Intelligent fault diagnosis of Wind Turbines via a Deep Learning Network Using Parallel Convolution Layers with Multi-Scale Kernels. (June 2020)

• Record Type: Journal Article
• Authors: Chang, Yuanhong; Chen, Jinglong; Qu, Cheng; Pan, Tongyang
• Is Part Of: Renewable energy. Volume 153(2020)
• Page Start: 205
• Page End: 213
• Publication Date: 2020-06
• Subjects: Intelligent fault diagnosis -- Deep learning -- Wind turbines -- Generator bearing -- Multiple scales; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2020.02.004
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: In recent years, the intelligent diagnosis technology of wind turbines has made great progress. However, in practical engineering applications, the operating states of wind turbine are various, accompanied by a large number of noise interference, which leads to the decline of the discrimination accuracy of intelligent diagnosis. In order to solve this problem, inspired by the Google team Inception model, this paper proposes a concurrent convolution neural network (C–CNN), the raw data is fed into the network without any prior knowledge, and the characteristics are learned directly and adaptively from the input. Even if the data is accompanied by noise, the model still has high accuracy and strong generalization ability. The model is composed of a CNN with multiple branches. Meanwhile, the convolutional layer of different branches selects the kernels with different scales in same level, thus improving the learning ability of entire network. In this paper, the feasibility of this method for fault diagnosis of bearings in wind turbines is demonstrated by three bearing datasets. The results show that the proposed method can extract discriminative features and classify bearing data accurately under the disturbance of different rotating speed, different load and random noise. Highlights: The C–CNN realize each layer has multi-scale kernels, so as to data features in a multi-angle and deep level. The C–CNN network not only has good classification accuracy, but also has good generalization ability. The model can balance the depth and width of the network, and thus control the growth of parameters and calculations. The feasibility of this method for fault diagnosis of bearings in wind turbines is demonstrated by three data sets.