CFD I_2

download CFD I_2

of 53

Transcript of CFD I_2

  • 7/27/2019 CFD I_2

    1/53

    The equations governing the steady 2-D Newtonian flow are

    or identically

    But this is just a theoretical example in which the flow is assumed to havenull thickness

    If we want to make a more adequate approach that takes into account the

    third dimension we have to use the Shallow Water equations (SSWW)

    governing equations

    shallow waterscomputational fluid dynamics I

    xfux

    p

    y

    uv

    x

    uu

    t

    u

    1

    yfvy

    p

    y

    vvx

    vut

    v

    1

    0, iiu ufuuu

    p

    t

    1

    0

    y

    v

    x

    u

    2,1i

  • 7/27/2019 CFD I_2

    2/53

    governing equations

    shallow waterscomputational fluid dynamics I

    The assumptions to be made are

    The distribution of the horizontal velocity along the vertical direction is assumed

    to be uniform

    An integration in height is carried out, and the horizontal velocity is taken as the

    mean value of the horizontal velocities along the vertical direction

    The main direction of the flow is the horizontal one, and only very small flows

    take place on vertical planes

    The acceleration in the vertical direction is negligible compared to gravity and a

    hydrostatic distribution of the pressure is assumed

  • 7/27/2019 CFD I_2

    3/53

    Integrating the continuity equation along the z-axis

    As the Leibniz rule to bring the derivatives into the integral sign gives

    it is obtained

    governing equations

    shallow waters. continuity eq.computational fluid dynamics I

    0 bh

    h

    h

    h

    hwhwdz

    y

    vdz

    x

    u

    bb

    0

    z

    w

    y

    v

    x

    u

    h

    hbH=h+hb

    h

    0 xh

    hux

    hhuudzxdzx

    u bb

    h

    h

    h

    h bb

    0 bbbh

    h

    bb

    h

    h

    hwhwy

    hhv

    y

    hhvvdz

    yx

    hhu

    x

    hhuudz

    xbb

  • 7/27/2019 CFD I_2

    4/53

    w(h), (vertical component of the velocity on the surface) is given by

    Substituting in the former equation

    Noting that , and taking and renaming the main velocities as

    the continuity equation is obtained as

    0

    th

    t

    hvdz

    yudz

    x

    b

    h

    h

    h

    h bb

    hvyh

    hux

    h

    t

    h

    dt

    dh

    hw

    0

    t

    hb

    uudzH

    u

    h

    hb

    1vvdz

    Hv

    h

    hb

    1

    0

    y

    vH

    x

    uH

    t

    h

    governing equations

    shallow waters. continuity eq.computational fluid dynamics I

  • 7/27/2019 CFD I_2

    5/53

    As the vertical acceleration is negligible, the third dynamic equation

    can be written as

    Integrating this equation in depth and assuming the atmospheric pressure

    to be zero it is obtained

    Deriving with respect tox and y

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

    01

    zf

    z

    p

    zfz

    w

    y

    w

    x

    w

    z

    p

    z

    ww

    y

    wv

    x

    wu

    t

    w

    2

    2

    2

    2

    2

    21

    dzz

    pdzf

    h

    h

    h

    hz

    bb

    phphphhf bbz

    x

    p

    x

    hfz

    1

    y

    p

    y

    hfz

    1

  • 7/27/2019 CFD I_2

    6/53

    The first dynamic equation results into

    Adding the continuity equation multiplied by u, it is obtained

    this is

    as

    2

    2

    2

    2

    2

    2

    zu

    yu

    xu

    xhff

    zuw

    yuv

    xuu

    tu zx

    2

    2

    2

    2

    2

    2

    z

    u

    y

    u

    x

    u

    x

    hff

    z

    w

    y

    v

    x

    uu

    z

    uw

    y

    uv

    x

    uu

    t

    uzx

    2

    2

    2

    2

    2

    22

    z

    u

    y

    u

    x

    u

    x

    hffz

    uw

    y

    uv

    x

    u

    t

    uzx

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

    z

    wuw

    z

    u

    y

    vuv

    y

    u

    x

    uu

    t

    u

    z

    uw

    y

    uv

    x

    u

    t

    u

    2

    2

  • 7/27/2019 CFD I_2

    7/53

    Integrating in depth the former expression

