Академический Документы
Профессиональный Документы
Культура Документы
By
Amr Gamal Mohammad Guaily
B.Sc. in Aerospace Engineering, 2002
The author wishes to express his gratitude to all those who provided help in
various ways at the different stages of this work.
i
ABSTRACT
ii
Table of Contents
ACKNOWLEDGMENTS i
ABSTRACT ii
List of Figures v
Nomenclature x
1.1 Introduction 1
2.1 Introduction 10
iii
2.7 Numerical Results 20
2.7.1 Planer Problems 20
2.7.1.1 Shock-Reflection Problem (SRP) 20
2.7.1.2 Supersonic Channel Flow (SCP) 24
2.7.1.3 Airfoil-Spoiler Configuration 33
2.7.2 Axisymmetric Problems 39
2.7.2.1 Nozzle Flow (Shock Free) 39
2.7.2.2 Nozzle Flow (With Shock) 45
2.7.2.3 Jet Flow (Nearly Fully Expanded) 49
2.7.2.4 Jet Flow (Fully Expanded) 53
2.7.2.5 Under Expanded Jet Flow (UEJF) 54
3.1 Introduction 60
References 104
iv
List of Figures
v
Figure 2-27 Grid used in airfoil-spoiler configuration 35
Figure 2-28 Zoom on the grid near the airfoil 36
Figure 2-29 Pressure coefficient contours (airfoil-spoiler problem) 36
Figure 2-30 Coefficient of pressure on the upper surface 37
Figure 2-31 Coefficient of pressure on the lower surface 37
Figure 2-32 Growth of wake in (airfoil-spoiler problem) 38
Figure 2-33 Physical nozzle and the used grid (shock free nozzle flow) 40
Figure 2-34 Mach number contours (shock free nozzle flow) 40
Figure 2-35 Pressure contours (shock free nozzle flow) 41
Figure 2-36 Radial velocity contours (shock free nozzle flow) 41
Figure 2-37 Pressure distribution (shock free nozzle flow) 42
Figure 2-38 Mach number distribution (shock free nozzle flow) 43
Figure 2-39 Temperature distribution (shock free nozzle flow) 43
Figure 2-40 Density distribution (shock free nozzle flow) 44
Figure 2-41 Velocity vector plot (shock free nozzle flow) 44
Figure 2-42 Zoom on the lower boundary (shock free nozzle flow) 45
Figure 2-43 Mach number contours (nozzle flow with shock) 46
Figure 2-44 Mach number distribution (Nozzle Flow With Shock) 46
Figure 2-45 Pressure contours (Nozzle Flow With Shock) 47
Figure 2-46 Pressure distribution (Nozzle Flow With Shock) 47
Figure 2-47 Convergence history (Nozzle Flow With Shock) 48
Figure 2-48 Velocity vector plot (Nozzle Flow With Shock) 48
Figure 2-49 Supersonic free jet main regions 50
Figure 2-50 Axisymmetric jet flow boundaries 50
Figure 2-51 Computational domain and center line Mach number
distribution (jet flow problem) 51
Figure 2-52 Axial Mach number distributions (jet flow problem) 52
Figure 2-53 Mach number contours (jet flow problem) 52
Figure 2-54 Mach Number Contours (Jet Flow Full Expanded) 53
Figure 2-55 Radial velocity Contours (Jet Flow Full Expanded) 54
Figure 2-56 Computational domain (UEJF) 55
Figure 2-57 Contour plot for the Mach number (UEJF) 56
vi
Figure 2-58 Axial Mach number distribution (UEJF) 56
Figure 2-59 Axial pressure distribution (UEJF) 57
Figure 2-60 Axial density distribution (UEJF) 57
Figure 2-61 Axial distribution of radial-velocity (UEJF) 58
Figure 2-62 Radial distribution of axial Mach number (UEJF) 58
Figure 2-63 Radial distribution of pressure (UEJF) 59
Figure 2-64 Radial distribution of radial-velocity (UEJF) 59
Figure 3-1 Transformation of a unite circle to an ellipse by S 63
Figure 3-2 Spring analogy for a patch of elements. 64
Figure 3-3 A node on the grid and its surroundings 66
Figure 3-4 Initial mesh (analytical test case). 69
Figure 3-5 Isocontours for the initial mesh (analytical test case). 69
Figure 3-6 Adapted mesh (analytical test case). 69
Figure 3-7 Adapted isocontours (analytical test case). 70
Figure 3-8 Magnification of grid (analytical test case). 70
Figure 3-9 The computation domain (SRP). 71
Figure 3-10 The initial and 1st adapted pressure (SRP) 72
nd rd
Figure 3-11 The 2 and 3 adapted pressure (SRP) 72
Figure 3-12 The initial grid and the and the pressure contours (SRP) 74
Figure 3-13 The 1st adapted grid and the pressure contours (SRP) 74
Figure 3-14 The 2nd adapted grid and the pressure contours (SRP) 75
Figure 3-15 The 3rd adapted grid and the pressure contours (SRP) 75
Figure 3-16 The 4th adapted grid and the pressure contours (SRP) 76
th
Figure 3-17 The 5 adapted grid and the pressure contours (SRP) 76
Figure 3-18 Zoom on the grid in the incident shock region SRP 77
Figure 3-19 Zoom on the grid in the reflected shock region (SRP) 77
Figure 3-20 Pressure distribution at y=0.5 (SRP) 78
Figure 3-21 Convergence history (SRP) 78
Figure 3-22 The initial grid and the Mach number contours (SRP) 79
st
Figure 3-23 The 1 adapted grid and Mach number SRP 79
Figure 3-24 The 2nd adapted grid and Mach number (SRP) 80
Figure 3-25 The 3rd adapted grid and Mach number (SRP) 80
vii
Figure 3-26 The 4th adapted grid and Mach number (SRP) 81
Figure 3-27 The 5th adapted grid and Mach number (SRP) 81
Figure 3-28 The initial grid and pressure (density adapted) SRP 82
Figure 3-29 The 1st adapted grid and pressure (density adapted) SRP 82
Figure 3-30 the 2nd adapted grid and pressure (density adapted) SRP 83
Figure 3-31 The 3rd adapted grid and pressure (density adapted) SRP 83
th
Figure 3-32 The 4 adapted grid and pressure (density adapted) SRP 84
Figure 3-33 The 5th adapted grid and pressure (density adapted) SRP 84
Figure 3-34 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 1st adp. SRP 85
Figure 3-35 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 2nd adp. SRP 85
Figure 3-36 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 3rd adp. SRP 86
Figure 3-37 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 4th adp. SRP 86
Figure 3-38 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 5th adp. SRP 87
Figure 3-39 Comparison of pressure adapted (upper) and density adapted
(lower) grid for the 1st adp. SRP 88
Figure 3-40 Comparison of pressure adapted (upper) and density adapted
(lower) grid for the 2nd adp. SRP 88
Figure 3-41 Comparison of pressure adapted (upper) and density adapted
(lower) grid for the 3rd adp. SRP 89
Figure 3-42 Comparison of pressure adapted (upper) and density adapted
(lower) grid for the 4th adp. SRP 89
Figure 3-43 Comparison of pressure adapted (upper) and density adapted
(lower) grid for the 5th adp. SRP 90
Figure 3-44 Comparison of pressure distribution at y=0.5 for SRP 90
Figure 3-45 Comparison of pressure adapted (upper) and density adapted
(lower) Mach number contours for the 1st adp. SRP 91
viii
Figure 3-46 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 2nd adp. SRP 91
Figure 3-47 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 3rd adp. SRP 92
Figure 3-48 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 4th adp. SRP 92
Figure 3-49 Comparison of pressure adapted (upper) and density adapted
(lower) pressure contours for the 5th adp. SRP 93
Figure 3-50 Comparison of pressure contours (SRP) 93
Figure 3-51 The initial grid and the corresponding pressure contours 95
Figure 3-52 The 1st adapted grid and the corresponding pressure cont. 95
Figure 3-53 The 2nd adapted grid and the corresponding pressure cont. 96
Figure 3-54 The 3rd adapted grid and the corresponding pressure cont. 96
Figure 3-55 The 4th adapted grid and the corresponding pressure cont. 97
Figure 3-56 The 5th adapted grid and the corresponding pressure cont. 97
Figure 3-57 Evolution of density contours during adaptation 98
Figure 3-58 Magnification of grid in reflected shock region 99
Figure 3-59 Convergence history 99
ix
Nomenclature
E Nodal error
g Adaptation parameter
H Hessian matrix
h The element edge length
I Functional to be minimized
k Spring constant
L2 Residual error
P Potential energy
Q Vector of unknowns
ℜ Residual vector (scalar)
T Transformation matrix
t Time
W Weight function
x, y Cartesian coordinates
Ω Solution domain
φ Basis function
δ Dirac delta function
ζ Local element coordinate
Γ Contours of the elements
Λ Diagonal matrix of the eigen-values
ω Relaxation parameter.
