Physically-Informed Neural Networks for digital twins of Flight Control Actuators
Research Groups 14-Meccatronica e servosistemi
Thesis type SIMULATION
Description "Physically-Informed Neural Networks" are currently perceived as a possible breakthrough technology for the definition of digital-twins of complex components with reduced computational requirements.
The student will perform a literature review on the subject and investigate the considered case study, an electro-hydraulic servo actuator for primary flight controls.
A physics-based, high-fidelity model of such case study will be provided.
The student will define the network topology and its training strategy, proceeding hence to replace parts of the physics-based model with the neural networks prepared during the first part of the work.
A combarison between the simulation behavior with such modification and the original physics-based model will be performed.
Expected results include the definition of a dynamic model for the flight control actuator able to maintaing a high complexity level with reduced computational load, and a first indication towards the possibility to deploy the produced model within real-time simulation environments.
Required skills Mechatronics, knowledge of Matlab/Simulink environment
Deadline 29/11/2023 PROPONI LA TUA CANDIDATURA