convection diffusion pd

Upload: sandyengineer13

Post on 03-Apr-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 Convection Diffusion PD

    1/23

  • 7/28/2019 Convection Diffusion PD

    2/23

    assume a known flow field

    xxxt jj

    j

    j

    I II III IV

    We know how to handle I, III and IV. Term II is new

    Consider thatu is given, and satisfies the continuity equation as:

    jju

    xt

  • 7/28/2019 Convection Diffusion PD

    3/23

    Steady1D convection and diffusion

    convection- ddddiffusion

    equation:

    dxdx

    u

    dx

    continuity constuu

    dx or0

    x)x)ww x)x)ee (Assume u is given)

    PPWW EE

    xxww ee

    xx

  • 7/28/2019 Convection Diffusion PD

    4/23

    Discretization using central differences`

    we

    wedx

    d

    dx

    duu

    .

    diffusionnet

    P

    W

    E (piece(piece--wise linear profile)wise linear profile)::

    1

    xxPPWW EE

    ww ee WPw

    PEe

    212

    WPwPEe

    wWPePE uu

    2121we

  • 7/28/2019 Convection Diffusion PD

    5/23

    Discretization using centraldifferencesRe-arran in the discretization e uation

    WWEEPP aaa

    www

    e

    e

    ee

    E

    Fu

    FD

    u

    xa

    22

    weWEwwee

    P

    wW

    FFaauu

    a

    xa

    22

    F = u represents the advection strength, while D = /x is the

    di usion stren th.

  • 7/28/2019 Convection Diffusion PD

    6/23

    Example: letue = uw = 4, e/x = w/x = 1

    LetE=200, W=100, findP

    ee u

    321

    2

    wwW

    E

    ua

    x

    2 weWEP uuaaa

    (?)50 P

    P

    Similarly, letE=100, W=200, then P= 250 (?)

  • 7/28/2019 Convection Diffusion PD

    7/23

    The central difference scheme is found to be suitable for flows in

    which diffusion strength is high compared to the advection

    strength

    The relevant non-dimensional number representing the relativestrength of the two transport mechanisms is called the Peclet

    number (P = uL/), where L is a representative length scale and

    is the diffusivity.

    The central difference scheme assumes equal influence of the a ues o wo a acen gr po n s n e erm n ng e convec veflux at the interface of the control volume between those grid

    points. This assumption will be valid only for low Peclet number

    .

    For very high Peclet number (Pe > 2)flows, the influence of theupstream node will be dominant at the interface, and this is the

    basis of the U wind method.

  • 7/28/2019 Convection Diffusion PD

    8/23

    In this scheme, the value ofat an interface is equal to that at the

    node on the upstream side of the interface.

    0if

    0if

    eEe

    ePe

    F

    F

    Accordingly, the advective flux at the interfaces e and w can be written as:

    0,0, eEePe uMaxuMaxu 0,0, wPwWw uMaxuMaxu

    e

    WWEEPP aaa wwww

    W

    eeeE

    FDux

    a

    uxa

    0,0,

    ,,The discretized equation iswritten as

    weWE

    w

    w

    e

    e

    P

    uuaa

    u

    x

    u

    x

    a

    0,0,

  • 7/28/2019 Convection Diffusion PD

    9/23

    Ex onential scheme exact solution The Upwind scheme is ideal for high Peclet number flows (Pe >

    2).

    , ,

    analytical solution for variation ofover the length of a one-dimensional control volume can be observed.

    -

    dx

    d

    dx

    du

    dx

    d

    x 0L

    o

    Lx

    Solution:

    1exp

    1exp

    P

    LP

    oL

    o

  • 7/28/2019 Convection Diffusion PD

    10/23

    Ex onential scheme contd In the exponential scheme, the solution above is used as the profile

    assumption forbetween successive grid points.

    respectively, in a local coordinate system where the origin corresponds to theface at which the flux of needs to be calculated, the discreization equation is:

    www

    ee

    e

    E

    DFF

    DFa

    )/exp(

    1)/exp(

    eeee FxuP )()(

    weWEPww

    FFaaa

    D

    1)/exp( ee

    volume length for any Peclet number.

    However, approximate representations ofvariations are requiredin the context of a numerical solution for com utational efficienc

    The Hybrid and the Power Law schemes are two such examples

    described subsequently.

  • 7/28/2019 Convection Diffusion PD

    11/23

    The H brid SchemeThe hybrid differencing scheme (Spalding, 1972) is based on a combination

    of central and upwind differencing schemes. The central differencing

    sc eme, w ic is suita e or sma Pec et num er, is emp oye or Pec et

    numbers (Pe < 2). The upwind scheme, which is more suitable for higherPeclet number, (Pe 2). The hybrid scheme is essentially a piecewise linear

    inter olation o the ex onential scheme. A summar o the scheme is:

    ,2For eEe PaP

    21,22For e

    e

    E

    e

    e

    P

    D

    aP

    0,2For e

    E

    eD

    aP

  • 7/28/2019 Convection Diffusion PD

    12/23

    The Power-law differencing scheme of Patankar (1980) is a more accurate

    approximation than the hybrid scheme. In the hybrid scheme, the influence of

    diffusion is switched off for Pe>2, which may be too soon. In the power lawscheme, diffusion is set to zero when cell Pe exceeds 10. If 0 < Pe < 10, the

