Wind turbine fault detection using multiwavelet denoising with the data-driven block threshold

Rapid expansion of wind turbines has drawn attention to reduce the operation and maintenance costs. Continuous condition monitoring of wind turbines allows for early detection of the generator faults, facilitating a proactive response, minimizing downtime and maximizing productivity. However, the weak features of incipient faults in wind turbines are always immersed in noises of the equipment and the environment. Wavelet denoising is a useful tool for incipient fault detection and its effect mainly depends on the feature separation and the noise elimination. Multiwavelets have two or more multiscaling functions and multiwavelet functions. They possess the properties of orthogonality, symmetry, compact support and high vanishing moments simultaneously. The data-driven block threshold selected the optimal block length and threshold at different decomposition levels by using the minimum Stein's unbiased risk estimate. A multiwavelet denoising technique with the data-driven block threshold was proposed in this paper. The simulation experiment and the feature detection of a rolling bearing with a slight inner race defect indicated that the proposed method successfully detected the weak features of incipient faults.