Multi-Objective Bayesian Optimization of Lithium-Ion Battery Cells

In the last years, lithium-ion batteries (LIBs) have become the most important energy storage system for consumer electronics, electric vehicles, and smart grids. A LIB is composed of several unit cells. Therefore, one of the most important factors that determine the performance of a LIB are the characteristics of the unit cell. The design of LIB cells is a challenging problem since it involves the evaluation of expensive black-box functions. These functions lack a closed-form expression and require long-running time simulations or expensive physical experiments for their evaluation. Recently, Bayesian optimization has emerged as a powerful gradient-free optimization methodology to solve optimization problems that involve the evaluation of expensive black-box functions. Bayesian optimization has two main components: a probabilistic surrogate model of the black-box function and an acquisition function that guides the optimization. This study employs Bayesian optimization in the design of cylindrical cells type 18650. The materials of the cathode, anode, and electrolyte are Nickel-Cobalt-Aluminum, graphite, and LiPF₆ in EC-EMC, respectively. The black-box functions are simulations of the cycling performance test in Simcenter Battery Design Studio. The design variables are the thickness and porosity of the coating of the LIB electrodes. The thickness is restricted to vary from 22 μm to 240 μm and the porosity varies from 0.22 to 0.54. Two objectives are optimized: maximization of the specific energy and minimization of the capacity fade. The probabilistic surrogate models of the specific energy and capacity fade are Gaussian process regression models. The acquisition function that solves the multi-objective optimization problem is the Euclidean-based expected improvement. The results show that Bayesian optimization can identify high-performance LIB cells employing a reduced number of function evaluations. The methodology identifies 32 Pareto designs with specific energies in the range of 67 to 202 mWhr/g and capacity fade in the range of 18.4 to 23.8%.