This tutorial shows how to perform a Restart of a stopped optimisation. Therefore, the Advanced User Settings are needed. More information regarding the background of a Restart here.
With this option a terminated optimization optimisation can be continued or a two-stage optimization with varified optimization parameters can be done. Instead . To realize a Restart instead of the Design spaceanother geometry can be chosen as the Start space. In this case the optimization optimisation starts not with the specified geometry although the whole Design space however this Design space is still the geometry limitation for the optimization.
...
the optimisation.
The Start space STL-file needs to be entered into the optimisation in S.I. units!
Start Space STL-file
The Start space has to for a Restart must be a STL-file. Therefore, a already optimized optimised geometry can be exported in the Post Processing. In this case iteration 22 from the symmetrical Bridge Bookshelf Tutorial will be used.
...
Attention: The STL-file for the Restart has to must be exported in SI-Units (Meters).
...
optimisation Setup
In the next step a new scenario Scenario should be created for the Restart optimisation. With the Advanced User Settings- available through a right click on the Scenario - the Start space STL-file can be selected. In this case the exported iteration 22 from an former optimization is selected. Furthermore, the symmetry constraints from the first optimization are added.
The most important part of a Restart is to add and adjust the Advanced User Settings especially the number of iterations, the resolution level and the stress goal percentage. More background information regarding this the Advanced User Settings here.
...
In this example the Bridge Bookshelf should be restarted as if the optimisation never stopped, so with the same optimization optimisation parameters as the first optimization optimisation excluding the already done iterations. This can be done when an optimization is terminated and restarted at another time. Therefore, the missing number of iterations settings for the parameters Strut Density: Medium and Shape Quality: Balanced are added to the Advanced User Settings and adjusted according to the number of iterations already calculated. Which number of iterations and which stress percent goal should be input for each resolution level can be found here.
In this case the optimization will restart at iteration 22, thus the following 42 iterations have to must be added to reach the total number of 64 iterations (Shape Quality: Balanced). Therefore, the UpSampleConfig in the picture above is added (18 iterations at resolution level 2, 20 iterations at resoultion level 1 and 4 iterations at resolution level 0)
...
The information are added in the Advanced User Settingsin the following style:
UpSampleConfig
fOptimizer_switchAddRemove=36
iteration
.level_2=18”
.level_1=20”
.level_0=4”
fOptimizer_stressPercentGoal
.level_2=40
.level_2=90
.level_2=100
The command fOptimizer_switchAddRemove needs to be active for the last 4-10 iterations for the quality of the surface smoothness, so in this case 42-6=36.
After adding these commands to the Advanced User Settings the optimization optimisation can be started. The first iteration of the restarted optimization optimisation will be the imported Start space.
...
Two-stage optimization
A seconed possibility is to perform a two-stage optimization. Instead of a simple restart with the same optimization parameters the parameters can be varied.
Exampels:
Skipping the coarse resolution level and calculate directly on the finest resolution level with an early iteration from a former optimization, if a very fine high-resoltion level is desired
Vary the value for the stress goal to keep the overall geometry with more or less material
Modify an optimization output (remove or add struts) and restart with this modified geometry
Creation of a lattice stucuture of an optimization output and restart to optimize the lattice
A two-stage optimization is useful for the following applications:
...
An optimization by means of a grid structure is to follow. See tutorial ...
...
A very fine high-resolution structure is desired. Then it is advisable to first perform an optimization up to a reduced structure (the optimization must not be finished). Then a second optimization with a start design is performed, which starts directly at a higher resolution level and receives more iterations in these levels to form the fine structure.
...
space
...
.