基于边缘计算的轴向柱塞泵磨损状态辨识方法研究  

Wear State Identification Method for Axial Piston Pumps Based on Edge Computing

作　　者：王丹丹[1] 黄伟迪[1] 张军辉[1] 赵守军[2,3] 于斌 刘施镐[1] 吕飞 苏琦 徐兵[1] WANG Dandan;HUANG Weidi;ZHANG Junhui;ZHAO Shoujun;YU Bin;LIU Shihao;LÜFei;SU Qi;XU Bing(State Key Laboratory of Fluid Power and Mechatronic Systems,Zhejiang University,Hangzhou 310027;Beijing Institute of Precision Mechatronics and Controls,Beijing 100076;Laboratory of Aerospace Servo Actuation and Transmission,Beijing 100076)

机构地区：[1]浙江大学流体动力基础件与机电系统全国重点实验室,杭州310027 [2]北京精密机电控制设备研究所,北京100076 [3]航天伺服驱动与传动技术实验室,北京100076

出　　处：《机械工程学报》2024年第4期189-199,共11页Journal of Mechanical Engineering

基　　金：国家重点研发计划(2019YFB2004504);国家自然科学基金(51835009,52105075);浙江省自然科学基金(LQ21E050022);航天伺服驱动与传动技术实验室开放基金(LASAT-20210104)资助项目。

摘　　要：判断轴向柱塞泵的磨损状态对维护轴向柱塞泵正常运行具有重要意义。然而,现有的轴向柱塞泵故障诊断方法大多为离线式和基于云计算的在线式,存在延时长和数据量大的问题,无法满足轴向柱塞泵磨损状态辨识的实时性需求。为了减少延迟时间和传输数据量,提出一种基于边缘计算的轴向柱塞泵磨损状态辨识方法。构造出一个集成信号采集、信号预处理、特征提取和磨损状态分类的边缘节点,用于正确和实时地辨识磨损状态。设置4种轴向柱塞泵滑靴副磨损状态作为故障源,并构建相应的磨损故障数据集。为了减少边缘节点计算量,在上位机中利用随机森林包外误差选择敏感特征值。为了实现磨损状态辨识,在上位机中训练磨损状态分类的特征选择人工神经网络模型,并将信号预处理和特征提取功能算法以及模型参数嵌入边缘节点。通过与其他方法的比较和在线磨损状态辨识试验证明所提方法的正确性和实时性。Wear state identification is of great significance to maintain the normal operation of axial piston pump.However,most existing fault diagnosis methods for axial piston pump are offline while those online are based on cloud computing,producing a long delay or a large amount of data,which cannot meet the real-time requirements of wear state identification.To cut down latency and data amount,a wear state identification method for axial piston pump based on edge computing is proposed.An edge node integrating the functions of data acquisition,data pre-processing,feature extraction and wear state classification is constructed to identify the wear state accurately and timely.Four kinds of wear states of axial piston pump are set as fault sources,and the corresponding wear fault dataset is established.To reduce the computational load of the edge node,the out-of-bag error of random forest is employed to select the sensitive features at the host computer.To classify the wear state,a feature-selected artificial neural network for wear state classification is trained at the host computer,and the data pre-processing and feature extraction algorithm together with model parameters are embedded into the edge node.The accuracy and real-time performance of the proposed method are demonstrated through comparisons with other methods and the online wear state identification experiment.

关 键 词：轴向柱塞泵 边缘计算 磨损状态辨识 特征敏感程度 人工神经网络 

分 类 号：TH322[机械工程—机械制造及自动化]