A Bio-Inspired Framework for Hyperparameter Optimization of an XGBoost Model in Main Bearing Fault Diagnostics

Suboptimal hyperparameter settings for an Extreme Gradient Boosting Tree (XGBoost) leads to an inferior model that undermines the effectiveness of condition monitoring systems. This study proposed an Artificial Bee Colony (ABC) for hyperparameter optimization of an XGBoost main bearing fault diagnostic model. We selected the most influential hyperparameters of the XGBoost for fine tuning through a novel hyperparameter scoring technique. A stochastic exploration strategy based on the randomized search method identifies promising regions. We then localized hyperparameter optimization using the ABC to these areas. The ABC-optimized XGBoost had accuracy, precision, and recall of 94.7, 95.1, and 94.7%, respectively, while also outperforming its standalone counterparts employed in comparable studies. The ABC- optimized XGBoost is a valuable resource for accurate main- bearing health state classification. The insights from this study does not only advance intelligent condition monitoring for wind turbine main bearings but also offer valuable strategies for optimizing extreme gradient boosting trees in large search spaces with constrained computational resources.