Scholarlink

<p>As a key component of mechanical equipment, the fault identification of rolling bearings is affected by the high dimensionality and complex redundancy issues brought by multi-source information. Therefore, a sparse deep non-negative matrix factorization (SDNMF) method based on manifold structure preservation is proposed, the L<sub>2</sub>, p norm (0 &lt; p ≤ 1) is introduced to enhance the resistance to noise and outliers, and a global regularization term is adopted to preserve the global characteristics of data. The k-nearest neighbor algorithm is used to construct a similarity matrix, ensuring that the local structure of the data remains unchanged during dimensionality reduction, and improving the quality of feature representation. In addition, sparse regularization is used to reduce model complexity. The experimental results demonstrate that compared with traditional methods, the classification accuracy has been improved about 15 %, and the recognition rate can remain above 90 % with high noise.</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

Bearing fault diagnosis using sparse deep non-negative matrix factorization with manifold structure preserving

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Hongdi Zhou,
  • Chenyu Huai,
  • Qi Tao,
  • Zhaoguang Zhang

摘要

As a key component of mechanical equipment, the fault identification of rolling bearings is affected by the high dimensionality and complex redundancy issues brought by multi-source information. Therefore, a sparse deep non-negative matrix factorization (SDNMF) method based on manifold structure preservation is proposed, the L2, p norm (0 < p ≤ 1) is introduced to enhance the resistance to noise and outliers, and a global regularization term is adopted to preserve the global characteristics of data. The k-nearest neighbor algorithm is used to construct a similarity matrix, ensuring that the local structure of the data remains unchanged during dimensionality reduction, and improving the quality of feature representation. In addition, sparse regularization is used to reduce model complexity. The experimental results demonstrate that compared with traditional methods, the classification accuracy has been improved about 15 %, and the recognition rate can remain above 90 % with high noise.