A Resource-Efficient Deep Neural Framework For Bearing Health Monitoring Using Current Sensor Analytics

Fault diagnosis of rolling bearings is a critical task in industrial motor systems, as bearing defects can lead to severe mechanical failures and costly downtime. Traditional vibration-based monitoring systems require additional hardware and complex signal processing techniques, making them expensive and difficult to deploy in practical environments. In contrast, current sensor-based fault diagnosis offers a more economical and convenient alternative, as motor current signals can be collected without installing extra sensors.This project presents LiteFDNet, a lightweight deep learning framework designed for efficient and accurate bearing fault diagnosis using motor current signals. Instead of directly processing high-dimensional raw signals, the proposed approach extracts meaningful time-domain statistical features to reduce computational complexity. A compact neural network architecture with residual and dense connections is implemented to enhance feature representation while maintaining low model complexity. Additionally, explainable feature selection techniques are applied to identify the most informative features contributing to fault classification.Experimental results demonstrate that LiteFDNet achieves high diagnostic accuracy while significantly reducing computational cost and inference time compared to conventional deep learning models. The proposed system is suitable for real-time industrial applications, particularly in resource-constrained edge computing environments