为了解决卷烟制叶丝段不同生产线同一牌号产品感官质量存在差异等问题,提出了一种基于改进Fisher判别分析和参数贡献度评价的工艺特征参数提取方法.采用改进Fisher判别分析提取制叶丝段优秀生产线和问题生产线同一牌号工艺参数数据的最大分离方向,进而沿最大分离方向计算不同生产线的相似度,并根据优秀生产线的相似度计算控制限.如果问题生产线相似度超出控制限的占比大于给定置信值,计算问题生产线各工艺参数的差异贡献度,将本次迭代中贡献度最大的工艺参数作为特征参数,通过迭代提取获得导致感官质量差异的所有工艺特征参数.基于某特定牌号对不同生产线的工艺参数数据进行实验验证,结果表明:该方法提取的工艺特征参数更加科学和完备,通过量化问题生产线工艺参数的差异贡献度,有效提高了不同生产线多参数差异分析的可靠性.该方法可为制叶丝段不同生产线相同工艺路径下的工艺参数优化提供技术支持.

In order to eliminate the blind spots of monitoring at idle running stage in batch process of cut strip processing section and prevent the mutual interference between several pieces of equipment working at abnormal status to communicate with the overall monitoring model at one time,a multi-stage distributed monitoring and trouble diagnosis method was proposed based on the strategy of dividing into blocks horizontally and into layers vertically. According to the different monitoring requirements at idle running stage and steady production stage,the monitoring models for Sirox heating and humidifying machine and KLD thin plate dryer at different stages were separately developed by principal component analysis to monitor equipment abnormality at different stages effectively. Contribution plot method was used to accurately identify the reasons causing abnormality. Validation was conducted on the basis of actual running data of cut strip processing section in Hangzhou Cigarette Factory,the results showed that the proposed method raised the ability of the monitoring models to capture batch process information,monitored and diagnosed equipment abnormalities at different stages accurately and effectively. This method provides a theoretical support for the state monitoring and trouble diagnosis of batch process of primary processing in smart cigarette factories.

In view of high misdiagnosis rate caused by the strong correlation between process variables in tobacco strip processing, a fault diagnosis method was proposed, which combined variant Fisher discriminant analysis (VFDA) with significant fault variable extraction. The VFDA contains orthogonal discriminant components, it prevents the singularity of intra-category scatter matrix by a two-step feature extraction and protects the verticality between discriminant components via data deflation. VFDA is used for extracting fault direction. The contribution of each variable to the faults was measured along the fault directions, those variables which influenced the faults heavily were referred to as fault variables, and those variables which did not affected the faults were referred to as general variables. Fault diagnosis models for fault variables and general variables were established separately for online equipment fault diagnosis. Off-line validation was conducted based on actual running data of tobacco strip processing equipments, the results showed that comparing with typical contribution plot fault diagnosis method, the proposed method was helpful to intensive understanding of the process and characteristics of faults and eliminating the influences of minor information via significant fault variable analysis and extraction. The variables causing faults were isolated timely and accurately, and the reliability of fault diagnosis of cigarette manufacturing equipments was effectively promoted. The proposed method provides theoretical support for the precise diagnosis of the equipment while it is not properly functioning.