Architecting Digital Twin-based Predictive Maintenance Systems

Summary

From bicycles to wind turbines, every physical system needs to be maintained. Depending on requirements on reliability, and a system’s cost of downtime, a system is normally maintained 1) once it fails or 2) according to a schedule. Recent advancements have enabled a third option: maintaining at exactly the right moment. By measuring a system’s vibrations, temperature, or other indicators, a predictive maintenance algorithm can propose the optimal maintenance event – tailored to an individual system. Building such predictive maintenance algorithm is a complex task. We must think about many different components that support data collection, processing, the algorithm itself, and integration into the maintenance planning software. This PhD research – in collaboration with several industrial partners – aimed to provide the bigger picture: What is needed to build a predictive maintenance system? The study contributed several guidelines for predictive maintenance system design and developed algorithms to estimate the optimal maintenance event.