    Taking into account that , it is obtained

    Cancelling terms

    xhhuxhhudzuxthhuudztb

    b

    h

    h

    h

    h bb

    222

    dzHx

    hffhwhuhwhu

    y

    hhvhu

    y

    hhvhuuvdz

    y

    h

    hzxbb

    h

    h

    bbb

    bb

    u

    hvy

    hhu

    x

    h

    t

    h

    dt

    dhhw

    dzHx

    hffhv

    y

    hhu

    x

    h

    t

    hhuhv

    y

    hhu

    x

    h

    t

    hhu

    y

    hhvhu

    y

    hhvhuuvdz

    y

    h

    hzxb

    bb

    bbb

    h

    h

    bbb

    bb

    u

    dzHx

    h

    ffuvdzydzuxudzt

    h

    hzx

    h

    h

    h

    h

    h

    h bbbb

    u

    2

    x

    hhu

    x

    hhudzu

    xt

    hhuudz

    t

    bb

    h

    h

    h

    h bb

    222

    2

    2

    2

    2

    2

    22

    z

    u

    y

    u

    x

    u

    x

    hff

    z

    uw

    y

    uv

    x

    u

    t

    uzx

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

  • 7/27/2019 CFD I_2

    8/53

    Taking mean velocities it is obtained

    The viscosity effects can be evaluated as

    where v is the turbulent viscosity

    Where are the shear stresses acting on the surface (due to the wind action)and on the bottom (due to the roughness of the channel)

    = Wind drag coefficient

    = Manning coefficient

    = Wind velocity components

    = Air density

    dzHx

    hff

    y

    uvH

    x

    Hu

    t

    uH h

    hzx b

    u

    2

    xxb

    bs

    h

    hH

    y

    u

    x

    udz

    2

    2

    2

    2

    u

    xx bs ,

    iaw

    s

    WWC

    i 34

    2

    h

    i

    b H

    uVgn

    Hi

    WC

    n

    iW

    a

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

  • 7/27/2019 CFD I_2

    9/53

    Developing the derivatives in the left hand side

    Taking into account the continuity eq.

    the former eq becames

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

    yHuvH

    yvuv

    yu

    xHuH

    xuu

    tHuH

    tu

    yuvH

    xHu

    tuH

    )(2 2

    2

    0

    y

    HvH

    y

    v

    x

    HuH

    x

    u

    t

    h

    y

    vH

    x

    uH

    t

    h

    vHy

    u

    y

    HvH

    y

    v

    x

    HuH

    x

    u

    t

    HuH

    x

    uuH

    t

    u

    y

    uvH

    x

    Hu

    t

    uH

    )(

    2

    vH

    y

    uH

    x

    uuH

    t

    u

    y

    uvH

    x

    Hu

    t

    uH

    0

    2

  • 7/27/2019 CFD I_2

    10/53

    The derivatives of the depth with respect tox and y are

    Carrying out the same operations for they dimension, and developing the

    derivatives taking into account the last expression it is obtained

    wherefc is the Coriolis factor

    34

    2

    2

    2

    2

    2

    h

    xaw

    cH

    uVgn

    H

    WWC

    y

    u

    x

    uvf

    x

    hg

    y

    uv

    x

    uu

    t

    u

    34

    2

    2

    2

    2

    2

    h

    yaw

    cH

    vVgn

    H

    WWC

    y

    v

    x

    v

    ufy

    h

    gy

    v

    vx

    v

    ut

    v

    governing equations

    shallow waters. dynamic eq.computational fluid dynamics I

    xhxhhxHb

  • 7/27/2019 CFD I_2

    11/53

    The shallow water equations result into

    with boundary conditions

    impermeability , (no slip)

    discharge

    contour stresses ,

    water level

    governing equations

    shallow waterscomputational fluid dynamics I

    34

    2

    2

    2

    2

    2

    h

    xaw

    cH

    uVgn

    H

    WWC

    y

    u

    x

    uvf

    x

    hg

    y

    uv

    x

    uu

    t

    u

    34

    2

    2

    2

    2

    2

    h

    yaw

    cH

    vVgn

    H

    WWC

    y

    v

    x

    vuf

    y

    hg

    y

    vv

    x

    vu

    t

    v

    0 yvH

    xuH

    th

    0Nu 0Tu

    QdsHuN0NN

    0TT

    thth 0

  • 7/27/2019 CFD I_2

    12/53

    If in the N-S dynamic equation we substitute the non-linear velocities by a known

    velocity field and the rest of the velocities by the a scalar unknown we arrive to

    the convection diffusion equation that rules the transport of substances byconvective and diffusive actions.

