A System-Level Approach for Enablement of Prognostics and Health Management - Utilizing Industrial AI

Industrial production systems depend on key assets to maintain process continuity and support essential operational tasks. To meet these requirements, assets must consistently deliver their intended functions. Failures can reduce system performance and result in downtime, production loss, and increased maintenance costs. Furthermore, severe failures may compromise asset integrity and introduce safety risks. These potential consequences highlight the importance of systematic health management to promote reliable and safe operation throughout the asset’s life cycle. Maintenance plays a vital role in ensuring asset functionality. However, relying solely on corrective or schedule-based maintenance approaches often proves insufficient. When maintenance occurs too late, early signs of degradation may remain undetected and progress into more severe failures. In contrast, performing maintenance too early leads to unnecessary work and higher costs. Prognostics and Health Management (PHM), which encompasses state detection, diagnostics, and prognostics, and maintenance decision support, enhances the ability to understand the asset condition and intervene proactively. The work described in this thesis aims at enabling PHM by establishing essential steps: system-level description of the asset and its functions, failure analysis, and detecting deviations in critical components. Assets are decomposed into subsystems and components to clarify their roles and interactions. Further, the thesis examines how asset functions can be impaired and identifies the components whose degradation is most critical to functional performance. These insights guide the selection of components that require condition monitoring. Methods are then developed for deviation detection from expected performance, forming the basis for state detection within a PHM framework. Additionally, a review of data-driven PHM approaches for mining processing machines highlights gaps in system-level health management and the use of different data modalities, further motivating the structured methods developed in this thesis. The contributions are demonstrated through two case studies in construction and mining, respectively. In the construction case study, 3D point-cloud data are used to identify missing or deviated scaffolding elements that compromise structural stability. In mining, image-based analysis is applied to a Rotary Vacuum Drum Filter (RVDF) to detect wire failures that can damage the filter cloth and reduce asset availability. Early fault detection supports structural integrity, reduces downtime and lowers maintenance costs. By combining asset structure analysis, failure analysis to identify critical components, and methods for deviation detection, this work delivers essential building blocks for the enablement of PHM. Future work will explore the development of diagnostics, prognostics models and multimodal data integration to further support maintenance decision-making and to strengthen health management capabilities.