552 / 2022-03-29 19:38:03

A Novel Graph Convolutional Network for High Voltage Circuit Breaker Mechanical Fault Diagnosis

fault diagnosis,high-voltage circuit breakers,graph convolutional network,dynamic adaptive K-nearest neighbor,small samples

In recent years, convolutional neural networks (CNNs) have achieved worth seeing results in mechanical fault diagnosis of high-voltage circuit breakers (HVCBs) due to their powerful classification capabilities. However, CNN not only relies on massive samples, but also it only utilizes the numerical information of vibration signals while ignoring node information, resulting in insufficient feature utilization and limited diagnostic accuracy. To address these issues, this paper proposes a novel graph convolutional network (GCN) for high-precision and robust fault diagnosis of HVCBs. First, this paper proposes a novel dynamic adaptive K-nearest neighbor method to convert vibration signals into graph signals. The dynamic adaptive K-nearest neighbor method is not only highly fault-tolerant to noisy signals, but also can prevent huge computational losses. Then, a multi-attention GCN is proposed to make full use of the nodes and numerical information representing the vibration signal of HVCBs to achieve adaptive classification. The introduction of multi-attention ensures that GCN pays attention to key node information, thereby extracting more discriminative features. The experimental results show that the diagnostic accuracy of the GCN proposed in this paper reaches 96.87%, and the diagnostic standard deviation is 0.27, which can achieve high-precision and robust diagnosis of mechanical faults of HVCBs, which has obvious advantages over traditional methods. In addition, the proposed GCN has strong fault tolerance for small samples, which provides a novel solution for the mechanical fault diagnosis of HVCBs.