【摘要】 近年来,基于深度学习的故障诊断方法取得了显著的成就。然而,传统的深度学习故障诊断方法都基于训练数据与测试数据来自相同的概率分布这一假设。在实际工业应用中,设备工作在复杂多变的工况下,难以保证训练数据与测试数据属于相同分布。当训练数据与测试数据属于不同工况时,训练后的模型在测试集中的准确率会明显下降。为了解决这个问题,提出了一种改进的深度迁移学习算法——联合领域自适应算法。通过最小化源域和目标域的边缘分布差异与条件边缘分布差异,联合领域自适应算法可以获得更强的知识迁移能力。将联合领域自适应算法与卷积神经网络(CNN)结合,提出了一种端到端的多工况故障诊断模型。最后,使用西储大学(Case Western Reserve University)的滚动轴承故障数据集进行实验。实验结果显示模型在多工况故障诊断中的表现优于传统的深度学习、数据驱动与迁移学习算法。更多还原

【Abstract】 In recent years,studies on deep learning for fault diagnosis have achieved remarkable results.However,they all share the same assumption that training data and testing data are from identical distribution.In the real world,this assumption is not applicable when working condition changes frequently.Classification accuracy will decline sharply when training data and testing data are from different distributions.In order to solve this problem,a novel transfer learning algorithm,named joint domain adaptation,is proposed in this paper.Joint domain adaptation minimizes the distance between marginal distributions as well as the distance between conditional distributions for stronger knowledge transfer.Combing joint domain adaptation and convolutional neutral network,an end-to-end fault diagnosis model for multi-working condition is established.The proposed model is tested on the roller bearing fault dataset from Case Western Reserve University.The result shows that the proposed model outperforms traditional data-driven method and traditional deep transfer learning method in the experiment.更多还原