x
Subscripts
i, j Nodal indices
∞ Free stream value
dim Dimensional value
t, n Normal and tangential coordinates
Superscripts
T Transposes
n Time level
e Element
nd Number of nodes per element
m Level of iteration
xi
Chapter 1 Introduction and Literature Review
1.1 Introduction
This research aims at building a least squares finite element Euler solver
for the transonic regime combined with an error estimate and a directionally
adaptive grid algorithm, allowing the use of anisotropic (stretched) elements.
Adaptive finite element methods place more fine scale elements where more
resolution is needed. Isotropic or shape regular adaptive methods use only ele-
ments with bounded aspect ratio (stretched elements are avoided). Anisotropic
adaptive methods fit high aspect ratio elements (highly stretched elements) along
the regions of rapid variation of the solution for situations like shocks or bound-
ary layers. Anisotropic adaptive methods give a bigger saving in terms of com-
putational cost (number of elements and degrees of freedom) than the isotropic
ones if stretched elements are placed appropriately [1].
The current interest in the area of high-speed flows has increased the need
for advanced computational fluid dynamics (CFD) codes, which have become
the primary tools for the prediction of aero-thermal loads. Such flows are charac-
terized by regions with steep directional gradients of flow variables, embedded in
regions where the flow variables vary more smoothly. One approach [2] for im-
proving the solution accuracy of such problems is to apply grid adaptation tech-
niques.
1
The essence of the method was certainly available for a long time, but the
above two names made it a practical proposition. In addition both names were
among the first to suggest and actually use the method for fluid dynamics prob-
lems.
In the finite difference method (FDM) the PDE is written in discrete form
involving the grid points. For the FEM the PDE is first recast in an integral form
over the entire domain and its residual minimized by several means. Both the
FDM and the FEM result in a large system of simultaneous linear algebraic equa-
tions. The dependent variable is solved for at all grid points simultaneously
through direct or iterative methods.
2
The FEM accounts for boundary conditions in an easy, straightforward
manner. Unlike the FDM, the FEM has no numerical boundary conditions. The
no penetration boundary condition is implemented by a Dirichlet condition even
with a curved boundary.
3
fied by adding a perturbation to the standard Galerkin test function. The added
perturbation creates an upwind effect by weighting more heavily the upstream
nodes. Another approach was proposed by Baruzzi et al.[6], where the Laplacian
of dependent variables was added to the continuity and momentum equations.
The amount of artificial viscosity was then controlled by a single parameter as
the coefficient of the Laplacians. They later extended this first-order artificial
viscosity method to second-order.
Another method is based on the least squares weighed residual method. The
method has very good stability properties due to its minimization nature, and has
been applied for the solution of a variety of problems. As one of the earliest ef-
forts in this field one can mention the technique presented by Polk and Lynn [7]
for the solution of unsteady gas dynamic equations, with elements that are con-
structed in both space and time. Fletcher [8] used the least-squares method to
solve the Euler equations for the subcritical compressible flows. The special fea-
ture of his method was to represent groups of variables rather than single vari-
ables. Application of the least squares method to a governing equation of the
form: L(Q)=f leads to the favorable result of a symmetric and positive definite
coefficient matrix, if L is a first-order differential operator.
Taghaddosi et al. [2] applied this approach combined with adaptive grid al-
gorithm to the Euler equations. Also Moussaoui [9] applied this approach to
solve both the compressible and incompressible flows using one formulation. If
L is a higher order operator, this property is completely lost during the integra-
tion by parts and moreover elements with higher order continuity requirements,
must be employed. Lynn and Arya [10] proposed to break down the higher order
system to its first order counterpart as a way of eliminating this disadvantage.
Pontaza et al. [11] used this approach to solve both the Euler and Navier-Stokes
equations for the compressible regime. Bolton and Thatcher [12] used the
LSFEM to solve the Navier-Stokes equations in the form of stress and stream
functions
4
1.6 Literature Review of the Adaptation Techniques
For a given mesh as shown in Figure 1-1, grid adaptation methods are
mainly composed of an adaptive strategy such as mesh movement (r-method) as
in Figure 1-2; mesh refinement/coarsening (h-method); as in Figure 1-3, higher-
order interpolation (p-method). Even when the same error estimate is used to
assess the accuracy of the solution, the resulting adapted grid depends strongly
on the selected adaptation strategy. Classical methods such as grid refinement
produce isotropic meshes in which the length scales in all directions are essen-
tially the same. These methods are optimal only for those flow regions possess-
ing nearly equal gradients in all spatial directions. As a result, directional flow
features such as shocks, contact discontinuities and boundary layers are not nec-
essarily adapted efficiently and the number of elements needed to represent them
may increase disproportionally with each isotropic refinement.
5
Figure 1-1 Original mesh to be adapted
6
The above approaches have primarily been used on unstructured meshes.
