the plane stress problem - uppsala...

27
The Plane Stress Problem Martin Kronbichler Applied Scientific Computing (Tillämpad beräkningsvetenskap) February 2, 2010 Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 1 / 24

Upload: others

Post on 01-Apr-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

The Plane Stress Problem

Martin Kronbichler

Applied Scientific Computing (Tillämpad beräkningsvetenskap)

February 2, 2010

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 1 / 24

Page 2: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Outline

Plane stress

FEM in 2DDiscrete interpolation

Triangular elementsRectangular elements

The algebraic problemSolution, postprocessing

Accuracy of FEM

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 2 / 24

Page 3: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Plane stress

A 2D problem: plane stress

Assumptions:I thin specimenI free to move normal to the plane

(x3 direction, i.e. direction ofunit vector e3)

I only loaded in the plane (x1, x2

direction)

Image from http://en.wikipedia.org.

Thus no stress normal to the plane, but specimen typically “bulges” ([sv.bukta]) in the normal direction so displacements is typically not zeronormal to the plane (u3 6= 0).

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 3 / 24

Page 4: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Plane stress

Plane stress modeling

No stress (force) normal to the plane ⇒ σ · e3 = 0⇒ third column of σ is zero, and, because of σ = σT, also the third row

σ =

σ11 σ12 0σ12 σ22 00 0 0

=(σ 00T 0

).

Constitutive law:

σ = λI tr ε+ 2µε =E

1 + ν

1− 2νI tr ε+

22ε

]⇒ ε =

(ε 00T ε33,

), where ε33 = − ν

1− νtr ε.

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 4 / 24

Page 5: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Plane stress

Plane stress modeling

No stress (force) normal to the plane ⇒ σ · e3 = 0⇒ third column of σ is zero, and, because of σ = σT, also the third row

σ =

σ11 σ12 0σ12 σ22 00 0 0

=(σ 00T 0

).

Constitutive law:

σ = λI tr ε+ 2µε =E

1 + ν

1− 2νI tr ε+

22ε

]⇒ ε =

(ε 00T ε33,

), where ε33 = − ν

1− νtr ε.

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 4 / 24

Page 6: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Plane stress

Plane stress modeling

Express constitutive law in terms of upper 2× 2 block,

σ =E

1 + ν

1− νI tr ε+ ε

]or

σ = λI tr ε+ 2µε, λ =Eν

(1 + ν)(1− ν)

Conclusion: for plane stress, compute in 2D with modified constitutivelaw

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 5 / 24

Page 7: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Plane stress

Plane strainAdjoint/dual problem to plane stress (observe: not the mathematical“adjointness”)

Assumptions:I no displacements orthogonal to the

plane (a thick specimen or a restrictedmovement)

I loads constant in direction orthogonalto the plane

Assumptions ⇒

ε =

ε11 ε12 0ε12 ε12 00 0 0

=(ε 00T 0

).

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 6 / 24

Page 8: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D

Finite Element Discretization for PlaneStress Problems

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 7 / 24

Page 9: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D

Finite Element (FE) discretiztation for 2D linear elasticity

Consider the equilibrium equation∫Ω

[λ tr(ε(v)

)tr(ε(u)

)+ 2µε(v) : ε(u)

]dΩ =

∫Ωv · f dΩ +

∫Γ1

v · g ds,

or, in abbreviated form,

a(v,u) = l(v) for all admissible v

FEM aims at satisfying this equation for asubset of admissible displacements

→ work on a discretization associated witha triangulation the domain, consisting ofnonoverlapping triangles (quadrilaterals alsocommon) An example triangulation

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 8 / 24

Page 10: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D

Finite Element (FE) discretiztation for 2D linear elasticity

Consider the equilibrium equation∫Ω

[λ tr(ε(v)

)tr(ε(u)

)+ 2µε(v) : ε(u)

]dΩ =

∫Ωv · f dΩ +

∫Γ1

v · g ds,

or, in abbreviated form,

a(v,u) = l(v) for all admissible v

FEM aims at satisfying this equation for asubset of admissible displacements

→ work on a discretization associated witha triangulation the domain, consisting ofnonoverlapping triangles (quadrilaterals alsocommon) An example triangulation

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 8 / 24

Page 11: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Discrete interpolation

Admissible displacements uh arecontinuous on Ω and polynomials on eachtriangle (or quadrilateral).

