310 likes | 323 Views
Learn about computational modeling of interfaces using level sets and key techniques like surface tracking, EScript & Finley, and advection methods. Includes examples and implications for studying phenomena like Rayleigh–Taylor instabilities. Find the software at www.esscc.uq.edu.au.
E N D
The Australian Computational Earth Systems Simulator (ACcESS) Modelling Deforming Interfaces using Level Sets by Hans Mühlhaus, Laurent Bourgouin and Lutz Gross
Overview Introducing Example What’s needed to model interfaces • Constitutive models • Surface tracking, level set, stress advection EScript & Finley Separation of Physics from Linear Algebra and Parallel Computing Applications • Advection, upwinding, implicit vs. explicit • Lava dome simulation,Subduction, Rayleigh-Taylor Instabilities
Moving Interface: a 1D Example v2 h=small j<0 x2 h=large j>0 h x1 We define: so that Update: Note that: Time integration:
Governing equations Stress Equilibrium Heat Equation Concentration advection: Temperature and concentration dependence of density:
Example for Rayleigh – Taylor Instabilities using level sets: Mantle Plumes
The General Case • Implicit representation of the interface by the zero level set of a smooth function φ • φis usually chosen as a • “signed” distance function ( ) • At each time step, φ is updated solving the (hyperbolic) advection equation:
Problems…… • Symmetric difference expressions like • (symm.) (non-symm.) • don’t work well in hyperbolic problems (upwinding etc!) • 2. Inhomogeneous velocity field causes loss • of distance function property ( ) of j
Problems……(cont.) • Symmetric difference expressions don’t • work well in hyperbolic problems (upwinding!) 1. Upwinding If v is constant:
Problems…… This can be transformed into a non-symmetric expression by adding…. We expect that the FE approx. of the PDE: is better conditioned than the original Hyperbolic problem
Generalisations….. Taylor-Galerkin: 2-step alternative to Taylor-Galerkin upwinding (very effective in the presence of diffusionterms….):
The Level Set Method: Solving the advection equation Test: A Gaussian is advected in a constant 1D velocity field. • Explicit • Implicit • Taylor Galerkin
Software can be downloaded fromwww.esscc.uq.edu.au, contact Ken Steube (esys@access.edu.au) If you need instructions re libraries etc Formulation Finley PDE: Example : Momentum and Heat equation
LinearPDE class General form (as relevant here): PDE: natural boundary condition Kronecker symbol: δij=0 for i=j and 0 otherwise
Helmholtz Class in mytools.py from esys.linearPDEs import LinearPDE import numarray class Helmholtz(LinearPDE): def setValue(self,kappa,omega,f,eta,g): ndim=self.getDim() # spatial dimension kronecker=numarray.identity(ndim) self._setValue(A=kappa*kronecker,\ D=omega,Y=f,d=eta,\ y=g)
Use the Helmholtz Class # Helmholtz class defined in mytools.py from mytools import Helmholtz mydomain=... mypde=Helmholtz(mydomain) mypde.setValue(kappa=10,omega=0.1,\ f=12,eta=0,g=0) u=mypde.getSolution()
2. Problem: Inhomogeneous velocity field causes loss of distance function property of j Previous test: No topological change in the solution Need for a new test with: and New test: shear flow • Mesh: 100x100 • Courant Number: 0.25 • 1000 steps forward • 1000 steps with -v
The Level Set Method: Solving the advection equation The shape gets “noisy”… Problem: φ looses its distance function property Reinitialisation needed!
The Level Set Method: Reinitialisation • Idea: • Rebuild a “signed” distance function ψ from the distorted function φ • Requirements: • The interface must not be changed • ψ must represent a distance function • Solution: • Solve to steady state the equation: • Rewritten as: with Interpretation: The “distance information” is carried by w, a unit vector pointing away from the interface.
Remarks on re-initialisation….. • During iteration (pseudo time integration) the vector w is established once and then kept constant • In the explicit solution of the advection problem for y we found that only alumped mass matrix discretisation works
The Level Set Method: Reinitialisation Same test as before, with reinitialisation
Level set cont. : Merger of small and large bubbles Parameters: Surface tension: Calculation, includes inertia, Courant Number=0.5, msh:30 by 45 8 node quad’s
Level set cont. : Calculation of curvature for C_0 continuity Projection: and Representation of surface tension b.c. as volume force: =distance in the direction of the normal of l smoothing length, related to the element size
Level set: Surface membrane shell, surface tension where Inserting yields where at equilibrium.
Remarks • Escript & Finley: Rapid development of simulation software; parallelised assembly and solution phase; separation of physics from linear algebra • Level set modelling of interfaces: distance function property crucial • Modelling of surface tension; example of higher order b.c.’s • Upwinding strategy dependent on element type • Re-initialisation strategy has an (undesirable) element of mystique…..