Abstract: Aiming at the problem that the health maintenance and state detection of wind turbine planetary gearboxes are difficult to diagnose, a fault diagnosis research method of the initial dilated convolutional neural network （IDCNN） that combines the initial net and dilated convolution is proposed in this paper. This method first constructs an initial dilated convolutional layer to expand the receptive field to enrich the learned fault features. Subsequently, in order to facilitate signal input and ensure rich information, a preprocessing method of transforming the one-dimensional original signal sequence into a two-dimensional matrix will be adopted. Finally, the generated two-dimensional signal is input into IDCNN for model training, and the model is evaluated with test data. The experimental results show that the proposed IDCNN method has high accuracy in the fault diagnosis of the planetary gearbox of the wind turbine. In the comparison results, the diagnosis accuracy of the proposed method is higher than that of the traditional deep learning method.

Key words: wind turbines, inception net, dilated convolutional neural network, planetary gearbox, fault diagnosis

摘要： 针对风力发电机行星齿轮箱的健康维护和状态检测难以诊断的问题,该文提出一种初始网与膨胀卷积相融合的初始膨胀卷积神经网络（IDCNN）的故障诊断研究方法。该方法首先构建初始膨胀卷积层以扩大感受野来使学习到的故障特征更加丰富。随后为了方便信号输入且确保信息丰富,将采用将一维原始信号序列转化为二维矩阵的预处理方法。最终将生成的二维信号输入到IDCNN中进行模型训练,并用测试数据对模型进行评估。实验结果表明,提出的IDCNN方法在风力发电机行星齿轮箱的故障诊断中精度高,在对比结果中该文提出方法的诊断精度要高于传统的深度学习方法。

关键词: 风力发电机, 初始网, 膨胀卷积神经网络, 行星齿轮箱, 故障诊断