Scholarlink

<p>The imbalanced data problem seriously restricts the engineering application of Prognostics Health Management (PHM) technology. Aiming at the limitations of existing data augmentation models in modeling complex vibration signals, an Improved Stochastic Differential Equation (ISDE) diffusion model with a hierarchical cross modal generative decoder is proposed. Firstly, the SDE diffusion theory is introduced into the task of generating industrial vibration signals, and a global–local collaborative modeling strategy is designed by integrating the potential function and scoring network to construct the drift term of the diffusion model, precise control of distribution evolution is achieved; Secondly, an adaptive diffusion mechanism is proposed to improve the robustness of the model to complex industrial noise; Finally, a hierarchical cross modal decoder is constructed, which combines multi-layer decoding structure and cross modulation mechanism to achieve hierarchical reconstruction of multi-scale features. The experimental results show that the proposed ISDE diffusion model can effectively generate data, and significantly improve fault diagnosis accuracy on both publicly available single fault datasets and self-built composite fault datasets when imbalance ratio reach to 20:1. The research results provide reliable technical support for the health management of industrial equipment, which has important theoretical value and engineering significance.</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

ISDE diffusion model with hierarchical cross modal generative decoder: a data augmentation method for fault diagnosis under imbalanced data

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • Wenchang Zhu,
  • Lu Cai,
  • Qiuhua Miao,
  • Yudong Cao,
  • Peng Huang,
  • Hongwei Fan,
  • Xiaoli Zhao

摘要

The imbalanced data problem seriously restricts the engineering application of Prognostics Health Management (PHM) technology. Aiming at the limitations of existing data augmentation models in modeling complex vibration signals, an Improved Stochastic Differential Equation (ISDE) diffusion model with a hierarchical cross modal generative decoder is proposed. Firstly, the SDE diffusion theory is introduced into the task of generating industrial vibration signals, and a global–local collaborative modeling strategy is designed by integrating the potential function and scoring network to construct the drift term of the diffusion model, precise control of distribution evolution is achieved; Secondly, an adaptive diffusion mechanism is proposed to improve the robustness of the model to complex industrial noise; Finally, a hierarchical cross modal decoder is constructed, which combines multi-layer decoding structure and cross modulation mechanism to achieve hierarchical reconstruction of multi-scale features. The experimental results show that the proposed ISDE diffusion model can effectively generate data, and significantly improve fault diagnosis accuracy on both publicly available single fault datasets and self-built composite fault datasets when imbalance ratio reach to 20:1. The research results provide reliable technical support for the health management of industrial equipment, which has important theoretical value and engineering significance.