Hyperparameter Optimization on CNN Using Hyperband for Fault Identification in Wind Turbine High-Speed Shaft Gearbox Bearing

Given the adverse operating regime of the wind turbine gearbox, fault recognition models for components need to be optimized for reliable operation. However, selection of optimal hyperparameter is challenging, given the need to balance accuracy with computational efficiency. Therefore, this study proposed a hyperband-optimized convolutional neural network model for robust fault identification in high-speed shaft bearing (HSSB) of wind turbine gearbox. Vibration signals from the HSSB were employed to train a convolutional neural network (CNN) model optimized using Hyperband. Results of the study show that hyperparameter optimization with hyperband significantly improved the diagnostic power of the CNN classifier, with accuracy, sensitivity, and specificity increasing by 12.8%, 29.3%, and 15.9%, respectively. Whereas the standalone model scored below 90% on all performance metrics, for all signal-to-noise ratios (SNRs) of the noise-induced test data, the optimized model scored 100% for accuracy, sensitivity, and specificity for all SNRs above 8dB in the noise-induced test data. Results of the study demonstrate that optimizing the hyperparameters of a CNN model with Hyperband improves its performance and immunity to noisy vibration signals characteristic of the HSSB's operating regime.