Transfer Learning-based methodology for State of Health estimation of EVs battery pack.

Global warming is a pressing environmental issue caused by the increase in greenhouse gas emissions, primarily carbon dioxide, in the Earth's atmosphere. This rise in temperature leads to a range of detrimental effects, including melting ice caps, rising sea levels, extreme weather events, and threats to biodiversity. Urgent action is essential to mitigate global warming and preserve the planet for future generations. Electric vehicles (EVs) have emerged as a promising response to the urgent global challenge of combating climate change and reducing the impacts of global warming. As the transportation sector is a significant contributor to greenhouse gas emissions, transitioning from conventional internal combustion engine vehicles to EVs offers several environmental benefits. Lithium-ion batteries (LIBs) have revolutionized the world of portable electronics and EVs. These rechargeable powerhouses offer high energy density, long cycle life, and low self-discharge rates. The widespread adoption of EV battery packs induces a crucial challenge: accurately estimating their real-time SOH. Which serves as a vital indicator of EV ageing. Nonetheless, estimating SOH remains a complex task due to the electro-chemical intricacies of LIBs and their non-linear charge and discharge behaviours. This thesis delves into the existing approaches utilized for estimating the SOH of EV battery packs and develops a real-time, and computationally-efficient machine learning technique. This method leverages a narrow time window of voltage, current, and state of charge measurements obtained during vehicle operation. Real driving session data acquired from a private EV fleet management company build a base for the creation of a training dataset. To overcome the scarcity of publicly available EV monitoring data, a synthetic dataset is generated by simulating multiple driving sessions using a Simulink-based EV model. This synthetic data are then compared with real data obtained from the company to evaluate the performance of the EV simulator. A transfer learning approach is employed to bridge the gap between real and synthetic data, improving the performance of machine learning algorithms in estimation. The transfer learning methodology has led to positive results. However, the real dataset has a poor range of SOH values, bringing to the necessity of amplifying this span to have a clear idea of actual performance and improve the leverage of transfer learning methodology. Even though, we firmly believe that this procedure possesses significant potential for improvement in terms of its generalization capability.