PhD scholarship in AI for Railway Asset Fleet Monitoring

In the field of engineering systems and data analytics, artificial intelligence (AI) techniques, such as machine learning (ML) and deep learning (DL) methods, are essential data-driven methods for fault detection and diagnosis (FDD). Previous studies have proposed FDD approaches that leverage heuristics-based rules and supervised learning-based models, but these approaches have drawbacks. Rule-based approaches require an engineer to update rules for every type of fault, while supervised learning-based approaches require finely labelled training datasets. Achieving high FDD classification accuracy through supervised learning requires a well-formed training dataset in the training phase. Additionally, both approaches are limited in that they can only identify trained types of faults and fail to recognize unseen types of faults. Imbalanced training datasets also invalidate most supervised learning-based FDD systems. Factors such as operating conditions can also impact the monitoring outcomes. Calibration of hardware and algorithm settings is often required for meaningful monitoring results, particularly when monitoring systems are deployed on different objectives. These robustness problems need further studies in the area of railway fleet monitoring and fault diagnosis.