基于领域对抗网络的轴承故障诊断方法研究

Fault diagnosis method for bearings based on domain adversarial neural networks

在使用传统机器学习方法进行机械设备故障诊断过程中,因运行工况复杂多变无法满足测试数据和训练数据的同分布,导致模型诊断性能不高。针对这一问题,提出了一种基于领域对抗网络的设备变工况故障诊断方法。在卷积神经网络基础上,建立了包含特征提取器、故障分类器以及领域判别器的诊断模型,对测试与训练样本进行了分析处理,通过最小化故障分类器损失和最大化领域判别器损失,实现了对机械设备的故障诊断过程;通过在轴承试验台上进行了故障诊断模拟实验,将该方法诊断结果与其他故障诊断方法结果进行了对比,验证了该诊断模型对故障的识别能力。研究结果表明:该方法取得了96%以上的平均诊断准确率,在诊断过程中具有不受训练样本和测试样本差异影响的效果。

Aiming at the performance limitation of the mechanical equipment fault diagnosis,a method of fault diagnosis for equipment based on domain adversarial neural networks(DANN)was proposed.The accuracy of diagnosis was effected by the distribution of training and testing data due to the complex and changeable operating conditions when using traditional machine learning methods.Based on the convolutional neural networks,the diagnosis model was established consisting of feature extractor,fault classifier and domain discriminator.The diagnosis was conducted by minimizing the loss of fault classifier and maximizing the loss of domain discriminator.The fault identification ability was proved in the diagnosis experiments of bearing fault with the comparison of other methods.The results indicate that the proposed method has an average accuracy higher than 96%,and it is not affected by the differences between training and testing data.