    The equations are

    or in 1D

    where is the quantity being transported, kis the diffusion coefficient, Ui is the

    known velocity field, and Q are the external sources of the quantity. These are

    also known as the Transport Equations

    governing equations

    convection-diffusion equationcomputational fluid dynamics I

    f

    zyxz

    W

    y

    V

    x

    U

    t

    2

    2

    2

    2

    2

    2

    0 QkU jjjjt ,,,

    0

    Q

    xk

    xxU

    t

  • 7/27/2019 CFD I_2

    13/53

    CFDI

    finite elements in fluids

    computational fluid dynamics I

    3. Finite Elements in Fluids

  • 7/27/2019 CFD I_2

    14/53

    There is no analytical solution for most engineering problems such as

    fluid flow

    The determination of the velocity and pressure field is required in a

    domain ofinfinite degrees of freedom

    The Finite Element Method (developed about 1950 for structures)

    substitutes the domain by another with a finite number of freedom

    degrees, thus an approximation of the solution is obtained

    Some important names in the finite element history are Courant, Turner,

    Clough, Zienkiewicz, Brookes, Hughes,

    Now it is used not only in structural mechanics but also in heat

    conduction, seepage flow, electric and magnetic fields, and of course

    in fluid dynamics

    finite elements in fluids

    general issuescomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    15/53

    finite elements in fluids

    general issuescomputational fluid dynamics I

    sms.avi

  • 7/27/2019 CFD I_2

    16/53

    finite elements in fluids

    general issuescomputational fluid dynamics I

    2D h zoom at the mine.avi

    3D H(x,y), water depth colour (only 600 days).avi

    largo modulos.avi

  • 7/27/2019 CFD I_2

    17/53

    finite elements in fluids

    general issuescomputational fluid dynamics I

    X

    Y

    0 100 200

    0

    50

    100

    150

    200

    250

    VEL

    1.78125

    1.6625

    1.54375

    1.425

    1.30625

    1.1875

    1.06875

    0.950002

    0.831252

    0.712501

    0.593751

    0.475001

    0.356251

    0.237501

    0.11875

    1.69975E-06

    9.60324E-07

    4.19696E-07

    8.11084E-08

    2D H (water level).avi

  • 7/27/2019 CFD I_2

    18/53

    finite elements in fluids

    general issuescomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    19/53

    The main way of solving continuum problems in the finite element method are the

    following

    The direct approach (matrix analysis), by using a direct physical reasoning to establish

    the element properties. Requires very simple basic elements (bars, pipelines,)

    Variational approach (e.g. Rayleigh-Ritz based method), in this method the stiffness

    matrix is obtained as a result of the resolution of a variational problem

    Weighted residual approach (e.g. Galerkin Method), as a result of weighting the

    differential equations and integrating them in the domain

    finite elements in fluids

    general issuescomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    20/53

    Main steps of the finite element method

    Subdivide the domain in a finite number of elements interconnected a the nodes,

    where the unknowns (p, u) are going to be determined

    It is assumed that the variation of the unknowns can be approximated by a simple

    function

    The approximation functions are defined in terms of the values of the field

    variables at the nodes

    When the equilibrium or variational equations has been obtained the new finite

    unknowns are introduced into the equations

    The system of equations is solved and the unknowns are determined at the nodes The approximat ion functions give the solution in the rest of the domain points

    Following, the fem solution of the one simple 1-D problem is to be considered

    on a 6-step basis

    finite elements in fluids

    general issuescomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    21/53

    In the traditional Rayleigh-Ritz methods the interpolating functions have to be

    defined over the enti re domain and have to satisfy the boundary conditions.

