Abstract
The commercial Gas Turbine Engine is vital to modern-day living. Industry is focused on ever more efficient engines to combat rising fuel prices, environmental concerns and commercial competition. Increased efficiency requires higher temperatures, which then results in higher rates of corrosion. One form of engine degradation is hot corrosion. The mechanisms behind this degradation are not yet fully known, nor are the products and reactants involved well defined. This results in conservative estimates of component lifespans being used. On-wing detection of hot corrosion could help understand how components have been affected by corrosion, and thus could help reduce waste in industry.Initial work was conducted using artificially corroded samples, to reduce some of the unknown variables that affect in-service components. Samples were analysed using Raman spectroscopy and Energy Dispersive X-Ray spectroscopy (EDX). Trends were observed across the sample set that could be linked to both temperature and duration, and by the combination of both EDX and Raman, previously unknown compounds could be identified on the sample surfaces. Further analysis proved difficult, as the collected Raman data could not be appropriately analysed, and identification of the collected spectra was difficult without either understanding the sample surface environments or having a database with which to compare.
A novel method of data analysis was therefore developed, which combined Non-Negative Matrix Factorisation (NMF) and Principal Component Analysis (PCA), to create a technique that produced orthogonal trends, describing how Derived Spectra are related to one another. These Derived Spectra are physically realistic, and as such can be readily compared to literature. This methodology was applied to two datasets, and proof of concept successfully demonstrated.
A final area of work looked at how the above could be integrated into an industrial workflow. A number of techniques for automating the process of identifying compounds based on their Raman spectra were discussed, and a method of producing physical maps from the trends generated by Non-Negative Assisted PCA was tested.
| Date of Award | 21 Jun 2022 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Sponsors | Rolls-Royce plc |
| Supervisor | Peter Flewitt (Supervisor), John C C Day (Supervisor) & Alejandro Jeketo (Supervisor) |