Upscaling Prognostics for Aerospace Structures

In the aerospace industry, traditional maintenance practices rely on time-scheduled servicing, often leading to unnecessary costs and inefficiencies as maintenance is performed without considering the actual condition of structures. This has driven growing interest in predictive maintenance (PdM), which uses prognostic health management (PHM) to optimise maintenance schedules based on real-time data. While significant progress has been made in prognostics for individual components, there is a gap in applying these techniques to larger, more complex aerospace structures. This thesis addresses that gap by focusing on upscaling prognostic models for aerospace structures. Specifically, it adapts the interoperability input-output model (IIM) to predict the remaining useful life (RUL) of higher-level structures using the training data of the lower-level components of that structure.

Current research on prognostics primarily addresses low-level structures, such as individual coupons, with a limited focus on larger structures. To address this gap, the thesis investigates upscaling methods, specifically system-level prognostics (SLP), which considers a structure as a combination of interconnected components. The IIM is chosen due to its transparency, low complexity, scalability and ability to model interdependencies, making it suitable for aerospace applications.

The methodology adapts the IIM for aerospace structures by incorporating data-driven base predictors, specifically the hidden Markov model (HMM) and support vector regression (SVR). These predictors generate RUL estimates, which are then used to train the IIM through offline and online algorithms. A physical specimen, designed to resemble a higher-level aerospace structure comprising lower-level components, is subjected to fatigue testing to collect real-world data, which is used for SLP.

The case study examines the IIM’s predictions using this specimens' data and compares them to results from the base predictors. It evaluates how the model responds to different training data and system behaviours, including sudden failure scenarios. The thesis concludes by validating the modified IIM's effectiveness for system-level prognostics in aerospace structures and offers insights for further refinement and application of this approach in real-world aerospace settings.