44
Adrian Egger Applications of the Scaled Boundary Finite Element Method in Linear Elastic Fracture Mechanics

SBFEM_Demo.pdf (PDF, 2.5 MB)

Embed Size (px)

Citation preview

Page 1: SBFEM_Demo.pdf (PDF, 2.5 MB)

Adrian Egger

Applications of the Scaled Boundary Finite Element Method in Linear Elastic Fracture Mechanics

Page 2: SBFEM_Demo.pdf (PDF, 2.5 MB)

Contents

Motivation

SBFEM computational steps

SBFEM vs. ABAQUS: A displacement based comparison

Crack related phenomena ABAQUS demo

xFEM demo

SBFEM demo

Numerical comparisons Edge crack under uniform tension

Edge crack under uniform shear

Slant crack under uniform tension

Conclusion

Questions12/10/2015 2Adrian Egger | FEM II | HS 2015

Page 3: SBFEM_Demo.pdf (PDF, 2.5 MB)

Motivation I

12/10/2015 3Adrian Egger | FEM II | HS 2015

FEM has been extendedto handle a multitude of problems:

Non-linearity Geometric

Material

Dynamics

Contact problems

Crack problems

Optimization problems

Inverse problems

http://gem-innovation.com/services/mesh-independent-fem/

Page 4: SBFEM_Demo.pdf (PDF, 2.5 MB)

Motivation II

12/10/2015 4Adrian Egger | FEM II | HS 2015

#DOF▲

Non linear

Dynamic

Cracks / Contact

Only partially alleviated by:

• Vectorization and parallelization of code

• Multiscale schemes, sub-structuring, …

• Fast multipole boundary element method (FMBEM)

• Extended finite element method (xFEM)

• Etc.

Page 5: SBFEM_Demo.pdf (PDF, 2.5 MB)

Conceptual Comparison of FEM, BEM and SBFEM

12/10/2015 5Adrian Egger | FEM II | HS 2015

FEM:

- High amount of DOF

- Crack surface discretized

- Discretization error in all

directions

BEM:

- Discretization on boundary only

- Crack surface discretized

- non-symmetric dense matrices

SBFEM bounded domain:

- Introduction of a scaling center

- Discretization on boundary only

- Crack surface not discretized

- Analytical solution in radial

direction

SBFEM unbounded domain:

- Introduction of a scaling center

- Discretization on boundary only

- Analytical solution in radial

direction

Page 6: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM Fundamentals I:

12/10/2015 6Adrian Egger | FEM II | HS 2015

Page 7: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM Fundamentals II:

Applying the principle of virtual work* yields 2 equations:

General solution for displacements is assumed of form:

12/10/2015 7Adrian Egger | FEM II | HS 2015

Boundary:

Domain:

where

* Derivation at end of presentation

Page 8: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM Fundamentals III:

Solution assumed as a power series: Notice the similarity to the mode superposition method!

8

Adrian Egger | FEM II | HS 2015

http://web.sbe.hw.ac.uk/acme2011/Handout_Scaled_boundary_methods_CA.pdf

12/10/2015

Page 9: SBFEM_Demo.pdf (PDF, 2.5 MB)

Leads to a Hamiltonian eigenvalue problem of [Z]: Introducing a new variable leads to a first order differential equation

Substituting the general solution into the obtained equation:

Having calculated the decomposition

SBFEM Fundamentals IV:

12/10/2015 9Adrian Egger | FEM II | HS 2015

with:

Displacements:

Forces:

boundeddomain

unboundeddomain

�������� = + �� ����

�������� = − Φ�� Φ����

Page 10: SBFEM_Demo.pdf (PDF, 2.5 MB)

Basic Analysis Procedure

12/10/2015 10Adrian Egger | FEM II | HS 2015

+

Reduction of problem dimension by 1

Semi-analytical solution in radial direction

Bounded and unbounded domains

-

Assembly of multiple coefficient matrices

Schur/Eigen decomposition

Fully populated stiffness matrix

Page 11: SBFEM_Demo.pdf (PDF, 2.5 MB)

Comparison to ABAQUS reference solution

12/10/2015 11Adrian Egger | FEM II | HS 2015

ABAQUS:#DOF = 50’000+

Real time = 11s

Time savings:

