Generalized Roughness Bearing Fault Diagnosis Using Time Series Analysis and Gradient Boosted Tree

Generalized roughness bearing faults are the most commonly reported ones in rotating machine failures. Early detection of bearing faults plays a vital role in minimizing operation losses. This paper aims to develop a real-time condition monitoring for bearing faults and for this purpose a sensory system to measure 3 axes vibration is developed. A time series model is proposed for vibration signals to extract features accurately. These features are classified using gradient boosted trees (GBT). GBT is an ensemble of trees and is optimal for parallel evaluation with relatively low cost. The proposed feature extraction algorithm also has a low computational cost. Overall, the proposed algorithm has very low latency and computational cost which enables fault diagnosis through embedded processors. This algorithm can diagnose five different faults with a promising accuracy of 98.5%, which is calculated for different load and speed conditions.