With the fast growth in wind energy, the performance and reliability of the wind power generation system has become a major issue in order to achieve cost-effective generation. Integration of condition monitoring system (CMS) in the wind turbine has been considered as the most viable solution, which enhances maintenance scheduling and achieving a more reliable system. However, for an effective CMS, large number of sensors and high sampling frequency are required, resulting in a large amount of data to be generated. This has become a burden for the CMS and the fault detection system.
This thesis focuses on the development of variable selection algorithm, such that the dimensionality of the monitoring data can be reduced, while useful information in relation to the later fault diagnosis and prognosis is preserved.
The research started with a background and review of the current status of CMS in wind energy. Then, simulation of the wind turbine systems is carried out in order to generate useful monitoring data, including both healthy and faulty conditions.
Variable selection algorithms based on multivariate principal component analysis are proposed at the system level. The proposed method is then further extended by introducing additional criterion during the selection process, where the retained variables are targeted to a specific fault. Further analyses of the retained variables are carried out, and it has shown that fault features are present in the dataset with reduced dimensionality.
Two detection algorithms are then proposed utilising the datasets obtained from the selection algorithm. The algorithms allow accurate detection, identification and severity estimation of anomalies from simulation data and supervisory control and data acquisition data from an operational wind farm.
Finally an experimental wind turbine test rig is designed and constructed. Experimental monitoring data under healthy and faulty conditions is obtained to further validate the proposed detection algorithms.