102x @ ~0.1‰

0.8 s

-0.1%

0.1%

0.2%

-0.1%

0.1%

0.2%

0.8 s

Page 12: SBFEM_Demo.pdf (PDF, 2.5 MB)

Cost of using higher order elements

12/10/2015 12Adrian Egger | FEM II | HS 2015

#DOF

tim

e0.8 seconds

360 DOF

Page 13: SBFEM_Demo.pdf (PDF, 2.5 MB)

Stress Intensity factors (SIFs)

12/10/2015 13Adrian Egger | FEM II | HS 2015

http://solidmechanics.org/Text/Chapter9_3/Chapter9_3.php

Analog to stress concentration factors in i.e. tunneling

In fracture mechanics predicts stress distributionnear crack tip and is useful for providing a failure criterion:

��� �, � =�

������ � + higher order terms

K = stress intensity factor

��� = function of load and geometry

Page 14: SBFEM_Demo.pdf (PDF, 2.5 MB)

What is being compared?

12/10/2015 14Adrian Egger | FEM II | HS 2015

ABAQUS xFEM SBFEM

Countour Integral Local enrichment Analytical limit in radial direction

Compiled (Fortran?) Scripted (Matlab) Scripted (Matlab)

Vectorized Non-vectorized Non-vectorized

Highly optimized Proof of concept code Proof of concept code

Total CPU time Matlab tic - toc Matlab tic - toc

Page 15: SBFEM_Demo.pdf (PDF, 2.5 MB)

ABAQUS: Contour Integral

Integral based ontractions and displacements

Requires information aboutcrack propagation direction

Cannot predict how a crack will propagate

12/10/2015 15Adrian Egger | FEM II | HS 2015

http://imechanica.org/files/Modeling%20fracture%20mechanics.pdfP.H. Wen, M.H. Aliabadi, D.P. Rooke, A contour integral for the evaluation of stress intensity factors, Applied Mathematical Modelling, Volume 19, Issue 8, August 1995, Pages 450-455, ISSN 0307-904X, http://dx.doi.org/10.1016/0307-904X(95)00009-9

Page 16: SBFEM_Demo.pdf (PDF, 2.5 MB)

ABAQUS DEMO

12/10/2015 16Adrian Egger | FEM II | HS 2015

Page 17: SBFEM_Demo.pdf (PDF, 2.5 MB)

Extended finite element method (xFEM) I

12/10/2015 17Adrian Egger | FEM II | HS 2015

FEM XFEM FEM XFEM

Goal: Separate geometry from mesh XFEM achieves this by locally enriching the FE approximation with

local partitions of unity enrichment functions

Page 18: SBFEM_Demo.pdf (PDF, 2.5 MB)

Extended finite element method (xFEM) II

xFEM aims to overcome the shortcomings of FEM

Does so by introducing two kinds of enrichment Jump enrichment

Tip enrichment

Achieves: Higher accuracy for

stresses at crack tip

Less remeshing required

Level set method used to efficiently track cracks

12/10/2015 18Adrian Egger | FEM II | HS 2015

courtesy of Kostas Agathos

Page 19: SBFEM_Demo.pdf (PDF, 2.5 MB)

xFEM: Jump enrichment

12/10/2015 19Adrian Egger | FEM II | HS 2015

Heaviside step function

courtesy of Kostas Agathos

Page 20: SBFEM_Demo.pdf (PDF, 2.5 MB)

xFEM: Tip enrichment

Analytical solution for the crack problem solvedby Westergaard (1939) using a complex Airy stress function

These can be spanned by the following basis, which are used as enrichment functions for the crack tip

12/10/2015 20Adrian Egger | FEM II | HS 2015

courtesy of Kostas Agathos

Page 21: SBFEM_Demo.pdf (PDF, 2.5 MB)

xFEM DEMO

12/10/2015 21Adrian Egger | FEM II | HS 2015

Page 22: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM: Analytical limit in radial direction

12/10/2015 22Adrian Egger | FEM II | HS 2015

Strain:

Stress:

Take limit as ξ→0; Singularity for eigenvalues -1 < λ < 0

By matching expressions with the exact solution:

where:

Page 23: SBFEM_Demo.pdf (PDF, 2.5 MB)

