HEALTH MONITORING FOR WIND TURBINE PITCH BEARINGS THROUGH SIGNAL PROCESSING AND DATA ANALYSIS

Abstract

In recent decades, wind energy has attracted substantial global attention within the academic community worldwide. Wind turbines can harness the kinetic energy of the wind and converting it into electrical energy, thereby offering a potential contribution to the grid, although the intermittency of wind necessitates supplementary power sources or storage solutions. Within wind turbines, the pitch bearing helps support the blade, allowing the blade to rotate to the appropriate wind receiving angle, thereby enabling the entire wind turbine to achieve a large power output. Failure of pitch bearings can result in inaccurate turbine control or even complete breakdown, resulting in a reduction in power production. The costs associated with assembling and repairing pitch bearings are substantial. Hence, the significance of health monitoring (HM) for wind turbine pitch bearings cannot be overstated. However, research on HM for wind turbine pitch bearings is limited. This is because it has around 10% failure rate in 20 years and initial studies may have predominantly centered on the overll performance of wind turbines, such as control systems and power generation efficacy. Consequently, components such as pitch bearings, as part of the system, might have been overlooked in certain research endeavors. Regarding the implementation of HM for wind turbine pitch bearings, several notable challenges are accompanied: 1.) low signal-to-noise ratio (SNR) due to significant background noise and the not full revolution with slow rotation speeds. 2.) limited fault features, i.e. subtle and intermittent, due to the dynamic nature of wind turbine systems (For example, existence of overturning moments may result in ball bearings transition from a four-point contact to a two-point contact with the inner and outer raceways.) 3.) significant disturbance factors due to operating under varying environmental conditions (e.g. blade flapping, dynamic wind loads, and mechanical vibrations) To address the challenges associated with HM for wind turbine pitch bearings, we foucs on three primary HM tasks, including fault detection, fault diagnosis, and spall size estimation. Firstly, a fault detection method was devised to detect early faults that may be difficult to diagnose using direct fault diagnosis methods. This method provides an additional layer of early fault detection to improve the reliability of the HM system. Secondly, we proposed two fault diagnosis methods, one based on statistical learning and the other on deep learning techniques. These methods aim to accurately identify and classify different fault types in the pitch bearings. Lastly, a spall size estimation method was proposed to estimate the extent of damage to the pitch bearings. This information is crucial for determining the remaining useful life (RUL) and scheduling appropriate maintenance activities. To validate the effectiveness of the proposed methods, experiments were conducted on a test-rig. The test-rig consisted of a naturally damaged wind turbine pitch bearing that had been in operation on a wind farm for over 15 years. The validation process confirmed the reliability and accuracy of the proposed HM techniques in real-world operating conditions.

Date of Award	31 Dec 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Wuliang Yin (Supervisor) & Long Zhang (Supervisor)