Physics-Based State of Health Estimation of Lithium-Ion Battery Using Sequential Experimental Design

State of health (SOH) estimation is a critical yet challenging task due to the complex degradation process of lithium-ion (Li-ion) battery. This paper proposes to combine physics-based modeling of Li-ion battery and sequential design of simulation experiments to build an accurate SOH estimator in a computationally efficient manner. A novel sequential backward optimization process is adopted to build a multivariate Gaussian process model that quantifies three degradation modes in a Li-ion battery cell: loss of lithium inventory and losses of active materials in the positive and negative electrodes. The sequential process for the design of simulation experiments is realized via the use of an acquisition function, the maximization of which gives rise to a new sample point in the design space for the next experiment. The acquisition function achieves an optimal balance between exploration of new regions in the design space with high prediction uncertainty and exploitation of challenging regions with high response nonlinearity. The preliminary results from COMSOL Multiphysics degradation scenario simulations show that the SOH estimator designed with the sequential sampling process can provide faster error decay in degradation estimation when compared to that without the sequential sampling process.