Abstract: The bearings of the wind turbine are subjected to alternating stresses and shock loads during operation, leading to nonlinear, non-stationary, and noisy vibration signals, thus rendering conventional feature extraction insufficient. Aiming at the inherent defects in the wind turbine bearings failure message processing and feature extraction for wind turbine bearing fault diagnosis, a novel method has been proposed on the basis of the optimal Variational Modal Decomposition combined with a random forest algorithm. Firstly, The technique utilizes the sooty tern optimization algorithm to conduct a search optimization of the values in the variables for the variational modal decomposition. Subsequently, the method with optimized parameters is employed to decompose the vibration signal of rolling bearing signals to obtain modal components. Finally, the peak value, kurtosis, and envelope entropy are applied to construct the fusion feature training set and input them into the random forest classifier to realize fault recognition. The results of the case analysis demonstrate the efficacy of the proposed methodology to identify faults in achieving a fault recognition accuracy of up to 100% for wind turbine bearing faults, facilitating accurate fault discrimination.

Key words: wind turbine, feature extraction, fault diagnosis, optimal variational modal decomposition, random forest algorithm