A Novel Spectral Sparse Classification Scheme with Applications to Intelligent Diagnostics of Wind Turbine Planetary Gearboxes

Intelligent diagnostics and prognostics can greatly promote the availability, durability, and reliability of wind turbines (WT), and meanwhile, save huge economic costs for the operation and maintenance of wind farms. As an essential transmission unit for energy conversions in wind turbines, the planetary gearbox is inevitably prone to unexpected damages because of the transient working loads and harsh operational environment. To tackle the challenge of planetary gearbox diagnostics, a novel spectral sparse classification (SSC) scheme is presented in this paper. The proposed SSC scheme achieves robust intelligent diagnostics via implementing three key procedures, namely, data augmentation, spectral dictionary design, and sparse classification. First, the vibrational data is augmented by exploiting the prediction translation-invariance of planetary gearbox signals to promote the sample volume and quality. Then, aiming at the strong dictionary reconstruction ability, the spectral features under various health states are incorporated to design the robust spectral dictionary. Finally, the sparse classification strategy with respect to the spectral dictionary is explored to accomplish robust recognition of planetary gearbox health states. The effectiveness and advantage of our developed SSC scheme for intelligent diagnostics have been experimentally validated with a planetary gearbox fault dataset. The comparison results with some state-of-the-art methods have also verified our SSC scheme with excellent classification accuracy and strong robustness to hyperparameters, thus showing a great prospect for our proposed SSC scheme with applications to WT planetary gearbox diagnostics.