Two-Step HNN-Based Pattern Recognition Combining DWT-Based Multi-Resolution Analysis for Rechargeable Cells Distinction

This article newly suggests two-step Hamming neural network (HNN)-based pattern recognition which combines the discrete wavelet transform (DWT)-based multi-resolution analysis (MRA) for providing an innovative and unique cell distinction. In contrast to the conventional methods, there are some remarkable features. First, the discharging/charging current signals (DCCSs) for reflecting load conditions with various dissimilar scenarios are used as representative patterns in the first-step HNN. Specifically, the usage of decomposed low- or high-frequency components (A5/D5) by the DWT-based MRA enables added characteristic parameters for providing the closest matches. Second, according to state-of-charge (SOC) range, especially in high SOC ranges, the allowable maximum charging currents in constant current/voltage (CC/CV) schemes are adjusted for avoiding over-charging. Third, the second-step HNN finally completes the recognition after closely matching several representative discharging/charging voltage signals applying the selected DCCS in the first step. Finally, simultaneous internal parameter extraction in the equivalent electrical circuit model during the second-step HNN provides useful information for verification of this article. Then, unnecessary experimental procedure and the time required for getting parameters of an arbitrary lithium-ion cell are abbreviated or reduced.