• Aucun résultat trouvé

A meshless method for the Reissner-Mindlin plate equations based on a stabilized mixed weak form using maximum-entropy basis functions

N/A
N/A
Protected

Academic year: 2021

Partager "A meshless method for the Reissner-Mindlin plate equations based on a stabilized mixed weak form using maximum-entropy basis functions"

Copied!
29
0
0

Texte intégral

(1)

A meshless method for the Reissner-Mindlin plate equations based on a stabilized mixed weak form

using maximum-entropy basis functions

J.S. Hale*, P.M. Baiz

11th September 2012

(2)

Introduction

Aim of the research project

Develop a meshless method for the simulation of Reissner-Mindlin plates that is free of shear locking.

Figure : 6th free vibration mode of SSSS plate

(3)

Existing Approaches (FE and Meshless)

I Reduced Integration (Many authors)

I Assumed Natural Strains (ANS) (eg. MITC elements, Bathe)

I Enhanced Assumed Strains (EAS) (Hughes, Simo etc.)

I Discrete Shear Gap Method (DSG) (Bletzinger, Bischoff, Ramm)

I Smoothed Conforming Nodal Integration (SCNI) (Wang and Chen)

I Matching Fields Method (Donning and Liu)

I Direct Application of Mixed Methods (Hale and Baiz)

The Connection

Many of these methods are based on, or have been shown to be

equivalent to, mixed variational methods.

(4)

Key Features of the Method

Weak Form

Design is based upon on a Stabilised Mixed Weak Form, like many successful approaches in the Finite Element literature.

Stabilised Mixed Weak Form (Brezzi and Arnold 1993, Boffi and Lovadina 1997)

a b (θ; η) + λαa s (θ, z 3 ; η, y 3 ) + (γ, ∇ y 3 − η) L

2

= f (y 3 ) (1a) ( ∇ z 3 − θ, ψ) L

2

− t ¯ 2

λ(1 − α t ¯ 2 ) (γ ; ψ) L

2

= 0 (1b)

(5)

Key Features of the Method

Basis Functions

Uses (but is not limited to!) Maximum-Entropy Basis Functions

which have a weak Kronecker-delta property. On convex node sets

boundary conditions can be imposed directly.

(6)

Key Features of the Method

Localised Projection Operator

Shear Stresses are eliminated on the ‘patch’ level using a localised

projection operator which leaves a final system of equations in the

displacement unknowns only.

(7)

The Reissner-Mindlin Problem

(8)

The Reissner-Mindlin Problem

Displacement Weak Form

Find (z 3 , θ) ∈ ( V 3 × R ) such that for all (y 3 , η) ∈ ( V 3 × R ):

Z

0

L(θ) : (η) d Ω + λ t ¯ −2 Z

0

( ∇ z 3 − θ) · ( ∇ y 3 − η) d Ω

= Z

0

gy 3 d Ω

(2)

or:

a b (θ; η) + λ ¯ t −2 a s (θ, z 3 ; η, y 3 ) = f (y 3 ) (3) Locking Problem

Whilst this problem is always stable, it is poorly behaved in the

thin-plate limit ¯ t → 0

(9)

Shear Locking

10

2

10

3

number of degrees of freedom

10

7

10

6

L

2

error in z

3

t = 0.002

t = 0.02

t = 0.2

(10)

Shear Locking

The Problem

Inability of the basis functions to represent the limiting Kirchhoff mode

∇ z 3 − η = 0 (4) A solution?

Move to a mixed weak form

(11)

Mixed Weak Form

Treat the shear stresses as an independent variational quantity:

γ = λ ¯ t −2 ( ∇ z 3 − θ) ∈ S (5)

Mixed Weak Form

Find (z

3

, θ, γ ) ∈ ( V

3

× R × S ) such that for all (y

3

, η, ψ) ∈ ( V

3

× R × S ):

a

b

(θ; η) + (γ; ∇ y

3

− η)

L2

= f (y

3

) (6a)

( ∇ z

3

− θ; ψ)

L2

− t ¯

2

λ (γ; ψ)

L2

= 0 (6b)

Stability Problem

Whilst this problem is well-posed in the thin-plate limit, ensuring

stability is no longer straightforward

(12)

Stabilised Mixed Weak Form

Displacement Formulation Locking as ¯ t → 0

Mixed Formulation Not necessarily stable

Solution

Combine the displacement and mixed formulation to retain the

advantageous properties of both

(13)

Stabilised Mixed Weak Form

Split the discrete shear term with a parameter 0 < α < ¯ t −2 that is independent of the plate thickness:

a s = αa displacement + (¯ t −2 − α)a mixed (7)

(14)

Stabilised Mixed Weak Form

Mixed Weak Form

Find (z

3

, θ, γ ) ∈ ( V

3

× R × S ) such that for all (y

3

, η, ψ) ∈ ( V

3

× R × S ):

a

b

( θ ; η ) + ( γ ; ∇ y

3

− η )

L2

= f (y

3

) (8a)

( ∇ z

3

− θ; ψ)

L2

− t ¯

2

λ (γ; ψ)

L2

= 0 (8b)

Stabilised Mixed Weak Form (Brezzi and Arnold 1993, Boffi and Lovadina 1997)

a b (θ; η) + λαa s (θ, z 3 ; η, y 3 ) + (γ, ∇ y 3 − η) L

2

= f (y 3 ) (9a) ( ∇ z 3 − θ, ψ) L

2

− t ¯ 2

λ(1 − α t ¯ 2 ) (γ ; ψ) L

2

= 0 (9b)

(15)

α independence

10

4

10

3

10

2

10

1

thickness ¯ t 10

3

10

2

10

1

e

H1

( z

3

)

α independence with fixed discretisation h = 1/8, α = 32.0

eH1 (z3 )

(16)

Eliminating the Stress Unknowns

I Find a (cheap) way of eliminating the extra unknowns associated with the shear-stress variables

γ h = λ(1 − α ¯ t 2 )

¯ t 2 Π h ( ∇ z 3h − θ h , ψ h ) (10) Figure : The Projection Π

h

represents a softening of the energy

associated with the shear term

(17)

Eliminating the Stress Unknowns

I We use a version of a technique proposed by Ortiz, Puso and Sukumar for the Incompressible-Elasticity/Stokes’ flow problem which they call the “Volume-Averaged Nodal Pressure” technique.

