Header menu link for other important links
How Does a Control Room Operator Identify the Process? Insights Using a Cognitive Engineering Approach
L Das, B Srinivasan,
Published in

Industrial accidents in the past such as those in BP Texas City Refinery (2005), Three Mile Island (1979) and BayerCrop Science (2008) that have caused loss of money and lives and environmental damage have served as unremitting impetus for process plants to enhance safety. As a result, industries have developed and equipped themselves with several technological advancements such as distributed and redundant sensing, sophisticated control algorithms and automation techniques, advanced alarm management systems and support and guidance systems. These technologies collect, process and present information regarding the process state at all times to plant operators, typically stationed in a distributed control system (DCS) room. Operators, the consumer of the above information have the responsibility of operating the process under normal conditions so as to maximize profits as well as handling abnormal situations to prevent accidents.

In carrying out these tasks, operators use their knowledge of process dynamics (including nominal operating values of variables, interrelationship between variables) along with fault diagnosing capability, which are consolidated in what is called the operator’s mental model of the process. Operators typically acquire their mental model through operator training systems (OTS) and/or on-the-job experience. An OTS consists of a simulated environment of the process wherein trainee operators are exposed to typical scenarios that might occur in the real plant and constitutes a major component of training programs. These systems are required to be able to accurately simulate process dynamics in a range of operating conditions in real time and research efforts have identified avenues for improving OTSs from a computational point of view (Patle et al., (2014)). For example, an expert guidance based intelligent training system was proposed in (Shin and Venkatasubramanian, (1996)) that enhances diagnosing skill of operators, and an assessment technique of operators during training was proposed based on the deviations of operator actions from predefined sequences of actions (Lee et al., 2000). These techniques emphasize on improving simulation capabilities of an OTS, enhancing diagnosing skills of operators and assessing the operator’s ability to follow standard guidelines.

The key element of a training program, i.e., the operator and the ultimate aim of a training program, i.e. imparting an adequate mental model of the process to the operator is not addressed explicitly in the above techniques. However, it is important to take into account the cognitive abilities of trainees in designing tasks so as to ensure a successful training program. With advancement of technology that have led to techniques such as model predictive control (MPC), the roles of operators also change (Kluge et al., (2014)) which further increases the importance of accounting operator’s abilities. Understanding the cognitive behavior of operators has been emphasized as being critical to enhancing their skills and abilities (Bullemer and Nimmo, (1994)). Previous work has also identified the importance of cognitive processes of operators during an OTS task in evaluating their ability (Sharma et al., (2016)).

In this work, we propose a framework that accounts for the operator’s abilities in designing training tasks and that is focused on imparting an adequate mental model through an online method of knowledge and capability assessment. We use eye tracking as a tool to peek into the cognitive processes of the operator during a training task and assess their mental models as they evolve during/across tasks. We follow a systems engineering approach to operator training that comprises three components – (1) Design suitable training tasks, (2) measure the operator’ response and analyze the same with respect to their actions to examine their mental models and (3) update their mental model during the task which can then be used to identify gaps for designing suitable tasks. This translates to an adaptive training program that accounts for the operator’s abilities during the task in an online manner and design tasks so as to ensure an adequate model imparted to the operator. The proposed framework is partly used with human subjects to assess their mental models during a simulated training exercise.

An ethanol production plant is simulated and training tasks are designed in three stages – (1) manipulation of valves to observe the effect on process variables, (2) tracking a process variable to a desired set point, and (3) tracking multiple process variables at the same time. The first stage is aimed at allowing the operator to identify the relation between manipulated valves and process variables while the latter two stages require the operator to identify appropriate valve(s) to achieve certain objective(s). Eye tracking data is collected during the tasks and the operator’s gaze data is analyzed to obtain association graphs at different stages of a task, representative of the operator’s mental model of the process (or the part of the process being targeted by a task). Association graphs are directed graphs derived from eye gaze data where nodes represent the variables the operator has gazed, and edges represent an association between two nodes. The strength of each edge is set to be proportional to the gaze transition probability between the two nodes, representing the extent to which an association is examined by the operator. Evolution of these mental models can be studied during a task, and across multiple tasks. Knowledge and capability can be assessed through their performance in unguided tracking tasks. This monitoring and assessment technique can be performed online and gaps in operator’s mental model identified at the end of each task so as to adaptively design subsequent training tasks that can ensure an adequate model imparted to the operator.

About the journal
Journal2017 AIChE Annual Meeting