Rotatory Machine Fault Diagnosis based on Improved One-Dimensional Residuals Convolutional Auto-encoding Network and Class Adaptation with Limited Samples


  • Received:2021-06-01 Revised:2021-08-31 Published:2021-09-17



  1. 1. Air Force Engineering University
    2. 空军工程大学
    3. 陕西省西安市空军工程大学防空反导学院发射系统教研部电力系统教研室
  • 通讯作者: 邵思羽

Abstract: Abstract: Aiming at the insufficient model training problem caused by limited rotating machinery sensor signal samples, a fault diagnosis framework that combines an improved residual auto-encoding network and a class-adaptive method was proposed to deal with the problem of limited training samples. Firstly, paired sampled were created by a small amount of labeled source domain data and unlabeled target domain data. An improved one-dimensional residual convolutional auto-encoding network was designed to extract features from two different distributions of original vibration signals. Secondly, the maximum mean difference was used to reduce the distribution difference, and the data space of the same fault category from different domains was mapped to a common feature space, which finally helped realize accurate fault diagnosis. The experimental verification is carried out through the experimental platform data set, and the results show that the proposed fault diagnosis framework is able to effectively improve the fault diagnosis accuracy of the micro-labeled target domain vibration data under different working conditions compared with the fine-tuning and domain adaptation methods.

Key words: One-dimensional Residual Convolutional Auto-encoding Network, Class Adaptation, Rotatory Machine Fault Diagnosis, Small Sample Learning, Maximize Mean Discrepancy

摘要: 针对旋转机械传感器信号样本有限影响深层网络模型训练学习的问题,本文提出一种结合改进残差自编码网络与类自适应方法的故障诊断模型应对小样本数据。首先将少量已标记的源域数据和微量未标记的目标域数据创建成对样本,并利用一维残差卷积自编码网络对两种不同分布的原始振动信号进行特征提取;其次,利用最大均值差异减小分布差异,将两个域同一故障类别的数据空间映射到一个共同的特征空间,最终实现准确的故障诊断。通过实验台数据集进行实验验证,结果表明,与微调、域自适应等方法相比,本文模型能够有效提高不同工况下微量标记的目标域振动数据的故障诊断准确率。

关键词: 一维残差卷积自编码网络, 类自适应, 旋转机械故障诊断, 小样本, 最大平均差异