Ball mill fault detection system based on autoencoder neural network

Ball mills play a critical role in modern industrial production, and issues with their bearings directly impact production efficiency and equipment maintenance costs. Traditional methods for bearing fault detection require using fault samples to learn the mechanism, but this can damage equipment and result in high costs. This paper proposes a bearing fault detection method based on autoencoder, which trains only on normal data to overcome the limitations of traditional approaches. The fault detector utilizes vibration signals from bearings to achieve real-time bearing state monitoring and fault early warning. Additionally, due to the complex composition of ball mills, multiple key components near the bearings need to be monitored. A single anomaly detector cannot locate or comprehensively analyze faults. Hence, this study proposes a ball mill fault detection system integrated with multiple anomaly detectors, capable of monitoring the bearing status in multiple parts of the ball mill, automatically localizing and warning of faults when they occur. Finally, Practical deployment at Wugang Resources Group validates that the autoencoder-based fault detector can identify bearing anomalies and demonstrates the ability of the integrated system for fault warning and localization.