Tool for fast prototyping – Physics-Informed Neural Network
In the framework of DEFACTO, a tool for fast prototyping using an AI/ML technique call physics-informed neural networks (PINN) has been created to provide the evolution of lithium concentration in active material particles in both electrodes during a discharge process as well as the discharge curve of the full battery, based on a Single Particle Model. It is available at GitHub following this link:
The model is trained using a NMC811 G-Si chemistry in the following range of geometrical parameters and C-rates:
– Negative electrode thickness in [5e-5, 2e-4] m
– Positive electrode thickness in [5e-5, 2e-4] m
– Negative electrode porosity in [0.2, 0.6]
– Positive electrode porosity in [0.2, 0.6]
– C-rate from 1C to 3C
but due to mathematical properties of the model in the training process, this PINN model is also usable outside the specified ranges in certain cases.
In GitHub repository, apart from the software itself, the user can also find a Jupyter notebook with indications for use and examples.