Control Loop Performance Diagnosis in the Presence of Uncertainties

点击翻译按钮获取中文摘要

Process control systems are critical because they allow for steady and safe process operations, consistent product quality, less waste, and better control of emissions. But the performance of control systems can deteriorate over time due to wear of instruments, variation of feed or product grade, and change of process dynamics and operating environment. Timely diagnosis for detection of the sources of performance deterioration is critical to maintenance of optimal performance. The main challenges to effective performance diagnosis come from stochastic uncertainty, epistemic uncertainty, monitor discretization uncertainty, and lack of historical feature data. The objective of this proposal is to build a general framework for control loop performance diagnosis to solve all the above issues. Our proposed methodology is based on evidence theory, specifically Dempster-Shafer theory, along with the use of continuous performance monitor outputs. Problems associated with practical engineering applications are also considered and will be solved under a unified framework. The solutions will be evaluated by pilot-scale experiments as well as by industrial applications. Our long-term goal is to establish an integrated controller asset advisory system that consists of control performance monitoring, control loop performance diagnosis, and decision making, so that process control system can be optimized with considerations such as economics, process constraints, and environmental impacts. Our research will establish a theoretical and methodological foundation for robust control loop performance diagnosis. Our proposed solutions will be applicable to a wide range of industries. By involving students in all aspects of this work, our research program will train personnel who are highly qualified in control, monitoring and diagnosis - areas that are highly relevant to Canadian industry.