This trend is mainly driven by the intrinsic ability of triangular elements in 2D
and tetrahedral elements in 3D to deal with arbitrarily complex geometries. Un-
structured adaptation algorithms can yield highly stretched grids as well as lo-
cally refined/coarsened meshes. In contrast, most refinement techniques for
structured grids avoid propagating the refinement to the boundaries by allowing
sides to have hanging nodes.
7
able from finite element interpolation theory. The Hessian of a selected solution
variable is computed and then modified to produce a positive definite matrix al-
lowing one to define a measure of error, namely a Riemannian metric. The edge-
based error estimate is thus expressed as the length of the edge of the elements in
this Riemannian metric. The construction of an anisotropic mesh may thus be
interpreted as being a uniform mesh in the defined metric. This metric introduces
and controls the magnitude as well as the direction of the grid anisotropy. A
mesh movement scheme is then applied as the adaptive strategy, which in con-
trast to the spring analogy technique used by Nakahashi and Deiwert [16], has no
constraint on grid orthogonality. This leads to a simple and efficient nodal redis-
tribution algorithm offering a greater range of point displacements. In this
method, the optimal grid for a fixed number of nodes is thus defined as one in
which the error is equidistributed over the edges.
The least squares finite element method is used to solve the compressible
Euler's Equations for both the Cartesian and axisymmetric flows in the non-
conservative form. The quality of the numerical results indicates the remarkable
performance of the used technique. This is quite evident from the final results
despite using coarse mesh. Also the robustness of the technique allows one to use
large time step to reach the steady state solution in few iterations, which saves
computational time considerably. But a disadvantage existed; namely the high
value of the inherent artificial viscosity in the method which forbidden the sharp
resolution of discontinuities, this disadvantages has been remedied.
8
thogonality constraints is used to equidistribute the lengths of the edges of the
elements of the edges of the elements in the defined metric. The adaptive proce-
dure has been proven to be effective on analytical test case.
The used technique enforces the solid wall boundary condition since it en-
forces the boundary condition as a Dirichlet type at each node using a transfor-
mation matrix for each node. This results in a faster convergence than other
methods and it can be used for both steady and unsteady flows unlike other tech-
niques.
The flow solver, combined with the proposed grid adaptation method is
then validated on a variety of problems. The quality of the numerical results in-
dicates the remarkable performance of the adaptive method, and demonstrates its
superiority to many existing techniques. This is quite evident in the final adapted
grids.
9
Chapter 2 Least Squares FEM for Inviscid
Compressible Flows
2.1 Introduction
The Euler equations for both Cartesian and axisymmetric flows can be
written in the nondimensional form as a first order system in terms of primitive
variables [2] as:
∂Q ∂Q ∂Q
+ Ax + Ay + α AaxisQ = 0 (2-1)
∂t ∂x ∂y
where
Q T = ( ρ ,u ,v , p ) is the vector of unknowns.
10
u ρ 0 0
1
0 u 0 ρ
Ax =
0 0 u 0
0 γ p 0 u
v 0 ρ 0
0 v 0 0
Ay = 1
0 0 v
ρ
0 0 γp v
v 0 0 0
y
0 0 0 0
Aaxis =
0 0 0 0
0 0 0 γv
y
with
α =1 for axisymmetric flow
α =0 for planar flow
11
ρ∞ Free stream density.
C∞ Free stream speed of sound.
Q n +1 = Q n + ∆Q
neglecting the higher order terms, and discratizing the unsteady term using
backward difference as:
∂ Q ∆ Q
≈
∂ t ∆ t
L ∆Q n +1 = −f (2-2)
where
∂ ∂ 1
L = A x
n
+ A n
y + ( I + A n
)
∂x ∂y ∆t
∂Q n
∂Q n
f = (A x
n
+ A n
y + α f ax is )
∂x ∂y
T
ρ nv n
γ p nv n
f ax is = 0 0
y y
12
∂u ∂v v ∂ρ ∂ρ ρ
( ∂x + ∂y + α y ) ∂x ∂y
+α
y
0
−1 ∂p ∂u ∂u
0
ρ 2 ∂x ∂x ∂y
A =
−1 ∂p ∂v ∂v
0
ρ 2 ∂y ∂x ∂y
∂p ∂p γp ∂u ∂v v
0 +α γ( + + α )
∂x ∂y y ∂x ∂y y
1
I (∆Q n +1 ) =
2 ∫∫Ω
(ℜ)T (ℜ)d Ω (2-3)
ne
∆Q n +1 ≈ ∆Q hn +1 = ∑ N j ∆Q jn +1 (2-4)
j =1
where
ne is the number of nodes per element.
∂ℜ
W = = LN
∂ (∆Q n +1 )
13
∫∫
Ω
(LN )T (L ∆Q n +1 + f )d Ω = 0 (2-5)
Using equation (2-4) in equation (2-5) results the linear algebraic equa-
tions:
[ K ] {∆ Q } = − {R } (2-6)
where
= ∫∫ ( L N )T ( L N )d Ω
e e
K ij i j
Ωe
rie = ∫∫ (L N i )T fd Ω (2-7)
e
Ω
For the Euler’s equation (2-1) the numerical viscosity inherent in the least
squares formulation can be demonstrated as follows [2]. The right-hand side
vector of equation (2-6) as defined in equation (2-7) is
ri e = ∫∫ ( L N i )T fd Ω
Ωe
T
∂N i ∂N i 1
= − ∫∫ A xn + A yn +( I + A n )N i
Ωe
∂x ∂y ∆t
n ∂Q n n ∂Q
n
(
x A + A y + α f ax is d Ω
)
∂x ∂y
14
T
∂N i ∂N i n ∂Q n n ∂Q
n
= − ∫∫ A xn + A yn A
x + A y d Ω + ...
e
Ωe
∂ x ∂ y ∂ x ∂ y
After integrating this equation by parts (Green’s formula), the first term on
the right-hand side yields the following term
T
n ∂ ∂ n ∂Q n n ∂Q
n
∫∫e ∂ x + + dΩ
n
A x A y A x A y
Ω
∂ y ∂ x ∂ y
N i (ξ , η ) = (1 + ξξ i )(1 + ηη i ) / 4 (2-8)
(ξi , ηi) are the local coordinate of the corner node “i” of any quadrilateral
element, while (ξ , η) are the element local axes as illustrated in Figure 2-1. The
integer index “i” varies between 1 and 4 according to the corresponding node.
15
Figure 2-1 Element shape in global and local coordinates
4
∆ Q = Σ ∆ Q iN i (2-9)
i=1
4
x = Σ xi N i
i =1
4 (2-10)
y = Σ yi N i
i =1
16
∂Ni ∂x ∂y ∂Ni ∂Ni
∂ξ ∂ξ ∂ξ ∂x
= ∂N = [J ] ∂∂Nx
∂ N i ∂x ∂y i i
(2-11)
∂η ∂η ∂η ∂y ∂y
The last relation equation (2-12) is very important since all calculus is done
in terms of the local coordinates.
Integration over each element is done numerically using Gauss-Legendre
quadrature of the 5th order accuracy. This method of integration involves evalua-
tion of the integrand at fixed points, and multiplication of the resulting value by a
corresponding weight, followed by summation ([4], [17]).
