Fault data generation of lithium ion batteries based on digital twin: A case for internal short circuit

Battery accidents are emerging for various kinds of applications, such as commercial electronics, electric scooters, electric vehicle and energy storage stations. Effective early warning algorithm is essential for mitigating the damage caused by battery failure. An early warning algorithm needs to be trained by failure data. However, building battery failure data (such as internal short circuit) by experiment consumes so much time and cost that it cannot match the development requirement of new product. This work demonstrates the validity of using digital twin technique to generate battery failure data to train the online early warning algorithm. The digital twin is formed by a highly reliable thermal-electrical coupled battery model. The influential factors, e.g., load current, short circuit position, equivalent short resistance and environment temperature, which determine the performance of the early warning algorithm were discussed. The warning threshold that may judge the fault warning rate and response speed of the algorithm was also analyzed. The digital twin helps us find that there is a temperature difference between the sensor location and the failure (short circuit) point. The temperature dif-ference enlarges under the cases that the failure point is far from the measurement point, thereby hindering the timely detection and alarm for fault. The temperature gap can be as high as 261.2 degrees C under the extreme case of 0.1 omega short circuit resistance. Therefore, the mismatch of the failure point and the measurement will give us the illusion that the failure occurs suddenly, or the so-called sudden death. The thermal runaway boundary for battery module varies with the ambient temperature and the short resistance, helping the determination of the threshold of the detection algorithm for thermal runaway. At last, a guidance on how to develop fault diagnosis algorithm with the help of digital twin to judgement of fault diagnosis threshold is summarized, with clear discussion of the accuracy, cost and efficiency of the method. The proposed method paves a way for the application of the digital twin on the fast development of battery management algorithms.

Automated summary

This paper investigates the validity of using digital twin technique to generate battery failure data for training online early warning algorithms. The authors build a reliable thermal-electrical coupled battery model to form the digital twin and discuss influential factors on the performance of the early warning algorithm. They also discover a temperature difference between the failure point and the measurement location, which hinders timely fault detection and alarm. Additionally, the paper provides guidance on developing fault diagnosis algorithms with the help of digital twin.