This technology estimates a battery's current capacity and State of Health (SoH) without complex physical models by training machine learning algorithms—such as FNN, CNN, RNN, and LSTM—on time-series patterns derived from the three key data points measured during charging: voltage (V), current (I), and temperature (T).
Existing electrochemical or equivalent circuit models rely on complex equations, leading to high computational costs and significant hardware resource consumption. Furthermore, relying on only one or two indicators, such as voltage, often results in low accuracy when estimating battery degradation in real-world usage environments.
This technology creates a matrix-style training dataset by pairing capacity with time-series measurements of voltage, current, and temperature changes that occur during constant-current and constant-voltage charging cycles. By utilizing models like LSTM and GLU, it learns degradation patterns that account for irregular previous discharge histories. Applicable to EV BMS, energy storage system management, and residual value assessment of used batteries, it enables the determination of replacement timing and reuse potential using only charging data, without the need for separate diagnostic equipment.
N/A
