摘要
产品质量在工业生产中至关重要,以往的质量相关监控方法不能正确区分操作条件改变引起的运行工况变化和真正的过程故障,不利于生产的连续性和经济性。将典型相关分析(canonical correlation analysis,CCA)与动态慢特征分析(dynamic slow feature analysis,DSFA)结合,构建新的优化目标函数,提取过程数据中与质量强相关的缓慢变量,将过程变量分解为与质量相关的子空间、与质量无关的子空间,并分别建立对应的稳态和动态监控指标。该方法运用实时获得的过程数据检测故障,并判断是否影响质量,可以区分正常工况偏差和真实故障,并具有良好的动态性能。最后,通过仿真实验对比以往的方法,验证了所提出方法的有效性。
Product quality is key-performance-indicator in industrial process monitoring.The previous quality-related monitoring algorithms only focus on the monitoring statistics of the steady-state characteristics of the system,and cannot distinguish between changes in operating conditions caused by changes in operating conditions and real process failures,which is detrimental to the continuity and economy of production.Canonical correlation analysis(CCA)and dynamic slow feature analysis(DSFA)are combined to construct a new optimization objective function to extract the slow variables and their first-order derivatives that are strongly related to quality in the process data.They are the steady-state information and dynamic information that reflect the essential changes of the system,respectively.The method decomposes process variables into quality-related subspaces and quality-unrelated subspaces.Then the corresponding steady-state and dynamic monitoring indicators are established in the two subspaces respectively.The method uses the process data obtained in real time to detect faults and judge whether they affect the quality.It can distinguish between normal operating conditions deviation and real faults and has good dynamic performance.Finally,the effectiveness of the proposed method is verified by comparing the previous methods through simulation experiments.
作者
邵远哲
赵忠盖
刘飞
SHAO Yuanzhe;ZHAO Zhonggai;LIU Fei(Key Laboratory for Advanced Process Control of Light Industry of the Ministry of Education,Jiangnan University,Wuxi 214122,China)
出处
《控制工程》
北大核心
2025年第11期2096-2104,共9页
Control Engineering of China
基金
国家自然科学基金资助项目(61833007)。
关键词
潜变量模型
慢特征分析
典型相关分析
质量相关
故障检测
动态慢特征分析
Latent variable model
slow feature analysis
canonical correlation analysis
quality-related
fault detection
dynamic slow feature analysis
