Scholarlink

<p>As gearbox induced excitations can lead to acoustic abnormalities in wind turbines, one main target of the development process of gearboxes for wind turbines are low-excitation levels. To achieve this excitation-relevant characteristics such as stiffness variation and transmission error of the gears are optimized over large parameter search spaces. In this paper it is proven that Machine Learning (ML) methods can predict these features depending on the geometry and flank modifications of a&#xa0;gear stage’s gears. Regarding the need of data for application of ML methods, Datasets based on gear calculations are used for training purposes within this work. The underlying data structure enables the usage of efficient gradient-boosting methods to predict the gear meshing characteristics. Here, the high prediction performance is used to evaluate large parameter spaces. Outside the range of the geometric quantities covered by the data set, the validity of the developed models must be evaluated carefully.</p><p>Applying the developed methods to the optimization of the flank modifications of an existing gear stage shows the effectiveness of these techniques regarding the prediction of geometry influence on gear-meshing.</p><p>Subsequently, the process is validated by a&#xa0;test run of the gearbox whereby an optimized excitation-behavior of the gearbox can be expected, and results are presented here.</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

Machine learning methods for the design of excitation-optimized gears in wind-energy gearboxes

  • 期刊论文

  • 发表日期：

    2025年2月11日

查看全文
  • Frederik Stachowske,
  • Julia Kowalski,
  • Kai Lubenow

摘要

As gearbox induced excitations can lead to acoustic abnormalities in wind turbines, one main target of the development process of gearboxes for wind turbines are low-excitation levels. To achieve this excitation-relevant characteristics such as stiffness variation and transmission error of the gears are optimized over large parameter search spaces. In this paper it is proven that Machine Learning (ML) methods can predict these features depending on the geometry and flank modifications of a gear stage’s gears. Regarding the need of data for application of ML methods, Datasets based on gear calculations are used for training purposes within this work. The underlying data structure enables the usage of efficient gradient-boosting methods to predict the gear meshing characteristics. Here, the high prediction performance is used to evaluate large parameter spaces. Outside the range of the geometric quantities covered by the data set, the validity of the developed models must be evaluated carefully.

Applying the developed methods to the optimization of the flank modifications of an existing gear stage shows the effectiveness of these techniques regarding the prediction of geometry influence on gear-meshing.

Subsequently, the process is validated by a test run of the gearbox whereby an optimized excitation-behavior of the gearbox can be expected, and results are presented here.