Abstract: An improved convolutional neural network fault diagnosis method based on ensemble empirical mode decomposition is proposed to extract weak fault features and low fault diagnosis accuracy of bearings. Firstly, we employ ensemble empirical mode decomposition to reduce the noise of the signal, and transform the denoised signal into a 2-dimensional (2-D) signal. Then, to solve the problem of parameter explosion of convolutional neural network (CNN) and the uncertainty of data characteristics, a batch normalization layer is added between the convolutional layer and the pooling layer of CNN to construct an improved convolutional neural network (ICNN). Finally, the weak fault dataset of the wind turbine bearings are taken as an example to verify that the proposed method has superior performance compared with other methods. The experimental results show that the proposed method can effectively extract fault features, and has high fault diagnosis accuracy and efficiency.

Key words: fault diagnosis, ensemble empirical mode decomposition, convolutional neural network, batch normalization