5 – SECTOR EXTERNO
7- ESTRUCTURA DE COMERCIALIZACIÓN Muestras y Materiales de Publicidad
Checkerboard instability refers to the formation of checkerboard-like patterns
made of alternating solid and void (or hard and soft) elements. This problem
was investigated by D´ıaz and Sigmund (1995) and Jog and Haber (1996).
They demonstrated that the reason for checkerboard formation is that these
patterns gain artificially high stiffness due to numerical problems in the mixed
formulation of the finite element discretisation. An example of checkerboard
patterns is given in Figure 2.23. This topology is obtained by the SIMP
method solving exactly the same problem already solved in section 2.5.3.
All the parameters have been selected similar to the original problem, but
instead of high order 9-node elements, here, 4-node elements have been used.
The homogenisation, the SIMP, the ESO, and the original BESO methods
Figure 2.23 The checkerboard instability in a result obtained by the SIMP method for the SCB problem. Instead of 9-node elements, 4-node elements have been used here.
element formulation which is used in all of these methods. However, there
are several ways to prevent formation of checkerboard patterns. Here some
of the well-known techniques will be introduced.
Using stable elements
Jog and Haber (1996) proposed a patch test to identify unstable elements
which can cause the checkerboards formation in topology designs. Also D´ıaz
and Sigmund (1995) presented some guidelines for choosing stable elements.
It has been shown in both papers that higher order finite elements can prevent
checkerboard formation. In the examples already presented in this chapter,
9-node and 8-node high-order finite elements have been used and no checker-
board pattern has been found in the final topologies. However in the SIMP
method this approach does not guarantee a checkerboard-free result if one
applies big penalty values (D´ıaz and Sigmund 1995). Also this approach
might not work with the ESO and the BESO methods. Furthermore, us-
effort compared to 4-node elements.
Special types of non-conforming finite elements may also help preventing
checkerboard formation. Jang et al. (2003) reported checkerboard-free results
using non-conforming 4-node finite elements with their nodes located in the
middle of the edges rather than corners.
filtering techniques
The use of filtering techniques to overcome checkerboard formation has been
first studied by Sigmund (1994). Although no rigorous proof was ever pro-
posed for this approach, it can successfully prevent checkerboard formation
with low computational effort. The filtering scheme is similar to filtering
techniques used in image processing. This technique was originally used in
the SIMP method. However Li et al. (2001) successfully applied a similar
technique to the ESO method. The linear filter used in the BESO algorithm
(2.34) is also capable of preventing checkerboard formation. As mentioned
before, in the BESO method, the filtering scheme plays an additional role of
assigning sensitivity numbers to voids as well.
The short cantilever beam problem is solved here using the SIMP method
with 4-node finite elements but sensitivities are filtered using a radius of twice
of the size of the elements. In the SIMP method the filtered sensitivities are
calculated using the following formula
d∂c ∂ρi = PN j=1ρj ∂c ∂ρjwij ρi PN j=1wij (2.38)
in Figure 2.24. Note the blurred boundaries in this figure. This is due to
the applied filtering scheme which assures a smooth transition from solids to
voids exists.
Figure 2.24 Preventing formation of checkerboard patterns using filtering techniques. The filtering radius is twice of the elements’ size.
An approach similar to filtering was proposed by Ghabraie (2005) to
overcome checkerboards instability in the SIMP method. In this approach
the sensitivities at nodes are calculated by averaging the sensitivities of the
elements connected to them. Then the sensitivities of the elements are re-
calculated by averaging the resulted nodal sensitivities.
Perimeter control
Another way to control checkerboards is to impose a constraint on the perime-
ter. Between two topologies with the same amount of material the one with
less holes has a shorter perimeter. By setting an upper limit on the perimeter,
one can thus control the complexity of the resulted topologies and prevent
checkerboards. This approach was first used by Haber et al. (1996). Perime-
Other techniques
Apart from the abovementioned techniques, several other approaches have
been proposed that can deal with the checkerboards problem. Guest et al.
(2004), Rahmatalla and Swan (2004), and Matsui and Terada (2004) used
nodal sensitivities instead of elemental ones and reported checkerboard-free
topologies using the SIMP and the homogenisation methods. Another ap-
proach was proposed by Poulsen (2002) in which an additional constraint
is added to prevent one-node connected hinges. A simple method was pro-
posed by Kim et al. (2000) which can control the number of holes in the
final topology produced by the ESO method and also can solve the checker-
boards problem. Using fixed-grid finite element in the ESO method has been
investigated by same researchers (Kim et al. 2003) and it was shown that
this approach is capable of preventing checkerboard patterns. To overcome
the checkerboards problem in the homogenisation method Fujii and Kikuchi
(2000) introduced a gravity control function to be added to the objective
function. This function is similar to the perimeter control function, however
it penalises the intermediate densities as well. It was shown through exam-
ples that this approach can solve the mesh dependency and local minima
problems as well.
It should be noted here that most of the techniques which are used to
deal with the mesh dependency problem can also overcome the checkerboards