    Meanwhile in the FEM the interpolat ing trial funct ions are defined on a finite elementbasis, being more versatile when the shape is not simple enough

    The limitation is that the FEM trial functions have to satisfy in addition some

    convergence conditions (continuity and completeness and compatibility)

    finite elements in fluids

    general issuescomputational fluid dynamics I

    '...as the nature of the universe is the most perfect and the work of the

    Creator is wiser, there's nothing that takes place in the universe inwhich the ratio of maximum and minimum does not appear. So there is

    no doubt whatsoever that any effect of the universe can be explained

    satisfactorily because of its final causes, through the help of the method

    of maxima and minima, as can be by the very causes taking place

    Leonhard Euler

    (Basel,1707- Saint Petersburg,1783)

  • 7/27/2019 CFD I_2

    22/53

    When using a variational approach, the aim is to find the vector function of

    unknowns, that makes a minimum or a maximum of the functional I (typically

    the energy)

    After the discretezation has been carried out in terms ofE smaller parts the

    piecewise approximation is introduced so that

    or in terms of the so called shape functions Ni

    where are the values of the unknowns at the nodes

    finite elements in fluids

    variational approachcomputational fluid dynamics I

    dSx

    gdVx

    FI

    ,...,,...,

    e

    aprox

    e

    ee

    NNe

    2211

    i

  • 7/27/2019 CFD I_2

    23/53

    Afterwards, the condition of extremezation ofIwith respect to i is imposed

    Adding all those element contributions it is obtained

    Assuming I to be a quadratic functional of the element equation results in

    finite elements in fluids

    variational approachcomputational fluid dynamics I

    0

    M

    i

    I

    II

    I

    2

    1

    E

    e i

    e

    i

    II

    1

    0

    eee

    e

    e

    PKI

  • 7/27/2019 CFD I_2

    24/53

    After the assembling process it is obtained

    where and

    After applying the boundary conditions the system is solved for the nodal

    unknowns i

    Once i are known, we can obtain other variables as a post-processing value

    finite elements in fluids

    variational approachcomputational fluid dynamics I

    PK

    E

    e

    e

    1

    KK

    E

    e

    e

    1

    PP

  • 7/27/2019 CFD I_2

    25/53

    Example. Find the velocity distribution of an inviscid fluid flowing trough a

    varying cross section pipe shown in the figure

    The governing equations are defined by finding the potential that minimizes the

    energy integral equation

    with the boundary condition u(x=0)=u0, where the cross section area is

    finite elements in fluids

    variational approach, examplecomputational fluid dynamics I

    dxdx

    dAI

    L

    0

    2

    2

    1

    LxeAA

    0

    u0

    A0

    L

    A1 A2

    1 2 3

    l(1) l(2)

  • 7/27/2019 CFD I_2

    26/53

    1st step. Discretization

    Divide the continuum into two finite elements. The values of the potential

    function at the three nodes will be the unknowns of the fem

    2nd step. Select an interpolation model, easy but leading to convergence

    The potential function will be taken as linear

    and can be evaluated at each element as

    where l(e) is the length of the e element

    bxax

    eeeee

    l

    xx 121

    finite elements in fluids

    variational approach, examplecomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    27/53

    3rd step. Derivation of stiffness matrices K(e) and load vectors P(e) by using

    a variational principle

    Deriving the interpolating function with respect to x it is obtained

    where the cross sectional areas can be taken for the first and second element

    as and

    where the nodal unknowns are respectively and

    eee

    l

    oe

    eeeel

    e

    eel

    e xl

    Adx

    lAdx

    dx

    dAI

    2

    2

    2

    1

    2

    12

    12

    2

    1

    2

    2

    0

    2

    12

    0

    2

    eeTe

    e

    e

    e

    ee

    e

    eeeee

    l

    A

    l

    AI K

    2

    1

    11

    11

    2

    1

    2

    2

    2

    1

    2112

    2

    1

    2

    2

    2

    10 AA

    2

    21 AA

    2

    11)(

    3

    22)(

    finite elements in fluids

    variational approach, examplecomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    28/53

    3rd step. (cont)

