ (x, t) = 0 on Γ
~n(~x).∇T (81)
where ~n(~x) is the normalized normal vector at the ~x point.
LSTC-LS-DYNA-ICFD-THE-1.1-1 25
1.5T n+1 − 2T n + 0.5T n−1 ∂T n+1 ∂ 2 T n+1
+ un+1
j − α = f n+1 (82)
∆t ∂xj ∂xj xj
The Galerkin method will now be applied to Equation (82) in order to find a weak
form that will be integrated over the prescribed domain
∂T n+1
Z Z
n+1 n n−1
Ni 1.5T − 2T + 0.5T dΩ + ∆tNi uj dΩ−
Ω Ω ∂xj
(83)
∂ ∂T n+1
Z Z
− ∆t Ni α dΩ = ∆t Ni f n+1 dΩ,
Ω ∂x j ∂x j Ω
n+1 n n−1
M 1.5T − 2T + 0.5T + ∆tS(uni )T n+1 − ∆tK(α)T n+1 = ∆tF, (84)
M(1.5T n+1 − 2T n + 0.5T n−1 ) + ∆tS(uni )T n+1 − ∆tK(α)T n+1 − τ (Sy (uni )Tproj
n − Su (uni )T n+1 ) = 0,
(85)
n
MTproj − C(uni )T n = ∆tF, (86)
26 LSTC-LS-DYNA-ICFD-THE-1.1-1
Figure 4: Vector of temperature unknowns when the fluid thermal solver and struc-
ture thermal solvers are coupled using a monolithic approach.
5.3 Summary
Figure (5) offers a simplified view of the interactions and couplings of the mechanical,
thermal and ICFD solvers.
LSTC-LS-DYNA-ICFD-THE-1.1-1 27
6 Free surface and mutliphase handling
6.1 The level set method
6.1.1 Introduction
Let us first start by introducing the implicit function φ whose zero isocontour, φ = 0
represents the interface. The implicit function φ is defined throughout the whole
computational domain while the isocontour defining the interface is one dimension
lower. As a convention the fluid domain where the Navier-Stokes equations will be
solved (see Section (4.1)) is defined by φ > 0 while the vacuum is defined by φ < 0.
Additionally, φ will be described as a distance function resulting in the level set
function
~ = 1,
|∇φ| (90)
This is shown in Figure (6) with the classical dam example where the fluid (φ > 0)
and the vacuum (φ < 0) are divided by the interface φ = 0. In the case of two phase
fluids, the problem is similar except that the Navier-Stokes equations are solved in
the two domains with a smoothing of the density and viscosity values at the interface.
28 LSTC-LS-DYNA-ICFD-THE-1.1-1
6.1.3 The convection equation
Now, let us suppose that the velocity of each points on the implicit surface is given
as V~ (~x) i.e V~ (~x) is known for every point ~x with φ(~x) = 0. The simplest way to move
all the points on the surface with this velocity would be to move the nodes located on
the interface by their velocity V through the grid (see Figure (6)). This Lagrangian
formulation would not be too hard to accomplish ([24]) if the connectivity does not
change and the surface elements are not distorted too much. However, even small
velocity fields could cause large distortion of the interface elements and the accuracy
of the method can deteriorate quickly if a frequent and regular re-meshing of the
domain is not applied. This need for frequent re-meshing implies higher computer
costs and less scalability when running with multiple processors.
In order to avoid these problems, the implicit function φ will be used both to represent
the interface and to evolve the interface (see Figure (6)). In order to define the
evolution of the implicit function φ, the simple convection equation will be used
∂φ ~ ~
+ V .∇φ = 0, (91)
∂t
Equation (91) defines the motion of the interface where φ(~x) = 0. It is an Eulerian
formulation of the interface evolution ([24]) since the interface is captured by the
implicit function φ as opposed to being tracked by the displacement of the interface
nodes as was done in the Lagrangian formulation.
