Due to Internet-of-Things (IoT) developments and sensor technology, monitoring critical components remotely and continuously is possible. However, the condition monitoring data is often limited, biased, and unlabelled, which limits the intelligence model from predicting the remaining useful life and supporting accurate maintenance actions.
This research aims to develop an interactive and adaptive framework for predictive maintenance. We focus on the rotating machinery, which is a critical component for many mechanical equipment systems with a delay-time degradation pattern. We will develop a maintenance framework that evolves over time by learning from past failures, adjusting strategies based on new data, and continuously optimizing policies. A maintenance framework will be defined by (i) continuously updating the system health assessment based on inspections and (ii) informing subsequent maintenance decisions. The model will
leverage the Continuous Learning (CL) paradigm, using accumulated sensor monitoring data and human knowledge related to component deterioration to update decision-makingmodels over time. This framework should achieve a high system availability and a low Total Cost of Ownership (TOC) in the long term.