    The minimal potential energy principle gives , if we take into account the

    external inflow

    where Q is the mass flow rate across section

    therefore, if we derive the functional Ifor each basic element

    or in matrix form

    eTeeeTeeeee

    e

    eeeee QQ

    l

    AI QK

    2

    1

    2

    22211

    12

    2

    1

    2

    2

    0

    i

    I

    02

    2121221

    1

    12

    2

    1

    2

    2

    111

    )()()( eeee

    eeeeeeee

    e

    e

    Ql

    AQQ

    l

    AI

    022 2122212122

    1

    2

    2

    221

    )()()( eee

    e

    eeeeeeee

    e

    e

    QlAQQ

    lAI

    0

    2

    1

    eeeeTeeeTe

    ii

    eI

    QKQK

    AuQ

    finite elements in fluids

    variational approach, example

    computational fluid dynamics I

  • 7/27/2019 CFD I_2

    29/53

    4th step. Assembly of the stiffness and load vectors

    Once we have obtained the matrices for all the basic elements as

    we can assemble the system to obtainQK

    11

    111

    11

    l

    AK

    11

    112

    22

    l

    AK

    0

    111uA

    Q

    23

    20

    uAQ

    22

    00

    3

    2

    1

    2

    2

    2

    2

    2

    2

    2

    2

    1

    1

    1

    1

    1

    1

    1

    1

    0

    0

    0

    uA

    uA

    l

    A

    l

    Al

    A

    l

    A

    l

    A

    l

    A

    l

    A

    l

    A

    finite elements in fluids

    variational approach, examplecomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    30/53

    5th step. Resolution of the system

    As we need a reference value for the potentials (u3

    is an unknown) we can set

    equal to 0

    Taking A(1) as 0.80 A0 and A(2) as 0.49 A0, and l

    (1)= l(2)=L/2, the system of two

    equations with two unknowns gives

    6th step. Computation of the results

    Once we have obtained the potentials, the velocities can be derived by using

    the equivalence

    which gives the velocities at elements 1 and 2 as

    Lu01 651. Lu02 0271.

    112

    ldx

    du

    0

    1

    251 uu .

    0

    2

    052 uu .

    finite elements in fluids

    variational approach, examplecomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    31/53

    In this method the FE equations can be directly obtained from the governing

    equations (or equilibrium equations)

    The discretization is made and the field variable is approximated as

    where i are constants and Ni(x) are linearly independent functions chosen

    such that the boundary conditions are satisfied

    A quantity R known as the residual or error is defined as

    The weighted function of the residual is taken as

    where f(R)=0 when R=0

    finite elements in fluids

    weighted residualscomputational fluid dynamics I

    GF

    n

    i

    ii xNx1

    ~

    ~~

    FGR

    0 dVRwfV

  • 7/27/2019 CFD I_2

    32/53

    There are several approaches to the weighted residuals method such as the

    collocation method, the Least Squares method and the most commonly used of

    all, the Galeking method

    In the Galerkin method the weighting functions are chosen to be equal to the

    trial functions and f(R) is taken as R

    with i=1,2,,n

    In the rest of the aspects the method is similar to the variational

    finite elements in fluids

    weighted residualscomputational fluid dynamics I

    0 dVRNV

    i

  • 7/27/2019 CFD I_2

    33/53

    Example. Find the velocity distribution of an inviscid fluid flowing trough a

    varying cross section tube shown in the figure

    The governing equations are given by the continuity equation

    with the boundary condition u(x=0)=u0, where the cross section area is

    finite elements in fluids

    weighted residualscomputational fluid dynamics I

    02

    2

    dx

    d

    LxeAA 1

    u0

    A1

    L

    A2 A3

    1 2 3

    l(1) l(2)

  • 7/27/2019 CFD I_2

    34/53

    1st step. Discretization

    Divide the continuum into two finite elements. The values potential function in

    the three nodes will be the unknowns of the fem

    2nd step. Select an interpolation model, easy but leading to convergence

    The potential function will be taken as linear

    and can be evaluated at each element as

    where l(e) is the length of element e

    bxax

    eeee

    l

    xx 121

    finite elements in fluids

    weighted residualscomputational fluid dynamics I

  • 7/27/2019 CFD I_2

    35/53

    This can also be obtained through the shape functions which have to be 1 at its

    node and zero at the others, that is

    this is

    (the same as obtained before)

    finite elements in fluids

    weighted residualscomputational fluid dynamics I

    xNxNx 2211

    elxN 11

    el

    xN 21

    l(e)