An example of one component of anadmissible displacement when uh is a linearfunction (first-order polynomial) at eachtriangle.

For piecewise-linear functions, the nodes are the vertices of thetriangulations → lowest-order linear Lagrange element or thethree-node triangle

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 9 / 24

Page 12: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Discrete interpolation

Admissible displacements uh arecontinuous on Ω and polynomials on eachtriangle (or quadrilateral).

An example of one component of anadmissible displacement when uh is a linearfunction (first-order polynomial) at eachtriangle.

For piecewise-linear functions, the nodes are the vertices of thetriangulations → lowest-order linear Lagrange element or thethree-node triangle

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 9 / 24

Page 13: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Each admissible displacement uh isexpressed as a weighted sum ofbasis functions φj(x):

uh =N∑

j=1

φj(x)uj .Basis function for continuous,

piecewise-linear functions

I The sum interpolates discrete displacement vectors uj located atnodes (interpolation points)

I A nodal basis function is a continuous function in the finite elementspace that is

I one at one node, andI zero at all other nodes

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 10 / 24

Page 14: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Lagrange elements Pp

Interpolation properties of nodal basis functions (•):I Polynomials of degree p on each triangle (element)I Continuous across each edge

Linear: P1

The three-node triangleu(x, y) = a1 + a2x+ a3y

Quadratic: P2

The six-node triangleu(x, y) = a1 + a2x+

a3y + a4x2 + a5xy + a6y

2

Cubic; P3

The ten-node triangle

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 11 / 24

Page 15: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Example basis functions

Functions within a single element:

For quadratic Lagrangian elements, there are two distinct types of nodalbasis functions:

Associated with mesh vertex Associated with edge midpoints

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 12 / 24

Page 16: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Example basis functions II

Basis functions on their support:

The basis function φi(x) when i correspondsto a corner node

The basis function φi(x) when icorresponds to an edge-midpoint node

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 13 / 24

Page 17: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Rectangular elements

A common element type for 2D elasticity.The displacements at the nodes (•) are interpolated by a product ofone-dimensional polynomials in each coordinate direction

The four-node rectangular element Q1.Bilinear, 4 coefficients

u(x, y) = a1 + a2x+ a3y + a4xy

(obtained as product of φ(x) = s1 +s2x and ψ(y) =t1 + t2y)

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 14 / 24

Page 18: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

Rectangular elements, II

The nine-node rectangular element Q2.Biquadratic, 9 coefficients

u(x, y) = a1 + a2x+ a3y + a4xy + a5x2 + a6y

2

+ a7x2y + a8xy

2 + a9x2y2

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 15 / 24

Page 19: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Discrete interpolation

The four-node Q1 on general quadrilaterals:I Linear on each edgeI Not in general of formq(x, y) = a0 + a1x+ a2y + a3xy!

I A quadrilateral is not the image of an affinemap of a rectangle!

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 16 / 24

Page 20: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D The algebraic problem

The algebraic problem

Problem (FE equation for linear elasticity and plane stress)Find displacement field uh, such that

a(vh,uh) = l(vh) for all vh ∈ V h. (2)

Form of an admissible displacement in the finite element subspace:

uh(x) =N∑

j=1

φj(x)uj =N∑

j=1

φj(x)(u1,je1 + u2,je2) =2N∑j=1

φj(x)uj (3)

where, for j = 1, . . . , N ,

u2j−1 = u1,j

u2j = u2,j

φ2j−1(x) = e1φj(x)

φ2j(x) = e2φj(x).

