基于叶片表面污垢预处理与CNN的风电机组叶片表面损伤识别

Wind Turbine Blade Surface Damage Identification Based on Blade Surface Dirt Pretreatment and CNN

为了消除叶片表面污垢对损伤识别的影响,提出了基于叶片表面污垢预处理与卷积神经网络(CNN)的故障诊断方法。利用鲁棒主成分分析方法将存在表面污垢的叶片图像进行预处理,将损伤特征提取过程转化为凸优化问题进行求解,获取叶片损伤特征,并使用无人机拍摄的实际叶片图像对CNN进行训练,利用训练好的CNN识别经过表面污垢预处理后的叶片图像,判定损伤的具体类型。结果表明:本文方法能够较为准确地识别含有表面污垢的叶片损伤,证明了其有效性。

To detect the damages hidden under the dirt on blade surface more effectively,a fault diagnosis method was proposed based on the pretreatment of dirt on blade surface and the convolutional neural network(CNN).The robust principal component analysis(RPCA)was used to preprocess the image of the blade with surface dirt,and then the damage feature extraction process was transformed into a convex optimization problem to be solved,thus to obtain the damage feature of the blade,while the CNN was trained with actual blade images taken by unmanned aerial vehicle(UAV).Finally,the trained CNN was used to identify the blade images preprocessed by RPCA and to determine the specific type of damages.Results show that the method proposed can help to indentify the damage of blades with surface dirt,which is proved to be effective by practical applications.