基于深度信念网络的风机叶片结构损伤识别研究

Structural damage identification of wind turbine blade based on deep belief networks

【目的】解决由于模态参数辨别的不确定性,以及虚假模态干扰造成的风机叶片结构损伤识别精度不高的问题.【方法】采用以深度信念网络提取的模态参数特征向量作为标识量的损伤检测方法.首先分别获取ANSYS仿真和实验条件下风机叶片的模态参数;然后利用深度信念网络提取模态参数特征向量作为损伤标识量,检测多种工况下的风机叶片损伤,并与传统BP神经网络方法进行对比;最后搭建实验平台,在实验条件下验证方法的有效性.【结果】基于深度信念网络的损伤识别方法相比传统BP神经网络精度更高,网络训练时间更长.【结论】将深度信念网络提取的模态参数特征向量作为BP神经网络训练的输入向量,可以减小噪声和虚假模态信息等因素对损伤识别结果的影响,提高损伤识别的精度.

【Objective】To solve the problem of the low identification accuracy of the wind turbine blade structure damage caused by the uncertainty of the modal parameter identification and spurious modes.【Method】A method of taking deep belief nets was adopted to extract features of modal parameters as damage signature.Firstly,the structural vibration modal parameters of the wind turbine blade were gotten under the condition of ANSYS simulation and experiments.Then,the characteristic vector of modal parameters were extracted as the signature for damage detection and identify the damage of the wind turbine blade under different situation.A comparison was conducted with the traditional BP neural network approach.Finally,the validity of the method were verified under the foundation of experiments.【Result】The damage identification method based on the deep belief nets was more accurate than the traditional BP neural network with longer training time.【Conclusion】It can reduce the influence of noise and false mode information on damage identification results,and improve the accuracy of damage identification,taking deep belief nets to extract features of modal parameters as damage signature.