17
∂Q
4- Robbins type in which c 1Q + c 2 = c 3 is specified, where,
∂n
c1 , c 2 , and c 3 are constants
18
u cos θ sin θ u n
=
v sin θ − cos θ ut
For nodes on solid wall boundary, a new nodal vector of unknowns is de-
fined by replacing the Cartesian components (u, v) by the normal and tangential
components in the global matrix [ K ] . The resulting matrix is nonsymmetrical.
1 0 0 0
0 cos θ sin θ 0
T=
0 sin θ − cos θ 0
0 0 0 1
Due to the discrete representation of the computational domain, the angle
at a typical wall node A (Figure 2-2) can have two different values. It is neces-
sary to assign a unique angle for such nodes. To accomplish this, a weighted av-
erage of the angles from the two adjacent wall edges can be used:
L 2θ1 + L1θ 2
θA =
L1 + L 2
19
2.7 Numerical Results
The effect of the inherent viscosity that is controlled by the time step, on
the shock resolution is shown in Figure 2-4, which shows the pressure contours,
when the time step changes from 0.15 to 0.05 (from top to bottom ∆ t=0.15, 0.1,
0.05). As expected oscillations near the shock start growing after reducing the
artificial viscosity (reducing the time step) and the shock becomes more smeared
by increasing it (increasing the time step), as indicated in Figure 2-5. Figure 2-6
20
shows the density contours while Figure 2-7 shows the Mach number contours
where the time step has changed from 0.15 to 0.05. This shows the effect of the
time step on the solution as the shock becomes more refined for smaller time
step. Figure 2-8 shows the convergence history using different time steps. As
expected as the time step decreased the solution reaches steady state slower.
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
21
Figure 2-5 Pressure distribution at y= 0.05 (SRP)
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
22
1
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 2-7 Mach number contours for different time steps, (SRP)
23
2.7.1.2 Supersonic Channel Flow (SCP)
where
b= (α 2
− 0.25 ) 2α
R= 0.25 + b 2
24
The boundary conditions at the inlet are:
On the walls the flow tangency condition is imposed and the exit boundary
is left free (supersonic exit).
The grid consists of 64 X 16 uniformly distributed bi-linear rectangular
elements with 16 elements in the y-direction, 22 elements on the bump, and 21
elements in each side of it as shown in Figure 2-10.
The effect of the inherent viscosity, controlled by the time step, on the
shock resolution is shown in the figures starting from Figure 2-11 to Figure 2-13,
which show the density contours, calculated with time steps ranging from 0.15 to
0.05.
Figure 2-14 to Figure 2-16 show the pressure contours, which assure the
important role of the time step on shock resolution. The leading- and trailing
edge shocks as well as the interaction of the trailing-edge shock with the re-
flected shock are qualitatively well captured.
The Mach number contours are plotted in the figures starting from Figure 2-17 to
Figure 2-19 to assure the ability of the scheme in capturing such a complex shock
structure.
Figure 2-20 and Figure 2-21 show the Mach number distribution of the up-
per and lower walls respectively compared to Eidelman et al. [22] which shows
good agreement. As expected oscillations near the shock start growing after re-
ducing the artificial viscosity and the shock becomes more smeared by increasing
it.
25
Figure 2-22 shows the convergence history using different time steps. As
expected as the time step decreased the solution reaches steady state slower.
The quality of the used method in applying the solid wall boundary condi-
tion is demonstrated in the figures starting from Figure 2-23 to Figure 2-24,
which show the velocity vector plot and a zoom on the lower wall. The tangency
condition is applied very well as seen in these figures.
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
26
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
27
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
28
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
29
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
30
Figure 2-20 Mach number distribution for the upper wall (SCP)
Figure 2-21 Mach number distribution for the lower wall (SCP)
31
Figure 2-22 Convergence history of the flow solver (SCP)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1.5 -1 -0.5 0 0.5 1 1.5
32
Figure 2-24 Zoom on the lower boundary (SCP)
Spoilers are widely used in aircraft as lateral control devices, as speed breaks,
and as lift dampers during landing. Despite their wide usage, very little theoreti-
cal information exists. Almost all design work is done by comparison with ex-
perimental results followed by wind tunnel tests of trail models. The flow field
past an airfoil with spoiler is complex. In fact the flow separated from the upper
airfoil surface, due to the adverse pressure gradient generated by the presence of
the spoiler, then reattaches to the spoiler surface at small angles of attack [23]. A
recirculating bubble is formed upstream of the spoiler hinge. The flow separates
again from the spoiler tip and converts into wake as a free shear layer. The flow
on the lower airfoil surface also leaves the trailing edge and converts into wake.
These shedding vortices make the wake highly turbulent and oscillatory. This
unsteady wake affects the effectiveness of the spoiler.
Extensive experimental investigations on steady spoiler characteristics
have been taken by several authors ([24], [25], [26],[27]). On the theoretical
side, Abdelrahman [24] developed a numerical scheme based on the steady state
incompressible Navier-Stokes equations. Choi et al. [28] studied the transient
response of an airfoil to a rapidly deploying spoiler using turbulent compressible
Navier-Stokes equations with a turbulence model.
33
Flow Description
The flow was examined using both water table and oil streaks on a flow-
splitter plate mounted on a wind tunnel model [23].
These examinations indicate that the wake has two main characteristics.
First it has an unsteady nature due to a shed vortex street. Second, despite this
periodic behavior there is a region of wake that remains nearly constant in shape
and closes a short distance downstream of the trailing edge. Figure 2-25 repre-
sents the flow on the splitter plate as investigated by Wentz and Ostowari [25] at
zero angle of attack, 40o splitter deflection angle, 2.2x106 Reynolds number, and
0.13 Mach number. It clearly shows two basic regions: first, an outer essentially
potential flow region, and second, a near wake region behind the spoiler. Also
Figure 2-25 shows that the near wake may be subdivided into two more parts.
The part upstream of the wing trailing edge shows very little fluid motion and
thus may be considered at constant pressure. The part downstream of the trailing
edge is characterized by a pair of vortices; the upper one rotating clockwise (for
left to right flow) and smaller one below it rotating counterclockwise
34
Figure 2-26 Airfoil-spoiler details
Figure 2-26 describes the details of the used configuration, while Figure 2-27
shows the O-type grid of [65x20] elements used to solve the problem of the air-
foil-spoiler configuration, followed be a zoom on the grid near the airfoil surface
in Figure 2-28, where the parameters in Figure 2-26 are as follows for NACA
23012:
d = 0.75 c, t= 0.01 c, δ = 30o, and L=0.15 c. The inlet Mach number used was
M= 0.2.
35
Figure 2-28 Zoom on the grid near the airfoil
Figure 2-29 shows the pressure coefficient isocontours, while Figure 2-30 and
Figure 2-31 show the distribution of the pressure coefficient over the upper and
lower surfaces compared to the experimental results of [24]. The coefficient of
pressure suddenly decreases to a negative value due to the presence of the
spoiler.
