A Hybrid Bearing Monitoring Model Integrating Cosine-Difference and Weightiness in Nearest Neighbors Algorithm

Abstract

This study presents a novel hybrid bearing condition monitoring model, CosWNN, which integrates cosine difference and weightiness within the k-nearest neighbors algorithm. The model addresses key challenges in vibration analysis, specifically the need for efficient computational resources and the scarcity of real-world faulty bearing data. By minimizing signal processing requirements and maintaining classification accuracy with limited data, CosWNN achieves an average accuracy of 77.1%, outperforming traditional nearest neighbors algorithms by 4.4% to 49.5%. Despite these advancements, the model's performance diminishes with fewer training samples, indicating the necessity for further optimization, including the adjustment of the quantity of nearest neighbors and the incorporation of data augmentation techniques. The study underscores the potential of CosWNN for robust bearing fault detection and its applicability in scenarios with constrained data and computational resources.