Bearing Fault Classification using Temporal Features for Wind Turbine Application: Harnessing Neural and Non-Neural Techniques

Wind turbines play a vital role in contemporary electrical grids as sources of renewable energy. Their continuous integration is essential for ensuring sustainable power generation, making the enhancement of turbine performance and longevity a top priority. Diagnosing bearing faults is critical but challenging due to complex vibration signals (nonlinear and non-stationary nature). In an effort to enhance the accuracy of bearing fault diagnosis, this study focused on identifying statistical time features (STFs) from run-to-failure vibration signal datasets. The research aimed to classify normal and three bearing fault variants using the STFs with principal component analysis (PCA). A comparative analysis was further conducted between neural and non-neural classifiers, employing 9 STFS firstly, then 5 reduced STFs by PCA for model training. Both the NN and non-NN models yielded superior results with the 9 STFs when compared to the PCA-transformed 5 STFS. Among the highest performing models, the Medium NN was chosen from the NN family classifiers and Ensemble Bagged Tree outperformed in its category of non-NN family architectures.