Convolutional Self-attention Network-based Fault Diagnosis Method of Mechanical Equipment
Deep learning-based intelligent faults diagnosis methods for mechanical equipment rely on a large amount of la-beled samples.However,it is usually difficult to obtain enough samples in engineering,which makes it difficult for deep learning methods to extract fault features completely and seriously affects the generalization and robustness of fault diagnosis models.There-fore,the paper proposes a fault diagnosis method of mechanical equipment under small samples based on convolutional self-atten-tion network and prior knowledge.The designed convolutional self-attention network can automatically learn sample features and fuse the sample multidimensional features obtained from prior knowledge during the training process,with the aim of reducing the number of samples required for model training and improving the fault diagnosis accuracy of mechanical equipment in the case of small samples.Finally,the proposed method is validated with a hydraulic screw pump dataset.The experimental results show that the proposed method achieves 97.5%fault diagnosis accuracy under small samples,and the performance is better than the current common deep learning methods.
deep learningprior knowledgeself-attention mechanismsmall samplefault diagnosi
国家科技期刊平台
NSTL
万方数据
