(504d) Model-Predictive Safety: Min-Max Optimization to Calculate the Most Aggressive Control Actions and the Worst-Case Uncertainties | AIChE

(504d) Model-Predictive Safety: Min-Max Optimization to Calculate the Most Aggressive Control Actions and the Worst-Case Uncertainties

Authors 

Soroush, M. - Presenter, Drexel University
Oktem, U., Near-Miss Management LLC
Seider, W., University of Pennsylvania
Arbogast, J. E., Process Control & Logistics, Air Liquide
Samandari Masooleh, L., Drexel University
In 2016 [1] we introduced a method of model-predictive safety (MPS) system design, and in 2017 and 2018 [2, 3] min-max optimization problems that should be solved offline to implement an MPS system in real time. The solutions of the optimization are (a) the most aggressive control action that minimizes each process-constraint index when uncertain model parameters take their nominal values, and (b) the most aggressive control action that minimizes each process-constraint index when uncertain model parameters take their worst-case values.

An MPS system [1] generates alarm signals that are predictive and systematically account for process nonlinearities and interactions, while typical existing functional safety systems generate reactive, non-interacting alarm signal(s) when a process variable exceeds a threshold. The MPS system design method allows for a systematic utilization of dynamic process models to generate alarm signals (alerts) for the predictive detection and proactive prevention of operation hazards (OHs) in real time. An MPS system uses a process model to predict the process safety status over a moving prediction horizon and to generate alarm signal(s) indicating the presence of a present or future OH with reasonable accuracy; it generates alarm signals that alert the process personnel to imminent and potential future OHs before the actual OHs occur.

In this paper, we propose and implement a particle-swarm optimization (PSO) method to solve the min-max optimization problems described in Ref. [2] for two process examples, a classical chemical reactor with series reactions and a free-radical polymerization reactor. Results from these two examples show that the PSO method reliably finds offline (a) the most aggressive control action that minimizes the infinity norm of each safety-constraint index over a prediction horizon, and (b) the most aggressive control action that minimizes the uncertainty-maximized infinity norm of each safety-constraint index over a prediction horizon. The performances of the MPS systems implemented with different prediction horizons are shown using numerical simulations.

References

[1] Mohseni Ahooyi, T., J.E. Arbogast, W.D. Seider, U.G. Oktem, and M. Soroush, "Model-Predictive Safety System for Proactive Detection of Operation Hazards," AIChE J., 62, 2024-2042 (2016).

[2] Soroush, M., J.E. Arbogast, and W.D. Seider, “Model-Predictive Safety System for Predictive Detection of Operation Hazards: Off-Line Calculation of Most Aggressive Control Actions and Worst-Case Uncertainties,” CAST Division 10 Plenary Session at the 2017 AIChE Annual Meeting, Minneapolis, MN (2017)

[3] Soroush, M., A.A. Shamsabadi, W.D. Seider, and J.E. Arbogast, "Implementation of Model-Predictive Safety Systems to Detect Predictively Operation Hazards in Non-Minimum-Phase Processes," 2018 AIChE Annual Meeting, Pittsburgh, PA (2018)