US 12,462,079 B2
Simulation by substitution of a model by physical laws with an automatic learning model
Gaël Goret, Réaumont (FR); Léo Nicoletti, Meylan (FR); Stéphane Pralet, Autrans (FR); and Cédric Bourrasset, Lodève (FR)
Assigned to BULL SAS, Les Clayes sous Bois (FR)
Filed by BULL SAS, Les Clayes sous Bois (FR)
Filed on May 19, 2021, as Appl. No. 17/324,491.
Claims priority of application No. 2005181 (FR), filed on May 20, 2020.
Prior Publication US 2021/0365616 A1, Nov. 25, 2021
Int. Cl. G06F 30/27 (2020.01); B64F 5/00 (2017.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01)
CPC G06F 30/27 (2020.01) [B64F 5/00 (2013.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01)] 8 Claims
OG exemplary drawing
 
1. A method for simulating the behavior of a system composed by an aerial vehicle in a physical environment, said method being implemented by a simulator comprising one or more processors operably connected to a non-transitory computer medium and a set of modules configured on said one or more processors, each simulating a portion of said system, wherein one module of said set of modules is configured to simulate at least one portion of said aerial vehicle comprising a turbojet engine of said aerial vehicle by implementing a trained multilayer neural network, said at least one portion corresponding to a physical modelling distinct from that of other portions of said aerial vehicle, wherein other modules of said set of modules are configured to implement simulations based on behavioral physical laws and wherein the trained multilayer neural network has been trained to predict output data from input data supplied by at least one of said other modules, said input data comprising a set of variables which characterize a stream flow in the turbojet engine, said output data comprising variables which characterize normal and parallel components of body forces of the turbojet engine being transmitted to at least one of said other modules, said method comprising:
receiving by the simulator configuration data of a design of the aerial vehicle;
performing by the simulator a simulation of the system and outputting results of the simulation;
assessing by the simulator the results of the simulation based on an objective function, said assessing comprising measuring an accuracy of the simulation and determining areas in a simulated system, wherein the accuracy of the simulation is insufficient according to an expected result, to automatically decide a strategy to trigger updates of parameters associated to said trained multilayer neural network;
when a decision is made, triggering by the simulator said updates of parameters associated to said trained multilayer neural network is made, according to said strategy, so as to improve an accuracy of subsequent simulations; and
iterating by the simulator the performing of a simulation using the updated trained multilayer neural network module and the assessing of the results of the simulation.