Scholarlink

The long-term efficiency and performance of wind turbines are crucial for a transition to more sustainable energy production. However, these structures are subject to complex dynamics influenced by various environmental and operational factors, particularly aerodynamic erosion, which can detrimentally impact turbine blades. The predictive modelling of these dynamics and the accurate assessment of the resulting damage are vital for efficient and reliable structural maintenance. This study introduces a Supervised Variational Autoencoder (SVAE) framework to model the wind turbine dynamics affected by blade aerodynamic erosion, potentially leading to rotor imbalance. The SVAE learns a compressed representation of these dynamics, while simultaneously classifying various types and levels of wind turbine blades’ damage. In this work, we show that the SVAE is able to handle multi-label classification problems, as each erodible zone of the blade has its own damage level. Through an empirical evaluation, using realistic data, we demonstrate the effectiveness of our approach in representing erosion-induced dynamics. We are able to accurately classify damage severity over the lifetime of the turbine. This methodology provides a robust approach for understanding wind turbine dynamics and facilitating accurate damage classification, showcasing adaptability to real use cases. The validation of the accuracy of the SVAE was conducted using realistic synthetic data obtained by simulating wind turbines subjected to erosion-induced imbalance. The results demonstrate the model’s efficacy in capturing dynamic states influenced by erosion and providing detailed damage classification. The application of this framework extends beyond the immediate scope of this study. Its adaptability to different scenarios and data biases renders this a promising tool for the renewable energy sector. Enhanced predictive capabilities regarding turbine behavior and damage classification will aid in the development of proactive maintenance strategies, improving the operational efficiency and longevity of wind energy systems.

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

Using Supervised Variational Autoencoders to Detect Aerodynamic Erosion in Wind Turbines

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kiran Bacsa,
  • Imad Abdallah,
  • Wei Liu,
  • Xudong Jian,
  • Eleni Chatzi

摘要

The long-term efficiency and performance of wind turbines are crucial for a transition to more sustainable energy production. However, these structures are subject to complex dynamics influenced by various environmental and operational factors, particularly aerodynamic erosion, which can detrimentally impact turbine blades. The predictive modelling of these dynamics and the accurate assessment of the resulting damage are vital for efficient and reliable structural maintenance. This study introduces a Supervised Variational Autoencoder (SVAE) framework to model the wind turbine dynamics affected by blade aerodynamic erosion, potentially leading to rotor imbalance. The SVAE learns a compressed representation of these dynamics, while simultaneously classifying various types and levels of wind turbine blades’ damage. In this work, we show that the SVAE is able to handle multi-label classification problems, as each erodible zone of the blade has its own damage level. Through an empirical evaluation, using realistic data, we demonstrate the effectiveness of our approach in representing erosion-induced dynamics. We are able to accurately classify damage severity over the lifetime of the turbine. This methodology provides a robust approach for understanding wind turbine dynamics and facilitating accurate damage classification, showcasing adaptability to real use cases. The validation of the accuracy of the SVAE was conducted using realistic synthetic data obtained by simulating wind turbines subjected to erosion-induced imbalance. The results demonstrate the model’s efficacy in capturing dynamic states influenced by erosion and providing detailed damage classification. The application of this framework extends beyond the immediate scope of this study. Its adaptability to different scenarios and data biases renders this a promising tool for the renewable energy sector. Enhanced predictive capabilities regarding turbine behavior and damage classification will aid in the development of proactive maintenance strategies, improving the operational efficiency and longevity of wind energy systems.