test on mls

Upload: xzfeng

Post on 05-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/2/2019 test on MLS

    1/15

    dataname: 11-18-11 w1BF s1 t7 z139

    bead1(0.6229, 0.3963, 0.3388)

    (7.7

    104,

    2.2

    104,

    0.0035)

    bead2(0.6249, 0.4593, 0.3262) (4.7 104,0.0022,0.0056)directly estimated normal strain: ex = 0.150, ey = 0.031, ez = 0.167maximum MLS estimated normal strain: e = 0.048

    = underestimated strain components.

    Figure 1: Displacement reconstruction of a set of real data.

    Figure 2: MLS-calculated strain component e11 at t=16 hours.

    J(x) =

    NI=1

    w(x, xI)(pT(xI)a(x) uI)2

    w(x, xI) = w(dI, dmI), dI =

    ||x

    xI

    ||uh(x) = pT(x)a(x) =

    I(x)uI

    1

  • 8/2/2019 test on MLS

    2/15

    minimization ofJ(x) in moving least square regression leads to

    a(x) = A1BU, where

    A = PTWP =

    N=1

    w(x, x)p(x)pT(x),

    B = PTW =w(x, x1)p(x1), , w(x, xN)p(xN)

    a,i(x) = (A

    1

    ,i B + A1B,i(x))U, where

    A1,i = A1A,iA1.

    Figure 3: [2]

    Visibility criterion ( [1] ): include MLS node P only if the evaluation point

    is visible from it discontinuous shape function.[4] proves that the discontinuous approximation generated by the visibility cri-terion converge at the same rate as continuous EFG approximation.

    Figure 4: Left: [3]; Right: [2]

    2

  • 8/2/2019 test on MLS

    3/15

    Ways to construct smooth shape function for non-concave bodies( [3] )

    A. diffraction method: wrap the domain of influence around a concave boundary.

    modified parameter:

    dI(x) = (s1 + s2(x)

    s0(x))s0(x)

    s0(x) = ||x xI||2, s1(x) = ||xw xI||2, s2(x) = ||x xw||2,

    derivative calculation:

    dw

    dxi=

    w

    dI

    dI

    xidI

    xi= (

    s1 + s2s0

    )s2

    xi+ (1 )(s1 + s2

    s0)s0

    xi

    s0xi

    = xi xIis0

    , s2xi

    = xi xcis2

    Figure 5: Top: [4]; Bottom: [3]

    3

  • 8/2/2019 test on MLS

    4/15

    B. transparency method: endow boundaries within the domain of influence

    with a varying measure of transparency.

    modified parameter:

    dI(x) = s0(x) + dmI (sc(x)

    sc), 2

    Figure 6: [3]

    Numerical results on infinite plate with a hole:

    Figure 7: [3]

    4

  • 8/2/2019 test on MLS

    5/15

    method efficiency accuracy implementationplain

    visibilitydiffraction

    Table 1: Interpolation time for 20 nodes.

    Numerical experiment:

    Test A)

    u1 = (x2

    1 + x2

    2) 0.1,u2 = (x1 + x2 sin(x3)) 0.1,u3 = x1x2x3

    0.1;

    Test B) Infinite 3D specimen with a hole under internal pressure

    Hole radius: a, internal pressure: p.

    (r) =a3p

    4

    1

    r3

    ui = (r)xi

    Tr = pa3

    r3

    T =pa3

    2r3

    Coupled FE-EFG method: allows for the use of the EFG method in thecrack region and the finite element method to handle complex geometries and

    essential boundary conditions.

    Enriched MLS basis representative for crack tip fields( [3]):

    pT(x) =

    1,x,y,z,

    r cos

    2,r sin

    2,r sin sin

    2,r cos cos

    2

    References

    [1] T. Belytschko. 1994 Belytschoko etal Element free galerkin methods.pdf.

    [2] Petr Krysl and Ted Belytschko. Element-free Galerkin method: Convergence

    of the continuous and discontinuous shape functions. Computer Methods in

    Applied Mechanics and Engineering, 148(3-4):257277, September 1997.

    5

  • 8/2/2019 test on MLS

    6/15

    Figure 8: Short term results for test A. Data number: 10000.

    [3] D Organ, M Fleming, T Terry, T Belytschko, and T Terry. Continuous mesh-

    less approximations for nonconvex bodies by diffracti0n and transparency.

    Computational Mechanics, 18, 1996.

    [4] N. Sukumar, B. Moran, T. Black, and T. Belytschko. An element-free

    Galerkin method for three-dimensional fracture mechanics. Computational

    Mechanics, 20(1-2):170175, July 1997.

    6

  • 8/2/2019 test on MLS

    7/15

    Figure 9: Long term results for test A. Data number: 10000.

    7

  • 8/2/2019 test on MLS

    8/15

    Figure 10: Short term relative L2 error for test B. Data number: 10000. Meshdistance to the sphere: 0.01.

    8

  • 8/2/2019 test on MLS

    9/15

    Figure 11: Short term relative L2 error for test B. Data number: 20000. Meshdistance to the sphere: 0.01.

    9

  • 8/2/2019 test on MLS

    10/15

    Figure 12: Long term relative error for test B. Left: Data number: 50000.Right: Data number: 10000. Point distance to the sphere: 0.001.

    10

  • 8/2/2019 test on MLS

    11/15

    Figure 13: Long term relative error for test B. Left: Data number: 50000.Right: Data number: 10000. Point distance to the sphere: 0.001. (Cont)

    11

  • 8/2/2019 test on MLS

    12/15

    Figure 14: Long term relative error for test A. Data number: 10000. Pointdistance to the sphere: 0.001.

    12

  • 8/2/2019 test on MLS

    13/15

    Figure 15: Relative error for test B. dI=0.1. Left: Data number: 100000. Right:Data number: 10000.

    13

  • 8/2/2019 test on MLS

    14/15

    Figure 16: Relative error for test B. dI=0.1. Left: Data number: 100000. Right:Data number: 10000(Cont).

    14

  • 8/2/2019 test on MLS

    15/15

    Figure 17: Convergence test for test B. Point distance to the sphere: 0.001.dI = 0.1

    15