TWiki> Main Web>ImmersedBoundary (2012-10-11, FlavienBillard)

Implementation of the Immersed Boundary method in Code_Saturne

The following describes a proposal for implementation of an immersed boundary method into Code_Saturne. It is following the approach detailed in . We consider here a simplified framework: 2-D and fixed geometry (.i.e. fixed mesh and lagrangian points)

Variables and arrays used

• nsupmax maximum number of points in the support of a lagrangian point
• nls number of lagrangian points
• xlag(nls), ylag(nls) x and y coordinates of the lagrangian points
• ds(nls) arc step length
• hx(nls), hy(nls) dilatation parameters in the x and y direction
• bmat(6,nls) matrix of the RKPM method (6 because we assume we're in 2-D)
• nsup(nls) number of points of the support of a given lagragian point
• supelt(nsupmax,nls) element index (i.e. Code_Saturne "iel") of the support
• epsmat(nls) coefficients

Steps of the method

• definition of the lagrangian points: nls, xlag, ylag, ds
• definition of the support for each lagragian point
• we find the nearest cell center I using findpt
• we define the neighborood as the set of cells which share a node with the cell I
• we define hx and hy using the definition given in Eq.32, Eq.33 and Eq.34 of 
• we calculate the support using the Eq.35 of  (we choose %$\varepsilon_x=0$% and %$\varepsilon_y=0$% for now)
• we then define nsup and supelt
• calculation of the RKPM matrix b for each lagrangian points
• we calculate the moment matrix M (terms rescaled by the dilitation factors to avoid ill-conditionning), it needs the regular_delta function and the list of powers as defined in Eq.28 of 
• we calculate the determinant of M
• if det(M)=0 we set b=[1 0 0 0 0 0]
• if det(M)<>0 we perform a Cholesky decomposition of M = LL'
• we find the vector y1 so that L*y1=[1 0 0 0 0 0]
• we find the vector b so that L'*b=y1
• b is multiplied by H (rescaling matrix defined in Eq.37 of )
• calculation of the epsilon coefficients
• calculation of the terms of the matrix A (as defined in Eq.44 of ): we only consider the 5 right-hand and left-hand neighbours of a given lagrangian point
• the system M*epsilon = [1 ..... 1] is solved. A Krylov-type iterative method is suggested for the resolution, no preconditionning required  (lagrangian points spacing ~ eulerian grid spacing).

Linear algebra

The following LAPACK subroutines are used:
• SDOT
• SPOFA (Cholesky decomposition)
• SPOSL (solves the real symmetric positive definite system using SPOFA)
• SAXPY (ax+y vector operation)
• SGEFA (LU decomposition)
• SGESL (solves a general system of linear equations, after factorization by SGEFA)

References

 Pinelli, A., I.Z. Naqavi, Piomelli, U. and Favier, J. Immersed-boundary methods for general finite-difference and finite-volume Navier-Stokes solvers. Journal of Computational Physics, Vol. 229, pp. 9073-9091 (2010)

Current Tags:
create new tag
, view all tags
Topic revision: r3 - 2012-10-11 - 14:47:12 - FlavienBillard
Main Web
16 Oct 2019

Manchester CfdTm
Code_Saturne

ATAAC
KNOO

DESider
FLOMANIA