Intelligent fault diagnosis of belt conveyor rollers using a polar KNN algorithm with audio features

Abstract

Belt conveyor rollers are critical components in industrial applications, where early fault detection is essential to maintaining operational efficiency and safety. Existing fault diagnosis methods, such as vibration- and vision-based approaches, often face limitations due to high costs, sensor degradation, and environmental interferences, particularly in complex settings like mines. This study proposes an intelligent fault diagnosis method using a polar K-nearest neighbor (PKNN) algorithm combined with audio signal features. The PKNN algorithm enhances the classic KNN model by integrating both distance and angular similarities, allowing it to capture subtle variations in audio signals indicative of roller faults. The proposed PKNN model was tested on 17 different audio datasets, demonstrating robust performance with 97.34% accuracy, 96.89% precision, 96.72% recall, and 96.70% F1 score. Comparative analyses revealed that PKNN outperformed conventional machine learning models and other audio, vibration, and vision-based diagnostic methods, achieving superior fault classification accuracy and adaptability even in high-noise environments. These findings indicate that the PKNN model offers a reliable, non-invasive, and cost-effective solution for real-time monitoring and fault diagnosis of belt conveyor rollers. Its high adaptability to challenging industrial environments underscores its potential for wide-ranging applications in automated conveyor system maintenance.