复杂工业系统的即插即用过程监测与控制理论方法研究

Due to the effects of the integration of global economic and the ever increasing demands on higher production quality, system performance and economic benefit, the modern industrial systems have become more and more complicated, which can be observed by more and more detailed task assignments of sub-systems with complicated system structures and high automation degrees. To ensure the safe and efficient operation of complex industrial systems, this project focuses on the process monitoring (PM) and fault-tolerant control (FTC) for complex industrial systems and their plug-and-play (PnP) design. To this end, based on the coprime factorization technique, the parameterization forms of observer based PM systems and stabilizing controllers are first studied. This study serves as an essential fundament for the constructed PnP process monitoring and control architecture with functionalized modules. The proposed PnP process monitoring and control architecture is an integrated design of process monitoring and control based on the pre-designed control systems, that allows the monitoring and control performance to be online optimized without modifying the structure or the parameters of the pre-designs. Then, with aid of the subspace identification and machine learning techniques, efficient data-driven design methods for the functionalized process monitoring and control modules are proposed using industrial process data. During the operation of industrial system, not only the process itself (aging, fault, unexpected disturbance, …) is monitored, but also the control performance is evaluated and supervised. Finally, reliable online learning and configuration approaches for the process monitoring and control modules are proposed, which allow the monitoring and control performance to be online optimized when a fault or performance degradation is detected. In short, the main objective of this proposal is to develop PnP process monitoring and control system for complex industrial systems while leaving the pre-designs untouched. This project aims to provide a more convenient and effective design and a solid theoretical basis for the practical application of process monitoring, performance optimization and FTC in complex industrial systems.

为保障工业系统的高效安全运行，本项目主要研究复杂工业系统的故障诊断与容错控制理论和技术及其即插即用设计。首先，借助互质分解技术研究故障诊断观测器与镇定控制器的不同参数化形式，在不改变已有监测与控制系统结构和参数的情况下，基于已有控制系统构建模块化的即插即用过程监测与控制系统设计框架。其次利用过程数据，基于子空间辨识、机器学习等技术完成各个监测与控制模块的数据驱动设计。使得在对工业系统进行实时监测的同时，对其控制性能也实时地进行评估和监控。最后，在工业系统发生故障或者控制性能下降的情况下，利用实时工业过程数据进行即插即用监测与控制系统的在线优化与快速重构，实现工业系统控制性能的实时优化与容错控制。本项目旨在不改变已有监测与控制系统结构和参数的情况下，设计即插即用过程监测与控制系统，为过程监测、系统性能优化以及容错控制在复杂工业系统中的实际应用提供更加方便有效的设计方法与坚实的理论依据。

在三年的项目执行时间内，研究内容全面按照计划执行，以保证和提高复杂工业系统运行的安全性、可靠性和高效性为立足点，开展对过程监测与控制系统的即插即用设计方法的研究。在即插即用过程监测与控制框架的基础上，基于大量离线/在线数据，研究了数据驱动的过程监测与控制系统设计问题，并对即插即用过程控制系统的关键参数进行在线辨识与实时优化，以保障复杂工业系统的健康稳定运行。项目在三年的执行时间内基本完成研究目标。针对实际工业系统，基于互质分解技术研究故障诊断观测器与镇定控制器的不同参数化形式，在不改变已有监测与控制系统结构和参数的情况下，设计了即插即用过程监测与控制框架。针对含有确定性扰动的工业系统，采用子空间方法实现了鲁棒的数据驱动过程监测方法。针对系统的稳定性，通过辨识系统闭环传递函数矩阵的乘法算子，实现了数据驱动的稳定裕度的估计，并基于此提出了面向系统控制性能指标的过程监测与容错控制方法。在即插即用过程监测与控制框架下，分别提出了分布式即插即用故障诊断方法、自适应数据驱动故障诊断方法以及在线跟踪性能优化方法等。上述研究成果分别在直流电机系统、三容水箱过控系统，以及工业轧钢系统中取得了较好的实验效果。