To meet the latest strike price, the cost of energy from wind turbines needs to decrease.One of the biggest cost contributors to wind energy is the operation and maintenance cost. If this cost is driven down, the cost of energy of wind will substantially decrease and the reliability of the wind turbine assets need to increase. For that reason, condition monitoring systems are installed in modern wind turbines. These systems collect data and in abnormal conditions trigger alarms that are an indication of a fault. Maintenance actions can be scheduled accordingly that way, and faulty components can be replaced before catastrophic failures and large downtimes occur.Therefore, the aim of this thesis is to utilise vibration and performance data collected from wind turbine gearboxes, in order to perform fault detection and diagnosis.The data is collected at various times prior to gearbox component failures and advanced signal processing techniques are applied to reveal fault signatures. Machine learning models are trained based on features extracted from vibration spectra and operational data separately, but also a combination of these two types of data is investigated. The output is fault detection and isolation on gearbox component level. Both unit-specific and fleet-based methods are examines. The models are trained on specific turbines,but the generalization to other turbines is also examined.The above will provide a exible but robust framework for the early detection of emerging wind turbine faults. This will lead to minimisation of the wind turbine downtime and increase of the wind turbines reliability and income through operational enhancement.
|Date of Award||13 Dec 2019|
- University Of Strathclyde
|Sponsors||University of Strathclyde|
|Supervisor||James Carroll (Supervisor) & Alasdair McDonald (Supervisor)|