New Method Uses Graph Networks and Ensemble Learning to Improve Bearing Fault Detection
• Proposes a new bearing fault detection method called BFDGE using graph neural networks and ensemble learning to address challenges in identifying anomalies mixed with normal data.
• Converts vibration signals into graph data to establish connections between signals, then uses a feature aggregation module and integrated learning strategy for detection.
• Achieves superior performance over other methods in detecting faulty samples mixed with normal ones, owing to the graph neural network's ability to aggregate features and ensemble learning enhancing robustness.
• Validated on public bearing fault datasets from Case Western Reserve University and Xi’an Jiaotong University, demonstrating higher AUC, accuracy, detection rate and lower false alarm rate.
• Involves various key parameters like number of base detectors, node neighbors and hidden layers which impact detection performance; experiments indicate optimal parameter values.