Development of signal processing-based methods for estimating the remaining useful capacity of lithium-ion batteries in battery management systems

Lithium-ion batteries are known as one of the high-energy-density technologies and have a wide range of applications for storing electrical energy. Compared to other types of batteries, they have significant advantages such as lightness, improved safety features, and high voltage output. They are widely used in various industries such as transportation, communications, aerospace, and military defense. In recent years, with the increasing use of electric vehicles and battery-powered devices, the demand for high-capacity and stable lithium-ion batteries is also increasing. In this context, evaluating battery capacity and predicting the remaining useful life of lithium-ion batteries becomes important. This study aims to estimate the remaining useful life of lithium-ion batteries. NASA's B0005-B0006-B0007 and B0018 battery datasets were adapted to the specified working standards and attribute data were created using voltage, current, temperature and time information collected during the discharge state. In addition, new attributes were generated using signal processing methods such as Empirical Wavelet Transform (EWT), Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD). Artificial Neural Networks (ANN) method was decided for prediction and the prediction process was performed with the Levenberg-Marquardt algorithm. Models were created with all the methods used and the prediction of these models was carried out. Performance values were evaluated using common metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). In the study using MATLAB, the success of all the methods was compared using test data from 90-100-110th discharge cycle data for B0005-B0006-B0007 batteries. In addition, the performance values of the methods are compared starting from 71-79-86th discharge cycle test data for B0018 battery. These results show that the proposed model is promising for battery capacity value estimation.