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 17 / 24

Page 21: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D The algebraic problem

Substitute (3) into FE equation and test with vh = φk, k = 1, . . . , 2N .Hence, (2) is equivalent to requiring

2N∑j=1

a(φk,φj)uj = l(φk) for k = 1, . . . , 2N,

which is a linear system Au = b , where

Akj = a(φk,φj) =∫

Ω

[λ tr(ε(φk)

)tr(ε(φj)

)+ 2µε(φk) · ε(φj)

]dΩ ,

bk = l(φk) =∫

Ωφk · f dΩ +

∫Γ1

φk · g ds,

u = (u1, u2, . . . , u2N )T = (u1,1, u2,1, u1,2, u2,2, . . . , u1,N , u2,N )T

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 18 / 24

Page 22: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D The algebraic problem

Practical implementation of elastic equations

Assembly of matrix A by a loop over all cells; deal.II open source finite element librarybased on C++, authors W. Bangerth, R. Hartmann, G. Kanschat,http://www.dealii.org, tutorial step-8

for (CellIterator cell = cell_begin; cell != cell_end; ++cell) phi.reinit (cell);cell_matrix = 0;for (int q_point = 0; q_point < n_q_points; ++q_point)

for (int i = 0; i < dofs_per_cell; ++i)for (int j = 0; j < dofs_per_cell; ++j)

cell_matrix(i,j)+= (lambda * trace(phi[u].gradient(i,q_point)) *

trace(phi[u].gradient(j,q_point))+2 * mu * phi[u].symmetric_gradient(i,q_point) *

phi[u].symmetric_gradient(j,q_point)) * phi.JxW(q_point);

cell->get_dof_indices(dof_indices);global_matrix.add (dof_indices, cell_matrix);

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 19 / 24

Page 23: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

FEM in 2D Solution, postprocessing

Solution and postprocessing

I Solve linear system Au = bI Direct: Gauss/Cholesky decomposition, COMSOL: UMFPACK (sparse

direct solver)I Iteratively: for larger size (> 500 000 degrees of freedom in 2D,> 100 000 in 3D) → CG with preconditioning, multigrid

I Postprocessing:I Translate components in u to displacements at certain pointsI Get stresses and/or strains from derivatives of (3), using kinematic and

constitutive relation

−→ Computer lab

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 20 / 24

Page 24: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Accuracy of FEM

Accuracy of FEM

On the accuracy of the discrete FEMsolutions

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 21 / 24

Page 25: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Accuracy of FEM

Error estimatesDiscretization errors measured in integral norms. If everything is “nice”, wehave for triangular or rectangular elements∥∥∥u− uh

∥∥∥L2(Ω)

=(∫

Ω

∣∣∣u− uh∣∣∣2 dΩ

)1/2

≤ Chp+1.

h: largest edge in mesh; p: element polynomial order; C depends on the(p+ 1)-th derivatives of u. Estimate requires some “niceness” conditions,typically

I Nondegenerate mesh refinements:router/rinner bounded as h→ 0 (or thatthe minimum (or maximum) angle isbounded away from 0 (or 180)

I Smooth solutions. Accuracy typicallyreduced in vicinity of reentrant cornerson the boundary (ω > 180)

routerrinner

!

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 22 / 24

Page 26: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Accuracy of FEM

Remarks

I Stresses are functions of derivatives of the displacement (constitutive& kinematic relation)

I Therefore: accuracy of stress approximations typically one orderlower than the displacement approximations

I In particular: stresses are piecewise constant for the P1 element(3-node triangle) → poor stress approximation

I Rule of thumb: P1 too inaccurate (P2 much better), see computerlab, task 1.1.

I Adaptive grid refinement (e.g. around reentrant corners) and errorestimates central aspects in CSM

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 23 / 24

Page 27: The Plane Stress Problem - Uppsala Universityuser.it.uu.se/~martinkr/AppliedScientificComputing/...Plane stress Plane stress modeling Nostress(force)normaltotheplane)˙e 3 = 0)thirdcolumnof˙iszero,and,becauseof˙=

Accuracy of FEM

P1 versus Q1

I Advantage Q1

Q1 (4-node rectangle) performs usually much better than P1 (bothhave error term h2!)Reason: the presence of the quadratic xy term for Q1 (stresses notconstant in element)⇒ Rectangular elements can give higher accuracy for the samenumber of degrees of freedom, particularly in 3D

I Disadvantage quadrilateralsMesh generation more difficult and less automatic for rectangularmeshes compared to triangles

Martin Kronbichler (TDB) The Plane Stress Problem February 2, 2010 24 / 24