Speaker
Description
There are current necessities of real-time simulation and satellites model exchanges with launchers. We are developping a geometrical model reduction (currently manualy done in the industry) complementary with a TMM reduction module (done by tools such as TMRT). A budget of faces for GMM is almost always imposed (Launcher or DSS constraints). The presented model reduction tool is physics unaware but physics driven. Here the focus is made on the thermal simulation. The first approach relies on the following assumptions :
• limited to radiative couplings, external fluxes (solar and albedo) will be considered next
• no conductive couplings in the resolution of the heat equation (focus on the radiative part for a steady state)
• Isothermo-optical properties (identical alpha, tau, rho for all mesh elements) so node couplings are accurate to view factors
First approach : geometric reduction of each physical node individually (Lindstrom-Turk edge collapse, preserving volume and shape boundaries of the surface mesh), then the assembled reduced model is given to an oracle (black box) returning the physical validity of the model (distorsion). If the new model is validated, the reduction step is repeated until the distorsion is too high.
An animation illustrating the progressive approximation of the geometry faithful to temperatures will be shown in this presentation.
As future works, we plan to achieve the following landmarks :
- Compare different geometric reduction metrics (volume preserving, normal orientations preserving, memoryaware reduction etc)
- Introduce external fluxes (solar and albedo) so the reduction process is receptive to shadows and masking (geometric algorithm only slightly affected since reduction process is physics unaware)
- Sensitivity of the algorithm for different thermo-optical properties (again should not imply many changes)
- Sensitivity to power injections
- Add the conductive part (implying bigger geometrical reduction since tolerance is higher)
- Results analysis to favor the use of AI (better reduction targeting, adaptive version)
External models to do more tests are welcome!