I A more general name might be the “Local Patch Projection”

technique.

(18)

Eliminating the Stress Unknowns

(19)

Eliminating the Stress Unknowns

For one component of shear (for simplicity):

(z 3 , x − θ 1 , ψ 13 ) L

2

− t ¯ 2

λ(1 − α ¯ t 2 ) (γ 13 ; ψ 13 ) L

2

= 0 (11) Substitute in meshfree and FE basis, perform row-sum

(mass-lumping) and rearrange to give nodal shear unknown for a node a. Integration is performed over local domain Ω a :

γ 13a =

N

X

i=1

R

a

N a {− φ i φ i ,x } d Ω R

a

N a d Ω

φ i

z 3i

(12)

(20)

Choosing α

The dimensionally consistent choice for α is length −2 . In the FE literature typically this paramemeter has been chosen as either h −1 or h −2 where h is the local mesh size.

Meshless methods

A sensible place to start would be ρ −2 where ρ is the local support

size.

(21)

2.4 2.6 2.8 3.0 3.2 3.4 3.6 log

10

(dim U)

− 2

− 1 0 1 2 3 4

log

10

( α ) α ∼ 1/ρ

2

Convergence surface for e

L2

(z

3

)

− 4.0

− 3.5

− 3.0

− 2.5

− 2.0

− 1.5

− 1.0

− 0.5

0.0

(22)

2.4 2.6 2.8 3.0 3.2 3.4 3.6 log

10

(dim U)

− 2

− 1 0 1 2 3 4

log

10

( α ) α ∼ 1/ρ

2

Convergence surface for e

H1

(z

3

)

− 3.0

− 2.5

− 2.0

− 1.5

− 1.0

− 0.5

0.0

(23)

Results - Convergence

10

2

10

3

10

4

dim U 10

5

10

−4

10

3

10

−2

10

1

e

Convergence of deflection in norms with α ∼ O(h

2

)

eL2 (z3 ) eH1 (z3 )

(24)

Results - Surface Plots

Figure : Displacement z

3h

of SSSS plate on 12 × 12 node field +

‘bubbles’, t = 10

4

, α = 120

(25)

Results - Surface Plots

Figure : Rotation component θ

1

of SSSS plate on 12 × 12 node field +

‘bubbles’, t = 10

4

, α = 120

(26)

Summary

A method:

I using (but not limited to) Maximum-Entropy basis functions for the Reissner-Mindlin plate problem that is free of

shear-locking

I based on a stabilised mixed weak form

I where secondary stress are eliminated from the system of equations a priori using “Local Patch Projection” technique Possible future work:

I Extension to Naghdi Shell model

I Investigate locking-free PUM enriched methods

(27)

Thanks for listening.

(28)

LBB Stability Conditions

Theorem (LBB Stability)

The discretised mixed problem is uniquely solvable if there exists two positive constants α h and β h such that:

a b (η h ; η h ) ≥ α h k η h k 2 R

h

∀ η h ∈ K h (13a)

ψ inf

h

∈S

h

sup

h

,y

3h

)∈(R

h

×V

3h

)

(( ∇ y 3h − η h ), ψ h ) L

2

( k η h k R

h

+ k y 3h k V

3h

) k ψ h k S

h0

≥ β h (13b)

(29)

LBB Stability Conditions

The Problem

I To satisfy the second condition 13b make displacement spaces R h × V 3h ‘rich’ with respect to the shear space S h

I If R h × V 3h is too ‘rich’ then the first condition 13a may fail as K h grows.

I Balancing these two competing requirements makes the

design of a stable formulation difficult.

Références

Documents relatifs

If the plate P is loaded in such a way that the solution of the Reissner–Mindlin model (2.12) is bending dominated, i.e., the corresponding Kirchhoff–love solution w 0 is not zero,

Undertaking numerical testing of new element designs that might have been tedious, lengthy and error prone with a traditional FE package where ‘elements’ (function spaces) and

Meshless methods such as the Element Free Galerkin method [2] have been applied to the solution of the Reissner-Mindlin plate equations.. Similarly to Finite Element methods,

As it is mentioned in the Introduction to this paper, in order to obtain the Reissner- Mindlin kinematics for a plate through the asymptotic analysis or Γ-convergence, we need

We consider a conforming finite element approximation of the Reissner-Mindlin eigenvalue system, for which a robust a posteriori error estimator for the eigenvector and the

We propose a new robust a posteriori error estimator based on H (div ) con- forming finite elements and equilibrated fluxes.. It is shown that this estimator gives rise to an

That modal synthesis, both applied on dis- placements and generalized stress fields parameters show a good convergence with our case (double plate with a thick- ness t = 1e − 3 m) as

It turns out that this enriched space can be embedded into the standard Hybrid High-Order (HHO) space for elasticity originally introduced in [21] (see also [20, Chapter 7] and [12]