State of Health Estimation of Cycle-Aged Cylindric LFP Batteries using ARMAX Modeling

The paper evaluates the efficacy of ARMAX modelling for the estimation of the Sate of Health (SoH) and State of Charge (SoC) of 40 Ah cylindric LiFePo battery cells under cycle aging. Experimental data of one LFP cell (the 'reference cell') are collected to train a set of ARMAX models for several SoH and SoC levels. The set of identified models is then used for subsequent SoH and SOC estimation by comparing their time-domain response with the real data of the cell under consideration. The ARMAX model with the highest similarity is then chosen, and its SOC/SOH levels are taken as corresponding estimates. The method is applied to a validation data set comprising measurements taken both on the reference cell and on a second cell (the'test cell') which is aged differently and whose data are not used for ARMAX identification. The results show a very good estimation accuracy when validation data is extracted from the reference cell. When estimating the SOH and SOC of the test cell, the accuracy is seen to decrease but remains overall acceptable, especially at intermediate SOC levels. This may suggest investigating more sophisticated merged ARMAX models that use training data from multiple cells to derive a more general estimator.