Damage evolution mechanism and early warning using long short-term memory networks for battery slight overcharge cycles

Slight faults can damage battery electrodes and electrolytes, leading to cumulative irreversible capacity loss and decreased performance, even a critical state of failure. In this study, the overcharge cycling of lithium-ion battery (Lithium, 2600 mAh, 3.7 V) is studied to reveal the damage evolution mechanism and establish a novel early warning method for slight faults. With the increase of cycles, the aggregation of the loss of active materials leads to the acceleration of capacity fading rate and the acceleration factor increases from 1 to 3.6 when the cut-off voltage attends 4.4 V. But these cells follow a similar damage evolution path to the normal cells during cycling. Based on the accelerating fading feature of fault cells, a capacity prediction model for early warning was developed. The batteries' capacity data are firstly smoothed by the Savitzky-Golay filter and then transferred to long short-term memory (LSTM) networks for training. The model can predict the capacity of overcharged cells well within a 2% error by optimizing the sizes of input and output data. And the slight overcharge fault can be early warned through a specific threshold of the root-mean-square deviation between the prediction and the norminal capacity degradation curve.