摘要
针对卷烟制丝过程由于多批次、数据不等长和产品多样性而导致的无法准确监测、诊断故障的问题,提出一种基于主元分析的监测与故障诊断方法。首先,通过制丝过程特性分析,将批次、时间和属性的三维数据按照属性展开成二维数据;其次,采用主元分析方法分别建立3个工段的监测模型,离线计算T2、SPE控制限;然后,分别采集3个工段的运行数据,根据产品牌号调用对应的监测模型在线计算T2、SPE统计量;最后,当任一统计量指标超出控制限时,采用贡献图进行故障诊断。基于实际运行数据的离线验证表明,该方法能够准确检测故障的发生并确定引起故障的原因变量。
In view of multi-batch, unequal-length data, product diversity of tobacco primary processing and difficulties in accurate monitoring and diagnosing faults, this paper presents a monitoring and fault diagnosis method based on principal component analysis. Firstly, on the basis of the characteristic analysis of tobacco primary processing, the three-dimensional data with batch, time and property is expanded into two-dimensional data by employing property expansion. Secondly, the monitoring models for three sections are established respectively using principal component analysis, and theT2 and SPE control limits are calculated off-line. Thirdly, theT2 and SPE statistics are on-line calculated by utilizing the same product grade monitoring model with the current collecting data. Finally, when there is any statistic index beyond the control limit, the fault is diagnosed using the contribution plot. Off-line testing results based on actual operation data show that the proposed method can more accurately detect the fault and determine the causal variable.
出处
《控制工程》
CSCD
北大核心
2017年第12期2435-2442,共8页
Control Engineering of China
基金
国家自然科学基金项目(61422306
61433005)
浙江省博士后科研项目择优资助项目(BSH1502045)
关键词
卷烟制丝过程
多批次
三维数据
主元分析
监测模型
故障诊断
Tobacco primary processing
multi-batch
three-dimensional data
principal component analysis
monitoring model
fault diagnosis