36
Exp. Num.
2
1.5
1
0.5
Cp
0
-0.5
-1
-1.5
-2 0 0.2 0.4 0.6 0.8 1
Exp. Num.
1.5
1
0.5
0
Cp
-0.5
-1
-1.5
-2
-2.5
0 0.2 0.4 0.6 0.8 1
X
37
Figure 2-32 Growth of wake in (airfoil-spoiler problem)
Figure 2-32 shows the growth of the wake behind the spoiler which in
qualitative agreement with the experimental results of [25].
38
2.7.2 Axisymmetric Problems
Nozzle flow will be considered here as it is one of the most important test
cases used to show code robustness. In particularly it tests the efficiency of the
flow tangency condition, which is considered one of the most difficult aspects
when talking about Euler's equations.
Nozzle Geometry
where A is the nozzle cross-sectional area and the radius, r=(A / Π)0.5.
The computational domain is discretized uniformly into [60 X 20] bi-linear
elements as shown in Figure 2-33. We consider flow at an inlet Mach number of
Min = 0.224 and specify the following boundary conditions at the inlet "subsonic
inlet" (ρ, u, v) = (1.0, 0.224, 0.0). The lower and upper boundaries are imperme-
able and the exit boundary is left free (supersonic exit).
Figure 2-34 shows the Mach number contours through the nozzle. Despite
solving the whole domain, the full symmetry is evident from the Mach number
contours as well as the pressure contours shown Figure 2-35. Figure 2-36 shows
the contours of the radial velocity, which assures the capability of the used tech-
nique to capture the symmetry of the flow.
39
0.2
0.15
0.1
0.05
0
r
-0.05
-0.1
-0.15
-0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Figure 2-33 Physical nozzle and the used grid (shock free nozzle flow)
0.15
0.1
0.05
-0.05
-0.1
-0.15
40
0.15
0.1
0.05
-0.05
-0.1
-0.15
0.15
0.1
0.05
-0.05
-0.1
-0.15
41
The flow accelerates (Mach number increases) as it goes through the noz-
zle as shown in Figure 2-37 which presents the pressure distribution of the nozzle
center line as well as the nozzle wall compared to the exact one dimensional so-
lution [29]. Figure 2-38 shows the Mach number distribution for both the nozzle
wall and the centerline compared to the one-dimensional solution. The wall ef-
fect is evident in the distribution of the wall Mach number as it starts to decrease
at an axial location of 0.7 due to the change in the area (acts like a compression
corner). Figure 2-39 shows the distribution of the temperature across the nozzle
in which again the two dimensional effect is clear. The density distribution is
shown in Figure 2-40.
42
Figure 2-38 Mach number distribution (shock free nozzle flow)
43
Figure 2-40 Density distribution (shock free nozzle flow)
To show how far the flow tangency condition is imposed, Figure 2-41
shows the velocity vector plot of the flow inside the nozzle. Figure 2-42 is a
zoom on the lower nozzle wall, which assures the robustness of the used method
in enforcing the solid wall boundary condition.
0.15
0.1
0.05
0
r
-0.05
-0.1
-0.15
44
Figure 2-42 Zoom on the lower boundary (shock free nozzle flow)
The previous nozzle shape is solved but with a predetermined back pres-
sure to generate a normal shock inside the divergent part.
So now we have a different exit boundary condition, which is a subsonic
exit. According to the theory of characteristics we should define only one de-
pendent variable (pressure) at that boundary. The used backpressure is = 0.5.
Figure 2-43 shows the Mach number contours of the resulting solution. In this
figure the two dimensional shock is evident and as expected the shock takes the
shape of a cap. Figure 2-44 shows the Mach number distribution for the center-
line and for the wall compared to the exact 1-D solution as illustrated in [29]. As
seen the shock is well captured despite using coarse grid [60X20].
45
0.15
0.1
0.05
-0.05
-0.1
-0.15
46
Figure 2-45 shows the pressure contours in which again the shock is evident
Figure 2-46 shows the pressure distribution, which assures the robustness of the
used method despite using coarse grid. Figure 2-47 shows the convergence his-
tory of the flow solver. In this figure we started the solution by large time step to
reach the steady state solution faster then the time step was reduced to refine the
solution. In this figure, the quadratic convergence feature is evident which as-
sures the robustness of the flow solver. Figure 2-48 shows a velocity vector plot,
which again assures the shock existence.
0.15
0.1
0.05
-0.05
-0.1
-0.15
47
Figure 2-47 Convergence history (Nozzle Flow With Shock)
0.3
0.2
0.1
-0.1
-0.2
-0.3
-0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
48
2.7.2.3 Jet Flow (Nearly Fully Expanded)
The current test case is that of a supersonic jet flow. An experimental investiga-
tion of flow of a moderate Reynolds-number and Mach-number axisymmetric,
near isentropically expanded, cold jet has been performed by Troutt and
McLaughlin [30], with the following jet conditions
To well capture the free jet phenomenon the solution domain is used as shown in
Figure 2-50
49
Figure 2-51 (top) shows the domain of solution used in the current study
which extended 70Rj in the axial direction and 20Rj in the radial direction using
clustering in the normal direction near the nozzle end to well capture the rapid
variations in the mixing layer region. Bi-linear elements are used with 60*26
elements in the axial and normal directions respectively.
50
Figure 2-51 Computational domain and center line Mach number distribution
(jet flow problem)
The centerline Mach number distribution for both the experiment and the
current work is shown in Figure 2-51 (bottom). The numerical results are in
good agreement with the experiment despite using coarse grid. The figure shows
the potential core length of the jet to be 10 radii while that of the experiment be-
tween 8 and 10 radii. The sonic point in the jet is reached between 22 and 24 ra-
dii while that of the experiment is between 18 and 20 radii.
Figure 2-52 shows the axial Mach number distributions for different radial
positions in which the jet decelerates and the outer stagnant air is accelerated.
Figure 2-53 shows the Mach number contours. In this figure we can see the ge-
neric features of the supersonic jet flow; e.g. the potential core region. The sym-
metry of the flow is depicted in the figure, which assures the quality of the used
method despite using coarse grid.
51
Figure 2-52 Axial Mach number distributions (jet flow problem)
52
2.7.2.4 Jet Flow (Fully Expanded)
The current test case is that of a supersonic full expanded jet flow with Mj = 2.1.
The importance of this test case is that we know the theoretical solution. To well
capture the free jet phenomenon the solution domain shown in Figure 2-50 is
used again. Figure 2-51 (top) shows the domain of solution used in the current
study which extended 70Rj in the axial direction and 20Rj in the radial direction
using clustering in the normal direction near the nozzle end to well capture the
rapid variations in the mixing layer region. Bi-linear elements are used with
40*20 elements in the axial and normal directions respectively. Figure 2-54
shows the contour plot for the Mach number, which supports the expected theo-
retical solution, i.e. the expected cylindrical contours. Figure 2-55 shows the con-
tour plot for the radial velocity. The symmetry of the solution is evident (the full
domain was solved not half) which assures the quality of the used technique.
