Machine Learning-Enhanced Electrical Circuit Model Parametrization for Battery Cells: Reducing Experimental Workload Through GITT Testing with Altair RapidMiner®

This study addresses the challenge of reducing the experimental workload involved in characterizing battery cell behavior as a function of state of charge and temperature. Galvanostatic Intermittent Titration Technique tests were carried out in a climate chamber across a wide temperature range, from -20 °C to 70 °C, with 10 °C intervals. The voltage and current response data collected from these tests were used to train several machine learning algorithms. The trained models could then be used to predict the cell voltage response every 5 °C from -15 °C to 55 °C. While the models were experimentally validated at 15 °C, 25 °C, and 35 °C, the predicted voltages across this range contribute to enhancing the characterization process. In particular, the inclusion of these predicted voltage profiles—combined with the experimental data collected every 10 °C from -20 °C to 70 °C—allows for the creation of more accurate lookup tables for the parameters of the equivalent circuit model. These parameters include the open circuit voltage, series resistance, and multiple resistor-capacitor pairs representing dynamic electrochemical behavior. This approach results in significantly improved parameter estimation compared to using only the original experimental dataset.