Effects of Stress Smoothing

12/10/2015 23Adrian Egger | FEM II | HS 2015

Exact solution

Page 24: SBFEM_Demo.pdf (PDF, 2.5 MB)

Error Estimator for Stress Intensity Factors

12/10/2015 24Adrian Egger | FEM II | HS 2015

Raw SIF

Recovered SIF

SIF error estimator

Calc

ula

ted

Err

or

in S

IF [

%]

Page 25: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM DEMO

12/10/2015 25Adrian Egger | FEM II | HS 2015

Page 26: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical experiments: Stress intensity factors

12/10/2015 26Adrian Egger | FEM II | HS 2015

1 2 3

Page 27: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical Experiment 1

12/10/2015 27Adrian Egger | FEM II | HS 2015

1.0%

1.0%

0.5%

0.1%

0.5%

0.1%

DOF106

SIF

K1

Page 28: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical Experiment 1

12/10/2015 28Adrian Egger | FEM II | HS 2015

1.0%

1.0%

0.5%

0.1%

0.5%

0.1%

seconds102

SIF

K1

Page 29: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical Experiment 2

12/10/2015 29Adrian Egger | FEM II | HS 2015

1.0%

1.0%

0.5%

0.1%

0.5%

0.1%

seconds102

SIF

K1

Page 30: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical Experiment 2

12/10/2015 30Adrian Egger | FEM II | HS 2015

1.0%

1.0%0.5%0.1%

0.5%0.1%

seconds102

SIF

K2

Page 31: SBFEM_Demo.pdf (PDF, 2.5 MB)

Numerical Experiment 3

12/10/2015 31Adrian Egger | FEM II | HS 2015

SIF

K2

SIF

K1

seconds102 seconds

102

0.5%

2.0%

Page 32: SBFEM_Demo.pdf (PDF, 2.5 MB)

Observations

Incredibly fast convergence of SBFEM using few DOF

Exeptional speed considering non-optimized code

Stress recovery methods dramatically improve accuracy atvirtually no additional computational cost

Always choose at least two elements per side

Three elements per side are better

Do not use two node elements if possible

12/10/2015 32Adrian Egger | FEM II | HS 2015

Page 33: SBFEM_Demo.pdf (PDF, 2.5 MB)

Conclusion

SBFEM combines many of the desireable characteristics ofFEM and BEM into one method with additional benefits ofist own: Analytical solution in radial direction:

Higher accuracy per DOF

permits elegant and efficient calculation of stress intensity factors

Stress recovery enhances results greatly Must only be performed on the boundary

Large workload can be performed in advance

No change necessary to solution process to extract crack related phenomena (i.e. SIFs and T-stress of various orders of singularity)

Dense and fully populated matrices: Higher order elements don’t (noticeably) impact performance

12/10/2015 33Adrian Egger | FEM II | HS 2015

Page 34: SBFEM_Demo.pdf (PDF, 2.5 MB)

Questions

12/10/2015 34Adrian Egger | FEM II | HS 2015

Page 35: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation I

12/10/2015 35Adrian Egger | FEM II | HS 2015

Page 36: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation II

Geometry transformation

Jacobian

Differential unit volumen

12/10/2015 36Adrian Egger | FEM II | HS 2015

Page 37: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation III

The linear differential operator L may thus be written as:

12/10/2015 37Adrian Egger | FEM II | HS 2015

with

Page 38: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation IV

Assuming an analytical solution in radial direction:

And therefore the strains and stresses become:

12/10/2015 38Adrian Egger | FEM II | HS 2015

Page 39: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation V

Setting up the virtual work formulation:

12/10/2015 39Adrian Egger | FEM II | HS 2015

Page 40: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation VI

12/10/2015 40Adrian Egger | FEM II | HS 2015

Page 41: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation VII

12/10/2015 41Adrian Egger | FEM II | HS 2015

Page 42: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation VIII

Introducing some substitutions

Leads to some significant simplifications

12/10/2015 42Adrian Egger | FEM II | HS 2015

Page 43: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation IX

12/10/2015 43Adrian Egger | FEM II | HS 2015

Page 44: SBFEM_Demo.pdf (PDF, 2.5 MB)

SBFEM derivation X

12/10/2015 44Adrian Egger | FEM II | HS 2015