53
Figure 2-55 Radial velocity Contours (Jet Flow Full Expanded)
The current test case is that of a supersonic under expanded jet flow with
Mj = 2.1 and pressure ratio pj/pamb. = 1.7. The importance of this test case is that
tests the capability of the scheme to capture the well-known cell-structure. To
well capture the free jet phenomenon the solution domain shown in Figure 2-56
is used with a [40 X 36] bi-linear elements which extended 50Rj in the axial di-
rection and 20Rj in the radial direction using clustering in the normal direction
near the nozzle end to well capture the rapid variations in the mixing layer re-
gion. Figure 2-57 shows the contour plot for the Mach number, which supports
the expected solution, i.e. the expected cell structure (since the jet flow has a
higher pressure value than the ambient, an expansion fan is created at the nozzle
exit. This expansion fan is reflected when it reaches the centerline which acts
like a wall. The reflected fan turns the slip line inwards while being reflected
again and when it reaches the centerline again, it is reflected again to meet the
slip line again and turns it outward, thus forming the first cell and so on) in con-
54
clusion the expansion fan and the slip line affects each other to make the slip line
takes the shape of a cell.
The symmetry of the solution is evident (the full domain was solved not
only one half) which assures the quality of the used technique. Figure 2-58
shows the axial Mach number distribution at different radial locations. As ex-
pected the Mach number increases till the pressure reaches the ambient pressure
as shown in Figure 2-59. Figure 2-60 shows the axial distribution of the density.
Figure 2-61 shows the axial velocity distribution at different radial location,.
Figures starting from Figure 2-62 to Figure 2-64show the radial variations of the
Mach number, pressure, and the radial velocity respectively, which again assures
the capability of the technique to capture the symmetry despite solving the whole
domain.
55
Figure 2-57 Contour plot for the Mach number (UEJF)
56
Figure 2-59 Axial pressure distribution (UEJF)
57
Figure 2-61 Axial distribution of radial-velocity (UEJF)
58
Figure 2-63 Radial distribution of pressure (UEJF)
59
Chapter 3 Directionally Adaptive Technique for FEM
3.1 Introduction
The current interest in the area of high-speed flows has increased the need
for advanced computational fluid dynamics (CFD) codes, which have become
the primary tools for the prediction of aero-thermal loads. Such flows are charac-
terized by regions with steep directional gradients of flow variables, embedded in
regions where the flow variables vary more smoothly. One approach for improv-
ing the solution accuracy of such problems is to apply grid adaptation techniques.
2
ζ d gq
Ee = (h e − ζ ) 2
(3-1)
2 dx e
where
ζ is the local element coordinate and he the element length.
60
A measure of the overall error in the element is then considered to be the
root-mean-square value of Ee
1
h e E e2 2 2
gq
1 2 d
Ee
RMS
= ∫ dζ = he (3-2)
he 120 dx
2
0 e
2
d gq
h
2
e 2
=c (3-3)
dx e
d 2 g h / dx 2 .
The above methodology can be extended to 2D based on the fact that each edge
of a 2D element can be considered as a 1D element ([18], [31], [32], [33]). So the
second derivative in equation (3-2) may be replaced with the Hessian matrix as
follows
∂2 g h ∂2 g h
∂x 2 ∂x ∂y
H = 2 (3-4)
∂ g ∂2 g h
h
∂y ∂x ∂y 2
61
The second derivative in equation (3-4) will vanish since we use a bi-linear
element. However, a mass lumping (center of mass) [18], can be applied to re-
cover an estimate of the second derivative. This yields the expression
∫A I g h , ij N I dA
g h , ij I = (3-5)
∫A I N I d A
where AI represents the elements sharing node I. After integration of equation (3-
5) by parts, the nodal values of the Hessian reduce to
∫Γ I g h ,i N I n j d Γ − ∫A I g h ,i N I , j dA
g h ,ij = (3-6)
I
∫A I N I dA
H = R (α )ΛR T (α ) (3-7)
where Λ is the diagonal matrix of the eigen-values of H and R is the matrix of the
eigenvectors. The transformation Λ is a scaling in the direction of the axes and R
is a rotation with angle α that the eigenvector corresponding to the smallest ei-
gen-value makes with the x-axis.
In order to obtain a symmetric, positive definite matrix, the Hessian is
modified by taking the absolute value of its eigenvalues. This results in
H = R (α ) Λ R T ( α ) = S ( α ) S T ( α ) (3-8)
lipse, rotated through an angle α , whose semi-major axis is the reciprocal of the
62
square root of the smallest eigenvalue, and semi-minor axis is the reciprocal of
the square root of the largest eigenvalue [18], as shown in Figure 3-1. Therefore
one can obtain a directionally stretched grid by mapping a uniform mesh using
the transformation S. However, in the current approach a mesh with edges of
equal length is sought in the transformed plane ST, where the length of a curve B
is given by
1
d (B ) = ∫
0
s ` ( l ) T H ( l ) s ` ( l ) d l (3-9)
63
3.2.2 Moving-Node Scheme
The adaptive strategy modifies the grid under the guidance of the error es-
timate to improve the quality of the numerical solution. Thus the use of an ap-
propriate adaptive scheme is crucial for achieving the desired directionally
adapted mesh. The used strategy relies on a node-moving scheme, also called
nodal redistribution. As illustrated in Figure 3-2, the mesh may be viewed as a
network of springs [15] whose stiffness constants represent the edge-based error
estimate.
The positions of the grid vertices may then be interpreted as the solution of
an energy minimization problem. This yields for each vertex I
m in P = m in ∑ ( x I − x J ) 2 k IJ (3-10)
xI I xI
J
where PI denotes the potential energy of the four springs sharing a node I and kIJ
are the associated four stiffness constants. These constants may be specified as
64
d (x I − x J )
k IJ = (3-11)
xI −xJ
where
x I − x J indicates the Euclidian norm
∑ (x Im +1 − x Jm +1)k =0
m
IJ (3-12)
J
−∑ (x mI − x Jm )k m IJ
∆x I = J
m
(3-13)
∑k IJ
J
xI
m +1
= x Im + ω∆x I (3-14)
65
tive or nearly zeroing Jacobian. The adaptive method uses the solution of one of
the scalar variables to adapt the mesh. Then goes back to the flow solver with the
adapted mesh. Each mesh adaptation followed by the flow solver is called one
adaptive cycle.
The previous scheme does not guarantee the smoothness of the resulting grid, i.e.
the resulting grid may contain elements with angles greater than 170o or less than
10o, which may cause ill-posedness of the resulting global matrix. So we pro-
pose to use additional diagonal elements to act as semi-torsional springs, or to
add a grid smoothening step, which guarantees the smoothness of the adapted
grid. Which is performed after each iteration of the adaptation scheme by simply
equating the gradient of the adapted grid lines at each point (see Figure 3-3) as in
equation (3-15) for the y direction and the same applies for the x direction.
