FreqMGCN-Net: An IoT-Integrated Multi-Parallel Graph Convolutional Network with frequency attention for motor bearing fault diagnosis

Abstract

Motor bearing fault diagnosis is crucial for predictive maintenance in industrial systems. Traditional methods struggle with complex vibration signals, especially under noisy conditions and with limited labeled data. The advent of Internet of Things (IoT) technology enables real-time data collection through IoT-enabled sensor networks, which enhances fault detection accuracy. This paper proposes a novel fault diagnosis method, FreqMGCN-Net, which integrates IoT-enabled sensor networks for continuous data acquisition. The method utilizes a frequency attention mechanism-enhanced CNN to extract key frequency-domain features from raw vibration signals, improving model accuracy and robustness. Additionally, the model integrates Multi-parallel Graph Convolutional Networks (MGCN) with Semi-supervised Meta-learning, enabling it to handle multi-sensor data from IoT devices and address challenges like few-shot learning and cross-domain generalization. Experimental results show that FreqMGCN-Net achieves 95.7% accuracy on the CWRU dataset and 93.4% on the Paderborn dataset, outperforming existing models in accuracy, F1-Score, and recall. These results demonstrate the effectiveness of FreqMGCN-Net for real-time motor bearing fault diagnosis in IoT-enabled industrial environments, particularly in noisy and data-scarce conditions.

Keywords

• Motor bearing fault diagnosis
• Frequency attention mechanism
• Multi-parallel graph convolutional networks
• Semi-supervised meta-learning
• Few-shot learning
• IOT