State of Health Estimation of Lithium Batteries for Automotive Applications with Artificial Neural Networks

This paper presents an algorithm based on Artificial Neural Networks (ANNs) for the estimation of the State of Health (SOH) in Lithium batteries. The method exploits a feed-forward pattern recognition classifier trained with datasets collected at different temperatures and at a predefined current mean value of the discharging profile. During the real implementation, the algorithm scans the time history of the battery load and analyses it on buffers of 60 seconds. Whenever the same conditions of the training dataset are encountered during the scanning (i.e. the same temperature and current mean value), the designed algorithm is enabled and provides an estimation of the SOH. The classifier acquires as input a set of predictors extracted from the direct measurement of characteristic parameters of the battery, namely voltage, temperature, capacity, energy and estimated State of Charge (SOC). The networks are trained and validated by means of a battery model based on look-up tables and previously characterized in a laboratory environment.