SOH Estimation of Lithium-Ion Batteries Based on ANNs and Synthetic Data

Estimating the State Of Health (SOH) of Lithium-Ion Batteries (LIB) is essential for optimal performance in applications such as electric vehicles (EV) and energy storage systems. However, this battery’s electrochemical parameter is impossible to measure directly. Many works have addressed estimating the battery SOH using different techniques. Recently, techniques such as machine learning have enabled data-driven predictions when a parameter cannot be measured. In this work, an artificial neuronal network (ANN) is proposed to estimate the battery’s SOH for an EV. The ANN was trained at the beginning with Ames PCoE battery data provided by NASA and complemented with synthetic data from a LIB electrochemical model. The electrochemical model describes the battery behavior with specific SOH. Therefore, a parameter estimation algorithm such as Particle Swarm Optimization (PSO) was used to figure out the model parameters for each SOH to generate new profiles and add more information to train the ANN. Finally, the ANN was implemented with a candidate battery to demonstrate its effectiveness and performance in out-range LIB operating conditions in which it was trained.