Predicting battery impedance spectra from 10-second pulse tests under 10 Hz sampling rate

Area of Science:

• Electrochemistry
• Machine Learning
• Battery Technology

Background:

• Onboard electrochemical impedance spectroscopy (EIS) for lithium-ion batteries is crucial for portable electronics and electric vehicles but faces challenges.
• High sampling rates and complex real-life usage profiles complicate accurate onboard EIS measurements.

Purpose of the Study:

• To develop a fast and accurate system for predicting electrochemical impedance spectroscopy (EIS) onboard lithium-ion batteries.
• To overcome limitations of high sampling rates and complex battery usage profiles for real-time monitoring.

Main Methods:

• Combines a fractional-order electric circuit model, a nonlinear model with physical meaning, with a median-filtered neural network machine learning approach.
• Utilizes over 1000 load profiles under varying states of charge and health for system verification.

Main Results:

• Achieved root-mean-squared-error (RMSE) bounded by 1.1 mΩ for 3-minute dynamic profiles and 2.1 mΩ for 10-second profiles.
• The system accommodates size-varying input data sampled as low as 10 Hz.

Conclusions:

• The proposed method enables fast and accurate onboard EIS prediction for lithium-ion batteries.
• Facilitates real-time detection of internal battery electrochemical characteristics using low-cost embedded sensors.