基于上游单点测量信息和机器学习的气动系统并联双气缸泄漏故障诊断  

Fault Diagnosis of Leakages in Parallel Double Pneumatic Cylinders Based on Machine Learning and Measured Information in a Single Upstream Point

作　　者：朱宏伟 王志文[1] 杨波 王虎 熊伟[1] ZHU Hong-wei;WANG Zhi-wen;YANG Bo;WANG Hu;XIONG Wei(Institute of Ship Electromechanical Equipment,Dalian Maritime University,Dalian,Liaoning 116026)

机构地区：[1]大连海事大学船舶机电装备研究所,辽宁大连116026

出　　处：《液压与气动》2023年第7期73-82,共10页Chinese Hydraulics & Pneumatics

基　　金：国家自然科学基金(51905066)。

摘　　要：利用少量传感器融合机器学习技术进行系统多故障诊断是实现气动系统低成本智能化故障诊断的潜在途径。以气动系统中常见的泄漏故障为例,探究了利用上游单点测量信息实现下游并联双气缸泄漏故障诊断的可行性。上游单点测量信息包括压力、流量和[火用]数据,预处理后的数据通过栈式自编码器(Stacked Autoencoder,SAE)进行特征提取,并将提取的特征送入高斯过程分类器(Gaussian Process Classifier,GPC)中进行学习分类。实验结果表明:通过机器学习模型学习分析上游单点测量信号来实现对下游并联双气缸泄漏故障的诊断和定位是可行的;在本实验同等条件下,基于[火用]数据的平均分类准确率达到100%,高于基于流量数据的98.99%和基于压力数据的77.38%。It is a potentially feasible approach to realize low-cost intelligent fault diagnosis of pneumatic system by using a few sensors and machine learning technology.In this study,the common leakage fault in pneumatic system is taken as an example to explore the feasibility of using the measured information in a single upstream point to achieve the leakage fault diagnosis of parallel double cylinders in downstream.Measured data of single upstream point includes pressure,flow rate and exergy data.The data features of the pre-processed data is extracted by stack autoencoder(SAE),and then are sent to Gaussian process classifier(GPC)for learning and classification.The experimental results show that it is feasible to diagnose and locate the leakage faults in the downstream parallel double cylinders by analyzing the measured data in a single upstream point with machine learning.Under the identical conditions in this experiment,the average classification accuracy of exergy data could reach 100%,higher than 98.99%of flow-rate data and 77.38%of pressure data.

关 键 词：气压传动 故障诊断 [火用] 自编码器 高斯过程分类 

分 类 号：TH138[机械工程—机械制造及自动化]