基于经验模式分解和一维密集连接卷积网络的电液换向阀内泄漏故障诊断方法

内泄漏作为电液换向阀常见的故障类型,其故障振动信号具有非平稳性、非线性等特点,且容易被其他信号淹没、破坏。对此提出了一种经验模式分解(Empirical Mode Decomposition,EMD)和一维密集连接卷积网络(Densely Connected Convolutional Networks,DenseNet)的电液换向阀内泄漏故障诊断方法。该方法首先利用EMD对振动信号进行分解得到一系列本征模态分量(Instrinsic Mode Function,IMF),并将IMF分量和原始振动信号依次进行并联堆叠;然后将并联堆叠信号作为一维密集连接卷积网络的输入进行特征的自动提取,并进行故障分类;最后通过DenseNet与传统的一维卷积神经网络(CNN)对比验证得出,该方法能准确、有效地对电液换向阀内泄漏故障进行诊断。

Fault Diagnosis for Internal Leakage of Electro-hydraulic Directional Valve Based on EMD and One-dimensional Densely Connected Convolutional Networks

As a common fault type of the electro-hydraulic directional valve, the internal leakage vibration signal has the characteristics of non-stationary and non-linear, and it is easy to be submerged and corrupted by other signals. For the feature, an empirical mode decomposition (EMD) and one-dimensional densely connected convolutional networks (DenseNet) method is proposed to diagnose the internal leakage in the Electro-hydraulic Directional Valve. Firstly, a series of intrinsic mode functions (IMF) are obtained by decomposing the vibration signals with EMD, and the IMF components and the original vibration signals are stacked in parallel successively. Then, the parallel stacked signals are used as the input of one-dimensional densely connected convolution network for feature extraction and fault classification. Finally, compared with DenseNet and the traditional one-dimensional convolution neural network, it can be concluded that this method can accurately and effectively diagnose the internal leakage fault in the Electro-hydraulic Directional Valve.