Multi-feature data-driven model for fast SOH assessment of batteries after their first life

In response to the increasing demand for sustainable energy solutions and the growing prevalence of Electric Vehicle technology, the effective management of retired batteries has become a critical concern. As these batteries reach the end of their operational life, accurately predicting their State of Health is imperative for optimizing reuse and recycling efforts. This study introduces a fast State of Health estimation model for retired batteries, which integrates three features: initial voltage, temperature, and internal resistance measured after a 10-second charge pulse. Two predictive models, Feedforward Neural Networks and Extra Tree Regressor , are proposed and evaluated at both the cell and module levels. Both models demonstrate high accuracy in predicting State of Health, yielding Mean Absolute Percentage Errors lower than 1.75 % at the cell level and 2.42 % at the module level. Experimental measurements from retired batteries are utilized for the training and testing of these models. By employing the proposed method, the time required for testing and the cost of assessing batteries at their end of life are significantly reduced compared to traditional State of Health assessment methods. Testing time transitions from hours to seconds, and the estimated testing cost is reduced by 60–82 % for cells and modules, respectively. This streamlined approach eliminates the need for pre-processing tests or climate chambers, enhancing the efficiency and practicality of battery evaluation processes.