SOH prediction model through OCV based on AI for high voltage battery

Lithium-ion batteries are representing nowadays the reference technology for the state-of-the-art energy storage system in hybrid and electric vehicles. Countless are the parameters to take into account for an optimal battery operation, from physical and chemical point of view, in addition to all the charging and discharging phases concerning the battery life and how it can be managed. A proper and accurate State of Health (SOH) prediction is needed to take right countermeasures and precautions defining a control strategy aimed to correctly exploit HV-Batteries along its whole life and extending it as well. From the customer point of view, having the proper idea of whenever the battery would be replaced is really important advantage that comes out from a proper SOH prediction. The more reliable the SOH is, the more accurate the DTE will be, meaning for the customer no undesired surprises in the cluster after 50.000km, 100.000km... This work investigates and exploits six Machine Learning (ML) algorithms, to predict the SOH value of vehicle batteries belonging to different km mileage. The performed analysis is of the open-circuit type, collecting and exploiting real OCV values from vehicle measurements, properly discharged at different SOC levels and OCV retrieved after a rest of two hours. The open circuit analysis allows to detach from the load applied to the battery, taking in account only the OCV evolution trend according to those vehicles at different km mileage, i.e. to batteries at different life stages. Considering a reasonable number of cells for all the batteries, the final idea is that one to link the OCV values to the energy (kWh) stored in the battery at every single measurement. In this way, it would be possible to correlate a Kappa factor to every vehicle, that holds the energy content with respect to OCV and SOC; the newer the vehicle, the higher the Kappa. The ratio between this actual factor for each vehicle, with respect to the Kappa belonging to a new vehicle, gives the SOH percentage at which the battery (and the vehicle) are standing. ML analysis performed to study, use and tune the six algorithms (LR, KNN, SVM, RF, CART and NN), improving their way of working together with predictive performance, according to overfitting and validation stages, comparing in the end results for the final SOH predictions.