An online unscented Kalman filter remaining useful life prediction method applied to second-life lithium-ion batteries

In electric vehicles (EVs), because of the high current demand, lithium-ion batteries (LiBs) degradation makes the EVs suffer from limitations in their maximum autonomy and acceleration. Thus, after a certain point, the LiBs cannot continue to operate in these applications. However, after the LiB is removed from the EV, it still has about 80% of its nominal capacity available. Therefore, an interesting alternative to not discarding these LiBs is to reuse them in applications with lower current demand, such as power backup systems, this process is known as second-life. In second-life applications, due to the high degradation state of the LiBs, the need to implement an algorithm to estimate the remaining useful life (RUL) is necessary as it provides an aid to preventive maintenance. Many methods can be applied to estimate the RUL of LiBs; nevertheless, many of them require a large amount of training data, or are not suitable for embedded applications. Also, due to the nature of second-life LiBs, the degradation curve of these LiBs can be very unpredictable, and estimating their RUL is a challenge. In this context, this work proposes a method that employs an unscented Kalman filter (UKF) and a degradation curve model to perform online estimations of the RUL of second-life LiBs. The proposed algorithm was validated using experimental data that consists of the degradation curve of six distinct second-life LiBs. During the validation of the algorithm, in the worst-case scenario, a mean absolute percentage error (MAPE) and R2 score, equal to 5.279% and 0.726, were obtained.