A comparison of machine learning approaches for the prediction of commercial electric vehicle battery faults

Context

Electric vehicles are considered a promising solution for lowering overall emissions in the transportation sector. Among them, electric commercial vehicles, such as buses, play an important role in cutting emissions. However, one of the major challenges these vehicles face is the battery system safety. If the faults cannot be detected on time, it can lead to accidents such as thermal runaway or even explosion. To prevent such events, fault prediction models can be developed to identify potential failures before they happen by analyzing data patterns from past fault occurrences. The predicted faults can then be used by vehicle controllers to apply suitable fault-tolerant control strategies, reducing the risk of critical system failures. The current state of the art provides only limited information about the commercial electric vehicle battery fault predictions using machine learning. The goal of this thesis is to provide a comparison of suitable ML approaches for real cases of battery electric and thermal faults. Additionally, there is a need for a dedicated evaluation framework to assess these predictive models. The commonly used metrics, such as model accuracy and precision, do fail to capture the real model prediction capabilities. Therefore, this work also aims to develop a more suitable evaluation methodology for validating predictive diagnostic models.

Goals

• State of the Art of the ML approaches used for predictive diagnostics in EVs
• Investigation of suitable ML approaches for the use cases of thermal and electric EV battery faults
• Development and implementation of suitable ML-based predictive diagnostic models
• Development of novel evaluation methodology for validation of predictive diagnostic models

Requirements

• Interest in the development of AI/ML systems
• Basic knowledge and experience in data processing
• Programming skills (Python/Matlab)