基于迁移学习的风电机组叶片损伤检测与分析

WIND TURBINE BLADE DAMAGE DETECTION AND ANALYSIS BASED ON TRANSFER LEARNING

针对风电机组叶片损伤成因复杂、故障识别效率低、精度不足等问题,提出一种基于迁移学习改进的DenseNet网络(DenseNet-TL)的风电机组叶片损伤检测方法。建立DenseNet-TL数学模型,提升特征提取能力,在该模型下对风电机组叶片图像进行识别分析,以确定叶片的损伤状态。以某风场数据集进行离线训练和测试,结果表明:与AlexNet、ResNet模型进行对比,该模型可有效节省训练时间、提高模型的泛化能力,训练准确度平均值达到90%以上,验证了该方法的有效性和精确性。

Aiming at the problems of complex causes of wind turbine blade damage,low fault identification efficiency,and insufficient accuracy,a wind turbine blade damage detection method is propsed based on improved DenseNet network improved by transfer learning is proposed.A mathematical model of DenseNet network improved by transfer learning(DenseNet-TL)is established to improve the feature extraction capability,and the recognition and analysis of wind turbine blade images are carried out under the model to determine the damage state of the blades.The offline training and testing is carried out with a wind farm dataset,and the results show that,compared with AlexNet and ResNet models,the model effectively saves the training time,improves the generalization ability of the model,and the average training accuracy reaches more than 90%,which verifies the validity and accuracy of the method.