Spiking Neural Networks for Accurate and Efficient State of Health Estimation of Lithium-Ion Batteries Across Varying Temperatures

Machine learning (ML) and deep learning (DL) have become essential tools in lithium-ion battery research, particularly for estimating the State of Health (SOH). However, conventional SOH estimation methods often rely on repeated charge/discharge cycles under strictly controlled laboratory conditions, limiting their applicability in real world scenarios. In this study, we present a comprehensive lithium-ion battery dataset developed by our team to support data driven approaches for battery diagnostics and predictive modeling. The dataset comprises nineteen lithium iron phosphate (LFP) cells with cycle lifetimes ranging from 500 to 2600 cycles and reflects realistic usage conditions, including non constant discharge currents and tests conducted at$25^\circ C$,$35^\circ C$, and$45^\circ C$. To demonstrate the utility of this dataset, we used a brain inspired Spiking Neural Network (SNN) referred to as SpikeSOH, a neuromorphic model that uses sparse, time coded spikes to mimic biological neurons. This approach provides temporal precision while reducing energy consumption. Our results show that the SNN-based model achieves an average Mean Absolute Error (MAE) of 4.5%, while also outperforming conventional deep learning models in computational efficiency, with an average inference time of 3.55$\mu \mathrm{s}$and an average energy consumption of 0.36 mJ. These characteristics make the model particularly suitable for integration into energy constrained battery management systems. By providing a realistic, high quality dataset and demonstrating the advantages of energy efficient neuromorphic models, this work advances accurate and scalable SOH estimation methods, helping safer and more reliable deployment of lithium-ion batteries in both first life and second life applications.