(172a) Human-in-the-Loop: Coordinated Decision-Making | AIChE

(172a) Human-in-the-Loop: Coordinated Decision-Making


Ghosh, S. - Presenter, Rensselaer Polytechnic Institute
Bequette, B. W., Rensselaer Polytechnic Institute
Chemical plants are highly integrated processes with advanced automation and fault-detection techniques. Although advanced automation systems, incorporating real-time optimization, process monitoring, and model predictive control have yielded significant economic benefits, humans remain important and critical components in these systems. Process operators must take active roles during startup, shutdown and abnormal situations. Unfortunately, poor decisions (at many levels in the decision-making hierarchy) can also lead to disasters, such as the BP Texas City refinery explosion and fire in 2005.

An integral part of this is a hierarchical structure which transcends beyond the plant and includes human at every level. In this work, an overview of Human-in-The-Loop (HiTL) approaches is presented. This starts with a general top-down analysis of a plant with humans at every level of operation. A brief discussion for the motivation of HiTL approaches in the industry is given to lay the foundation for the work [1-2]. A detailed discussion of a HiTL Supervisory Model Predictive Control (MPC) algorithm is given with simple simulation studies aimed at usage of this tool for operator training. The examples used are Single-Input Single-Output (SISO) and Multi-Input Multi-Output (MIMO) cases where the operator either shares the control decision with the MPC or controls a subset of the outputs while the remaining is controlled by the MPC. The MPC optimization objective is to minimize the set-point change input to regulatory PID controllers by penalizing the deviation from upper-layer output setpoints and human suggested inputs. Depending on the input/output pairs chosen for manual and MPC, the state-space matrices are modified. For shared control of input/output pairs, a tuning parameter is also included to weigh the human inputs. The MPC modifications discussed here are used to explore how different operator decisions ranging from closing control loops to taking supervisory decisions can affect the plant.

On a second study, we present results of an in-house experiment where human subjects are tested for their supervisory decision-making capabilities by using a dynamic simulation of the Vinyl Acetate Monomer benchmark problem [3]. Operator inputs are crucial to a plant during the startup/shutdown phases. This is a complicated MIMO problem which is further aggravated by alarm flooding and information transfer during shift changes. The experiment involves a simple setup mimicking these conditions under laboratory conditions to explore important human factors and how better control-room designs are possible.

Concluding remarks on using the HiTL supervisory MPC algorithm in the human experiments and future directions of the work are provided.


[1] S. Ghosh and B. W. Bequette, “Process Systems Engineering and the Human-in-the-Loop--the Smart Control Room,” Ind. Eng. Chem. Res., vol. 59, no. 6, pp. 2422–2429, 2019.

[2] S. Ghosh and B. W. Bequette, “A Framework for the Control Room of the Future: Human-in-the-loop MPC,” IFAC-PapersOnLine, vol. 51, no. 34, 2019.

[3] M. L. Luyben and B. D. Tyréus, “An industrial design/control study for the vinyl acetate monomer process,” Comput. Chem. Eng., vol. 22, no. 7–8, pp. 867–877, 1998.