基于卷积自注意力网络的机械设备故障诊断方法

Convolutional Self-attention Network-based Fault Diagnosis Method of Mechanical Equipment

基于深度学习的机械设备故障智能诊断方法依赖于大量有标签的样本数据。然而,工程应用中通常难以获取足够的样本数据,这使得深度学习方法难以充分挖掘故障特征,严重影响故障诊断模型的泛化性和鲁棒性。基于此,论文提出一种基于卷积自注意力网络和先验知识的机械设备小样本故障诊断方法。所设计的卷积自注意力网络能够自动学习样本特征,并在训练过程中融合样本多维度特征先验知识,从而减少诊断模型训练所需的样本量,提高小样本情况下机械设备故障诊断精度。最后,以液压螺杆泵为对象,对论文所提出的方法进行实验验证。实验结果表明,所提出的方法在小样本情况下故障诊断精度可达97.5%,性能优于目前常用的深度学习方法。

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.