Effective algorithms for real-time wind turbine condition monitoring and fault-detection

Abstract

Reliable condition monitoring (CM) can be an effective means to significantly reduce wind turbine (WT) downtime, operations and maintenance costs and plan preventative maintenance in advance. The WT generator voltage and current output, if sampled at a sufficiently high rate (kHz range), can provide a rich source of data for CM. However, the electrical output of the WT generator is frequently shown to be complex and noisy in nature due to the varying and turbulent nature of the wind. Thus, a fully satisfactory technique that is capable to provide accurate interpretation of the WT electrical output has not been achieved to date. The objective of the research described in this thesis is to develop reliable WT CM using advanced signal processing techniques so that fast analysis of non-stationary current measurements with high diagnostic accuracy is achieved. The diagnostic accuracy and reliability of the proposed techniques have been evaluated using data from a laboratory test rig where experiments are performed under two levels of rotor electrical asymmetry faults. The experimental test rig was run under fixed and variable speed driving conditions to investigate the kind of results expected under such conditions. An effective extended Kalman filter (EKF) based method is proposed to iteratively track the characteristic fault frequencies in WT CM signals as the WT speed varies. The EKF performance was compared with some of the leading WT CM techniques to establish its pros and cons. The reported experimental findings demonstrate clear and significant gains in both the computational efficiency and the diagnostic accuracy using the proposed technique. In addition, a novel frequency tracking technique is proposed in this thesis to analyse the non-stationary current signals by improving the capability of a continuous wavelet transform (CWT). Simulations and experiments have been performed to verify the proposed method for detecting early abnormalities in WT generators. The improved CWT is finally applied for developing a new real-time CM technique dedicated to detect early abnormalities in a commercial WT. The results presented highlight the advantages of the improved CWT over the conventional CWT to identify frequency components of interest and cope with the non-linear and non-stationary fault features in the current signal, and go on to indicate its potential and suitability for WT CM.