    1

    l(e)

    eeel

    x

    l

    x

    l

    xx 12121 1

    fi i l i fl id

  • 7/27/2019 CFD I_2

    36/53

    3rd step. Derivation of stiffness matrices K(e) and load vectors P(e) by using

    equilibrium. Obtaining of a weak form

    The integral of the weighted residual is

    integrating by parts

    00 2

    2

    dxdxd

    wel

    i

    finite elements in fluids

    weighted residuals

    computational fluid dynamics I

    dxdxddv

    2

    2

    dxdv

    iwu idwdu

    00000

    00 2

    2

    dxdxdwdxddxdwdxldlwdwdxddxdwdxdxdw

    il

    i

    e

    eii

    ll

    i

    l

    i

    ee

    e

    e

    12

    00 uwulwdx

    dx

    d

    dx

    dwi

    e

    i

    li

    e

    fi it l t i fl id

  • 7/27/2019 CFD I_2

    37/53

    finite elements in fluids

    weighted residuals

    computational fluid dynamics I

    This is

    The elementary matrices result into

    As

    where

    As the derivatives are

    the elementary matrices result into

    0 eee PK

    122

    121

    02211

    00 uwulwdx

    dx

    dN

    dx

    dN

    dx

    dwdxNN

    dx

    d

    dx

    dwi

    e

    i

    li

    li

    ee

    dx

    dx

    dN

    dx

    dN

    dx

    dN

    dx

    dNdx

    dNdx

    dNdx

    dNdx

    dN

    dxdx

    dN

    dx

    dN

    dx

    dNdx

    dNe

    2212

    2111

    21

    2

    1

    K

    2

    1

    u

    ue

    P

    Ldx

    dN 11 Ldx

    dN 12

    11

    111

    11

    11

    0

    22

    22

    e

    l

    ee

    eee

    ldx

    ll

    lle

    K

    2

    11)(

    3

    22)(

    fi it l t i fl id

  • 7/27/2019 CFD I_2

    38/53

    4th step. Assembly of the stiffness and load vectors

    Once we have obtained the matrices for all the basic elements as

    we can assemble the system to obtainQK

    11

    1111

    1

    lK

    11

    1112

    2

    lK

    0

    01u

    P

    2

    20

    uP

    2

    0

    3

    2

    1

    22

    2211

    11

    0

    110

    1111

    011

    u

    u

    ll

    llll

    ll

    finite elements in fluids

    weighted residuals

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    39/53

    5th step. Resolution of the system

    6th step. Computation of the results

    As can be seen, the system of equations obtained by the weighted residuals

    method is the same as in the variational method except for the absence of the

    density (which can be removed as it is a constant), and the cross section areas.

    The areas are not present in the second formulation as the system is solved invelocities and not in flow rates. To avoid this fact a two dimensional model

    should be considered.

    finite elements in fluids

    weighted residuals

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    40/53

    Finite elements = Piecewise approximation of the solution by dividing the

    region into small pieces

    This approximation is usually made in terms of a power series (polynomial) which

    is easy to integrate and easy to be improved in accuracy by increasing the order,

    fitting in this way the shape of the polynomial to that of the solution (see figure)

    When the polynomial is of higher order (bigger than one) the midside and/or

    interior nodes have to be used in addition to the corner nodes

    Some other approximations such as Fourier series could also be used

    Problems involving curved boundaries can be solved using isoparametricelements which are not straight-sided

    finite elements in fluids

    discretization

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    41/53

    finite elements in fluids

    discretization

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    42/53

    The mesh can be improved by

    Subdividing selected elements (h-refinement)

    Increasing the order of the polynomial of selected elements (p-refinement) Moving node points (r-refinement)

    Defining a new mesh

    In higher order elements the midside and/or interior nodes have to be used in

    addition to the corner nodes in order to match the number of nodal degrees offreedom with the number of constants

    As it will be shown a different interpolation for the velocity and pressure

    unknowns is required for fem in fluids

    Basic elements to be considered

    Triangular linear

    Quadrilateral linear

    Triangular linear (natural)

