| dc.description.abstract | Wind Energy’s presence continues to increase as the world puts focus into transitioning to cleaner energy sources. As with any energy generation method, maintaining the lowest LCOE is essential. Operations and Maintenance are a major cost drivers in Wind Energy and understanding the reliability characteristics of Wind Turbines is essential in predicting these costs. Reliability models are only as good as the data put into them, so failure data must be tracked in a consistent and deliberate fashion as the amount of failure data continues to increase with aging turbines. In this paper the information model surrounding Main Drive Train Component failures at Ørsted is mapped and evaluated using the Universal Modeling Language and then proposals are then made to improve it. On the collection and storage side, solutions are presented to automate and improve data collection, processing, and storage. On the application side the Competing Risk Reliability Model is implemented for all main components with sufficient data and its functionality is demonstrated through an analysis of the Siemens 4MW Gearbox. Work is also done relating operating conditions to reliability through linear regression, artificial neural network and random forest tree machine learning models. While the three models implemented have low overall predictive power they all suggest that low temperatures and turbulence intensity negatively affect gearbox reliability. Both the Competing Risk Model and sensitivity analysis are completely reliant on historical turbine data, so it is imperative that the information model surrounding main component failures accept, process, and store the data in a consistent and deliberate fashion. As more data becomes available with aging turbines, the framework outlined in this paper will allow Ørsted to optimally leverage its data in reliability predictions. | |