Novel Taxonomy and Approaches for the Identification of Frequently Occurring Regularities in Degradation Processes of Engineering Systems

Abstract

The trend is shifting toward hybrid methods that incorporate prior knowledge into data-driven methods to address challenges in diagnostics and prognostics such as limited data, interpretability, and complex system behavior. While system-specific prior knowledge facilitates accurate, physically plausible modeling, the resulting hybrid model is typically tightly coupled to an individual engineering system. In contrast, general prior knowledge—such as fundamental physical laws or broadly applicable degradation knowledge—supports scalable, transferable models across various engineering systems. This opens the door to more adaptable approaches for diagnostics and prognostics, but the potential remains underexplored. To address this, a taxonomy is proposed that defines prior knowledge as frequently occurring regularities with four levels of validity, enabling hybrid methods to be characterized by their expected transferability. Two approaches are introduced and applied, both aimed at systematically identifying such regularities: one driven by expert knowledge, the other by data. Expert interviews further validate both the taxonomy and the identified regularities, establishing a foundation for developing transferable hybrid methods between various engineering systems.