基于多种小波变换的一维卷积循环神经网络的风电机组轴承故障诊断

为解决在复杂工况下风力发电机组轴承故障诊断虚警率高的问题,提出一种端到端的混合深度学习框架--基于多种小波变换的一维卷积循环神经网络。首先,通过多种小波变换得到多个时-频矩阵,以充分提取信号特征;再通过一种扩展的LSTM,对多通道时-频矩阵不同时间步信息进行提取,捕获时-频数据时空特征;最后,通过全局池化层和分类层对故障状态进行分类。实验结果表明:在复杂工况下,多种小波变换的一维卷积循环神经网络对风力发电机组轴承故障识别率能够达到95%以上。

Bearing Fault Diagnosis of Wind Turbine Based on Multi-wavelet-1 D Convolutional LSTM

To solve the problem of high false alarm rate of wind turbine bearing fault diagnosis under complex conditions, an end-to-end hybrid deep learning framework is proposed. One-dimensional convolutional recurrent neural network based on multiple wavelet transforms (multi-wavelet-1D Convolutional LSTM, Mw-1DConvLSTM). Firstly, multiple time-frequency maps are obtained by multiple wavelet transforms to fully extract the signal features. Then, an extended LSTM is used to extract different time step information of multi-channel time-frequency maps, and time-space characteristics of time-frequency data are captured. Finally, the fault state is classified by the global pooling layer and the classification layer. The test results show that under complex conditions, Mw-1DConvLSTM can achieve more than 95% fault identification of wind turbine bearing faults.