This paper addresses the need for simple and cost-effective methods that detect the state-of-health (SOH) of batteries in vehicle applications solely based on data readily available from the battery management system without any knowledge of battery properties, prior laboratory measurements or additional equipment.
A power-optimized lithium-ion battery cell is operated in an emulated hybrid electric vehicle (HEV) environment on board a conventional heavy-duty truck. The HEV operation of the battery cell depends on the driving pattern of the truck within set limits. Beyond the HEV operation, the performance of the battery cell is periodically measured with on-board standard pulse and capacity tests. On basis of the battery operating data collected in the field test, support vector machine-based battery models are built. From a model input of current, temperature, state-of-charge, and current history, the model accurately estimates the battery voltage despite the tough HEV operating conditions with high current pulses. This data-driven battery model is used to estimate the battery cell’s charge and discharge resistance as well as capacity, i.e. the performance measures verified with the standard tests. These SOH indicators can be predicted by the model with adequate accuracy for on-board SOH detection and are followed throughout the one-year field test period.