Condition Monitoring of Wind Turbine Gearboxes Using Vibration-Based Methods

From 2003 to 2012, the aggregated downtime of gearboxes is the longest for wind turbines. The maintenance costs of gearboxes are the highest in all wind turbines. This thesis is motivated by the increasing demand for new repair methodologies in wind turbine gearboxes and the need for the monitoring of these expensive equipment. The aim is to distinguish the damaged state data from the undamaged state data. There are several steps to monitor the condition of the gearboxes, including operational evaluation, data acquisition, feature extraction and feature discrimination. The operational evaluation is to justify the performance of condition monitoring and define the damage. The data acquisition is to select the excitation methods, the sensor types, number and locations, and the hardware of acquisition. In this paper, the Welch's method and the principal component analysis (PCA) method are adopted to extract the features. The outliner analysis, multi-layer perceptron (MLP) and radial basis function (RBF) are applied to discriminate the data respectively to compare with their advantages and disadvantages to others. The results show that the data are successfully processed to extract the features and classify whether the data are damaged or not by these methods. However, each method has its own drawbacks.