Dynamic estimation of continuum and noncontinuum states for process control monitoring

在过去的20年里，过程控制监测领域取得了重大进展。已有的研究几乎全部集中在连续值系统或连续统系统上。然而，在现实中，许多系统或问题可能采取非连续形式或状态，例如控制阀中的“粘滞”与“不粘滞”。即使对于具有连续状态的过程，人们也可能只对控制性能监测问题中的非连续表示感兴趣，如“好”、“坏”或“最优”。因此，在过程控制监测中需要同时考虑连续状态和非连续状态。因此，本研究的长期目标是开发一种新的基于状态估计的过程控制监测框架，更具体地说，同时解决过程故障检测和控制性能监测问题。短期目标是研究和开发连续状态和非连续状态的动态估计方法，并使用这些方法来解决化工过程中某些突出的故障检测和控制性能监控问题。尤其是滚动视界估计方法，将用于连续和非连续状态的动态估计。一旦建立了新的框架并开发了算法，就可以在该框架下解决具有连续状态和非连续状态的过程控制监测问题，这意味着许多过程故障检测和控制性能监测问题可以转化为状态估计问题，并且很容易得到解决。此外，许多看起来非常脱节的过程控制监控问题可以通过统一的方式解决。这一研究成果不仅为解决不同监测问题提供了理论基础，而且为在同一计算框架下开发工业应用的计算算法提供了便利。

The field of process control monitoring has gained significant progress over the last 20 years. Existing studies have almost exclusively been devoted to continuous-valued or continuum system. In reality, however, many systems or problems may take noncontinuum form or state, such as "stiction" vs "no stiction" in a control valve. Even for processes with continuum state, one may be interested in only the noncontinuum representation such as "good", "bad" vs "optimal" in control performance monitoring problem. There is therefore a need to consider both continuum and noncontinuum states in process control monitoring.The long-term objective of this research is therefore to develop a novel framework based on state estimation for process control monitoring, more specifically, simultaneous process fault detection and control performance monitoring problem. The shorter-term objective is to investigate and develop dynamic estimation methods for both continuum and noncontinuum states and use these methods to solve certain outstanding fault detection and control performance monitoring problems for chemical processes. The moving horizon estimation approach, in particular, will be used for dynamic estimation of continuum and noncontinuum states. Once the new framework is established and algorithms are developed, process control monitoring problems with both continuum and noncontinuum states are solved under the developed framework.Success of this work implies that many process fault detection and control performance monitoring problems can be cast as state estimation problem and be readily solved. In addition, many process control monitoring problems that appear quite disjoint can be solved in a unified way. This research outcome thus not only provides a theoretical foundation but also facilitates development of computation algorithms for industrial applications under the same computation framework for solving different monitoring problems.