Scholarlink

<b>Purpose:</b> <p>Accurate bearing fault diagnosis ensures reliable mechanical system operation. However, varying working conditions cause distributional shifts in bearing vibration signals. Recently, domain adaptation (DA)-based diagnostic models have garnered significant attention to solve this difficulty. Unfortunately, when a source domain (SD) contains more bearing categories than a target domain (TD), traditional DA (TDA) models may exhibit poor performance. Additionally, irrelevant bearing categories increase the risk of distributional misalignment in a partial domain adaptation (PDA) diagnostic setting.</p> <b>Method:</b> <p>A novel diagnostic model called Manifold Structure Smoothness Dynamic Alignment (MSSDA) for PDA bearing fault diagnosis is proposed. First, model smoothness is instantiated as a structural preservation within the TD. This promotes learning reliable domain-invariant knowledge for MSSDA. Then, MSSDA adaptively implements distribution alignment by quantitatively evaluating the relative contributions of marginal and conditional distributions. Finally, an adaptive diagnostic model is learned through these two steps with the principle of structural risk minimization.</p> <b>Result and Conclusion:</b> <p>Experimental results indicate that the average accuracy exceeds 99% for single-device diagnostic tasks and over 90% for cross-device tasks, outperforming other typical DA models in TDA and PDA settings. The diagnostic results validate that the MSSDA has superior cross-condition diagnostic ability.</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

Manifold Structure Smoothness Dynamic Alignment: A Novel Partial Domain Adaptation Model for Cross-Condition Bearing Fault Diagnosis

  • 期刊论文

  • 发表日期：

    2025年7月7日

查看全文
  • Hongdi Zhou,
  • Haitao He,
  • Bo Li,
  • Xixing Li,
  • Yijie Cai

摘要

Purpose:

Accurate bearing fault diagnosis ensures reliable mechanical system operation. However, varying working conditions cause distributional shifts in bearing vibration signals. Recently, domain adaptation (DA)-based diagnostic models have garnered significant attention to solve this difficulty. Unfortunately, when a source domain (SD) contains more bearing categories than a target domain (TD), traditional DA (TDA) models may exhibit poor performance. Additionally, irrelevant bearing categories increase the risk of distributional misalignment in a partial domain adaptation (PDA) diagnostic setting.

Method:

A novel diagnostic model called Manifold Structure Smoothness Dynamic Alignment (MSSDA) for PDA bearing fault diagnosis is proposed. First, model smoothness is instantiated as a structural preservation within the TD. This promotes learning reliable domain-invariant knowledge for MSSDA. Then, MSSDA adaptively implements distribution alignment by quantitatively evaluating the relative contributions of marginal and conditional distributions. Finally, an adaptive diagnostic model is learned through these two steps with the principle of structural risk minimization.

Result and Conclusion:

Experimental results indicate that the average accuracy exceeds 99% for single-device diagnostic tasks and over 90% for cross-device tasks, outperforming other typical DA models in TDA and PDA settings. The diagnostic results validate that the MSSDA has superior cross-condition diagnostic ability.