Abstract:Recently, data-driven methods, especially deep learning, outperform other methods for rolling element bearing (REB) fault diagnosis. Nevertheless, most research work assumes that REB dataset is unbiased. In the real industry applications, the dataset bias exists with REB owing to varying REB working conditions and noise interference. Recently proposed adversarial discriminative domain adaptation (ADDA) is an increasingly popular incarnation to solve dataset bias problem. However, it mainly devotes to realizing domain alignments, and ignores class-level alignments; it can cause degradation of classification performance. In this study, we propose a new REB fault diagnosis model based on improved ADDA to address dataset bias. The proposed diagnosis model realizes domain- and class-level alignments in dataset bias scenario; it consists of two feature extractors, a domain discriminator, and two label classifiers. The feature extractors and domain discriminator are trained in an adversarial manner to minimize the domain difference in feature extractors. The domain discrepancy in label classifier is reduced by minimizing correlation alignment (CORAL) loss. We evaluate the proposed model on the Case Western Reserve University (CWRU) bearing dataset and Paderborn University bearing dataset. The proposed method yields better results than other methods and has good prospects for industrial applications.

Key words:rolling element bearing (REB), dataset bias, adversarial discriminative domain adaptation (ADDA), correlation alignment (CORAL) loss

摘要：近年来，基于数据驱动的方法，特别是深度学习方法在滚动轴承故障诊断方面的表现优于其他方法。然而，大多数研究工作都假设滚动轴承故障数据集是无偏差的。在实际工业应用中，由于工作条件的变化和外部噪声的干扰，导致滚动轴承故障数据集存在偏差。对抗判别域自适应方法是一种解决数据集偏差问题的流行方法。然而，该方法主要实现领域对齐，而忽略了类级对齐，会导致分类性能的下降。本研究提出了一种基于改进的对抗判别域自适应的滚动轴承故障诊断模型，该模型实现了数据集偏差的域级和类级对齐。由两个特征提取器、一个域判别器和两个标签分类器组成。以对抗的方式训练特征提取器和域判别器，最小化特征提取器的域差异，通过最小化相关对齐损失来减少标签分类器中的域差异。在凯斯西储大学滚动轴承数据集和帕德博恩大学滚动轴承数据集上对所提模型进行了模型性能评估。实验结果表明所提模型性能优于其他方法，具有良好的工业应用前景。

关键词:滚动轴承，数据集偏差，对抗判别域自适应，相关对齐损失