Project: Fault Prediction Using Digital Twin Techniques

Key information:

Abstract: 

Our safety and economy rely on countless systems working without fault. Dependable systems must be inspected, repaired, and replaced regularly to guarantee their function. As the uncertainty of system health increases, so do the maintenance and expense. Fault prediction algorithms aim to reduce this uncertainty, improving safety and lowering costs.
Industrial implementation has requirements that black box algorithms struggle to fulfil: explainable decisions, guaranteed behaviour, and accountability. This research attempts to meet these requirements by investigating algorithms that construct theories of a system's dynamics through the composition of mathematical objects.
Computational implementations of measure spaces, topologies, and abstract algebras allow the manipulation of industrial data in a mathematically rigorous format. Exploring the construction of fault prediction algorithms from these elements is motivated by the supposition of improving trust and performance. Generating system models with nested layers of abstraction can provide higher-level explanations of an algorithm's decisions, while their provable behaviour provides quantified uncertainty. These features facilitate the incorporation of algorithms into industrial decision-making for high-risk applications.