Scholarlink

<p>This paper introduces a geometry-based method to model wind turbine states and predict faults using a deep convolutional neural network (DCNN). Initially, 3D-point cloud models are constructed in spaces defined by wind speed, power, and various auxiliary variables. Subsequently, the geometry feature of the point cloud in the 3D space is extracted, forming a 3D surface represented in the R, G, or B channels. For identification purposes, this 3D surface is transformed into a 2D image, with grayscale used to depict height. Additionally, the grayscale representing external environmental information sets the background of the 2D image. Consequently, the resulting stacked RGB image model encapsulates the necessary dynamic behavior and operating states of the wind turbine system. Finally, the DCNN, trained using the stacked image model through Xception-based transfer learning, identifies operating states and predicts faults. This paper’s method, which relies on the geometric distribution characteristics of sampling points rather than horizon characteristics, offers a novel perspective on wind turbine state identification and fault prediction. Experimental results demonstrate that the proposed method achieves exceptionally high accuracy in identifying the wind turbine’s operating state and predicts faults with up to <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(93.875\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>93.875</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> accuracy.</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

Geometry-based image modeling method for intelligent state identification and fault prediction of wind turbines

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • Jinman Luo,
  • Xiaoxia Li,
  • Yuqing Li,
  • Haiji Wang,
  • Pu Zhang

摘要

This paper introduces a geometry-based method to model wind turbine states and predict faults using a deep convolutional neural network (DCNN). Initially, 3D-point cloud models are constructed in spaces defined by wind speed, power, and various auxiliary variables. Subsequently, the geometry feature of the point cloud in the 3D space is extracted, forming a 3D surface represented in the R, G, or B channels. For identification purposes, this 3D surface is transformed into a 2D image, with grayscale used to depict height. Additionally, the grayscale representing external environmental information sets the background of the 2D image. Consequently, the resulting stacked RGB image model encapsulates the necessary dynamic behavior and operating states of the wind turbine system. Finally, the DCNN, trained using the stacked image model through Xception-based transfer learning, identifies operating states and predicts faults. This paper’s method, which relies on the geometric distribution characteristics of sampling points rather than horizon characteristics, offers a novel perspective on wind turbine state identification and fault prediction. Experimental results demonstrate that the proposed method achieves exceptionally high accuracy in identifying the wind turbine’s operating state and predicts faults with up to \(93.875\%\) 93.875 % accuracy.