摘要
提出采用主元分析方法分析化工过程积累的数据,进而实现化工过程故障的诊断与识别。首先,通过PCA方法对正常工况数据进行训练,获得统计量T2与SPE的控制限阈值;然后计算故障工况数据的统计量,并与控制限进行比较,超过控制限阈值即判断为故障工况数据,计算故障发现率并作为故障诊断能力的评价标准;最后计算变量的统计量贡献率,识别出引起故障的主控变量。对田纳西-伊斯曼过程进行案例研究,选择正常工况数据集和6种故障模式数据集,当提取主元个数为31时,对6种故障均有很高的诊断能力,识别出的主控变量也与该故障实际的工艺扰动监测值相对应。
A principal component analysis method was proposed to analyze the accumulated data of chemical process and diagnose and identify the faults there.Firstly,having PCA method adopted to train the normal condition data so as to obtain control limit thresholds of both statistics T2 and SPE,and then having statistics’rate of contribution of the normal condition data calculated and compared with the control limit,those exceeded the control limit threshold would be judged as fault condition data and having the fault discovery rate calculated as the evaluation standard of fault diagnosis ability;and finally,having the main control variables which incurred the faults identified to calculate contribution rate of the variables’statistics so as to recognize main control variables which caused the faults.Analyzing a case of Tennessee-Eastman chemical process showed that,while selecting a normal working condition data set and six fault-mode data sets and the number of principal components extracted is 31,it has a high diagnostic ability for the six kinds of faults,and the main control variables identified also correspond to the faults’monitoring value of the actual process disturbance.
作者
刘丽云
国蓉
牛鲁娜
栗月姣
胡海军
LIU Li-yun;GUO Rong;NIU Lu-na;LI Yue-jiao;HU Hai-jun(College of Photoelectronic Engineering,Xi’an Technological University;Sinopec Research Institute of Safety Engineering;School of Chemical Engineering and Technology,Xi’an Jiaotong University)
出处
《化工自动化及仪表》
CAS
2020年第5期398-406,449,共10页
Control and Instruments in Chemical Industry
基金
国家重点研发计划资助项目(2017YFF0210400)。
关键词
故障诊断
故障识别
化工过程
田纳西-伊斯曼过程
主元分析
故障发现率
大数据
fault diagnosis
fault identification
chemical process
Tennessee-Eastman process
principle component analysis
fault discovery rate
big data