As in Section (4.3), the Galerkin method will be applied to Equation (91) in order
to find a weak form that will be integrated over the prescribed domain
∂φn+1
Z Z
i
Ni 1.5φn+1
i − 2φni + 0.5φn−1
i dΩ + ∆tNi unj dΩ = 0, (92)
Ω Ω ∂xj
resulting in the matrix form
n+1 n−1
n
M 1.5φi − 2φi + 0.5φi + ∆tS(uni )φn+1
j = 0, (93)
LSTC-LS-DYNA-ICFD-THE-1.1-1 29
Dam problem : Initial state
Figure 6: Free surface dam breaking problem : description of the different interface
tracking methods
30 LSTC-LS-DYNA-ICFD-THE-1.1-1
n+1 n n−1 n n+1 n n n n+1
M 1.5φi − 2φi + 0.5φi + ∆tS(ui )φj − τ Sy (ui )ψj − Su (ui )φj = 0,
(94)
LSTC-LS-DYNA-ICFD-THE-1.1-1 31
7 Anistropic/Isotropic Generalized Flow Through
Porous Media
7.1 Generalized Porous Media Model: isotropic media
A generalization of the Navier-Stokes equations (see [23]) that will allow the definition
of sub-domains with different permeability/porosity by means of the *ICFD PART
LSDYNA keyword was developed. The OSS|SUPG stabilizing Finite Element Method
for the spatial approximation and the second order Fractional Step Method for the
time integration was adopted.
Being , κ the porosity and the permeability of the porous media respectively, we can
define:
= void volume/total volume, (96)
and being ui the volume averaged velocity field defined in terms of the fluid velocity
field (uif ) as
ui = uif , (97)
the generalized flow equations of momentum and mass conservation can be stressed
as follow (Einstein sum convention is used for i = 1, 2, 3),
∂ui
= 0, (98)
∂xi
2
ρ ∂ui ∂ ui uj 1 ∂(p) µ ∂ ui
+ =− + + ρgi − Di , (99)
∂t ∂xj ∂xi ∂xj ∂xj
where Di are the forces exerted to the fluid by the porous matrix. For the isotropic
model the porous forces are a function of the matrix porosity and its permeability.
For the isotropic model two well-known models are available:
µui 1.75ρ|u|
- Model 1: Ergun correlation Di = +√ √ ui , (100)
κ 150 κ3/2
µui F ρ|u|
- Model 2: Darcy-Forchheimer Di = + √ ui . (101)
κ κ
32 LSTC-LS-DYNA-ICFD-THE-1.1-1
7.1.1 Model Parameters limits
• If the porous media is highly permeable (very high porosity): → 1, κ →
∞ ⇒ Di = 0,
∂ 2 ui
∂ui ∂ ∂p
Navier-Stokes is recovered ρ + (ui uj ) = − +µ + ρgi ,
∂t ∂xj ∂xi ∂xj ∂xj
LSTC-LS-DYNA-ICFD-THE-1.1-1 33
cpeff is the porous media effective specific heat capacity, cpfl the fluid specific heat
capacity, cps the solid (or porous matrix) specific heat capacity, ρfl the fluid density,
and ρs the solid (or porous matrix) density. keff is the porous media effective thermal
conductivity, kfl the fluid thermal conductivity, and ks the solid (or porous matrix)
thermal conductivity. For adiabatic porous media problems cpeff = ρfl cpfl and keff =
kfl .
pb pa
Δx
Figure 7: Experimental Porous Slab
For instance, the P–V experimental data are defined as in Table 4 (see also Fig-
ure (8)).
So that, we can fit that experimental curve with a quadratic polynomial of the form
∆p(u) = αu2x + βux , being ux the velocity along the x-axis. For the example above,
the fitting curve has the following expression:
From the governing equations (98), we can write an expression for the dynamics
of this experiment where pressure gradient is in equilibrium with Darcy and Forch-
34 LSTC-LS-DYNA-ICFD-THE-1.1-1
Table 4: Pressure – Velocity experimental data
u[m/sec] ∆p[N/m2 ]
0 0.0
2. 9.5
8. 134.0
12. 297.0
26. 1371.5
30. 1822.5
34. 2337.5
38. 2916.5
heimer terms (we also neglected the effect of gravity forces). So that,
∂p µ F ρ|u|
= −Di = − ui − √ ui (109)
∂xi κ κ
LSTC-LS-DYNA-ICFD-THE-1.1-1 35
slab probe of ∆x thickness obtaining an expression for ∆p across the probe:
µ F ρ
∆p = −Di ∆x = − ux ∆x − √ u2x ∆x. (111)
κ κ
We recall that for this test along x-axis the ∆p (or the pressure drop) decreases as
∆x increases, so using the experimental data in equation (108) and the theoretical
model in equation (111) we can write (∆p)experimental = (−∆p)theory , so that:
µ F ρ
2u2x + 0.75ux = ux ∆x + √ u2x ∆x (112)
κ κ
From the last equation (eq. (112)) we obtain the following relationships:
µ
0.75ux = ux ∆x (113)
κ
F ρ
2u2x = √ u2x ∆x (114)
κ
that is
µ
0.75 = ∆x (115)
κ
F ρ
2 = √ ∆x (116)
κ
From the first relationship (eq. (113) or eq. (115)) we can solve for the permeability
κ (please note those relationships are independent of the velocity ux ):
µ
κ= ∆x (117)
0.75
being the fluid viscosity µ and the probe thickness ∆x known from the experiment.
