Impact of Data Window on Prediction of Battery Aging and Swelling

Pannala, Sravan

Lithium-ion batteries degrade over time, based on their usage and these changes can be observed in the terminal voltage and cell expansion curves with respect to SOC. Physics-based battery models augmented with degradation modes can describe the changes in battery electrical and mechanical response as the battery ages and can be tuned to predict the observed rate of change. Tuning these models is computationally intensive, because it requires multiple simulations of the battery aging process over the entire lifetime. Accelerated simulations that use an adaptive inter-cycle extrapolation algorithm [1] can reduce simulation time and thus enable use of optimization algorithms to automatically tune degradation models [2].

In this work, we investigate how the amount of aging data used to train the model impacts the prediction of future battery aging assuming the battery continues to operate under the same conditions. The electrode-specific state of health (eSOH) parameters are computed from the reference performance testing performed at periodic intervals throughout the battery's lifetime. These four parameters include the capacities of positive and negative electrodes C_p and C_n, and the corresponding electrode stoichiometric windows y₀ and x₁₀₀. These 4 parameters extracted from a cell aged at C/5 constant current charging rate were used for training the model.

The aging model includes SEI growth in the negative electrodes which causes a loss of cyclable lithium/loss of lithium inventory (LLI) and loss of electrode active material (LAM) due to particle cracking at both the negative and positive electrodes. Simulation results show us that a shorter training window of three eSOH reference points (every 25 cycles) over (Cycles 41 to 98) over-predicts battery degradation because the rate of degradation during these initial cycles is larger than the degradation which happens later in the battery's life. The battery degradation parameters that are predicted (and shown in the figure) include battery capacity, maximum reversible swelling in the battery, loss of active material in the negative electrode and loss of lithium inventory. If we increase the model training window to five eSOH data points (Cycles 41 to 175), the prediction of battery degradation is much better because the model now accounts for both the initial higher rate of degradation and subsequent lower rate of degradation.

References:

V Sulzer et al. Accelerated Battery Lifetime Simulations Using Adaptive Inter-Cycle Extrapolation Algorithm, Journal of Electrochemical Society (2021)
S. Pannala et al. Methodology for Accelerated Inter-Cycle Simulations of Li-ion Battery Degradation with Intra-Cycle Resolved Degradation Mechanisms, American Controls Conference (2022)