Predicting battery impedance spectra from 10-second pulse tests under 10 Hz sampling rate
Xiaopeng Tang,
Xin Lai,
Qi Liu,
Yuejiu Zheng,
Yuanqiang Zhou,
Yunjie Ma,
Furong Gao
Affiliations
Xiaopeng Tang
Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong SAR 999077, China
Xin Lai
Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong SAR 999077, China; School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
Qi Liu
Department of Physics, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong SAR 999077, China; Corresponding author
Yuejiu Zheng
School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
Yuanqiang Zhou
Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong SAR 999077, China
Yunjie Ma
School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
Furong Gao
Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong SAR 999077, China; Guangzhou HKUST Fok Ying Tung Research Institute, Guangzhou, Guangdong 511458, China; Corresponding author
Summary: Onboard measuring the electrochemical impedance spectroscopy (EIS) for lithium-ion batteries is a long-standing issue that limits the technologies such as portable electronics and electric vehicles. Challenges arise from not only the high sampling rate required by the Shannon Sampling Theorem but also the sophisticated real-life battery-using profiles. We here propose a fast and accurate EIS predicting system by combining the fractional-order electric circuit model—a highly nonlinear model with clear physical meanings—with a median-filtered neural network machine learning. Over 1000 load profiles collected under different state-of-charge and state-of-health are utilized for verification, and the root-mean-squared-error of our predictions could be bounded by 1.1 mΩ and 2.1 mΩ when using dynamic profiles last for 3 min and 10 s, respectively. Our method allows using size-varying input data sampled at a rate down to 10 Hz and unlocks opportunities to detect the battery’s internal electrochemical characteristics onboard via low-cost embedded sensors.