From the second relationship (eq. (114) or eq. (116)) we can solve for the Forchheimer
factor F : √
2 κ
F = (118)
ρ∆x
being the fluid density ρ known from the experiment (κ was already computed using
the previous relationship). We recall that the porosity parameter is also known from
the experiment and it is just a geometrical quantity ( = void volume/total volume).
In summary: users can provide the P–V curves from experiments and they will be
used to compute the permeability κ and the Forchheimer factor F . For this purposes,
we will allow the setting of the P–V data using a LOAD_CURVE. So, we will introduce
in LS-DYNA a new porous model to put in the 4th card of the ICFD_MAT keyword
pointing to the load curve in which the experimental data (P–V curve) is defined.
36 LSTC-LS-DYNA-ICFD-THE-1.1-1
8 Turbulence models
8.1 The RANS model
8.1.1 Introduction
Reynolds-averaged Navier-Stokes (RANS) equations are the most classic approach to
turbulence modeling. An time averaged version of the governing equations is solved,
which introduces new apparent stresses known as Reynolds stresses. This adds a
second order tensor of unknowns for which various models can provide different
levels of closure. The incompressible solver provides a k- model which is one of
the most widely used turbulence models in CFD. This model adds two additional
variables, the turbulent kinetic energy k, and the turbulent dissipation and provides
two additional equations that close the model and account for history effects like
convection and diffusion of turbulent energy. Since RANS models consider a time
averaged version of the governing equations, such models will be more adequate
where transient effects are not important and a stationary solution can be reached.
This technique for decomposing the instantaneous motion is referred to as the Reynolds
decomposition. Substitution of Equations (119) and (120) in (10) and (11) yields
Equations (121) and (122) can now be averaged to yield an equation expressing
momentum conservation for the average motion. By noting that the average of a
LSTC-LS-DYNA-ICFD-THE-1.1-1 37
derivative is the same as the derivative of the average and that the time average of
the fluctuating quantity is zero, resulting in the following Reynolds-averaged Navier
Stokes equations (or RANS equations)
∂ ūi ∂(ūi ) ∂(P̄ ) ∂ ∂ ūi 0 0
ρ( + (ūj ) ) = − + µ − ρui uj in Ω, (123)
∂t ∂xj ∂xi ∂xj ∂xj
∂(ūi )
= 0 in Ω. (124)
∂xi
As can be seen from Equations (123) and (124), the time averaged equations are
very similar to the instantaneous equations but for the apparition of a new apparent
viscous stress term, ρu0i u0j owing to the fluctuating velocity field generally referred to
as the Reynolds stresses. The aim of the different RANS models will be, under given
hypotheses, to provide additional equations that allow the solving of the system.
