The cost of wind energy has decreased over the last decade as technology has matured and the industry has benefited greatly from economies of scale. That being said, operations and maintenance still make up a significant proportion of the overall costs and needs to be reduced over the coming years as sites, particularly offshore, get larger and more remote. One of the key tools to achieve this is through enhancements of both SCADA and condition monitoring system analytics, leading to more informed and optimised operational decisions. Specifically examining the wind turbine generator and highspeed assembly, this thesis aims to showcase how machine learning techniques can be utilised to enhance vibration spectral analysis and SCADA analysis for early and more automated fault detection. First this will be performed separately based on features extracted from the vibration spectra and performance data in isolation before a framework will be presented to combine data sources to create a single anomaly detection model for early fault diagnosis. Additionally by further utilising vibration based analysis, machine learning techniques and a synchronised database of failures, remaining useful life prediction will also be explored for generator bearing faults, a key component when it comes to increasing wind turbine generator reliability. It will be shown that through early diagnosis and accurate prognosis, component replacements can be planned and optimised before catastrophic failures and large downtimes occur. Moreover, results also indicate that this can have a significant impact on the costs of operation and maintenance over the lifetime of an offshore development.
|Date of Award||29 Sep 2021|
- University Of Strathclyde
|Sponsors||EPSRC (Engineering and Physical Sciences Research Council)|
|Supervisor||James Carroll (Supervisor) & Alasdair McDonald (Supervisor)|