Use of Ultrasound to Characterise and Estimate State-of-Charge, State-of-Health, and Temperature In Lithium-Ion Batteries

Lithium-ion batteries have seen rapid adoption since the 1990s, a trend set to continue with global decarbonisation efforts. Electric vehicles represent a major driver of this growth. These batteries operate through the movement of lithium-ions between electrodes during charge and discharge. Current battery management systems do not exploit these internal changes directly, instead relying on charge counting, voltage, and impedance measurements.

Recent research has introduced ultrasound as a method to monitor internal changes by tracking variations in the ultrasonic signal, which reflect changes in electrode material properties. This has been applied to estimate state-of-charge and state-of-health, though the relationship between SOH and ultrasonic response remains unclear due to conflicting findings. This thesis addresses these inconsistencies and investigates the under-explored impact of temperature on the ultrasonic signal.

Long-term cycling experiments were used to track ultrasonic time-of-flight drift as a function of battery degradation, separating this from cyclical variations due to charge-discharge. The time-of- flight drift direction was suggested to be dependent on cathode chemistry. The ultrasonic method showed high sensitivity to SOH fluctuations across the battery lifecycle.

A thermal cycling method combined with a global sensitivity analysis was used to assess the effects of temperature (10 to 50°C) and state-of-charge (0 to 100%) on the ultrasonic signal. Results showed both variables independently affect the signal, with temperature having the dominant effect.

Two regression models---linear and Gaussian Process---were trained to predict SOC and temperature from ultrasonic data. The linear model demonstrated consistent stability and accuracy across varying noise levels, but was significantly influenced by data formatting.

This thesis focusses on the effects of temperature and state-of-charge on the ultrasonic signal to improve the accuracy of state-of-charge monitoring and has demonstrated the use of machine learning to predict both variables.