流程工业过程非稳定工况的故障检测与诊断方法研究

There are several operating modes and transition phases in modern process industry. Under these non-stationary conditions, it's diffcult to apply traditional fault detection and diagnosis strategies. In this project, monitoring methods based on local density and distribution probability will be developed to solve the multimode problem. For the new unknown operating modes, a semi-supervised strategy will be developed by integrating process knowledge. To cope with process slow dynamics, the monitoring model can be updated by integrating recursive procedures into the moving window strategy. Using neighbohood information, a more accurate monitoring model can be achieved. To deal with the transition phases between adjacent modes, two different fault detetion methods are proposed based on fuzzy clustering methods and local regression methods, respectively. Based on process knowledge, computational intelligence will be used to isolate and diagnose fault causes. The study will provide new insights and effective fault detection and diagnosis methods for the non-stationary process industry.

现代流程工业生产过程不仅具有多种生产工况，还普遍存在开停车及工况间的过渡过程，此外季节变化和设备老化等现象会导致过程参数发生缓慢变化，这些非稳定工况给过程的故障检测与诊断带来很大困难。本课题针对多工况故障检测问题，研究基于局部密度信息和分布概率信息的监控策略，降低特征提取算法的复杂度，研究融合过程工艺知识的半监督故障检测模型，增强模型对未知工况的学习能力；针对过程参数缓变问题，研究基于邻域信息和移动窗口的递归更新算法，简化自适应过程的计算复杂度；针对过渡过程，研究基于模糊聚类的离线建模算法和基于局部信息的实时建模策略，提高故障检测率；针对非稳态工况的流程工业过程，分析过程的工艺机理，研究基于计算智能的故障诊断策略。本课题的研究将为存在非稳态工况的流程工业过程提供故障检测与诊断策略，并通过结合典型的工业过程，形成理论和方法的应用推广。

本项目利用基于数据驱动的过程监控方法，通过系统分析流程工业过程中生产数据的特征，深入研究对非稳定工况进行更为精确有效的故障检测与诊断方法，并取得了一些具有创新性的研究成果。针对多工况过程故障检测与诊断，提出了离线模态划分方法、单一模态建模方法、在线结果整合方法以及建立全局模型的方法，提高故障检测与诊断模型精度。针对缓变过程故障检测与诊断，提出了结合移动窗策略以及基于邻域信息的实时建模方法，提高故障检测率与诊断精度。针对过渡过程，提出了稳定模态与过渡模态的划分方法以及基于局部信息的建模策略，能够对过渡模态进行准确划分并进行故障检测。相关研究成果总结整理后已分别在国内外专业期刊、学术会议发表。