This thesis presents a comprehensive exploration of battery state of health (SOH) forecasting and end of life (EOL) prediction using neural network architectures. A combination of neural networks and ordinary differential equations, namely neural-ODEs, augmented neural-ODEs, and predictor-corrector recurrent ODEs are studied. Their performance is analyzed against established recurrent neural networks such as long short-term memory and gated recurrent units. Similar architectures are further utilized with larger datasets involving a wide range of fast-charging cycling conditions. New approaches to loss and feature engineering are introduced to address the challenge of extracting informative features and accurately regressing SOH and EOL from large amount of laboratory data. An automatic...[ Read more ]

This thesis presents a comprehensive exploration of battery state of health (SOH) forecasting and end of life (EOL) prediction using neural network architectures. A combination of neural networks and ordinary differential equations, namely neural-ODEs, augmented neural-ODEs, and predictor-corrector recurrent ODEs are studied. Their performance is analyzed against established recurrent neural networks such as long short-term memory and gated recurrent units. Similar architectures are further utilized with larger datasets involving a wide range of fast-charging cycling conditions. New approaches to loss and feature engineering are introduced to address the challenge of extracting informative features and accurately regressing SOH and EOL from large amount of laboratory data. An automatic tool for feature extraction is proposed. Finally, this thesis explores the integration of physical and data-driven models for accurately modeling battery states. Hybrid models show the best performances compared to standalone models. Two cycling datasets, including one collected in-house, are used for evaluation. The findings offer promising applications for battery diagnostics and prognostics in battery management systems, with implications on battery performance and safety.