7.4 Phase 3 - Model development
121
An overview of the categories along with a list of relevant parameters for each category
is given in Figure 88. The numerical studies generally follow the same procedure, which
is illustrated in Figure 91. This procedure begins with the definition of a benchmark
model parameter set. This corresponds to selecting reasonable dummy values for all relevant model parameters as given in Figure 88. Therefore, the benchmark parameter set
is defined based on experience (i.e. from modelling database) and it may require some
trial and error to find suitable parameters. In parallel, a certain number of increments or
variants is defined for the model parameter to be studied. For instance, this could be
mesh size increments or different material models. Subsequently, the benchmark model
parameter set is implemented together with the FE-geometry representation and the
simulation is performed n times, depending on the number of increments or variants for
the studied model parameter. The obtained simulation results are then evaluated based
on the requirements of Phase 1 and the model parameter, which best suits the requirements, is selected. The numerical study sub-step concludes when all relevant model parameters are defined. In the following, the three categories of numerical studies are described in more detail based on the detailed honeycomb example.
Numerical parameters
There are general numerical parameters, which greatly influence the accuracy of the simulation results and at the same time the computational
effort. It is therefore common practice to perform convergence analyses, in order to determine suitable parameter settings. In case of the explicit integration method, mass
scaling and increased loading rate are standard methods to reduce the computational
time of quasi-static simulations. Mass scaling achieves this by increasing the nodal mass
of the model in order to increase the critical time step, while an increased loading rate
leads to a reduced time period of the simulation. Therefore, both parameters reduce the
number of required time increments. However, they also influence the kinetic energy of
the system. Excessive mass scaling and loading rate lead to unstable and inaccurate results. This is illustrated exemplary in Figure 92, where one of the performed sensitivity
studies for the detailed honeycomb investigation is illustrated based on flatwise compression. The implemented benchmark parameters in terms of mesh size and boundary
conditions are described in section 4.1.3 from page 44 onwards. The model scale was
defined as depicted in Figure 37 on page 48 and a SL isotropic material model was used
as benchmark. Four increments of mass scaling were studied, while the results are given
as stress-strain curves in Figure 92. It can be seen that a time increment of 1e
-4 s leads
to inaccurate results. Therefore, the time increment should have an upper limit of about
1e
-5 s. In case of implicit time integration similar sensitivity studies may be required in
order to determine a reasonable amount of damping for stabilization of non-linear simulations. Regardless the time integration method, sensitivity studies regarding the mesh
size (mesh convergence) are always recommended. Regarding the honeycomb core of
121
An overview of the categories along with a list of relevant parameters for each category
is given in Figure 88. The numerical studies generally follow the same procedure, which
is illustrated in Figure 91. This procedure begins with the definition of a benchmark
model parameter set. This corresponds to selecting reasonable dummy values for all relevant model parameters as given in Figure 88. Therefore, the benchmark parameter set
is defined based on experience (i.e. from modelling database) and it may require some
trial and error to find suitable parameters. In parallel, a certain number of increments or
variants is defined for the model parameter to be studied. For instance, this could be
mesh size increments or different material models. Subsequently, the benchmark model
parameter set is implemented together with the FE-geometry representation and the
simulation is performed n times, depending on the number of increments or variants for
the studied model parameter. The obtained simulation results are then evaluated based
on the requirements of Phase 1 and the model parameter, which best suits the requirements, is selected. The numerical study sub-step concludes when all relevant model parameters are defined. In the following, the three categories of numerical studies are described in more detail based on the detailed honeycomb example.
Numerical parameters
There are general numerical parameters, which greatly influence the accuracy of the simulation results and at the same time the computational
effort. It is therefore common practice to perform convergence analyses, in order to determine suitable parameter settings. In case of the explicit integration method, mass
scaling and increased loading rate are standard methods to reduce the computational
time of quasi-static simulations. Mass scaling achieves this by increasing the nodal mass
of the model in order to increase the critical time step, while an increased loading rate
leads to a reduced time period of the simulation. Therefore, both parameters reduce the
number of required time increments. However, they also influence the kinetic energy of
the system. Excessive mass scaling and loading rate lead to unstable and inaccurate results. This is illustrated exemplary in Figure 92, where one of the performed sensitivity
studies for the detailed honeycomb investigation is illustrated based on flatwise compression. The implemented benchmark parameters in terms of mesh size and boundary
conditions are described in section 4.1.3 from page 44 onwards. The model scale was
defined as depicted in Figure 37 on page 48 and a SL isotropic material model was used
as benchmark. Four increments of mass scaling were studied, while the results are given
as stress-strain curves in Figure 92. It can be seen that a time increment of 1e
-4 s leads
to inaccurate results. Therefore, the time increment should have an upper limit of about
1e
-5 s. In case of implicit time integration similar sensitivity studies may be required in
order to determine a reasonable amount of damping for stabilization of non-linear simulations. Regardless the time integration method, sensitivity studies regarding the mesh
size (mesh convergence) are always recommended. Regarding the honeycomb core of