Scholarlink

<p>The bearings of reactor coolant pumps play a vital role in ensuring the smooth operation of the spindle, and the oil film thickness is an important indicator reflecting the operating condition of the bearing. However, the internal environment of the reactor coolant pump is characterized by high temperature, high pressure, and strong corrosion, making it difficult to install high-precision sensors to monitor the oil film thickness of the bearings in real time. To address this issue, a noise-adaptive soft-sensing framework is proposed, introducing a robust stochastic configuration network with a mixture of Gaussian (RSCN-MoG) to overcome the limitations of conventional soft sensors. The RSCN-MoG model employs an adjusted objective function combined with the expectation–maximization (EM) algorithm to adapt to any continuously distributed noise, representing a novel mechanism for handling mixed Gaussian and non-Gaussian disturbances in industrial measurement. Comprehensive experiments on benchmark datasets and a nuclear pump testbed demonstrate that the proposed RSCN-MoG model achieves a root mean squared error (RMSE) improvement of approximately 7.443% over the other models on benchmark datasets, and an average improvement of 13.266% across various noise conditions in the nuclear pump testbed. Consequently, the RSCN-MoG model exhibits strong robustness and generalization performance, achieving accurate and noise-resilient estimation of oil film thickness in reactor coolant pump bearings.</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

Reactor coolant pump bearing oil film thickness modeling based on robust stochastic configuration network with mixture of Gaussian

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Wei Wang,
  • Shujiang Li,
  • Wei Fu

摘要

The bearings of reactor coolant pumps play a vital role in ensuring the smooth operation of the spindle, and the oil film thickness is an important indicator reflecting the operating condition of the bearing. However, the internal environment of the reactor coolant pump is characterized by high temperature, high pressure, and strong corrosion, making it difficult to install high-precision sensors to monitor the oil film thickness of the bearings in real time. To address this issue, a noise-adaptive soft-sensing framework is proposed, introducing a robust stochastic configuration network with a mixture of Gaussian (RSCN-MoG) to overcome the limitations of conventional soft sensors. The RSCN-MoG model employs an adjusted objective function combined with the expectation–maximization (EM) algorithm to adapt to any continuously distributed noise, representing a novel mechanism for handling mixed Gaussian and non-Gaussian disturbances in industrial measurement. Comprehensive experiments on benchmark datasets and a nuclear pump testbed demonstrate that the proposed RSCN-MoG model achieves a root mean squared error (RMSE) improvement of approximately 7.443% over the other models on benchmark datasets, and an average improvement of 13.266% across various noise conditions in the nuclear pump testbed. Consequently, the RSCN-MoG model exhibits strong robustness and generalization performance, achieving accurate and noise-resilient estimation of oil film thickness in reactor coolant pump bearings.