    . ,

    ,10For eEe PaP

    1.01,010For 5 eee

    E

    e

    e

    PPD

    aP

    1.01,100For 5 ee

    E

    e

    a

    PDaP

    ,or e

    eD

  • 7/28/2019 Convection Diffusion PD

    13/23

    yxtyx

    nn u

    PPWW EE

    ww ee yy

    vJ

    x

    y

    yyxx

    SSxxz=1z=1

  • 7/28/2019 Convection Diffusion PD

    14/23

    Discretized unstead 2-D convection-diffusion equation (fully implicit)

    oldoldold yxc baaaaTa SSNNWWEEPP

    PSNWEP

    ppCp

    x

    yxSaaaaa

    t

    etc.

    e c.

    nn

    ee

    n

    n

    w

    w

    e

    e

    FP

    FP

    xxx

    .assumptionprofileondependssne

    PA

  • 7/28/2019 Convection Diffusion PD

    15/23

    Scheme Formula for

    Central difference 1- 0.5P S

    PA

    pw n

    Hybrid 0, 1- 0.5P

    Power law 0, (1- 0.1P)5

    Exponential P/ [exp(P)-1]

  • 7/28/2019 Convection Diffusion PD

    16/23

    False diffusion Commonly misinterpreted and controversial topic

    Common view: - Central diff. is second order accurate

    - Upwind causes severe false diffusion

    Ta lor series central di . : Truncation error TE x 2

    Upwind: can be shown that TE(x)

    ,

    to forward or backward difference)

    ommon notion: Upwin ess accurate. However, since

    ~ x variation is exponential, Taylor series approx. is good

    only for smallx (not good for high Peclet number)

    Upwind better in these circumstances

  • 7/28/2019 Convection Diffusion PD

    17/23

    False diffusion (continued) Compare coefficients:

    Central diff. Upwind

    2

    e

    eE

    F

    Da 0,eeE FDa , e . e e

    in Upwind i.e. replaced by2

    xu

    Flaw in argument: we have assumed that central diff.

    scheme is the most perfect

    Conclusion:Small Peclet No.: central accurate; UPWIND over-predicts

    us on; power aw, y r , exponen a en o cen ra

    Large Peclet No.: central fails, others are correct

  • 7/28/2019 Convection Diffusion PD

    18/23

    False diffusion (continued)

    Make = 0. Any effect of diffusion will be false diffusion.

    100 C, Hot 100 C, Hot

    0 C, Cold 0 C, Cold

    0 = 0

  • 7/28/2019 Convection Diffusion PD

    19/23

    Uniform flow in x directionNN

    Hot

    PPWW EE

    ColdSS

    UPWIND SCHEME

    Since = 0, and Fn = Fs = 0 aN= aS= 0.

    lso, a = 0 UPWIND

    aP= aW P= W

    ,

    NO FALSE DIFFUSION

  • 7/28/2019 Convection Diffusion PD

    20/23

    Uniform flow at 45 to the grid lines

    NNUPWIND (or high Peclet No.)

    PPWW EE

    ,

    N, E downstream neighbours

    SSHot

    Cold

    Since = 0, and aE= aN= 0.

    Also, aW= aS P= 0.5 W+ 0.5 S

    ARTIFICIAL (FALSE) DIFFUSION OCCURED

  • 7/28/2019 Convection Diffusion PD

    21/23

    Quadratic Upstream Interpolation for

    Higher order differencing scheme: more accurate

    minimise false diffusion: can take into account corner nodes

    3 - point quadratic interpolation

    E EEPW

    WW

    WWWW

    ww

    PPWW EE EEEE

    ee

    at interface is obtained from a quadratic function passing

    Control volume Control volume boundary

    through the two surrounding nodes (one on each side of the

    face) and a node on the upstream side

  • 7/28/2019 Convection Diffusion PD

    22/23

    Let ue > 0, uw >0

    = =for uniform grid, quadratic interpolation gives:

    w W P- WW

    e= (6/8)P+ (3/8) E- (1/8) W

    For the diffusion terms, use gradient of the appropriate parabola

    aaaa

    aaa 1163

    weWWWEP

    wewwee

    FFaaaa 8888

    Stability: Conditionally stable (high Pe, negative coeff. possible)

  • 7/28/2019 Convection Diffusion PD

    23/23

    MODULE 4 Review Questions

    Define Peclet number in a convection-diffusion problem. Why is the significance of Peclet

    number in a convection-diffusion problem? Why does the Central difference scheme fail for

    high Peclet number problems?

    Show that the ex onential scheme is a ro riate from a h sical reasonin stand oint.

    Show that the upwind scheme can introduce false diffusion in low Peclet number problems.

    The power law scheme is a close appropriation of the exponential scheme, and is therefore

    regarded as a very accurate scheme. However, the exponential scheme is a result of analytical

    .

    problems is you use power law scheme.

    What is false diffusion? Can false diffusion occur because of the gridding system? What are

    the remedies for false diffusion? For various Peclet number values (Pe = -20, -7, 0, 5, 10), calculate the aE/De from the power

    law scheme and from the exponential scheme, and express the difference as a percentage of

    the exact aE/De value. Although you might get a rather large percentage difference for some

    values of Pe, the power law scheme is expected to give good results. Why?

    Why is the QUICK scheme more accurate than conventional schemes? Does it alwaysconverge?