基于NAKF和DBN的液压管路故障智能诊断方法  

Hydraulic pipeline intelligent diagnosis method based on NAKF and DBN

作　　者：姚存治[1] 张明真[1] 张尚然 王冠群 YAO Cun-zhi;ZHANG Ming-zhen;ZHANG Shang-ran;WANG Guan-qun(School of Artificial Intelligence,Zhengzhou Railway Vocational and Technical College,Zhengzhou 451460,China;Hebei Petroleum University of Technology,Chengde 067000,China)

机构地区：[1]郑州铁路职业技术学院人工智能学院,河南郑州451460 [2]河北石油职业技术大学,河北承德067000

出　　处：《机电工程》2022年第5期587-595,共9页Journal of Mechanical & Electrical Engineering

基　　金：河南省科技厅科技发展计划软科学研究项目(212400410187);郑州铁路职业技术学院校级教育教学改革研究与实践项目(2020JG01)。

摘　　要：针对航空液压管路故障识别困难的问题,提出了一种基于非线性自适应卡尔曼滤波器(NAKF)和深度信念网络(DBN)的液压管路智能故障诊断方法。首先,在传统卡尔曼滤波器(KF)的基础上,利用最小二乘法修正构造的Sigma点,消除高斯分布对Sigma点影响,提出了非线性自适应卡尔曼滤波器,并用其对仿真信号进行了降噪处理;然后,对液压管路实测振动信号中的随机噪声进行了去除,对深度信念网络模型参数进行了设计,并将液压管路数据集输入到深度信念网络模型中进行了训练;最后,基于同一样本数据,分别采用支持向量机(SVM)和反向传播神经网络(BPNN)等模型进行了训练处理,利用分类准确率等两个指标,对3种故障诊断模型进行了综合评估,对3种模型分类性能进行了对比分析。研究结果表明:采用NAKF-DBN智能故障模型得到的液压管路故障诊断准确率能达到99.72%,SVM模型和BPNN模型等浅层网络的平均故障诊断准确率不高于95%,而未经非线性自适应卡尔曼滤波器滤波的深度信念网络的诊断准确率仅有86.58%;该结果验证了NAKF-DBN模型对于液压管路故障识别的有效性,可以为航空液压管路的智能化诊断提供新思路。Aiming at the problem that difficulty of identifying the fault of aviation hydraulic pipeline,an intelligent fault diagnosis method of hydraulic pipeline based on nonlinear adaptive Kalman filter(NAKF)and depth belief network(DBN)was proposed.Firstly,on the basis of the traditional Kalman filter(KF),the least square method is used to modify the Sigma points constructed,the influence of Gaussian distribution on Sigma points was eliminated,and the nonlinear adaptive Kalman filter was proposed.Then,the random noise of vibration signals measured in aviation hydraulic pipeline was removed,the parameters of the deep belief network model were designed,and the hydraulic pipeline data set was input into the deep belief network model for training.Finally,based on the same sample data,support vector machine(SVM)and back propagation neural network(BPNN)were used for training and processing respectively.Classification accuracy were used to comprehensively evaluate the classification performance of the three fault diagnosis models.The results show that the accuracy of NAKF-DBN hydraulic pipeline intelligent fault model can reach 99.72%,the average accuracy of traditional support vector machine model and back-propagation neural network model is less than 95%,and the accuracy of DBN network without NAKF filtering is even lower,only 86.58%.The effectiveness of NAKF-DBN model for hydraulic pipeline fault identification is verified,which provides a new idea for intelligent diagnosis of aviation hydraulic pipeline.

关 键 词：液压传动回路 支持向量机 反向传播网络 深度信念网络 非线性自适应卡尔曼滤波器 智能故障模型 

分 类 号：TH137.7[机械工程—机械制造及自动化] V263.6[航空宇航科学与技术—航空宇航制造工程]