多 源 特 征 节 点 的 图 神 经 网 络 轴 承 变 转 速 故 障 诊 断

Here,aiming at problems of signal non-stationarity,difficult feature extraction and difficult fault classification in bearings under variable rotating speed condition,a graph neural network model based on multi-source and multi-feature nodes was proposed. This method could use adaptive weighting algorithm combined with Gram angle difference field to optimize Gram diagram,and realize combination of vibration signals and rotating speed information. Then,Swin Transformer mechanism was improved to realize image feature extraction,construct a structural graph and input it into graph convolutional neural network model for fault diagnosis. Thereby,the correctness rate of bearing fault diagnosis was improved under variable rotating speed working condition. The experimental results showed that under variable rotating speed working condition,fault diagnosis correctness rates of convolutional neural network,long short-term memory network,Transformer and traditional graph neural network deep learning model are lower,but the proposed method can obtain 99. 9% correctness rate in Ottawa public dataset and 99% correctness rate in a self-testing dataset.