112
7 Virtual testing approach for sandwich panel joints
mechanisms. This is also evident in the reviewed literature where most introduced models for sandwich joints correspond to one of these three levels of detail. The quasi-3Dcontinuum models have the lowest computational cost and are therefore most suited
for preliminary design and optimization studies. However, the geometry representation
is limited to axisymmetric joint designs. This is not the case for honeycomb sandwich
inserts, which have an irregular interface between potting and core. This effect can be
modelled with true 3D-continuum and detailed models. The advantage of the detailed
models is their capability to model the buckling of the cell walls due to core shear accurately. This yields better results when high shear deformation beyond core damage initiation occurs. Furthermore, the accurate geometry representation is considered to be
superior in multiaxial stress states of the core. Due to the homogenization, 3D-continuum models are limited in representing these effects, as it was shown in the performed
bending study (section 5.1.2) In addition, detailed models enable to perform design studies with cellular cores for instance for the development of novel core geometries. Expectedly, detailed models require by far the highest modelling and computational effort.
Therefore, the expected computing time of the eventually implemented virtual test
strongly correlates with the core modelling approach. This is also shown in the performed bending study, where the detailed core models required 5-10 times more computational time. Therefore, detailed models generally yield computing times in the range
of several hours. They are therefore less suitable for extensive optimization studies. In
this case 3D-continuum models are the preferred choice, since they can be optimized to
run in a matter of minutes. However, this comes at the cost of reduced degree of damage
modelling. In addition, 3D-continuum models are generally sufficient for homogenous
cores such as foam and in case of honeycomb when the core deformation after damage
initiation is limited.
Selection of time integration
The second step of Phase 2 addresses the definition of the time integration scheme for
the virtual test. As discussed in 2.3.1 both, implicit and explicit integration, have their
specific field of application where they are most suited. However, virtual tests often
stand in between. In case of quasi-static tests, the implicit method is generally favorable.
However, with increasing non-linearity of the simulation, implicit solving encounters increasing difficulty with regards to convergence. As depicted in Figure 17 on page 20, single non-linear effects up until rupture can often still be represented. However, virtual
testing of sandwich panel joints often involves multiple non-linear effects from the different constituents. Furthermore, including the fixture in the simulation often requires
implementing a penalty contact, thus further adding a non-linear effect beyond the material behavior of the constituents. A high degree of non-linearity generally corresponds
to the number of non-linear effects to be reflected in the model. This is where the robustness of explicit solvers is advantageous. As rule of thumb it has been established
7 Virtual testing approach for sandwich panel joints
mechanisms. This is also evident in the reviewed literature where most introduced models for sandwich joints correspond to one of these three levels of detail. The quasi-3Dcontinuum models have the lowest computational cost and are therefore most suited
for preliminary design and optimization studies. However, the geometry representation
is limited to axisymmetric joint designs. This is not the case for honeycomb sandwich
inserts, which have an irregular interface between potting and core. This effect can be
modelled with true 3D-continuum and detailed models. The advantage of the detailed
models is their capability to model the buckling of the cell walls due to core shear accurately. This yields better results when high shear deformation beyond core damage initiation occurs. Furthermore, the accurate geometry representation is considered to be
superior in multiaxial stress states of the core. Due to the homogenization, 3D-continuum models are limited in representing these effects, as it was shown in the performed
bending study (section 5.1.2) In addition, detailed models enable to perform design studies with cellular cores for instance for the development of novel core geometries. Expectedly, detailed models require by far the highest modelling and computational effort.
Therefore, the expected computing time of the eventually implemented virtual test
strongly correlates with the core modelling approach. This is also shown in the performed bending study, where the detailed core models required 5-10 times more computational time. Therefore, detailed models generally yield computing times in the range
of several hours. They are therefore less suitable for extensive optimization studies. In
this case 3D-continuum models are the preferred choice, since they can be optimized to
run in a matter of minutes. However, this comes at the cost of reduced degree of damage
modelling. In addition, 3D-continuum models are generally sufficient for homogenous
cores such as foam and in case of honeycomb when the core deformation after damage
initiation is limited.
Selection of time integration
The second step of Phase 2 addresses the definition of the time integration scheme for
the virtual test. As discussed in 2.3.1 both, implicit and explicit integration, have their
specific field of application where they are most suited. However, virtual tests often
stand in between. In case of quasi-static tests, the implicit method is generally favorable.
However, with increasing non-linearity of the simulation, implicit solving encounters increasing difficulty with regards to convergence. As depicted in Figure 17 on page 20, single non-linear effects up until rupture can often still be represented. However, virtual
testing of sandwich panel joints often involves multiple non-linear effects from the different constituents. Furthermore, including the fixture in the simulation often requires
implementing a penalty contact, thus further adding a non-linear effect beyond the material behavior of the constituents. A high degree of non-linearity generally corresponds
to the number of non-linear effects to be reflected in the model. This is where the robustness of explicit solvers is advantageous. As rule of thumb it has been established