基于深度神经网络的液压泵泄漏状态识别

Recognition of hydraulic pump leakage status based on deep neural network

针对液压信号的高度复杂性以及难以识别的特点,构建了一种基于堆栈稀疏自编码器和Softmax的深度神经网络来对液压泵泄漏状态进行识别。利用小波变换和希尔伯特-黄变换提取液压信号的低维特征,并输入深度神经网络。通过堆栈稀疏自编码器的逐层学习对特征进行优化并提取出高维特征,然后使用Softmax进行识别。实验结果表明,堆栈稀疏自编码器能够有效地提取液压泵泄漏状态的高维特征,构建的深度神经网络可有效地识别液压泵泄漏状态,识别精度达到了97.6%。此外与支持向量机、极限学习机、卷积神经网络以及长短期记忆网络相比,深度神经网络具有更好的识别效果。

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.