辽宁科技学院学报

针对实际工程应用中风力发电机组运行工况复杂，测试信号受噪声干扰严重进而导致现有智能诊断方法判别准确率低的问题，文章提出一种基于LReLU-BN-CNN的风力发电机组轴承故障诊断方法。为解决卷积神经网络中ReLU激活函数会使神经元完全失活，进而导致部分特征损失的问题，文章引入了Leaky ReLU激活函数；同时，为进一步提高深层神经网络的训练效率，采用数据批量归一化的处理方式。实验结果表明，提出的LReLU-BN-CNN方法在风力发电机轴承故障诊断中分类精度达到98.01%；对比结果表明，文章所题的方法在故障分类精度上要优于其他深度学习方法。

In practical engineering applications, operating conditions of the wind turbine are complex, and the test signal is seriously perturbed by noise, eventually leading to problems of low identification accuracy of the intelligent diagnosis method. This paper proposes a research method called LReLU-BN-CNN for wind turbine bearings fault diagnosis. In order to solve the problem of partial feature loss due to the complete deactivation of neurons in the ReLU activation function of the convolutional neural network, the Leaky ReLU activation function is introduced. In order to further improve the training efficiency of the deep neural network, the new method also adopts the processing treatment of data batch normalization. The experimental results show that the proposed LReLU-BN-CNN method has a classification accuracy of up to 98.01% in the fault diagnosis of wind turbine bearings. Comparisons further show that the fault classification accuracy of the current method is better than other deep learning methods.