66
3.2.4 The Grid Adaptation Procedure
The grid adaptation procedure may be summarized in the following steps:
------------------------------------------------------------------------------------------
Read a background mesh and the corresponding solution
Compute H on the background mesh
Current mesh is initialized by an initial mesh guess (optional)
Move the nodes of the current mesh as follows
DO m=1,MAXITER
DO inod=1,NNODE
DO iedge=1,NEDGE
Determine H by interpolating on the background mesh (optional)
Compute springs constants by numerical integration of (3.11)
ENDDO
Find new position of inod
Move inod to its new position
Check quality of elements sharing node I
ENDDO
Grid smoothness (optional)
If (MAXDISP .<=. tolerance) exit the m loop
ENDDO
-----------------------------------------------------------------------------------------
where
MAXITER: maximum number of iterations.
NNODE: total number of nodes.
NEDGE: number of edges surrounding the current adapted node.
inod: counter over the NNODE.
iedge: counter over the NEDGE.
MAXDISP: maximun allowable displacment.
tolerance: allowable tolerance.
67
3.3 Numerical Results
(
g ( x,y ) = tan −1 1000 ( x 4 y 4 − 0.25 ) ) (3-16)
68
Figure 3-4 Initial mesh (analytical test case).
Figure 3-5 Isocontours for the initial mesh (analytical test case).
69
Figure 3-7 Adapted isocontours (analytical test case).
70
3.3.2 Shock Reflection Problem (SRP)
The second test case is the shock reflection problem discussed previously
in Chapter 2 (section 2.7.1.1). Figure 3-9 shows the computational domain
again. This example tests certain features of the algorithm including the resolu-
tion of a system of two oblique shocks and their proper angles. This test case
will be solved with and without the proposed smoothening step. Figure 3-10
shows the initial grid (a) and the corresponding pressure contours (b). Also in
this figure the 1st adapted grid is shown (c) as well as the corresponding pressure
contours (d). Figure 3-11 shows the results for the 2nd and 3rd adaptation cycles.
In these figures one can see that the resulting grids are not adapted well to the
required solution. Therefore in the following figures we will use the proposed
smoothening step.
71
1
(a)
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
(b)
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
(c)
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
(d) 0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
(a) 0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
0.5
(b)
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
(c) 0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
1
(d) 0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
72
The initial coarse mesh composed of [61 x 21] nodes and the corresponding iso-
contours are shown in Figure 3-12 . Using ∆t=0.1, the results are adapted through
five cycles using the pressure or the density as the adaptation parameter as speci-
fied in the title of each figure. The artificial viscosity is reduced beginning at the
third cycle by reducing the time step to 0.05, since its amount was very high for
the size of the grid near the shock. Figure 3-13 through Figure 3-17 show the
pressure contours and the grids after each adaptation. The improvement in the
shock resolution after adaptation is quite evident.
Figure 3-18 and Figure 3-19 show how the elements are re-oriented to be
aligned with the shock, creating very high aspect ratio elements, which assures
the robustness of the technique. The pressure distribution at (y=0.5) is shown in
Figure 3-20. The adapted solution captures the shock more sharply. The conver-
gence history is shown in Figure 3-21. Each jump in the Figure corresponds to an
adaptation cycle. The quadratic convergence of the Newton linearization is quite
evident. The Mach number contours for the initial solution and for the adapted
solution are shown in Figure 3-22 through Figure 3-27, which again assures the
importance of the grid adaptation to sharply capture shocks despite using coarse
mesh.
73
1
0.8
0.6
y
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
0.8
0.6
y
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
Figure 3-12 The initial grid and the and the pressure contours (SRP)
0.8
0.6
y
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
0.8
0.6
y
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
Figure 3-13 The 1st adapted grid and the pressure contours (SRP)
74
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-14 The 2nd adapted grid and the pressure contours (SRP)
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-15 The 3rd adapted grid and the pressure contours (SRP)
75
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-16 The 4th adapted grid and the pressure contours (SRP)
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-17 The 5th adapted grid and the pressure contours (SRP)
76
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35
Figure 3-18 Zoom on the grid in the incident shock region SRP
0.56
0.54
0.52
0.5
0.48
0.46
0.44
Figure 3-19 Zoom on the grid in the reflected shock region (SRP)
77
Figure 3-20 Pressure distribution at y=0.5 (SRP)
78
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-22 The initial grid and the Mach number contours (SRP)
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-23 The 1st adapted grid and Mach number SRP
79
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-24 The 2nd adapted grid and Mach number (SRP)
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-25 The 3rd adapted grid and Mach number (SRP)
80
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-26 The 4th adapted grid and Mach number (SRP)
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-27 The 5th adapted grid and Mach number (SRP)
The problem of reflected shock is re-solved using the density as the adapta-
tion parameter in order to know how far the effect of the adaptation parameter is.
See Figure 3-28 through Figure 3-33 for the evolution of the grid and the solution
during adaptation, which again supports the importance of the adaptation.
81
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-28 The initial grid and pressure (density adapted) SRP
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-29 The 1st adapted grid and pressure (density adapted) SRP
82
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-30 the 2nd adapted grid and pressure (density adapted) SRP
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-31 The 3rd adapted grid and pressure (density adapted) SRP
83
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-32 The 4th adapted grid and pressure (density adapted) SRP
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-33 The 5th adapted grid and pressure (density adapted) SRP
84
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
85
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
86
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-39 through Figure 3-43 also show the resulting grids using the
pressure and the density as the adaptation parameter. From these figures it can be
seen that the density-adapted grids are more refined and oriented to the shock
direction. The difference between the adapted results using density and pressure
as an adaptation parameter is evident in Figure 3-44. To complete the compari-
son, Figure 3-45 through Figure 3-49 show the Mach number contours for the
pressure-adapted case (top) and the density adapted case (bottom). From these
figures one can conclude that the density as an adaptation parameter is recom-
mended since it produces less oscillations during adaptation cycles.
87
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
88
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
89
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
3.5
Exact
press adp.
dens.adp
3 no adp
2.5
1.5
0.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
90
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
91
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
92
1
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
0.8
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Figure 3-50 shows how important is the role of the adaptation. This figure shows
the pressure contours for reference [34] (middle) and the pressure contours for
the current work (top). The importance of this figure lies in that reference [34]
uses a [121 X 81] grid i.e. (9801) nodes, while in the current work the used grid
is [61 X 21]; i.e. (1281) nodes only, which saves computational time considera-
bly. Also shown in this figure (bottom) are the pressure contours for [11].
93
3.3.3 Supersonic Channel Problem (SCP)
A more challenging problem for this algorithm exists in the analysis of su-
personic flow past the parabolic arc bump discussed previously in Chapter 2
(section 2.7.1.2). The mesh used for this problem consists of 1024 elements and
1088 nodes evenly distributed in a domain which has a bump height of 4%. This
problem creates an interesting shock interaction behind the bump. The shock
from the leading edge of the bump is reflected down off the ceiling of the tunnel
and crosses, then combines, with the shock formed at the trailing edge of the
bump.