2
ρu0i u0j = 2µt Si,j − ρkδi,j (125)
3
2
k
µt = ρCµ (126)
1 ∂ ūi 1 ∂ ūk
Si,j = − δi,j (127)
2 ∂xj 3 ∂xk
where µt is the eddy viscosity, Si,j is the mean strain rate, k the mean turbulent
kinetic energy and the turbulent dissipation. The k − model then provides the
two mandatory transport equations in order to close the model
∂ρk ∂(ρkūi ) ∂ µt ∂k
+ = (µ + ) + µt S 2 − ρ, (128)
∂t ∂xi ∂xj σk ∂xj
2
∂ρ ∂(ρūi ) ∂ µt ∂
+ = (µ + ) + C1 µt S 2 − C2 ρ , (129)
∂t ∂xi ∂xj σ ∂xj k k
38 LSTC-LS-DYNA-ICFD-THE-1.1-1
with the following default constants that can be changed by the user through the
ICFD CONTROL TURBULENCE card:
Applying this decomposition to Equation (10) yields after filtering the resulting
equation
LSTC-LS-DYNA-ICFD-THE-1.1-1 39
e
∂Uei ∂(U
ei ) ∂ Pe ∂ ∂ Ui
ρ( + (Uj )
e )=− + µ + ρτi,j in Ω (138)
∂t ∂xj ∂xi ∂xj ∂xj
with the extra stress term due to the sub grid scales (SGS)
τi,j = U ej − U
ei U ]i Uj (139)
Subgrid-scale turbulence models usually employ the Boussinesq hypothesis in order
to express the SGS stress as a turbulent viscosity µsgs
e ei
∂Uei ∂(Uei ) ∂ P
e ∂ ∂ Ui ∂ U
ρ( + (U
ej ) )=− + µ + µsgs in Ω (140)
∂t ∂xj ∂xi ∂xj ∂xj ∂xj
40 LSTC-LS-DYNA-ICFD-THE-1.1-1
where A+ is a constant and y + a non-dimensional wall distance for a wall-bounded
flow defined as
u∗ y
y+ = (145)
ν
q
τw
where u∗ = ρ
the friction velocity and τw = µ( ∂u
∂y
) is the shear wall stress.
LSTC-LS-DYNA-ICFD-THE-1.1-1 41
where
x t c l
x̃ = , t̃ = , k̃jn = kjn , c= , (149)
l τ c(j) τ
n
pni = ijm ζjn km n
, qin = ijm ξjn km (150)
ζin , ξin , ωn ∈ N (0, 1); kin ∈ N (0, 1/2), (151)
where l, τ are the length and time scales of turbulence, ijm is the permutation
tensor and N (M, σ) is a normal distribution with mean M and standar devia-
tion σ. kin and ωn represent a sample of n wavenumber vectors and frequencies
of the modeled turbulence spectrum
1/2
2
E(k) = 16 k 4 exp(−2k 2 ). (152)
π
3. Scale and transform vi generated in the previous step to obtain the turbulent
flow field ui
wi = ci vi , (153)
ui = aik wk . (154)
√
The input of this procedure are the turbulent intensities Iu , Iv , Iw (i.e. rij = uavg ·
diag{Iu , Iv , Iw } in the transformed principal axes) and the turbulence lenght scale l.
These are introduced in the LS-DYNA model using ICFD CONTROL TURB SYNTHESIS
keyword in combination to VAD=4 in the card ICFD BOUNDARY PRESCRIBED VEL.
42 LSTC-LS-DYNA-ICFD-THE-1.1-1
Figure 9: Flow around Ahmed body with synthetic turbulent inlet boundary condi-
tions.
In [59]: # u_avg -> mean longitudinal (y axis) wind speed, per node, [m/s]:
u_avg = 23.6
#turbulence intensity Iu != Iv != Iw
Iu = 0.1
LSTC-LS-DYNA-ICFD-THE-1.1-1 43
Iv = 0.1
Iw = 0.1
urms = u_avg*Iu
vrms = u_avg*Iv
wrms = u_avg*Iw
44 LSTC-LS-DYNA-ICFD-THE-1.1-1
In [62]: print "begin simulation..."
un = np.zeros([3,timev.size]) # 3x4000
# initialize seed:
random.seed()
for n in range(0,N):
omegamn = random.gauss(0,1)
knj = normalrandomvec(0,0.5) # 1x3
Zetan = normalrandomvec(0,1) # 1x3
Xin = normalrandomvec(0,1) # 1x3
pni = np.cross(Zetan.transpose(),knj.transpose()) # 1x3
qni = np.cross(Xin.transpose(),knj.transpose()) # 1x3
knjtil[0,0] = knj[0]*c/cn[0]
knjtil[1,0] = knj[1]*c/cn[1]
knjtil[2,0] = knj[2]*c/cn[2]
un[0,:] = un[0,:] +
+ pni[0]*cos(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:]) +
+ qni[0]*sin(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:])
un[1,:] = un[1,:] +
+ pni[1]*cos(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:]) +
+ qni[1]*sin(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:])
un[2,:] = un[2,:] +
+ pni[2]*cos(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:]) +
+ qni[2]*sin(np.inner(knjtil.T,xtil.T) + omegamn*timetil[:])
LSTC-LS-DYNA-ICFD-THE-1.1-1 45
#print(n)
uxt[0,:] = cn[0]*math.sqrt(2./N)*un[0,:]
uxt[1,:] = cn[1]*math.sqrt(2./N)*un[1,:]
uxt[2,:] = cn[2]*math.sqrt(2./N)*un[2,:]
begin simulation...
end simulation...
