Machine learning approach for lifespan prediction of second-life lithium-ion batteries

• Machine learning approach is presented for lifespan prediction of second life of LIBs. • Adaptive Moment Estimation is used to align the predicted with actual cycle life. • Real data of various LIBs extracted from NASA are used in the model. • The proposed model is compared with other recent models. • The effectiveness of the proposed model is verified under different LIB conditions. Second-life lithium-ion batteries (LIBs) offer technical, environmental, and economic benefits. In this context, this paper presents a machine learning model using the time interval of an equal charging voltage difference and the time interval of an equal discharging voltage difference features with Adam optimization to predict second-life SoH lithium-ion battery. The analysis uses data from commercially available 18650 Li-ion cells with a capacity of 2 Ah, tested under ambient temperature. The cells (B0005, B0006, and B0007) are subjected to repeated charge-discharge cycles using constant current protocols and electrochemical impedance spectroscopy to simulate accelerated aging and track internal parameter changes. These datasets train a machine learning model using the Adaptive Moment Estimation (Adam) optimizer, a machine learning optimizer with a unique feature engineering approach and a customized machine learning model aiming to closely align the predicted cycle life with the actual lifespan of the batteries. The model’s performance is benchmarked against other optimization and machine learning techniques, including the quasi-Newton method, linear regression, support vector regression, gradient descent (GD), stochastic GD, adaptive gradient, and root mean square propagation. Testing over 168 lifecycle samples, the Adam-based model achieves the lowest root mean square error, recording values of 5e-04 for battery B0005, 5.245e-04 for B0006, and 5.725e-04 for B0007. These results demonstrate the superior accuracy of the proposed method in predicting the lifespan of second-life LIBs, providing a promising pathway for improving the reliability and efficiency of lifespan predictions for second-life LIBs.