Fig. 1: EV dataset and challenges in fault detection.
From: Realistic fault detection of li-ion battery via dynamical deep learning
a The data used in this study contains vehicles from three manufacturers, aliased Dahu, Socea and Naobop. Each dot represents the amount of data between the first charging record and the last charging record of a single vehicle. The x-value and y-value of a dot indicate the distance traveled and the time elapsed in records. b Sampled charging segments show that real-world charging patterns are diverse and irregular. Charging modes can be categorized into fast charging and regular charging based on the level of the current. c Normal and abnormal EVs are poorly differentiated with canonical features such as the variance in cell voltage, current, or temperature, as an AUROC (area under receiver operating characteristic) around 0.5 can be achieved by random guesses. Higher AUROC value indicates greater prediction power. d A detailed view of LiB charging snippets from two sampled EVs. The comparison suggests that there is no simple feature to detect EV fault.