In [63]: um = max(uxt[0,:])
vm = max(uxt[1,:])
wm = max(uxt[2,:])
subplot(311)
plot(timev,uxt[0,:]/u_avg,color="red",linewidth=1)
axis([0, 3, -um/u_avg, um/u_avg])
title(’Turbulence Syntheis’)
xlabel(’time [secs]’)
ylabel(’$u_x$’,fontsize=20)
grid(True)
subplot(312)
plot(timev,uxt[1,:]/u_avg,color="blue",linewidth=1)
axis([0, 3, -vm/u_avg, vm/u_avg])
xlabel(’time [secs]’)
ylabel(’$u_y$’,fontsize=20)
grid(True)
subplot(313)
plot(timev,uxt[2,:]/u_avg,color="green",linewidth=1)
axis([0, 3, -wm/u_avg, wm/u_avg])
xlabel(’time [secs]’)
ylabel(’$u_z$’, fontsize=20)
grid(True)
# simulated fluctuations
oc = Oct2Py()
46 LSTC-LS-DYNA-ICFD-THE-1.1-1
octave.addpath(’/shang2_2/rpaz/LS-DYNA/DOCUMENTS/IMPLEMENT/SINTESIS/gen_spectru
fff,pf = oc.gen_spectrum(uxt,timev)
# theoretical fluctuations
kmax = 100
k = np.arange(0.1,kmax,0.5) # 400,
#k = f/u_avg
Ek = 16.*math.sqrt(2./math.pi)*(k/u_avg)**4.*(np.exp(-2.*(k/u_avg)**2.))
fig,ax = subplots()
ax.loglog(fff[0,:],pf[0,:],"b",linewidth=1)
ax.loglog(k,Ek,"r",linewidth=1)
ax.axis([1, 1000, 1.e-10, 100])
ax.set_xlabel(’$k$ [1/hz]’,fontsize=20)
ax.set_ylabel(’$E(k)$’, fontsize=20)
ax.set_title(’Spectra’, fontsize=20)
ax.legend(["Simulated Spectrum","Theoretical Spectrum"]);
ax.grid(True)
LSTC-LS-DYNA-ICFD-THE-1.1-1 47
In [64]: subplot(311)
plot(timev,u_avg+uxt[0,:],color="red",linewidth=1)
axis([0, 3, u_avg-um, u_avg+um])
title(’Turbulence Syntheis’)
xlabel(’time [secs]’)
ylabel(’$U_x$ [m/s]’,fontsize=12)
grid(True)
subplot(312)
plot(timev,uxt[1,:],color="blue",linewidth=1)
axis([0, 3, -vm, vm])
xlabel(’time [secs]’)
ylabel(’$U_y$ [m/s]’,fontsize=12)
grid(True)
subplot(313)
plot(timev,uxt[2,:],color="green",linewidth=1)
axis([0, 3, -wm, wm])
xlabel(’time [secs]’)
ylabel(’$U_z$ [m/s]’, fontsize=12)
grid(True)
48 LSTC-LS-DYNA-ICFD-THE-1.1-1
LSTC-LS-DYNA-ICFD-THE-1.1-1 49
9 The Volume Mesher
9.1 Introduction
The ICFD solver uses an automatic volume mesher for the fluid domains. This
greatly simplifies the pre-processing stage. For this feature a good quality body-
fitted surface mesh has to be provided. For FSI simulations, the solver uses an
ALE approach for mesh movement. In the cases where FSI simulations result in
large displacements the solver can automatically re-mesh to keep an acceptable mesh
quality. This section will focus on giving the basic steps of how the volume mesh is
built as well as describe some local refinement and re-meshing tools. The meshing
technique adopted by the solver is inspired by ([18]).
