Scholarlink

<p>In the industrial production process, the operating status of machinery changes frequently, rendering the deep model trained by the data under one working condition unable to achieve satisfying results on the data under another working condition. Aiming at the above difficulties, a domain adaptive transfer method based on 1D large-convolution DenseNet (AT-1DLD) is put forward, which can effectively implement cross-domain bearing fault diagnosis under variable working conditions. Initially, as the feature extractor, the well-designed 1D large-convolution DenseNet is employed to excavate the fault feature information from the raw vibration signal data. Then, the joint maximum mean discrepancy (JMMD) metric is taken to reduce the data distribution discrepancy between the source and target domains. Additionally, the conditional domain discriminator is applied to implement the domain adjustment from a global perspective. Ultimately, the network model can extract domain-invariant feature representation from the two-domain samples, achieving better results for fault diagnosis in the target task. To verify the effectiveness of the model, the proposed AT-1DLD method is compared with multiple methods on two rolling bearing datasets. The experimental results demonstrate the superiority and generalization of the AT-1DLD method for cross-domain fault diagnosis of bearings under variable working conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive cross-domain fault diagnosis method for rolling bearing based on 1D large-convolution DenseNet

  • 期刊论文

  • 发表日期：

    2025年5月25日

查看全文
  • Xiaoshan Lin,
  • Chenhui Qian,
  • Quansheng Jiang,
  • Yehu Shen,
  • Shilei Wu,
  • Qixin Zhu

摘要

In the industrial production process, the operating status of machinery changes frequently, rendering the deep model trained by the data under one working condition unable to achieve satisfying results on the data under another working condition. Aiming at the above difficulties, a domain adaptive transfer method based on 1D large-convolution DenseNet (AT-1DLD) is put forward, which can effectively implement cross-domain bearing fault diagnosis under variable working conditions. Initially, as the feature extractor, the well-designed 1D large-convolution DenseNet is employed to excavate the fault feature information from the raw vibration signal data. Then, the joint maximum mean discrepancy (JMMD) metric is taken to reduce the data distribution discrepancy between the source and target domains. Additionally, the conditional domain discriminator is applied to implement the domain adjustment from a global perspective. Ultimately, the network model can extract domain-invariant feature representation from the two-domain samples, achieving better results for fault diagnosis in the target task. To verify the effectiveness of the model, the proposed AT-1DLD method is compared with multiple methods on two rolling bearing datasets. The experimental results demonstrate the superiority and generalization of the AT-1DLD method for cross-domain fault diagnosis of bearings under variable working conditions.