Abstract
This paper investigates waveform-level voltage difference features for the early indication of cell-to-cell voltage reversal in two series-connected lithium-ion cells for electric-vehicle battery management systems (BMSs). The key observation is that a local sign reversal of the cell-to-cell voltage difference can appear inside the constant-current (CC) discharge waveform before the cycle-mean voltage difference changes sign. A positive-fraction feature is introduced to quantify this local waveform reversal and is interpreted as a candidate BMS-oriented indication layer rather than as a universal alarm criterion. The analysis combines a Kirchhoff decomposition of the cycle-mean voltage difference with waveform matrix analysis and physical consistency checks based on principal component analysis (PCA), voltage slope behavior, equivalent-circuit-model sensitivity, and auxiliary temperature signals. In the present dataset, the first 5% crossing of the introduced positive-fraction feature occurs 44 cycles before the cycle-mean voltage reversal, and the five-cycle persistent onset precedes it by 39 cycles; this lead stays between 29 and 46 cycles across the persistence and detection threshold settings examined. The onset location is consistent with the low-slope graphite-staging region of the loaded-voltage curve, where the open-circuit voltage (OCV) slope compensation is weak and the resistance-related offset can dominate locally. The reference event used to quantify this lead is the cycle-mean voltage reversal itself, which is an observable benchmark defined on the scalar that a BMS already computes rather than a ground-truth onset of imbalance; the two cells are measurably unequal from the first cycle. The onset is shown to be insensitive to the resampling grid, the interpolation rule, the detection threshold and realistic voltage noise and channel offsets, and to produce no false onset anywhere in the 998-cycle record. The result is a waveform-level diagnostic observable quantity that converts a hidden local sign reversal into an interpretable cycle-level indication for BMS-oriented screening and monitoring. Because this study rests on a single two-cell series connection, it is presented as a proof-of-concept case study, and replication on independent cell pairs is required before the behavior can be regarded as general.