The Decimate tool reduces the number of mesh elements while
preserving key features of the geometry, simplifying complex models without sacrificing
accuracy.
The Decimate tool extracts surface result data from result
files and writes the h3d and json files
required for PhysicsAI training. All decimated h3d and
json files with KPI vectors are stored in the simulation
directory as h3d files.Figure 1.
Note: Since current PhysicsAI support is limited to surface
field data due to high compute resources and limitation of submitting training
on multiple GPUs, current functionality will decimate just surface data and
write out h3d files. For example, Polyhedral data will be
converted to triangulated data and fields will be mapped onto nodes.
Figure 2.
Using the Decimate Inputs window
HyperMesh CFD will search result files from within the
folder defined in the Simulation directory field of
the Decimate inputs window. You can define search terms
using in the Simulation file keyword and
Result file extension fields. If no keywords are
defined, search will default to file extension.
If the Consider subdirectories for file check box is
engaged, subdirectories will also be considered in search.
Any folders or sub directories with name having Folder keyword to skip will
not be looked for searching result files.
Decimation Routine Parameters
Maximum threads
Each decimation job runs on single thread. This thread number defines
how many decimation jobs will be submitted simultaneously.
Note: This should be selected carefully based on
memory footprint required for each result file. If a result file is
too large it can use all memory available. In this case, set max
threads to 1, performing each decimation job one by one.
Part to skip
Parts with names containing keywords defined in the Parts to
skip fields will not be written out. For example, in
PhysicsAI training and prediction, it is preferable to run the tool on
car bodies and exclude volumetric boundary domain like tunnel walls,
ground, belts, etc.
Decimation features angle
Defines a constraint on mesh decimation. The amount of decimation is
highly dependent on this parameter. Smaller values like 0.5 degrees will
preserve most shapes as is, whereas a higher value may distort the
shape, while heavily reducing the model node count.
Decimation factor
Defines how aggressively the model node count is reduced. For example, a
0.5 factor reduce node count by 50%.
KPI Extraction
Extract KPIs
To train PhysicsAI with KPIs, a KPI vector must be extracted alongside
to the json file. Engage the Extract
KPIs checkbox to extract KPIs during decimation. If
turned on, KPIs will be extracted from either UFX results or from a
user-defined csv file, and create a
json file in the output folder.
KPI data file
In this field, direct the tool to a csv file. In
this csv file, the first column must contain the
name of the result file (with extension) and the second column the KPI
vector. All KPI vectors should have a name in the header row.
