The goal is to develop a novel surrogate model based on a Physics-Informed Neural Network (PINN) for next-generation electric propulsion system diagnostics and remaining life predictions. The PINN model includes an electric powertrain system consisting of battery, speed controller and motor, and will aim at detecting faults and predicting remaining useful life of those critical components.

This work changes the way physics is embedded in ML, and instead uses neural network “blocks” within physical and empirical models, since the physics driving the system dynamics is partially known. The neural network blocks aim at learning the relationships between variables where reliable models or physical knowledge are unclear or subject to high degree of uncertainty.