基于CVA及因果分析的复杂系统故障诊断方法研究

Fault diagnosis techniques have significant effects on ensuring safety and effectiveness of complex industries nowadays. Whereas current mainstream methods, taking principal component analysis (PCA) and partial least square (PLS) as instances, show low accuracy and efficiency on diagnosing faults as the result of the limitation on modeling accuracy and parsimony, which leads to difficultly determine faulty variables as well as the types and root causes of faults. The limitation not only exerts negative effects on production efficiency, but also brings tremendous threats to stability and security of industrial production. This project aims to improve the accuracy and efficiency of fault diagnosis under the feature of huge volume, high correlation and strong dynamics of process data. State-space model based canonical variate analysis (CVA) techniques, which can effectively capture dynamic information from process data, are proposed as the basic framework for data-driven fault diagnosis methods. The proposed methods contains (i) CVA-based contribution methods for diagnosing faulty variabls; (ii) CVA-based discriminant analysis for classifying faults; and (iii) CVA methods in conjunction with causality analysis for diagnosing root causes of faults. The effectiveness of the proposed approaches is demonstrated on the Tennessee Eastman process and catalytic cracking units. The project will also push forward the application of CVA methods in fault diagnosis.

故障诊断技术对保障复杂工业系统的安全生产与高效运行发挥着至关重要的作用。然而以PCA和PLS为代表的故障诊断方法在建模精确性和简约性方面所存在的局限性降低了其诊断故障的准确性和快速性，从而导致故障变量、故障发生的种类和根源难以及时有效确定。这不仅对生产效益带来巨大影响，更给工业生产安全带来巨大威胁。本课题针对复杂工业过程数据量巨大、相关度高且动态性强的特点，以提高故障诊断的准确率和快速性为目的，以充分提取过程数据中的动态模型信息为出发点，提出以基于状态空间模型的规范变量分析(CVA)技术为基本框架的数据驱动故障诊断方法。这些方法包括：(1) 基于CVA贡献分析的故障变量诊断方法；(2) 基于CVA判别分析的故障分类方法；(3) 综合CVA和因果分析的故障溯源方法。本课题的研究成果将在田纳西-伊斯曼仿真过程和催化裂化实际装置中进行验证。本研究将推动CVA方法在故障诊断领域中的应用。

随着工业生产效率等需求的日益增长，现代工业过程规模和结构不断趋向复杂化。这些过程一旦发生故障会造成巨大的经济损失并对生产安全形成威胁。因此，利用故障诊断技术确保复杂过程的安全生产和高效运行具有重要意义。本课题利用具有良好数据建模性能的规范变量分析(Canonical Variate Analysis, CVA)方法于故障诊断领域，较好解决了复杂工业过程对过程监控在快速性和准确性方面的需求。具体研究成果包括：（1）提出了一种基于规范变量分析(CVA)的贡献图方法，有效增加了故障变量诊断的准确性；（2）提出了基于CVA判别分析的监督式故障分类方法，提高了故障分类的准确性和快速性；（3）提出了综合CVA和因果分析的无监督式故障溯源方法，有效弥补了工业实践中不能事先获知故障数据集情况的应用需求。课题组在IEEE Trans. on Control Systems Technology, IEEE Trans. on Industrial Informatics, Journal of Process Control等公开发表12篇学术论文，其中被SCI检索9篇，被EI检索11篇。