Acoustic Enhanced Machine Learning Modeling of Li Metal Batteries

Higher energy density batteries are crucial for efficient energy storage. Anode-free Li-metal batteries have high energy-density, and we used mechanical information from the batteries to model their capacity degradation. Electrochemical and acoustic data were collected as the cells underwent formation and cycling. Ultrasound transmission through the cell provided acoustic data, with modulations in the received waveforms containing operando mechanical information. Raw waveforms or derived acoustic variables were used as features to train a random forest regressor with capacity data as the labels. Our model, using the derived variables, showed a mean R² of 0.998 across 33 cycles, accurately inferring the capacity for each cycle from the same cycle’s acoustics. This result indicates the modeling power of the acoustic information. Furthermore, we can perform highly accurate future predictions. This displays how chemo-mechanical acoustic data can predict future electrochemical degradation. Future predictions become more accurate with the derived variables, indicating that our variables successfully extract the valuable information from the raw waveforms. Highly predictive models, based on low-cost, operando, mechanical measurements provide valuable insight into battery degradation, aiding in the development of high-energy-density batteries.