深度对抗迁移学习的故障诊断方法研究

Research on Fault Diagnosis Method of Deep Adversarial Transfer Learning

针对实验室环境下容易获取大量有标签故障类型数据,而在实际工况条件下很难或无法获取大量带标签数据的问题,提出机械设备故障的深度对抗迁移诊断方法(MAAN)。该方法将实验室环境中积累的故障诊断知识迁移应用于工程实际装备,通过融合时域与频域数据获取更全面的故障信息,在特征提取层利用残差网络深度提取故障特征,对抗层采用最大化域分类损失用于对齐源域与目标域的边缘分布和条件概率分布,最小化类别预测损失用于机械设备的故障分类实现无监督迁移学习。实验结果表明,此模型在无标签的目标数据集中有较高的分类精度,在一定条件下可以有效解决数据集缺少标签的难题,即实现机械故障诊断的智能诊断。

Aiming at the problem that it is easy to obtain a large amount of labeled fault type data in a laboratory environment, but it is difficult or impossible to obtain a large amount of labeled data under actual working conditions, a deep anti-migration diagnosis method(MAAN) for mechanical equipment failure is proposed. This method transfers the fault diagnosis knowledge accumulated in the laboratory environment to the actual engineering equipment, through the fusion of time domain and frequency domain data to obtain more comprehensive fault information;in the feature extraction layer, the residual network is used to deeply extract fault features. The adversarial layer uses maximizing domain classification loss to align the marginal distribution and conditional probability distribution between the source domain and target domain, and minimizing the category prediction loss to implement unsupervised migration learning for fault classification of mechanical equipment. Experimental results show that this method and its network models have high classification accuracy in unlabeled target data sets. Under certain conditions, this method can effectively solve the problem of lack of labels in data sets, that is, realize intelligent diagnosis of mechanical fault diagnosis.