A novel method to predict the stiffness evolution of in-service wind turbine blades based on deep learning models
A novel method to predict the stiffness evolution of in-service wind turbine blades based on deep learning models
复制标题
DOI：
10.1016/j.compstruct.2020.112702
复制
发表时间：
2020-11-15
影响因子：
6.3
通讯作者：
Leng, Jinsong
中科院分区：
文献类型：
--
作者：
Liu, Hongwei;Zhang, Zhichun;Leng, Jinsong
由于风力涡轮机长期在复杂的环境中运行，叶片的疲劳性能会受到风、光照、湿度、温度等因素的影响。对于风力发电机叶片制造商来说，在交货前确定其疲劳极限是必要的，疲劳加速实验通常需要大量的人工和实验费用。作为一种机器学习范式，深度学习关注的是数据固有的层次模型，在计算机视觉、语音识别、自然语言处理等方面取得了显著的成功。为了减少疲劳试验的时间和成本，研究了一种基于训练的风力机叶片疲劳时间序列刚度预测方法。基于卷积神经网络、长短期记忆网络和混合网络等深度学习方法，结合疲劳历史数据，获得了随疲劳寿命变化的叶片疲劳试验剩余刚度。结果表明，所建立的模型可以直接从原始刚度数据中学习特征，并依次完成剩余刚度预测。在所有刚度数据中加入不同信噪比的高斯白噪声，验证了模型对刚度预测的可行性。
Since wind turbines operate in a complex environment for long term, the fatigue behavior of the blades can be influenced by wind, illumination, moisture, temperature, and so forth. For wind turbine blade manufacturers, the determination of their fatigue limit before delivery is necessary and fatigue acceleration experiments usually require a lot of labor and experimental costs. As a machine learning paradigm, deep learning focuses on the inherent hierarchical models of data and has achieved notable success in computer vision, speech recognition, natural language processing, etc. Aimed at reducing the time and the costs during fatigue tests, this paper studies a training-based method for wind turbine blade stiffness prediction using time series stiffness data under fatigue tests. Based on deep learning methods including convolutional neural network, long-short term memory network and the hybrid network, the residual stiffness of the blade with fatigue life under fatigue tests is obtained by combining the fatigue historical data. The obtained results show that the developed models can learn features directly from raw stiffness data and complete the residual stiffness prediction in succession. White Gaussian noise with different signal-to-noise ratios is also added to all stiffness data to demonstrate the models' feasibility of stiffness prediction.