基于单类数据迁移的控制阀粘滞检测方法  

A Stiction Detection Method for Control Valve Based on One-class Data Transfer

作　　者：刘梦杰 赵春晖[1,3] 王文海 LIU Meng-jie;ZHAO Chun-hui;WANG Wen-hai(State Key Laboratory of Industrial Control Technology,Zhejiang University,Hangzhou 310027,China;NGICS Platform,College of Control Science and Engineering,Zhejiang University,,Hangzhou 310027,China;State Key Laboratory of Synthetical Automation for Process Industries,Northeastern University,Shenyang 110819,China)

机构地区：[1]浙江大学控制科学与工程学院工业控制技术国家重点实验室,浙江杭州310027 [2]浙江大学控制科学与工程学院NGICS大平台,浙江杭州310027 [3]东北大学流程工业综合自动化国家重点实验室,辽宁沈阳110819

出　　处：《控制工程》2022年第4期645-654,共10页Control Engineering of China

基　　金：国家杰出青年科学基金资助项目(62125306);NSFC-浙江两化融合联合基金资助项目(U1709211);工业控制技术国家重点实验室自主课题(ICT2021A15);流程工业综合自动化国家重点实验室开放课题(2020-KF-21-07)。

摘　　要：控制阀粘滞检测能提高回路控制性能诊断的效率与准确率,为工厂改善回路控制性能提供指导。现有粘滞检测方法未考虑回路被控对象的动态特性变化时,控制阀粘滞在回路数据上的表征不一致问题。针对此问题,提出了一种基于单类数据迁移的控制阀粘滞检测方法。该方法仅需使用待测回路的非粘滞数据对模型进行迁移,实现了与回路适配的粘滞特征提取模型的构建。首先,使用仿真模型生成的类别完备的数据对模型进行预训练;然后,结合待测回路的非粘滞数据对残差网络进行微调;最后,使用单分类算法得到控制阀粘滞检测结果。仿真数据集和开源工业数据集的实验结果验证了所提方法的有效性。Stiction detection of control valve can improve the efficiency and accuracy of loop control performance diagnosis,and provide guidance for factories to improve loop control performance.The existing stiction detection methods do not consider the inconsistent representation of control valve stiction on the loop data when the dynamic characteristics of the loop change.Aiming at this problem,a stiction detection method for control valve based on one-class data transfer is proposed in this paper.Only the non-stiction data of the loop to be tested are used to transfer the model,so as to establish a stiction feature extraction model adapted to the loop.Firstly,the data with complete category generated by the simulation model are used to pretrain the model.Secondly,the simulation data combined with the non-stiction data of the loop to be tested are used to fine tune the residual network.Finally,the one-class classification algorithm is used to obtain the stiction detection results.Experimental results on simulation datasets and open-source industrial datasets verify the effectiveness of the proposed method.

关 键 词：控制阀 粘滞检测 迁移学习 单分类算法 

分 类 号：TH134[机械工程—机械制造及自动化] TP273[自动化与计算机技术—检测技术与自动化装置]