Identifying defects in wind turbine high-speed shaft bearings through integrated signal processing and AI-driven diagnostic methods

Researchers in the wind turbine sector have focused on condition monitoring due to challenging operating environments and high maintenance costs. This work suggests a hybrid signal processing and data-based strategy to construct a distinct integrated defect detection method for high-speed shaft bearing (HSSB) of wind turbines. The aim of condition monitoring is to increase the system reliability of a wind turbine. The study’s feature extraction techniques are grounded in the fundamentals of signal processing: time-domain metrics capture amplitude fluctuations caused by defect initiation; frequency-domain features utilize their ability to detect spectral changes indicative of damage; and wavelet transformations are employed for their capacity to localize transient occurrences in non-stationary signals. The extracted features are sorted using the proposed feature selection algorithm, designed to prioritize discriminative features before being input into a feedforward neural network model to predict the operational condition of bearings. The new fault diagnostic strategy is evaluated using bearing run-to-failure test data collected in the field. Results show that the proposed methodology outperforms traditional data-driven methods, achieving a sensitivity of 99.01% and a specificity of 98.99%. Using the Case Western Reserve University Dataset, the algorithm attained a sensitivity of 100%.