摘要
考虑到深度神经网络具备优异的故障识别性能,针对柱塞泵压力与流量信号特征提取难度大问题,设计了一种综合运用小波变换与希尔伯特-黄变换来实现的特征提取方法,建立了RBM-BP网络来达到优化原始特征的作用,利用高级融合特征诊断柱塞泵泄漏状态。高级特征散点表明,深度置信网络在学习原始特征方面表现出了较强学习能力,实现原始特征的抽象提取,确保高级特征能够更准确完成柱塞泵内泄分级与诊断过程。研究结果表明:所有正常泄漏样本都被准确预测,微弱泄漏与严重泄漏都出现了1个样本发生错误预测情况。相比较SSAE与H-ELM,RBM-BP在各层中都表现出比更低的识别。RBM-BP方法获得了比SSAE与H-ELM更高的准确率,准确率波动性也更小,表明RBM-BP模型达到了更优的稳定性,表现出了对柱塞泵内泄状态更强辨识能力与稳定性。
Considering the depth of fault identification of the neural network has outstanding performance,and the pressure and flow of hydraulic pump signal feature extraction is difficult problem,we design a comprehensive use of the wavelet transform and Hilbert huang transform to realize feature extraction method,established a deep belief network to achieve the effect of original op⁃timization feature,USES advanced fusion diagnosis of hydraulic pump leakage.The scatter of advanced features shows that the deep confidence network has a strong learning ability in learning the original features,and the abstract extraction of the original features is realized to ensure that the advanced features can more accurately complete the classification and diagnosis process of the internal leakage in the hydraulic pump.The results show that all the normal leakage samples are accurately predicted,and one sample is mispredicted in both the weak leakage and the serious leakage.Compared with SSAE and H-ELM,RBM-BP shows lower recognition ratio in each layer.The accuracy of RBM-BP method is higher than that of SSAE and H-ELM,and the fluctu⁃ation of accuracy is smaller,which indicates that RBM-BP model achieves better stability,showing stronger identification abili⁃ty and stability of the leakage state in the hydraulic pump.
作者
李丹
朱渔
李晓明
张建国
LI Dan;ZHU Yu;LI Xiao-ming;ZHANG Jian-guo(Department of Information Engineering,Shangrao Polytechnic,Jiangxi Shangrao 334001,China;School of Information Engineering,Yichun Vocational and Technical College,Jiangxi Yichun 336000,China;School of Mechanical Engineering,Nanchang University,Jiangxi Nanchang 330031,China;Jiangxi Bochuan Automation Technology Co.,Ltd.,Jiangxi Pingxiang 337000,China)
出处
《机械设计与制造》
北大核心
2023年第9期173-176,共4页
Machinery Design & Manufacture
基金
江西省教育厅科学技术研究项目(GJJ191671)。
关键词
柱塞泵内泄
深度置信网络
故障诊断
辨识能力
准确率
Hydraulic Pump Internal Leakage
Deep Confidence Network
Fault Diagnosis
Recognition Ability
Accuracy
