基于WICA-WGNPE的高斯-非高斯联合指标间歇过程监控  

Batch process monitoring with Gaussian and non-Gaussian joint indicator based on WICA-WGNPE

作　　者：惠永永 赵小强[1,2,3] 

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050 [2]甘肃省工业过程先进控制重点实验室,兰州730050 [3]兰州理工大学国家级电气与控制工程实验教学中心,兰州730050

出　　处：《仪器仪表学报》2018年第1期190-199,共10页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61763029)项目资助

摘　　要：许多工业生产过程数据很难满足单一的分布,一般是高斯和非高斯成分共存的混合分布,并且在过程数据的提取中,会隐藏表征故障的有用信息。为了解决高斯与非高斯的混合多分布问题,充分提取过程数据的有效信息,提出一种基于WICAWGNPE的高斯和非高斯联合指标间歇过程监控方法。首先采用WICA算法提取非高斯成分;然后对提取非高斯成分后的残差使用WGNPE方法来得到高斯成分;最后基于非高斯-高斯两步策略确立监控模型并求得统计量,对非高斯与高斯统计量进行加权,从而得到混合模型的联合指标以实现过程监控。将该算法用于青霉素发酵过程和半导体实际工业过程来验证所提算法的有效性。Many industrial production process data are difficult to meet single distribution,usually mixed multi-distributions contain both Gaussian and non-Gaussian components,and the useful information that characterizes faults is hidden in the process data extraction.In order to solve the problem of Gaussian and non-Gaussian mixed multi-distribution and to fully extract the useful information in process monitoring data,a batch process monitoring method with Gaussian and non-Gaussian joint indicator based on WICA-WGNPE（Weighted Independent Component Analysis-Weighted Global Neighborhood Preserving Embedding） is proposed in this paper.Firstly,the WICA algorithm is used to extract the non-Gaussian component.Then,the WGNPE method is used to extract the Gaussian component from the residual after extracting the non-Gaussian component.Finally,the monitoring model is established based on two-step strategy of non-Gaussian-Gaussian,and the statistics are obtained.Through weighting non-Gaussian and Gaussian statistics,the joint indicator of the mixed model is obtained and the process monitoring is achieved.The proposed algorithm was applied in the penicillin fermentation process simulation platform and real semiconductor industrial process to verify the effectiveness of the proposed algorithm.

关 键 词：间歇过程 过程监控 高斯和非高斯 全局-局部 联合指标 

分 类 号：TH165.4[机械工程—机械制造及自动化] TP206.3[自动化与计算机技术—检测技术与自动化装置]