Monitoring wind turbine condition by the approach of Empirical Mode Decomposition

An efficient condition monitoring system is indispensable to a large offshore wind turbine (WT) as it suffers higher reliability risk being exposed to extreme running environment and subject to constantly variable loadings, however difficult to access for fault repair. Today, the majority condition monitoring techniques for WT are borrowed from other industry fields where they achieve success. However, to date these techniques have not proved entirely satisfactory in wind industry. The reasons are various. But one of the main reasons is lack of a proper approach to the accurate analysis of WT signals, which are non-stationary in both time and frequency. The inaccurate analysis of WT signals results in frequent spurious alarms, which cause unnecessary shut down of machines and seriously disturb the normal production of wind farms. Aim at improving this situation, a new technique is developed in this work through analyzing the total power signals measured from the terminals of the WT generator by using the approach of empirical mode decomposition (EMD). In comparison with those conventional Fourier transform-based techniques that are being popularly used today in wind industry, the EMD is more ideal for processing the non-stationary, nonlinear WT signals attribute to its intrinsic locally adaptive property. Additionally, the computational algorithm of the EMD is more efficient than that of previous wavelet analysis, which enables the EMD more suitable for use in online condition monitoring systems. The proposed approach has been experimentally validated on a deliberately designed WT test rig with a 3-phase induction generator. It has been proved that the proposed strategy is valid for detecting both drive train mechanical and generator electrical faults occurring in all types of WTs whether geared or direct-drive.