Battery safety issue is a key factor that hindered the reuse of retired batteries on new energy vehicles, while charge state, voltage and temperature are important parameters to judge the safety state of batteries. Based on this, a joint fault diagnosis of battery based on real vehicle data was proposed. Firstly, the data was obtained from the real vehicle data platform, and the feature extraction was carried out by data preprocessing and dung beetle optimizes K-nearest neighbor (DPO-KNN) algorithm. Then, the extracted features were input into the established autoregressive integrated moving average model, autoregressive integrated moving average model (ARIMA) fault diagnosis model to realize real-time diagnosis and accurate location for low and overvoltage of battery cells. Finally, voltage, state of charge (SOC) and temperature were combined to determine whether there was a risk of triggering thermal runaway, and different alarm levels were issued according to the degree of danger. It can not only accurately locate the abnormal position of the monomer, but also alarm 9 s in advance, effectively preventing the risk of thermal runaway of the battery. It achieved a good effect in engineering practical application, and lied a foundation for the future step recycling of power batteries and the research and development of safety warning platform.