A model predictive early warning system for disturbance type faults

故障检测与诊断（FDD）系统是现代工艺装置的重要组成部分。传统的FDD系统只有在实际信号或处理后的信号超过阈值后才会产生告警。虽然这是一种强有力的方法，但在某些情况下，对系统进行任何修改可能为时已晚。该研究计划的目标是将FDD系统的预测特征整合到早期预警中，为操作人员提供足够的时间来采取纠正措施。在本文中，我们主要研究扰动型断层。过程系统容易受到干扰：其中一些是可测量的，而另一些是不可测量的。扰动逐渐干扰过程系统；干扰的影响可以在过程中完全感受到它的影响之前预测出来。我们提出了一个新的FDD框架，该框架将估算进入过程系统的未知干扰，预测其对系统的影响，并在认为影响重大时发出警报。提出的预测故障检测与诊断框架包括两个模块：(1)同步输入与状态估计（SISE）模块和(2)移动视界预测与可行性分析模块。进入过程系统的干扰，以及系统内的其他状态，通常是无法测量的。SISE模块将估计未知状态和进入系统的干扰。期望最大化算法将用于迭代估计状态和未知扰动的大小。EM算法具有良好的统计性能和收敛性。基于估计的扰动大小、系统模型和可用的测量，将使用移动视界预测器预测系统的未来状态。最后，将进行可行性分析，以确定执行器是否有足够的能力来抵消干扰效应，并将输出保持在运行约束范围内。如果没有可行的解决方案，则会发出告警。新型FDD系统比传统报警系统更早发出报警。它将通过提供故障的早期指示，使过程工业以及许多其他工业更安全，并提高运行效率。通过该研究项目，将培养2名博士和2名硕士研究生。他们将获得过程监测和控制的理论和应用知识，并获得工业标准分布式控制系统（DCS）和监控和数据采集（SCADA）系统的实践培训。这些技能在加拿大的工业中非常抢手。因此，拟议的研究计划将对加拿大经济作出重大贡献。

Fault detection and diagnosis (FDD) systems are an integral part of modern process plants. Traditionally, FDD systems generate an alarm only after the actual signal or the processed signal has crossed the threshold. Though this is a robust approach, in some cases it may be too late to make any amendments to the system. The goal of this research program is to incorporate predictive features in FDD systems for early warning, to give operators sufficient time to take corrective actions. In the proposed research, we focus on disturbance-type faults. Process systems are prone to disturbances: some of them are measured while others are unmeasured. Disturbances perturb process systems gradually; the effect of a disturbance can be predicted before its impact is fully felt in the process. We propose a new framework for FDD that will estimate an unknown disturbance entering into a process system, predict its impact on the system, and finally issue an alarm if the impact is deemed significant. The proposed predictive fault detection and diagnosis framework consists of two modules: (i) simultaneous input and state estimation (SISE) module, and (ii) a moving horizon prediction and feasibility analysis module. Disturbances entering process systems, as well as other states within the system, are usually unmeasured. The SISE module will estimate both the unknown states and the disturbance entering the system. An expectation maximization (EM) algorithm will be used to iteratively estimate the states and the unknown disturbance magnitude. The EM algorithm has excellent statistical properties and guarantees convergence. Based on the estimated disturbance magnitude, system model, and available measurements, the future states of the system will be predicted using a moving horizon predictor. Finally, feasibility analysis will be carried out to determine if there is sufficient capacity in the actuators to counteract the disturbance effects and keep the outputs within the operational constraints. An alarm will be issued if there is no feasible solution. The novel FDD system is expected to issue an alarm earlier than the traditional alarm systems. It will make process industries, as well as many other industries, safer and improve the efficiency of operation by providing early indications of faults. Through this research program, two PhD and two Master's students will be trained. They will gain knowledge on the theory and applications of process monitoring and control and hands-on training on industry standard distributed control systems (DCS) and supervisory control and data acquisition (SCADA) system. These are highly sought after skills in Canadian industries. Thus, the proposed research program will contribute significantly to Canadian economy.