P2
P1
P3
P4
Figure 10: Violation of the Delaunay criterion, the point P4 is inside the circumcircle
of the triangle P1 P2 P3
50 LSTC-LS-DYNA-ICFD-THE-1.1-1
any gaps or open spaces between the surface boundaries. The nodes on the
boundary of two neighbor surfaces have to be uniquely defined (no duplicate
nodes) and should match exactly on the interface. Each surface element has
to be orientable i.e a normal vector defining the interior/exterior faces of the
surface element can be associated to it. If one of those conditions is not met,
the solver will return an error.
• Step two : Using the initial surface nodes, the solver will then join the surface
nodes in order to build an initial volume mesh. The tetrahedras or triangles
in 2D thus created all need to respect the Delaunay criteria.
• Step three : The solver will progressively add nodes to the volume mesh.
Every time a new node is added, its ”host” tetrahedra or triangle will be
divided in smaller tetrahedras or triangles respecting the Delaunay criteria.
This procedure will be repeated until the volume elements have reached their
desired size based on a linear interpolation of the surface sizes that define the
volume enclosure. Figure (11) shows an example of the different steps leading
to the construction of the initial volume mesh.
• For the MESH SIZE SHAPE card, a local element size can be defined by the
user in specific zones corresponding to given geometrical shapes (box, sphere,
LSTC-LS-DYNA-ICFD-THE-1.1-1 51
Step 1
Step 2
Step 3
52 LSTC-LS-DYNA-ICFD-THE-1.1-1
cylinder). This element size will be used as reference in the specified zone when
progressively adding nodes and building the final volume mesh (Step three).
This zone does not necessarily need to be entirely defined in the volume mesh.
However, in cases where a boundary surface mesh is defined in this zone, it
is advised to use a similar element size for the shape zone and the boundary
surface mesh in order to keep a good quality mesh.
Figure (12) shows an example of the two methods and the resulting volume mesh
using the same rectangular shape and element size in both cases.
LSTC-LS-DYNA-ICFD-THE-1.1-1 53
Using the MESH SIZE SHAPE card Using the MESH SIZE card
Step 1 Step 1
Step 2 Step 2
Step 3 Step 3
Figure 12: Locally imposing the mesh size : comparison between the MESH SIZE
card and the MESH SIZE BOX card
54 LSTC-LS-DYNA-ICFD-THE-1.1-1
||e||
η= (155)
||φ||
where φ can be pressure, velocity, temperature or even the level set function and e
can be written as
e = φ − φ̂ (156)
where e is the difference between the exact solution φ and the approximate finite
element solution φ̂. In order to estimate the exact solution, an L2 projection method
will be employed.
For every nodal value, it is therefore possible to associate an error value. As a result,
if too many nodes reach the critical value set by the user, the solver will apply a
remeshing and refine in the zones where the error was high.
LSTC-LS-DYNA-ICFD-THE-1.1-1 55
9.6.2 Terminal information
During the analysis, several additional information is also displayed on the terminal
window in cases of moving mesh or adaptivity if the maximum level of output infor-
mation has been triggered in the ICFD CONTROL OUTPUT card. The following
figure shows the maximum outputs that can appear during the run. Most of this
information is mainly for debugging purposes but it can give some hints on the mesh
quality to users.
The Minimum Jacobian(x) gives some indication on the mesh distortion compared
to its initial position. The Minimum Jacobian(u) should always stay positive as a
negative value means an inversion of an element. The Maximum Derivative gives also
some indication on the mesh distortion, a very large value could mean a very deformed
element that may lead to a non-convergence. The following two Adapt ratios(rat(1),
rat(2), reno) lines only appear in cases where the ICFD CONTROL ADAPT SIZE
or ICFD CONTROL ADAPT cards are being used and give some information on
the nodal and element deformation respectfully. rat(1) symbolizes the percentage of
nodes that need remeshing, rat(2) the fluid volume percentage of these nodes and
a positive value of reno indicated a heavily distorted element. If one of those three
values reach a certain criteria, a remeshing will be applied. The Adapt ratios(E/tlm,
E1/tlm, Total) only appears in cases where the ICFD CONTROL ADAPT card is
used. It gives some information on the volume percentage of the fluid domain that
needs to be refined (E/tlm), the volume percentage of the fluid domain that can be
coarsened (E1/tlm) as well as the sum of the two values T otal. If the T otal value
reaches a certain criteria, a remeshing will be applied.
