Due to multiscale character covering space distances from 10-6 up to 100 m of electrochemical cells and batteries, their mathematical modelling covering all aspects from chemical, electrical up to thermal problems are extremely demanding. Number of mathematical approaches are available in literature but coupled chemical, electric and thermal computations of cells and batteries are very time consuming and therefore impossible to be used for optimisation or dynamic control. The paper presents development of the low order time dependent model of Lithium-ion electrochemical cell using machine learning techniques. The Lithium-ion electrochemical cells are commonly used to build traction batteries in automotive industry. Therefore there is a great need to develop models that would in-fly predict electric and thermal parameters of the cell and the whole battery during its operation. In the first step an unsteady three-dimensional coupled thermal and electrochemical model of a cell was developed. The model covers heat transfer and current flow inside the cell and heat sources due to electrochemical reactions and current flow. This full scale CFD model was validated against experimental data. Having the validated numerical model a low order model based on the artificial neural network (ANN) has been developed. Finally, the low-order ANN model was validated against artificial data obtained with detailed 3D unsteady model. The developed ANN low order model is capable of predicting crucial cell parameters with reasonable accuracy.
Keywords
- Lithium-ion cells
- Computational fluid dynamics (CFD)
- Machine learning
- Model order reduction
- State prediction
Acknowledgment
The work was supported by the Silesian University of Technology through the Rector’s grant number 08/060/SDU/10-07-03 RGP24.