Meshfree Methods#

Brief Introduction#

Excellent overviews of meshfree methods are found in [belytschko1996], [fries2004] and [liu2003].

As reviewed by Wallstedt [wallstedt2009], one of the methods developed to circumvent the limitations of the FEM when very large displacements are required (e.g. extrusion or intrusion simulations where remeshing is mandatory), is the Smooth Particle Hydrodynamics (SPH) Method developed by Lucy [lucy1977] and Monaghan [monaghan1998] and coworkers to model astrophysics problems and later extended to solid mechanics by Libersky, Petschek and Randles (see [libersky1990] and [randles1996]). These first models suffered from instability and lack of convergence, which as corrected by Swegle [swegle1995], Johnson and Beissel [johnson1996], Dilts [dilts2000] and others. Despite correcting the stability and increasing the accuracy of the SPH, these developments introduced some inconvenience, such as keeping track of additional stress points.

As also reviewed by Wallstedt [wallstedt2009], another branch of development based on the Diffuse Element Method (DEM) of Nayroles and coworkers [nayroles1992]. Belytschko, Lu and Gu (see [belytschko1994], [lu1994] and [belytschko1994_2]) reformed the original proposal in terms of Moving Least Squares (MLS) [1]. This introduced the Element Free Galerkin (EFG) method which achieved substantial improvements in accuracy.

As stated by [wallstedt2009] the use of MSL-based methods is:

\(\checkmark\) useful for new classes of problems for which FEM is ill-suited

\(\times\) limited regarding the application of essential boundary conditions

\(\times\) substantially larger computational cost relative to FEM, mainly due to the nearest-neighbor searches

Several MLS-based methods appeared soon, including:

Other meshfree methods:

  • Diffuse Element Method (DEM) ([nayroles1992])

  • Element Free Galerkin (EFG) ([belytschko1994], [lu1994] and [belytschko1994_2]). While the DEM suffers from a number of problems due to some simplifications, the EFG may be considered a fixed version, being a very popular meshfree method [fries2004]

  • Generalized Interpolation Material Point (GIMP) ([bardenhagen2004])

  • Weighted Least Squares (WLS) (in the GIMP framework) ([wallstedt2009]). Does not require the nearest-neighbor searches and integrates over a background as the EFG


Gereral Characteristics#

Based on the excellent review of [fries2004], some interesting aspects of meshfree methods will be listed in this section. The author will try to collect the aspects that seem relevant to judge whether further investigation on these methods will be carried out in order to seek a future implementation in compmech.

  • Computational effort: usually higher than mesh-based counterpart models:

    • meshfree shape functions are more complex than polynomial-like functions used in mesh-based methods, requiring many more integration points (collocation meshfree methods don’t require integration, but this evokes accuracy and stability problems)

    • at each integration point the following steps are usually required: neighbor search, solution of small systems of equations and small matrix-matrix and matrix-vector operations in order to determine the derivatives


      this seems to be a perfect task for CUDA implementations…

    • the resulting system of equations has in general a larger band-width for meshfree methods than for mesh-based (like finite elements) [belytschko1996]

  • Hard to impose essential boundary conditions since you don’t have the Kronecker delta property in meshfree methods, which will guarantee that a given approximation function has a determined value at a given location