Abstract:Due to the high complexity, it is hard to recognize hydraulic signals. To solve this problem, a deep neural network is formulated for recognition of hydraulic pump leakage status, which is based on the stacked sparse autoencoder and Softmax. The lowlevel features are extracted by the wavelet transform and the HilbertHuang transform. These features are put into the deep neural network. Through the layerbylayer learning of stacked sparse autoencoder, the lowlevel features are optimized and the highlevel features are obtained. Then, Softmax is used to recognize other features. Experimental results show that the stacked sparse autoencoder can effectively extract the highlevel features of hydraulic pump leakage status. The formulated deep neural network can distinguish the pump leakage status and the recognition accuracy is 976%. In addition, compared with extreme learning machine, support vector machine, convolutional neural networks and long shortterm memory, the deep neural network has better recognition effectiveness.