Gas-Informed Machine Learning Framework for Stage Classification and Early Forecasting of Battery Degradation

Area of Science:

• Electrochemistry
• Materials Science
• Chemical Engineering

Background:

• Traditional battery health prediction relies on electrical signals (capacity, voltage, impedance).
• These signals are chemically nonspecific and often lag behind actual failure onset.
• Early identification of battery degradation stages is crucial for reliable performance.

Purpose of the Study:

• To demonstrate operando gas evolution via OEMS for early battery degradation identification.
• To develop models for distinguishing degradation stages and forecasting capacity.
• To establish gas evolution as a chemically grounded complement to electrical monitoring.

Main Methods:

• Online Electrochemical Mass Spectrometry (OEMS) to measure gas evolution (CO, CO2, C2H4).
• Principal Component Analysis (PCA) and K-means clustering for label-free stage classification.
• A single-shot forecasting model fusing capacity/SOH history with OEMS data.

Main Results:

• The stage classifier achieved a 3.6% misclassification rate, identifying nonlinear transitions 32.0 ± 17.2 cycles before the Bacon-Watts knee.
• The forecaster achieved 1.7-2.9% RMSE at 100-cycle horizons on held-out cells.
• Gas evolution signatures enabled earlier degradation detection and more accurate long-horizon capacity forecasting.

Conclusions:

• Operando gas evolution provides SOC-resolved chemical signatures for earlier degradation identification.
• Gas evolution monitoring, combined with electrical data, enhances battery health management.
• This approach motivates integrating compact off-gas sensing for cycle-level battery management.