xfem as an alternative for the classical h-refinement€¦ · 2 the xfem as an alternative for...

34
1 XFEM as an Alternative for the Classical h-refinement by Safdar Abbas & Thomas-Peter Fries USNCCM July 17, 2009

Upload: others

Post on 13-Aug-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

1

XFEM as an Alternative for the Classicalh-refinement

bySafdar Abbas & Thomas-Peter Fries

USNCCMJuly 17, 2009

Page 2: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

2

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 3: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

3

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 4: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

4

Enrichment Functions for High Gradient Solutions

• Motivation: Standard FEM (No Stabilization)

Page 5: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

5

Enrichment Functions for High Gradient Solutions

• Motivation: Stabilized FEM (SUPG Stabilization)

Page 6: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

6

Enrichment Functions for High Gradient Solutions

• XFEM approximation.

– Standard finite element approximation.

– Enrichment.

W

W

s

• Instead of stabilization and/or refinement we want to enrich the

approximation space.

Page 7: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

7

Enrichment Functions for High Gradient Solutions

• Enrichment Functions

– Weak discontinuity Abs-Enrichment

– Strong discontinuity Sign/Heaviside Enrichment

Page 8: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

8

Enrichment Functions for High Gradient Solutions

• Enrichment functions–Regularized sign function [Patzak and Jirasek, 2003] ( C4 continuous at ).

Page 9: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

9

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 10: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

10

Optimal Set of Enrichment Functions

Can be integrated

Can be captured

by FEM

Interpolation Problem:Interpolated function fInterpolating functions Ψ = [ψ1, ψ2, ψ3]

Find ∫ ωuh = ∫ ωf, for ω ∈ Ψ, where uh = ∑ ψiui= ΨΤuminimize the error |f – uh|

Page 11: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

11

Enrichment Functions for High Gradient Solutions

• An optimal set of 7 enrichment functions.

• Enrichment functions are relative to the element size.

Page 12: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

12

Enrichment Functions for High Gradient Solutions

• An optimal set of 7 enrichment functions.

• Enrichment functions are relative to the element size.

Page 13: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

13

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 14: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

14

Stationary High Gradient Developing Over Time

• First test case: In-stationary Burger’s Equation with stationary high gradientthat develops over time.

– Time-Stepping for the temporal discretization.

– Non-linear term is linearized using Newton-Raphson iterations.

– Diffusion coefficient is very small.

– No stabilization is used.

– Position of the highest gradient is known and stationary.

Page 15: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

15

Stationary High Gradient Developing Over Time

XFEM Results

(No Stabilization)

Page 16: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

16

Stationary High Gradient Developing Over Time

Page 17: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

17

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 18: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

18

Moving High Gradient (position known a priori)

• Second test case: In-stationary linear advection equation with moving high

gradient specified as an initial condition.

– Time stepping is not fully appropriate.

– Equation is discretized using Space-Time discretization with

Discontinuous-Galerkin in time.

– No stabilization is used.

– Position of the highest gradient known a priori at each time.

Page 19: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

19

Moving High Gradient (position known a priori)

XFEM Results

(No stabilization)

Page 20: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

20

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 21: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

21

Moving High Gradient (position not known)

• Third test case: In-stationary Burger’s Equation with unknown position ofthe highest gradient.

– Level-set function is transported using transport equation for the level-set.

– Equation is discretized using Space-Time discretization withDiscontinuous-Galerkin in time.

– Non-linear term is linearized using Newton-Raphson iterations.– Diffusion coefficient is very small.– No stabilization is used.– Position of the highest gradient in each time step is found iteratively by

a strong coupling loop.

Page 22: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

22

Moving High Gradient (position not known)

Strong Coupling

Solution ofBurger’s Equation

Solution ofTransportEquation

Solution ofBurger’s Equation

Solution ofTransportEquation

tn tn+1

Page 23: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

23

Moving High Gradient (position not known)

XFEMResultsSpace-Timeview

Page 24: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

24

Moving High Gradient (position not known)

FEMResults

Page 25: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

25

Moving High Gradient (position not known)

XFEMResults

Page 26: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

26

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori).

• Conclusions.

• Future outlook.

Page 27: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

27

Moving High Gradient in 2D (position known a priori).

• Fourth test case: A high gradient scalar function transported in a circular

velocity field.

Page 28: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

28

Moving High Gradient in 2D (position known a priori).

2d Advectionequation

(No stabilization)

Page 29: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

29

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions.

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 30: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

30

Conclusions

• A complete range of gradients is captured using an optimal set of high

gradient enrichment functions.

• No oscillations are observed near the high gradient.

• Solution quality is better than that achieved from stabilization without

refining the mesh.

Page 31: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

31

The XFEM as an Alternative for h-Adaptivity

• Enrichment functions for high gradient solutions.

• Optimal set of enrichment functions

• Stationary high gradient developing over time.

• Moving high gradient (position known a priori).

• Moving high gradient (position not known).

• Moving high gradient in 2D (position known a priori)..

• Conclusions.

• Future outlook.

Page 32: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

32

Future Outlook

• Using the optimal set of enrichment functions to simulate the cohesive

cracks in quasi-brittle materials.

Page 33: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

33

Financial support from the DeutscheForschungsgemeinschaft (German ResearchAssociation) through grant GSC 111 is gratefullyacknowledged.

Acknowledgements

Page 34: XFEM as an Alternative for the Classical h-refinement€¦ · 2 The XFEM as an Alternative for h-Adaptivity •Enrichment functions for high gradient solutions. •Optimal set of

34

Thanks for your Attention