56 LSTC-LS-DYNA-ICFD-THE-1.1-1
References
[1] I. C. A. Smirnov and S. Shi, Les of bubble dynamics in wake flows, Com-
puters & Fluids, 34 (2005), pp. 351–373.
[2] S. S. A. Smirnov and I. Celik, Random flow generation technique for large
eddy simulations and particle-dynamics modeling, Journal of Fluids Engineering,
123 (2001), pp. 359–371.
[4] H. G. Castro and R. R. Paz, A time and space correlated turbulence syn-
thesis method for large eddy simulations, Journal of Computational Physics, 235
(2013), pp. 742–763.
[5] P. Causin, J.-F. Gerbeau, and F. Nobile, Added-mass effect in the design
of partitioned algorithms for fluid-structure problems., Comp. Meth. Appl. Mech.
Engrg.,, 194 (2005), pp. 4506–4527.
[7] , Numerical study of slightly viscous flow, Math Comput, 57 (1973), p. 785.
[8] R. Codina, Pressure stability in fractional step finite element method for in-
compressible flows,, Journal of computational physics, 170 (2001), pp. 112–140.
[9] R. Codina and S. Soto, Pressure stability in fractional step finite element
method for incompressible flows,, Journal of computational physics, 170 (2001),
pp. 112–140.
[10] R. Codina and O. C. Zienkiewicz, Cbs versus gls stabilization of the incom-
pressible navier-stokes equations and the role of the time step as stabilization
parameter,, Communications in numerical methods in engineering, 18 (2002),
pp. 99–112.
LSTC-LS-DYNA-ICFD-THE-1.1-1 57
[12] W. E and J. Liu, Projection method i: Convergence and numerical boundary
layers,, SIAM J. Numer. Anal., 32 (1995), pp. 1017–1057.
[15] C. Felippa and K. Park, Staggered transient analysis procedures for coupled-
field mechanical systems: formulation., Comp. Meth Appl. Mech. Eng., 24
(1980), pp. 61–111.
58 LSTC-LS-DYNA-ICFD-THE-1.1-1
[23] P. N. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method for
Fluid Dynamics,, Elsevier, 2005.
[24] S. Osher and R. Fedkiw, Level Set Methods and Dynamic Implicit Surfaces,,
Springer, 2003.
[25] F. D. Pin, The Meshless Finite Element Method Applied to a Lagrangian Par-
ticle Formulation of Fluid Flows, PhD thesis, Facultad de Ingenieria y Ciencias
Hidricas (FICH) Instituto de Desarrollo Tecnologico para la Industria Quimica
(INTEC) Universidad Nacional del Litoral, 2003.
[26] S. Piperno and C. Farhat, Partitioned procedures for the transient solution
of coupled aeroelastic problems- part ii, Comp. Meth Appl. Mech. Eng., 190
(2001), p. 3147:3170.
[28] S. Rugonyi and K. Bathe., On the analysis of fully-coupled fluid flows with
structural interactions-a coupling and condensation procedure., nt. J. Comput.
Civil Struct. Engrg., 1 (2000), pp. 29–41.
[29] S. Rugonyi and K. Bathe, On finite element analysis of fluid flows fully
coupled with structural interactions., Comp. Model. Simulat. Engrg., 2 (2001),
pp. 195–212.
[30] J. Shen, On error estimates of projection methods for the navier-stokes equa-
tions: second-order schemes,, Math. Comp., 65 (1996), pp. 1039–1065.
[33] P. L. Tallec and J. Mouro., Fluid structure interaction with large structural
displacements., Comp. Meth. Appl. Mech. Engrg., 190 (2001), p. 3039.
[34] E. Van Driest, On turbulent flow near a wall, Journal of Aero. sci., 23 (1952),
pp. 1007–1011.
LSTC-LS-DYNA-ICFD-THE-1.1-1 59
[35] E. Walhorn, A. Kolke, B. Hubner, and D. Dinkler, Fluid-structure
coupling within a monolithic model involving free surface flows, Computers and
Structures, 83 (2005), pp. 2100–2111.
[36] O. Zienkiewicz and J. Zhu, A simple error estimator and adaptive procedure
for practical engineering analysis, International Journal for Numerical Methods
in Engineering, 24 (1987), pp. 337–357.
60 LSTC-LS-DYNA-ICFD-THE-1.1-1
A Consistent units for ICFD
LSTC-LS-DYNA-ICFD-THE-1.1-1 61