Scholarlink

<p>In order to solve the modeling errors inherent in finite element models of stator end windings for large-turbogenerator caused by many uncertainties such as simplification, idealized connection and material property uncertainty, a parametric correction method based on quadratic-polynomial response surface model and innovatively weighted-Kendall coefficient clustering was proposed. In this paper, the end winding of a 600 MW turbogenerator is taken as the research object, and the dynamic-response approximate model is constructed by fitting a small number of typical samples, and the correction efficiency is improved by avoiding multiple calls of finite element software while the accuracy is guaranteed. First, the natural frequency and modal assurance criteria (MAC) of the end winding were taken as the key indexes of dynamic response, and rod stiffness, radial support stiffness, binding stiffness, rod mass and radial support mass were selected as design variables. Samples were obtained through central composite design. The model corresponding to the experimental sample was established in ANSYS Workbench software, and its natural frequency and modal assurance criteria were calculated. The response surface method based on quadratic polynomial was used to construct the proxy model of the response, and the objective function and constraints of each design parameter were determined, and then particle swarm optimization was used to optimize each parameter. Secondly, in order to solve the coupling problem between the design parameters and the responses, the correlation coefficient index of weighted-Kendall coefficient based on the target-response error and the sensitivity of the design parameters is innovatively proposed. The hierarchical clustering algorithm is used to reasonably determine the number and position of the modified parameters to ensure that the modified parameters have appropriate decoupling ability. The average error of natural frequency and modal assurance criteria decreased from 5.62 % and 1.12 % to 2.26 % and 0.93 %.</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

Intelligent identification and location of model deviation for large generator simulation-sensitivity cluster analysis

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Meishu Xiong,
  • Ting Wang,
  • Wei Liao,
  • Yang Zhao,
  • Congying Deng,
  • Sheng Lu

摘要

In order to solve the modeling errors inherent in finite element models of stator end windings for large-turbogenerator caused by many uncertainties such as simplification, idealized connection and material property uncertainty, a parametric correction method based on quadratic-polynomial response surface model and innovatively weighted-Kendall coefficient clustering was proposed. In this paper, the end winding of a 600 MW turbogenerator is taken as the research object, and the dynamic-response approximate model is constructed by fitting a small number of typical samples, and the correction efficiency is improved by avoiding multiple calls of finite element software while the accuracy is guaranteed. First, the natural frequency and modal assurance criteria (MAC) of the end winding were taken as the key indexes of dynamic response, and rod stiffness, radial support stiffness, binding stiffness, rod mass and radial support mass were selected as design variables. Samples were obtained through central composite design. The model corresponding to the experimental sample was established in ANSYS Workbench software, and its natural frequency and modal assurance criteria were calculated. The response surface method based on quadratic polynomial was used to construct the proxy model of the response, and the objective function and constraints of each design parameter were determined, and then particle swarm optimization was used to optimize each parameter. Secondly, in order to solve the coupling problem between the design parameters and the responses, the correlation coefficient index of weighted-Kendall coefficient based on the target-response error and the sensitivity of the design parameters is innovatively proposed. The hierarchical clustering algorithm is used to reasonably determine the number and position of the modified parameters to ensure that the modified parameters have appropriate decoupling ability. The average error of natural frequency and modal assurance criteria decreased from 5.62 % and 1.12 % to 2.26 % and 0.93 %.