基于CEEMDAN与t-SNE的轴向柱塞泵典型故障诊断方法

A Typical Fault Diagnosis Method for Axial Piston Pump Based on CEEMDAN and t-SNE

掘进机在非均匀土壤工况下运行会使得其液压驱动系统受到较强的冲击负载,易导致其核心动力源液压泵发生故障。针对轴向柱塞泵内的典型磨损故障的特征分离提取与辨识分类,提出了一种基于CEEMDAN方法的故障振动信号时频域特征提取方法,对泵壳体采集的振动信号进行时频域分解,并构建70维度特征空间对状态特征进行表征描述;通过特征降维融合的策略,采用t-SNE对高维特征进行解析融合,将70维的特征空间降维至2维空间,并显著提升了分类器的训练效率和分类准确性;采用SVM方法训练3类典型磨损故障的分类器模型,并通过测试集验证,模型的辨识准确性达到96.7%,显著优于不进行降维处理的高维分类器辨识模型。

The hydraulic driving system of the tunneling machine is subjected to strong impact load when it runs in non-uniform soil condition,which is easy to cause the failure of its core power source hydraulic pump.For the characteristic separation,extraction and identification classification of typical wear faults in axial piston pump,a time-frequency domain feature extraction method of fault vibration signals based on CEEMDAN method was proposed.The vibration signals collected from pump shell were decomposed in time-frequency domain,and 70 dimensions’feature space was constructed to characterize the state features.Through the feature dimension reduction fusion strategy,t-SNE was used for analytic fusion of high-dimensional features,which reduced the dimension of 70-dimensional feature space to 2-dimensional space,and significantly improved the training efficiency and classification accuracy of the classifier.The SVM method was used to train the classifier model for three types of typical wear faults,and through the test set verification,the identification accuracy of the model reached 96.7%,which was significantly better than the high-dimensional classifier identification model was not processed by dimensionality reduction.