The initial coarse mesh composed of [65 x 17] nodes and the correspond-
ing isocontours are shown in Figure 3-51. Using ∆t=0.1, the results are adapted
through five cycles. The artificial viscosity is reduced beginning at the second
cycle by reducing the time step to 0.05, since its amount was very high for the
size of the grid near the shock. Figures starting from Figure 3-52 to Figure 3-56
show the pressure contours and the grid after each adaptation. The improvement
in the shock resolution after adaptation is quite evident. The density contours, for
the initial solution and for the adapted solution, are shown in Figure 3-57, which
again assures the importance of the grid adaptation to sharply capture shocks de-
spite using coarse mesh. Figure 3-58 shows a magnification of the grid at the
zone of shock reflection at the upper wall, which demonstrates the robustness of
the algorithm in reorienting the elements in the shock direction with very high
efficiency. Finally Figure 3-59 shows the convergence history of the flow solver,
each jump corresponds to an adaptation cycle, the quadratic convergence is evi-
dent in the figure also, which assures the flow solver and adaptation algorithm
robustness.
94
1
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-51 The initial grid and the corresponding pressure contours
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-52 The 1st adapted grid and the corresponding pressure cont.
95
1
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-53 The 2nd adapted grid and the corresponding pressure cont.
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-54 The 3rd adapted grid and the corresponding pressure cont.
96
1
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-55 The 4th adapted grid and the corresponding pressure cont.
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
0.8
0.6
0.4
0.2
0
-1.5 -1 -0.5 0 0.5 1 1.5
Figure 3-56 The 5th adapted grid and the corresponding pressure cont.
97
1
2nd adp.
dt = 0.05 0.5
0
-1.5 -1 -0.5 0 0.5 1 1.5
1
3rd adp.
dt = 0.05
0.5
0
-1.5 -1 -0.5 0 0.5 1 1.5
1
4th adp.
dt = 0.05
0.5
0
-1.5 -1 -0.5 0 0.5 1 1.5
98
0.95
0.9
0.85
0.8
0.75
0.7
0.65
99
Chapter 4 Summary and Conclusions
In chapter 2, the least squares finite element method for the compressible
Euler’s equation has been presented. First, the Least squares finite element ad-
vantages have been introduced in section 2.1. Then, the least squares formula-
tion has been presented in details in section 2.2. An interpretation of the inherent
artificial viscosity has been presented in section 2.3. The finite element ap-
proximation has been presented and the problem details as shape function and
local coordinates have been reviewed in section 2.4. The boundary condition,
one of the most important aspects in inviscid flows has been presented in section
2.5 using a very strong technique in applying the no-slip condition. In section
2.6 the solution method has been presented.
100
To show how robust and stable is the used technique, in section 2.7, a variety of
a well-known problems concerning the planar and Axisymmetric flows have
been solved, analyzed, and compared to exact or experimental results showing
the superiority of the used technique over the existing ones specifically in that it
uses a coarse mesh to capture discontinuities.
4.2 Conclusions
The least squares finite element method has been used to solve the com-
pressible Euler's Equations for both the Cartesian and Axisymmetric flows in the
non-conservative form. The quality of the numerical results indicates the remark-
able performance of the used technique. This is quite evident from the final re-
sults especially for the nozzle flow as well as the jet flow problem despite using
coarse mesh. Also the robustness of the technique allows one to use large time
101
step to reach the steady state solution in few iterations, as done in the shock re-
flection problem (∆t=0.15), which saves computational time considerably. But a
disadvantage has been encountered; namely, the large inherent artificial viscosity
in the method, which has prevented the sharp resolution of discontinuities, this
disadvantage, has been remedied. To remedy this disadvantage in the least-
squares finite element method, an adaptive algorithm with directional features,
using an edge based error estimate on quadrilateral meshes, has been used. The
error of the numerical solution is measured by its second derivatives and the re-
sulting Hessian tensor is used to define a Riemannian metric. A mesh movement
strategy with no orthogonality constraints is used to equidistribute the lengths of
the edges of the elements in the defined metric. The adaptive procedure has been
proven to be effective for an analytical test case.
Dealing with the boundary conditions, the used technique in treating the
solid wall boundary is very robust since it enforces the boundary condition as a
Dirichlet type at each node using a transformation matrix for each node this re-
sults in a faster convergence than other methods and it can be used for both
steady and unsteady flows unlike other techniques.
The flow solver, combined with the proposed grid adaptation method is
then validated on shock-reflection problem, capturing the system of two shocks
with high resolution and the correct angle despite using coarse mesh. The meth-
odology is also tested on the supersonic channel flow problem. The quality of the
numerical results indicates the remarkable performance of the adaptive method,
and demonstrates its superiority to many existing techniques. This is quite evi-
dent from the final adapted grids especially for the shock-reflection and super-
sonic channel problems where elements are strongly aligned with the shocks.
102
4.3 Recommendations for future work
103
References
[4]. W.G. Habashi 'The FEM in Fluid Mechanics' Coarse Notes, Concordia
University, Canada, ENGR 626.
[6]. G.S. Baruzzi, W. G. Habashi and M.M. Hafez, “Finite Element Solu-
tions of the Euler Equations for Transonic External Flows”, AIAA J. 29
(1991) 1886-1893.
[7]. J.F. Polk and P.P. Lynn, “A Least Squares Finite Element Approach to
Unsteady Gas Dynamics.”,' J. Comp. Phy. 12 (1978) 3-10.
104
[8]. C.A.J. Fletcher, “A Primitive Variable Finite Element Formulation for
Inviscid Compressible Flows”,' J. Comp. Phy. 33 (1979) 301-312.
[10]. P.P. Lynn and S.K. Arya, “Use of the Least Squares Criterion in the Fi-
nite Element Formulation”, Int. J. Num. Meth. Eng. 6 (1973) 75-88.
[11]. J.P. Pontaza, Xu Diao, J.N. Reddy, and K.S. Surana ”Least-Squares Fi-
nite Element Models of Two-Dimensional Compressible Flows”, Finite
Element in Analysis and Design 40 (2004) 629-644.
105
[17]. G. F. Carey and J. T. Oden, “Finite Elements Computational Aspects”,
Vol. III, Prentice-Hall, Inc. new Jersey, 1984.
106
[26]. M. D.Mack, H. C. Seetharam, Kuhn, W.G., and Bright, J. T. “Aerody-
namics Spoiler Control Devices”, AIAA Conf. Paper 79-1873, 1979.
[30]. T.R. Troutt and D.K. McLaughlin, “Experiments on the Flow and
Acoustic Properties of a Moderate-Reynolds-Number Supersonic Jet”, J.
Fluid Mech. 116 (1982) 123-156.
[34]. Z.-C. Zhang and S.-T. Yu, “Shock Capturing without Riemannian
Solver- A Modified Space-Time CE/SE Method for Conservation
Laws”AIAA Paper 99-0904,1999.
107