CutFEM on hierarchical B-Spline Cartesian grids with applications to fluid-structure interaction
1. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
CutFEM on hierarchical B-Spline Cartesian grids
with applications to fluid-structure interaction
Dr. C. Kadapa, Dr. W. G. Dettmer and Prof. D. Peri´c
Zienkiewicz center for Computational Engineering
Swansea University, Swansea, UK.
06-June-2016, ECCOMAS 2016, Crete Island, Greece.
6. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Fictitious Domain/Distributed Lagrange multiplier method
Advantages
Equal-order interpolation for velocity and
pressure
Works very well for thin structures
Issues
Zeroes on the matrix diagonal
Zeroes on the matrix diagonal due to
Lagrange multipliers
Unnecessary DOF when bulky solids are
present
8. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
What is CutFEM?
Advantages
Clean interfaces.
B-Splines of any order, along
with hierarchical refinement.
Stabilised formulation for fluids.
Fewer DOF.
9. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
What is CutFEM?
Advantages
Clean interfaces.
B-Splines of any order, along
with hierarchical refinement.
Stabilised formulation for fluids.
Fewer DOF.
Issues
Integration of cut cells.
Ill-conditioned matrices due to
small cut cells.
10. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Formulation
Incompressible Navier-Stokes
Find velocity, v = (v1, v2, v3), and pressure, p, such that
ρf ∂vf
∂t
+ ρf
(vf
· ∇)vf
− µf
∆vf
+ ∇p = gf
in Ωf
(1a)
∇ · vf
= 0 in Ωf
(1b)
vf
= vs
on Γf
D (1c)
σf
· nf
= tf
on Γf
N (1d)
where,
σf
= µ∇vf
− p I (2)
13. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Formulation (contd..)
Ghost penalty (Burman et al. [1])
Jump operator for a scalar valued problem:
gs(w, v) :=
k
j=1 F ∈F
h2(j−1)+s
F
[Dj
w][Dj
v] ds
[Djz] normal derivative of z, of order j, on face F
and k is degree of B-Splines.
Jump operator for a vector valued problem:
Gs(w, v) :=
d
i=1
k
j=1 F ∈F
h2(j−1)+s
F
[Dj
wi][Dj
vi] ds
Figure: Ghost-penalty
operators are applied on blue
coloured edges.
The ghost-penalty operator is given as,
Bf
GP({wf
, q)}, {(v, p)}) = γu
GP µ G1(wf
, vf
) + γp
GP
1
µ
g3(q, p) (9)
In this work, only the non-zero derivative of highest-order is considered.
14. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Integration of cut-cells - Subtriangulation
Advantages
Exact integration for boundaries with straight
edges.
Easy when boundaries are discretised with
straight edges.
Fewer Gauss points.
Disadvantages
Difficult for generic geometries in 3D. Need to
use constrained tetrahedralisation.
Difficult when boundaries are discretised with
curved elements or NURBS.
15. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Integration of cut-cells - adaptive integration
Subtriangularion: nGP = 420; % error = 1.55e-6.
(a) Level-3 adaptive integration
102
103
104
105
106
Number of Gauss points in cut-cells
10-5
10-4
10-3
10-2
10-1
100
abs(%errorinarea)
(b) nGP for different levels
16. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Integration of cut-cells - adaptive integration - merging
Subtriangularion: nGP = 420; % error = 1.55e-6.
(a) Level-3 adaptive integration
102
103
104
105
106
Number of Gauss points in cut-cells
10-5
10-4
10-3
10-2
10-1
100
abs(%errorinarea)
without merging
with merging
(b) nGP for different levels
17. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Integration of cut-cells - binary tree
(a) quad-tree (b) binary-tree
Figure: Adaptive integration - quad-tree Vs binary-tree.
18. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Integration of cut-cells - adaptive integration
(a) Level-3 adaptive integration
102
103
104
105
106
Number of Gauss points in cut-cells
10-5
10-4
10-3
10-2
10-1
100
abs(%errorinarea)
quad-tree without merging
binary-tree without merging
quad-tree with merging
binary-tree with merging
(b) nGP for different levels
Binary subdivision helps to reduce number of Gauss points in cut-cells by
20%. (In 3D, for sphere, the reduction is about 30%.)
20. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
CutFEM - Basic scheme
W. G. Dettmer, C. Kadapa, and D. Peri´c. Ingredients for an
immersed interface methods for FSI. COUPLED Problems, Venice,
Italy, May-2015.
W. G. Dettmer, C. Kadapa, and D. Peri´c. A stabilised immersed
boundary method on hierarchical b-spline grids. Submitted for
publication.
24. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Flow past a cylinder - Re = 100
0 50 100 150 200
Time
1.0
1.1
1.2
1.3
1.4
1.5
1.6
Dragcoefficient(CD)
0 50 100 150 200
Time
−0.4
−0.3
−0.2
−0.1
0.0
0.1
0.2
0.3
0.4
Liftcoefficient(CL)
Figure: CD and CL for Re = 100 with Q2 B-Splines with ∆t = 0.1.
Data CD CL St
Liu et al. [4] 1.35 ±0.339 0.165
Le et al. [3] 1.37 ±0.323 0.160
Kadapa et al. [2] 1.39 ±0.339 0.166
Present 1.37 ±0.331 0.167
Table: Comparison of CD, CL and St.
25. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Pressure relief valve
19111
5
3.2
pin
pout
noslip
noslip
1.0 0.50.5
2.1
0.4
g
r
Table: DOF comparison
Level FDM CutFEM % of FDM
2 99720 72030 72.23
3 149352 95289 63.80
4 327648 168360 51.38
33. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Flow past a sphere in 3D
Figure: Triangles = 776. DOF = 178084 for Q1.
34. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Flow past a sphere in 3D
0 20 40 60 80 100 120 140 160
Reynolds number (Re)
0
1
2
3
4
5
Dragcoefficient(CD)
Shirayama
Schlichting
Analytical
CutFEM-Q1
CutFEM-Q2
Figure: Drag coefficient (CD) versus Reynolds number (Re).
36. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Summary and Conclusions and Future work
Summary and Conclusions
Implemented and tested CutFEM on hierarchical B-Spline grids.
Higher-order spatial discretisations for the background grid.
Demonstrated the performance with several examples.
The scheme is stable, accurate and robust.
Can capture large deformations and topological changes effectively.
Future work
Fine tuning ghost-penalty parameters
Covering/uncovering
Completely closed scenarios
Parallelisation
Fluid-flexible solid interactions.
37. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
Acknowledgements
We thank Schaeffler Technologies AG & Co. KG, Germany, for funding
this project.
Thank you
38. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
References
E. Burman, S. Claus, P. Hansbo, M. G. Larson, and A. Massing.
CutFEM: Discretizing geometry and partial differential equations.
International Journal for Numerical Methods in Engineering,
104:472–501, 2014.
C. Kadapa, W. G. Dettmer, and D. Peri´c.
A fictitious domain/distributed Lagrange multiplier based
fluidstructure interaction scheme with hierarchical B-Spline grids.
Computer Methods in Applied Mechanics and Engineering,
301:1–27, 2016.
D. V. Le, B. C. Khoo, and J. Peraire.
An immersed interface method for viscous incompressible flows
involving rigid and flexible boundaries.
Journal of Computational Physics, 220:109–138, 2006.
C. Liu, X. Sheng, and C. H. Sung.
Preconditioned multigrid methods for unsteady incompressible flows.
Journal of Computational Physics, 139:35–57, 1998.
39. Motivation Formulation Adaptive integration Numerical examples Summary and Conclusions
References (contd..)
H. Schlichting.
Boundary-Layer Theory
McGraw-Hill, USA, 1979.
S. Shirayama.
Flow past a shpere: Topological transitions of the vorticity field
AIAA Journal, 30:349–358,1992.