Overview
- Date:Starts 13 December 2024, 14:00Ends 13 December 2024, 17:00
- Location:Campus Johanneberg: HA3
- Opponent:Associate professor Simona Onori, Stanford University, USA
- ThesisRead thesis (Opens in new tab)
Transportation electrification is critical to mitigating climate change, with lithium-ion (Li-ion) batteries playing a pivotal role in the shift to low-carbon energy sources. Given that batteries can account for up to 50% of an electric vehicle’s cost, optimizing their lifespan and performance is critical for cost-effective operation. Batteries though, degrade in ways that are inhomogeneous, nonlinear, and dependent on multiple factors. This makes accurate aging diagnostics and prognostics essential for ensuring their safe, efficient use. Diverse operating conditions, complex aging mechanisms, unpredictable usage profiles, and cell-to-cell variations pose significant challenges. At the same time, battery performance, including energy and power, is influenced not only by health state but also by conditions such as temperature, State of Charge (SoC), and applied current.
This thesis presents a series of machine learning (ML) frameworks developed using field data from vehicles and laboratory cycling data. One proposed framework is a battery capacity estimation algorithm that integrates multiple ML models with a Kalman filter, accommodating the diverse usage profiles of electric vehicles (EVs) in real-world scenarios. To reduce warranty costs, a histogram-based usage-related ML framework is developed, combining offline global models with online cell-specific models to track and predict future aging. Additionally, a remaining useful life (RUL) prediction model improves accuracy by combining usage and time-series data is developed as well.
Beyond aging diagnostics, the thesis proposes a method to extract relationships between battery performance indicators (PIs) and various influencing factors like temperature, SoC, and aging, using a neural network-based framework. Lastly, it introduces an online method to estimate battery plating potential, enabling faster charging while minimizing lithium plating risks to extend the lifetime of the battery. Collectively, these contributions provide practical tools for diagnostics, prognostics, and control, advancing safer, more efficient, and cost-effective use of Li-ion batteries in EVs.
This thesis presents a series of machine learning (ML) frameworks developed using field data from vehicles and laboratory cycling data. One proposed framework is a battery capacity estimation algorithm that integrates multiple ML models with a Kalman filter, accommodating the diverse usage profiles of electric vehicles (EVs) in real-world scenarios. To reduce warranty costs, a histogram-based usage-related ML framework is developed, combining offline global models with online cell-specific models to track and predict future aging. Additionally, a remaining useful life (RUL) prediction model improves accuracy by combining usage and time-series data is developed as well.
Beyond aging diagnostics, the thesis proposes a method to extract relationships between battery performance indicators (PIs) and various influencing factors like temperature, SoC, and aging, using a neural network-based framework. Lastly, it introduces an online method to estimate battery plating potential, enabling faster charging while minimizing lithium plating risks to extend the lifetime of the battery. Collectively, these contributions provide practical tools for diagnostics, prognostics, and control, advancing safer, more efficient, and cost-effective use of Li-ion batteries in EVs.