    Triangular quadratic

    finite elements in fluids

    discretization

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    43/53

    The FEM is an approximation that converges to the exact solution as the element

    size is reduced if:

    i. The field variable and its derivatives must have representation as the element

    size reduces to zero

    For example, second derivatives cannot be represented with linear functions

    Then the elements are said to be complete

    ii. The field variable and its derivatives should be continuous within the element (Cr

    piecewise differentiable, where ris the maximum order of derivatives within the

    integrand)

    (The polynomials are inherently continuous and satisfy this requirement)

    The field variable and its derivatives, up to the r-1-th, must be continuous at the

    element boundaries

    Then the elements are said to be compatible or conforming

    finite elements in fluids

    discretization, convergence

    computational fluid dynamics I

    dxdx

    dr

    r

    finite elements in fluids

  • 7/27/2019 CFD I_2

    44/53

    If we had for instance, flat penthouses as interpolating functions, the

    interpolating surface would be discontinuous (would break and split up)

    Still, there are many fem basic elements that not verifying the former properties

    still provide meaningful solutions (such as the checker board pressure mode)

    finite elements in fluids

    discretization, convergence

    computational fluid dynamics I

    5 . 00 1 0 .0 0 1 5 .0 0 2 0 .0 0 2 5 .0 0 3 0 .0 0 3 5 .0 0 4 0 .0 0 4 5 .0 0 5 0 .0 0 5 5 .0 0

    5.00

    10.00

    15.00

    20.00

    25.00

    30.00

    35.00

    40.00

    45.00

    50.00

    55.00

    0 . 00 1 0 .0 0 2 0 .0 0 3 0 .0 0 4 0 .0 0 5 0 .0 0 6 0 .0 0 7 0 .0 0 8 0 .0 0 9 0 .0 0 1 0 0 .0 0

    0.00

    10.00

    20.00

    30.00

    40.00

    50.00

    60.00

    70.00

    80.00

    90.00

    100.00

    finite elements in fluids

  • 7/27/2019 CFD I_2

    45/53

    Let the basic linear triangular element connecting nodes 1, 2,and 3 be

    The equation that gives the

    surface (plane) is

    (1)

    that leads to the following

    equations

    yxyx 321 ,

    131211yx

    232212 yx

    333213 yx

    1

    2

    3

    22 , yx

    11, yx 33 , yxx

    y

    finite elements in fluids

    discretization, triangular linear b.e.

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    46/53

    The solution of the former system gives

    (2)

    where

    substituting (2) in (1) and rearranging terms it is obtained

    3322111 21

    aaaA

    33221122

    1 bbb

    A

    33221132

    1 ccc

    A

    132 yyb 321 yyb 231 xxc

    312 xxc

    123 xxc 213 yyb

    23321 yxyxa

    31132 yxyxa

    12213 yxyxa

    finite elements in fluids

    discretization, triangular linear b.e.

    computational fluid dynamics I

    33

    22

    11

    1

    1

    1

    2

    1

    yx

    yx

    yx

    A

    finite elements in fluids

  • 7/27/2019 CFD I_2

    47/53

    The interpolating function results

    where

    (3) The shape functions take the value of 1 at its node and cero at the rest

    These expressions are complicated and depend onx andy

    332211 ,,,, yxNyxNyxNyx

    2332233211112

    1

    2

    1xxyyyxyxyx

    Aycxba

    AyxN ,

    3113311322222

    1

    2

    1xxyyyxyxyx

    Aycxba

    AyxN ,

    1221122133332

    1

    2

    1xxyyyxyxyx

    Aycxba

    AyxN ,

    finite elements in fluids

    discretization, triangular linear b.e.

