A Data-Driven approach to Prognostics and Health Management of Aircraft Structures for Predictive Maintenance in Operations and Sustainment

<p dir="ltr">This thesis makes significant research contributions towards the development of structural Prognostics and Health Management (PHM) methodologies with the development of diagnostics and prognostics algorithms for fleet aircraft sustainment and predictive maintenance operations. This research is an industry collaboration with the Australian Defence Science and Technology Group (DSTG), as well as interfacing with other Defence organisations, including the Capability Acquisition and Sustainment Group (CASG). </p><p dir="ltr">Unlike prior research, this thesis introduces novel contributions that develop PHM in a direction towards condition-based maintenance at the fleet level to account for Configuration, Role and Operating Environment (CR&E) delta. This research has developed diagnostic and prognostic approaches that utilise on board aircraft sensors to detect and quantify structural non-linearities (e.g., control surface freeplay). This marks an important step in enabling effective in service sustainment that accounts for varying CR&E of aircraft. Furthermore, to support subsequent decision-making, the research implements Natural Language Processing (NLP) of unstructured textual maintenance records and tools for visualisation using Mixed Reality (MR) devices and software. This end to-end process flow provides a pathway for the realisation of improved efficiency, timely decision-making and task performance. </p><p dir="ltr">This thesis presents novel methodologies and frameworks addressing key gaps in the literature on aircraft sustainment through advancements in Prognostics and Health Management (PHM). Current research in PHM lacks robust diagnostics and prognostics for non-linear structural degradation in aircraft, particularly in operational conditions. Additionally, conventional maintenance practices are limited by reliance on manual analysis of unstructured textual reports and insufficient integration of emerging technologies, such as Augmented Reality (AR) and digital twins, for effective damage evaluation and predictive maintenance. </p><p dir="ltr">The thesis first develops a PHM framework for aircraft control surface freeplay analysis, introducing a two-phase process: supervised Machine Learning (ML) for diagnostic severity detection and Particle Filter (PF) based Remaining Useful Life (RUL) estimation. Results for diagnostics of nonlinearities (e.g., freeplay) achieve high accuracy on limited time-series data, while prognostics results demonstrate RUL predictions are possible with good confidence. Addressing limitations in existing literature, a synthetic dataset and numerical model is developed for freeplay degradation, enabling better understanding of the nonlinear damage failure modes. To complement numerical models, an experimental freeplay test rig with a remote sensing system is developed to monitor buffet-induced structural damage in aircraft structures. A three-step data analysis framework incorporating signal processing, feature extraction and predictive modelling demonstrates the effectiveness of neural networks and PF methods in diagnosing and predicting damage under realistic load conditions. Combined with surrogate datasets of real-world aircraft, these numerical and experimental approaches are verified for performance in diagnosing and predicting nonlinear damage in aircraft structures. </p><p dir="ltr">The thesis further incorporates AR for Non-Destructive Evaluation (NDE) of damage hotspots, validated through a proof-of-concept using a decommissioned aircraft and digital twin models. This approach provides an alternative process to visual inspections of aircraft damage and demonstrates enhanced accuracy and efficiency in maintenance tasks, but also highlights the challenges in digital twin workflows for operational scalability. Lastly, the novel NLP framework is introduced to exploit unstructured maintenance and pilot reports, which are under-utilised and limited by labour-intensive manual analysis. This NLP framework uses statistical and ML techniques to provide a novel approach to identifying early lead indicators to aircraft major defects. </p><p dir="ltr">The contributions made in this thesis collectively tackled the full, end-to-end process chain for data-driven predictive maintenance. From data generation with on-board monitoring sensors, to ML algorithms for lead indicator identification and damage progression modelling, this thesis extends the state-of-the-art for multiple PHM steps, through not only covering freeplay prognostics, but extending this into visualisation, not previously seen in literature. </p><p dir="ltr">This thesis provides the necessary methodologies and frameworks to develop prognostics for aircraft control surface anomalies, NLP for automatic risk rating of fleet anomalies in underutilised unstructured maintenance reports and a Mixed Reality Non-Destructive Evaluation (MR-NDE) approach to aircraft sustainment. The outcomes of this thesis will support engineers and maintainers in the field to be more proactive in maintenance operations and help reduce overall costs through effective aircraft sustainment practices.</p>