Prof Tang Xiaopeng, Assistant Professor of the Science Unit.
Prof Tang Xiaopeng (2nd from the right), Assistant Professor of the Science Unit, wins the Best Paper Award at the IEEE-China Conference on System Simulation Technology and its Applications 2024.
Lithium-ion batteries are widely used daily inelectronic appliances, and with the growing popularity of electric vehicles globally, batterysafety has become a mounting concern. The development of battery monitoring and fault diagnosis technologies is therefore crucial. Prof Tang Xiaopeng, Assistant Professor of the Science Unit at Lingnan University, in collaboration with researchers from the University of Shanghai for Science and Technology, Shanghai University of Engineering Science, and Tongji University, has proposed a small-sample-learning-based,deep learning model to accurately predict the battery electrochemical impedance spectroscopy(EIS), making it a low-cost, very efficient and highly accurate battery assessment method. The co-authored paper "Online multi-scenario impedance spectra generation for batteries based on small-sample learning" was recently published in the prestigious international academic journal "Cell Reports Physical Science".
EIS measures the battery’s impedance characteristics against the frequency. It is widely used to evaluate the battery’s health, ageing mechanisms, power capability, temperature, etc.However, traditional measurement techniquesrequire expensive professional instruments and complex management systems, and are time-consuming, and thus not widely applied. The industry has also employed deep learning technology to develop battery monitoring techniques, but substantial differences in temperature, ageing, and performance in different batteries increase reliance on large datasets and hinder accessibility.
After in-depth analysis and research, Prof Tang and his research team have developed an EIS prediction method adopting the small-sample-learning technique, which uses virtual simulation techniques to mitigate the issue of limited data. For deep learning training, they generate a large set of samples covering variousbattery-using scenarios, including different battery chemistries, degrees of ageing, remaining capacities, and temperatures. The team then used a small amount of data (no more than 30 groups) from actual batteries to fine-tune the deep-learning model for the final online prediction of a battery's impedance. The typical error of the developed method is lower than five per cent, outperforming most big-data algorithms of its kind.
Prof Tang Xiaopeng, Assistant Professor ofthe Science Unit, stated the breakthrough was the ability to build an extremely accurate deep learning model using a very small number of battery samples, surpassing most similar algorithms that involve big data. He said "By simply uploading battery data online, we can automatically check the battery's electrochemicalimpedance and health status, removing the time and geographical constraints of traditional instrumentation. In addition, the new method reduces the time required to monitor battery performance, and cuts testing costs drastically, which significantly improves the safety and reliability of battery systems in the community."
Taking the increasingly popular electric vehicles as an example, Prof Tang went on to explain “Electric vehicles contain thousands of batteries, and the fault of a single cell will influence the performance of the entire battery pack, and may even lead to safety issues. This new technique lowers the limit of using EIS in industry, providingfast, cheap tools to analyse the mechanism of battery failure. It can improve the batteries’ risk-alarming capability and the electric vehicle’s safety.”
Another paper co-authored by Prof Tang Xiaopeng on lithium-ion batteries won the Best Paper Award at the IEEE-China Conference on System Simulation Technology and its Applications 2024 (CCSSTA 2024). It was one of only four papers to win out of 278 submissions to CCSSTA 2024, covering topics such asaerospace, robotics simulation, information processing and decision-making, intelligent control and decision-making, and renewable energy. The paper proposes a semi-supervisedlearning approach to screen retired batteries,increasing their reuse rate.
