Bearing fault diagnosis based on graph attention neural network

Rolling bearing vibration signals often exhibit overlapping features, lack of smoothness, and high susceptibility to noise interference, which can confuse fault-related information. To address this issue, a fault identification method for rolling bearings based on a Graph Attention Network (GAT) is proposed. Initially, segmented raw vibration signals are transformed into graph - structured data using the K-nearest neighbors (KNN) algorithm, in which each data point is connected to its K nearest neighbors. Subsequently, GAT is applied to extract features from the constructed graphs. The extracted features are then input into a classifier to perform fault classification. Experimental results show that the proposed method realizes superior classification accuracy and enhanced noise robustness.