    computational fluid dynamics I

    finite elements in fluids

  • 7/27/2019 CFD I_2

    48/53

    For anA element matrix equal to

    The integrals are

    As the integrand is a constant there is no need to integrate numerically

    dxdy

    y

    N

    y

    N

    x

    N

    x

    NA

    jiji

    ij

    e

    A

    discretization, triangular linear

    computational fluid dynamics I

    dxdy

    y

    N

    y

    L

    x

    N

    x

    N

    y

    N

    y

    N

    x

    N

    x

    N

    y

    N

    y

    N

    x

    N

    x

    N

    yN

    yN

    xN

    xN

    yN

    yN

    xN

    xN

    yN

    yN

    xN

    xN

    y

    N

    y

    N

    x

    N

    x

    N

    y

    N

    y

    N

    x

    N

    x

    N

    y

    N

    y

    N

    x

    N

    x

    N

    e

    e

    333323231313

    323222221212

    313121211111

    A

    e

    dxdy

    xxyysim

    xxxxyyyyxxyy

    xxxxyyyyxxxxyyyyxxyy

    Ae

    e

    2

    12

    2

    21

    12312113

    2

    31

    2

    13

    1223213231231332

    2

    23

    2

    32

    24

    A

    finite elements in fluids

  • 7/27/2019 CFD I_2

    49/53

    The basic element matrix results

    That now can be assembled

    in the stiffness matrix to yield

    discretization, triangular linear

    computational fluid dynamics I

    2

    12

    2

    21

    12312113

    2

    31

    2

    13

    1223213231231332

    2

    23

    2

    32

    4xxyysim

    xxxxyyyyxxyy

    xxxxyyyyxxxxyyyyxxyy

    Ae

    e A

    6

    2

    9

    10

    9

    8

    7

    6

    5

    4

    3

    2

    1

    10

    9

    8

    7

    6

    5

    4

    3

    2

    1

    f

    f

    f

    ff

    f

    f

    f

    f

    f

    finite elements in fluids

  • 7/27/2019 CFD I_2

    50/53

    The need of integrating the shape functions and their derivatives overthe domain leads to the use of the natural (local) coordinates, which

    allows for an element based integration that simplifies the calculations

    The natural triangular system of referenced is defined with the lineardependent coordinatesL1, L2, andL3

    whereAi is the area defined by the point P and the opposite side

    The shape functions for this triangular linear element are

    A

    AL 11 AAL 22

    A

    AL 33

    1

    3

    2P

    A2 A1

    A3

    ii LN 321 ,,i

    discretization, triangular linear b.e.

    computational fluid dynamics I

    1321 LLL

    finite elements in fluids

  • 7/27/2019 CFD I_2

    51/53

    The shape functions are in fact as seen in (3)

    Or in matrix form

    1

    3

    2P

    A2 A1

    A3

    discretization, triangular linear b.e.

    computational fluid dynamics I

    A

    A

    yx

    yxyx

    ALN ikk

    jj

    ii2

    2

    1

    11

    2

    1

    jkkjjkkjii xxyyyxyxyxAyxLyxN 21

    ,,

    y

    x

    xxyyyxyx

    xxyyyxyx

    xxyyyxyx

    AL

    L

    L 1

    21

    12211221

    31133113

    23322332

    3

    2

    1

    33

    22

    11

    1

    1

    1

    2

    1

    yx

    yx

    yx

    A

    finite elements in fluids

  • 7/27/2019 CFD I_2

    52/53

    The derivatives ofL1,L2 andL3 being

    1

    3

    2P

    A2 A1

    A3

    discretization, triangular linear b.e.

    computational fluid dynamics I

    Ayy

    xL

    2321

    Ayy

    xL

    2132

    A

    xx

    y

    L

    2

    312

    A

    xx

    y

    L

    2

    231

    Ayy

    xL

    2213

    A

    xx

    y

    L

    2

    123

    finite elements in fluids

  • 7/27/2019 CFD I_2

    53/53

    For natural coordinates in triangles the same procedure can be usedexcept for the fact that one of the three coordinates is lineardependant and can be dropped from the integration leading to a

    change in the integration limits

    where the jacobian determinant is

    and the integral is

    Axxyyxxyy

    AyL

    xL

    yL

    xLJ

    21

    41 231331322

    1221

    121

    0

    1

    02112

    1

    0

    1

    0

    11 1dLdLLLgdLdLyxfJdxdyyxf

    LL

    ,,,

    1321 LLL

    discretization, triangular quadratic

    computational fluid dynamics I

    121

    0

    1

    021

    1

    2 dLdLLLgAdyxfL

    ,,