This presentation summarizes three applications of artificial intelligence methods to real-time simulation of systems : evaluation of the deflection of a sheet metal as a function of the order of welding operations, evaluation of the porosity rate in a 3D printed part, and evaluation of real-world behavior from a simulation.
For the welded sheet metal, the data comes entirely from finite element simulations that take one hour of calculation each. The model extracted from these simulations allows the evaluation of all the deflections of the sheet in real time in order to explore the combinatorics and target the optimal sequence of welding operations (e.g. to minimize the deflection at certain points).
For 3D printing of titanium, the data is completely experimental : from the trajectory performed by the laser and the scan of the finished product with the tomograph, a regression model is built. It allows to evaluate the porosity rate in the different parts of the part, and thus to know in advance its mechanical quality.
For the third application, which allows the evaluation of the real behavior of a system from its simulations, the experimental and simulation data are mixed in a regression process. The resulting model allows the simulations to be deviated from the reality observed during the tests. In other words, within the same simulation process, the simulated theory is corrected by the accumulated experience.