Data-driven Wind Turbine Condition Monitoring

Abstract

Compared with traditional power plants, wind turbines are typically distributed over a broad remote region as well as are exposed to harsh and variable weather conditions. Thus, frequent unexpected failures often occur on wind turbines. Advanced condition monitoring systems for automatically and contentiously examining wind turbine conditions as well as issuing early alarms of impending failures are highly desired in the wind energy industry. Although condition monitoring approaches based on installing various additional sensors were introduced in the literature, the majority of reported methods are instructive and often require extra capital investment.

The wide deployment of supervisory control and data acquisition (SCADA) systems in wind farms have allowed the collection of large volumes of wind turbine operational parameters. In addition, recent applications of unmanned aerial vehicles (UAVs) introduce an emerging opportunity to inspect surface conditions of wind turbines remotely. In this dissertation, two types of data, SCADA data and UVA-taken images, are employed to develop data-driven models for realizing the wind turbine condition monitoring. To build data-driven models particularly fitting into considered problems, both existing supervised and unsupervised learning algorithms are extended and applied. Meanwhile, statistical process control charts are utilized to derive boundaries for detecting wind turbine anomalies.

Two research directions of wind turbine condition monitoring are studied, monitoring wind turbine power generation performance and monitoring conditions of wind turbine major sub-systems, such as, gearboxes and rotor blades. In the wind turbine power generation performance monitoring, a novel data-driven framework is developed to generate power curve profiles and recognize abnormal ones automatically. The condition monitoring of major sub-systems of wind turbines includes the identification of impending gearbox and blade failures and the automatic detection of wind turbine blade surface cracks.

The effectiveness of the approaches presented in this dissertation has been validated with real cases collected from the industry.

Date of Award	30 Aug 2017
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Zijun ZHANG (Supervisor)