A multivariate data driven approach for modelling the cognitive behaviour of control room operator using eye tracking

In an attempt to improve process safety, today’s plants utilize automation and control strategies along with sophisticated safety management regimes. Despite these advancements, accidents continue to occur, indicating that process safety is at least as important today as it was in the 1970s and 80s. Statistics along with numerous surveys indicate that human error is one of the major cause for accidents in process industries. Advancements in automation transfers the responsibilities to various individuals and does not eliminate the need for human operators. Therefore, it is important to understand the role of human operators in modern plants and develop methodologies to prevent human errors. Traditionally, human errors are qualitatively considered during HAZOP analysis and human failures quantified using error probabilities. However, this view-point abnegates the cognitive abilities of operator in handling a situation. Recent studies in process industries have utilized eye tracking measures to qualitatively understand the behaviour of control room operators. Results from these studies not only revealed the existence of a specific eye gaze pattern in certain participants (completed the task using only one manipulated variable) but also patterns in the eye gaze behaviour indicated that not all successful operations were alike; nearly 12% of the successful operators closely resembled operators who had failed to control the task [1]. Also, gaze patterns and pupillometry can be used to identify the various cognitive tasks performed during different phases of the task [2]. These works however provide only qualitative information about the eye gaze behaviour. Visual entropy is one of the widely used measures in several safety critical domains (aviation, healthcare) for quantification of eye gaze movements. Visual entropy reveals the degree of randomness in the scanning behaviour. Relatively low values of entropy e represents highly focused visual attention while scattered gaze behaviour results in higher value of visual entropy. This quantitative measure has been used understand the mental workload of operators, their expertise level and situation awareness [3,4,5]. Apart from visual entropy, there are several other quantitative measures (known in the literature) that could provide different perspectives on the cognitive behaviour of the operators. For instance, pupil dilations has the potential to provide real-time information about the mental workload of control room operators [6]. Visual entropy provides indication about the scanning behaviour of the operator (on the HMI) while pupil size variations help understand their perceived workload. Also, both these measures are correlated and provide information about the cognitive behaviour of operator. For instance, during high workload the operator will focus on large number of variables (high visual entropy) as the cause of abnormality is not known while little/no pupillary dilations with low visual entropy indicates the high confidence level of the operator in handling the situation. This work attempts to develop a multivariate approach to model the behaviour of control room operator using visual entropy and pupillary dilations. Experimental studies using a simulated ethanol plant were conducted on 44 participants with control engineering background. Process variables, alarm details, mouse clicks, eye movement and pupil measurements were conducted during these experiments. Eye gaze data and pupil measurements were utilized to compute visual entropy and pupil dilation measure. A multivariate data driven measure is developed from these measures to quantify the cognitive behaviour of control room operators. Results from application to experimental studies reveal the potential of the propose measure in modelling the cognitive behaviour of process control room operators.by Punitkumar Bhavsar, Babji Srinivasan and Rajagopalan Srinivasa