Weakly Supervised Learning for Predictive Maintenance

Abstract

With the advent of Industry 4.0, Predictive Maintenance (PdM) has garnered a lot of interest, both academically and in the industry. This thesis will be developing and using machine learning methods for PdM, using real world event-log data gathered from hybrid marine vessels, equipped with electric propulsion systems. The methods that will be used were chosen for their abilities to solve particular problems, such as data imbalance through the use of Balanced Random Forest, weakly labelled data through the use of Multiple Instance Learning, and maintaining interpretability through the use of interpretable pre-processing techniques, such as window aggregation.