基于频谱分析和卷积神经网络的高速轴向柱塞泵空化故障诊断

针对高速轴向柱塞泵不同空化故障等级诊断依赖人工特征提取、识别准确率低的问题,提出了一种融合振动信号频谱分析和卷积神经网络的诊断方法。采集不同空化等级情况下柱塞泵壳体振动信号,对连续的振动数据进行切片并作频谱分析,获得频谱图作为数据集;利用二维卷积神经网络对不同空化等级的信号频谱图进行分类。为提高所提方法的鲁棒性,采用带通滤波的方法抑制频谱图中的噪声频率。试验结果表明:对于不同信噪比的振动信号输入,均能准确地识别出柱塞泵的空化故障等级。

Cavitation Fault Diagnosis of High-speed Axial Piston Pump Based on Spectrum Analysis and Convolutional Neural Network

Traditional fault diagnosis of high-speed axial piston pumps relies on hand-crafted feature extraction and suffers from low recognition accuracy for different levels of cavitation intensity. Therefore, a diagnostic method that combines spectrum analysis of vibration signal and convolutional neural network (CNN) is proposed. Vibration signals are collected on the pump housing under different levels of cavitation intensity. The continuous vibration data is segmented into a series of data frames, followed by a spectral analysis to generate spectrograms which are fed to a 2D CNN model for cavitation classification. To improve the robustness of the proposed method, the noise-related frequencies in the spectrograms are suppressed by band-pass filtering method after spectrum analysis. The results show that the proposed model can achieve high recognition accuracy of cavitation conditions for input vibration signals with different signal-to-noise ratios.