Scholarlink

<p>Mechanical seals are widely used in critical marine equipment, and their seal faces are susceptible to wear degradation under particle-containing seawater conditions. Accurate wear-state identification is essential for fault diagnosis and preventive maintenance. This study proposes a diagnostic framework combining the two-dimensional Higuchi fractal dimension (TDHFD) with a human memory optimization-based kernel extreme learning machine (HMO-KELM). The proposed TDHFD extends the length–scale principle of conventional Higuchi fractal dimension from one-dimensional signals to two-dimensional wear images by constructing a three-dimensional gray-level surface, multi-scale directional chains, and local curvature weights. This enables more effective characterization of subtle wear textures, including scratches, plowing marks, and local gray-level transitions. The HMO algorithm is introduced to optimize the key parameters of KELM, improving classification stability and generalization. The framework is validated using both noise-contaminated texture images and mechanical seal wear images obtained from accelerated degradation tests. Results show that TDHFD provides stronger noise resistance and better feature separability than conventional fractal descriptors. Combined with HMO-KELM, the proposed method achieves a classification accuracy of 90%, demonstrating its potential for mechanical seal wear-state diagnosis.</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

Two-Dimensional Higuchi Fractal Dimension and Its Application in Mechanical Seal Fault Diagnosis

  • 期刊论文

  • 发表日期：

    2026年8月4日

查看全文
  • Jiajun Qin,
  • Ziyou Zhou,
  • Gongye Yu,
  • Rong Chen

摘要

Mechanical seals are widely used in critical marine equipment, and their seal faces are susceptible to wear degradation under particle-containing seawater conditions. Accurate wear-state identification is essential for fault diagnosis and preventive maintenance. This study proposes a diagnostic framework combining the two-dimensional Higuchi fractal dimension (TDHFD) with a human memory optimization-based kernel extreme learning machine (HMO-KELM). The proposed TDHFD extends the length–scale principle of conventional Higuchi fractal dimension from one-dimensional signals to two-dimensional wear images by constructing a three-dimensional gray-level surface, multi-scale directional chains, and local curvature weights. This enables more effective characterization of subtle wear textures, including scratches, plowing marks, and local gray-level transitions. The HMO algorithm is introduced to optimize the key parameters of KELM, improving classification stability and generalization. The framework is validated using both noise-contaminated texture images and mechanical seal wear images obtained from accelerated degradation tests. Results show that TDHFD provides stronger noise resistance and better feature separability than conventional fractal descriptors. Combined with HMO-KELM, the proposed method achieves a classification accuracy of 90%, demonstrating its potential for mechanical seal wear-state diagnosis.