Process diagnostics and event-driven control for safety-critical chemical processes and plants

When automatic control systems are used to protect safety-critical manufacturing processes and chemical plants, they must cope with major challenges under abnormal situations to avoid product and equipment damage, avert financial loss, and most importantly, prevent life-threatening situations. Whether the abnormal condition is due to sensor measurement errors, actuator failure (e.g., a stuck control valve), or leakage, fouling, or other equipment malfunction, it is critical that the abnormal event be detected, diagnosed, and that proper control action be taken before the situation escalates into a safety threat.

The circumstances become even more challenging when humans (process operators and engineers) are called to take action, situations in which a clear indication of what happened is vital so that appropriate operator intervention can be made. This research effort intends to address these needs and will advance and transform industrial practice for chemical and other manufacturing process safety.

The research team of this project will develop diagnostic algorithms that can quickly detect and identify the root cause of an abnormal situation in a chemical process or entire plant and will formulate automatic control algorithms that apply the necessary corrective action to prevent a potential disaster. The research team also will test their theories on process stability and the control system software they develop on a laboratory scale reactor system that (safely) can emulate an industrial reactor.

The development of such diagnostic and control algorithms will be accomplished through fundamental advancements in the field of systems science, particularly in data-driven modeling and feedback control methods. This project will support graduate and undergraduate education, including students from underrepresented groups, and will generate important safety related content for chemical process design and process control courses.

Control of safety critical manufacturing processes in the face of potential abnormal events requires the automated detection that an unusual event has taken place, unambiguous identification of which sensor, actuator, or process equipment component has failed, and reconfiguration of the control system to respond to the failure to maintain process stability and safety. To satisfy these requirements, the specific research objectives of the project are to: (a) develop algorithms for chemical process and plant diagnostics, based on either a first-principles model or data-driven statistical model; (b) perform experimental testing of the diagnostic algorithms in an industrially relevant but lab-scale chemical reactor, where a highly exothermic and potentially explosive reaction takes place (but in a safe manner); and (c) develop event-driven control algorithms based on the diagnosis of the abnormal event from alarm data, as well as dynamic analysis of its effect on the ability to enforce safety constraints.

To meet objectives (a) and (c), system-theoretical tools will be employed, advancing methods and results in nonlinear functional observers, statistical inference, and constrained nonlinear control. Meeting objective (b) will involve modeling, simulation, and diagnostic algorithm development for chemical reactors, as well as experimental implementation and testing.

The innovative features of the project consist of: (i) advances in the theory and algorithms for process and plant diagnostics, based on either first-principles models or statistical models, along with an experimental study that will serve as a paradigm for future fault diagnosis applications in real chemical processes; and (ii) advances in dynamic analysis and model-based fault-tolerant control in the face of possible abnormal events that can pose safety threats. The results of the proposed work will be broadly disseminated to researchers in academia and industry by presentations at domestic and international meetings, in scholarly refereed journal publications, and through a dedicated web site for the project.

The PI also will make freely available all case studies and implementations of the proposed work, and will create a web tool for easy access.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.