Internal short circuit early detection of lithium-ion batteries from impedance spectroscopy using deep learning

Detecting the early internal short circuit (ISC) of Lithium-ion batteries is an unsolved challenge that limits the technologies such as consumer electronics and electric vehicles. Here, we develop an accurate and fast ISC detection method by combining electrochemical impedance spectroscopy (EIS) with a deep neural network (DNN). We achieve zero false positives for ISC detection of the normal battery and an ISC detection average percentage accuracy of 97.5% over the full life cycle of the battery with the equivalent resistance for ISC from 200 Ω to 10 Ω. We also demonstrate the universality of the proposed methods by the other battery. Based on the distribution of relaxation times and sensitivity methods, we further reduce the required EIS measurement time and improve computational efficiency by choosing the most sensitive EIS spectrum to ISC. Our results demonstrate the value of the EIS spectrum in battery management systems.