Fault diagnosis of rotating machinery is essential for maintaining system performance and ensuring the operation safety. Deep learning (DL) has been recently developed rapidly and achieved remarkable results in fault diagnosis. However, the temporal information from time-series signals is ignored by convolutional neural networks (CNNs) based methods. Besides, the robustness against the noise is essential to methods for fault diagnosis. Therefore, a novel method based on recurrent neural networks (RNNs) is proposed to identify fault types in rotating machinery in this paper. One-dimensional time-series vibration signals are first converted into two-dimensional images. Then, Gated Recurrent Unit (GRU) is introduced to exploit temporal information of time-series data and learn representative features from constructed images. A multilayer perceptron (MLP) is finally employed to implement fault recognition. Experimental results show that the proposed method achieves the best performance on two public datasets compared with existing work and exhibits the robustness against the noise.

旋转机械的故障诊断对于维持系统性能和确保操作安全至关重要。深度学习（DL）最近得到了迅速发展，并在故障诊断中取得了显著成果。但是，基于卷积神经网络（CNN）的方法会忽略来自时序信号的时间信息。此外，抗噪声的鲁棒性对于故障诊断方法至关重要。因此，本文提出了一种基于递归神经网络的新方法来识别旋转机械的故障类型。首先将一维时间序列振动信号转换为二维图像。然后，引入门控循环单元（GRU）以利用时序数据的时间信息并从构造的图像中学习代表性特征。最后，采用多层感知器（MLP）来实现故障识别。实验结果表明，与现有工作相比，该方法在两个公共数据集上均具有最佳性能，并且具有抗噪声的鲁棒性。