Prediction of compression force evolution over degradation for a lithium-ion battery

This study proposes a method to predict the evolution of compression force during the degradation of a lithium-ion battery under packed conditions. The total compression force comprises irreversible and reversible forces. The former is estimated using a multivariate machine learning method, whereas the latter is estimated by combining machine learning and phenomenological modeling. For predicting the irreversible force, impedance-related features are extracted and their correlations with the evolution of the irreversible force are quantitatively analyzed using Grey relational analysis. Subsequently, features with high Grey relational grades are employed as representative health indicators for multivariate inputs of Gaussian process regression. For predicting the reversible force, the force evolution during the charge/discharge period is predicted using a phenomenological force model. The equivalent stiffness used in this model is separately estimated depending on the state of charge (SOC) to account for the inherent characteristics of phase transition and different degradation behaviors. The evolution of equivalent stiffness under high SOC shows nonlinearity but weak evolution characteristics, whereas those under low and medium SOCs show linearity but strong evolution characteristics. Finally, the proposed method is used to enable control and design for two potential applications: estimations of the state of health-dependent SOC and separator compression.