Barry Van Tassell, Thomas Hodson, Shaurjo Biswas, Andrew Hsieh, Daniel A. Steingart, Greg Davies, Kevin W. Knehr
Ultrasonic analysis was used to predict the state of charge and state of health of lithium-ion pouch cells that have been cycled for several hundred cycles. The repeatable ultrasonic trends are reduced to two key metrics: time of flight shift and total signal amplitude, which are then used with voltage data in a supervised machine learning technique to build a model for state of charge (SOC) prediction. Using this model, cell SOC is predicted to ∼1% accuracy for both lithium cobalt oxide and lithium iron phosphate cells. Elastic wave propagation theory is used to explain that the changes in ultrasonic signal are related to changes in the material properties of the active materials (i.e., elastic modulus and density) during cycling. Finally, we show the machine learning model can accurately predict cell state of health with an error ∼1%. This is accomplished by extending the data inputs into the model to include full ultrasonic waveforms at top of charge.