Quantifying situation awareness of control room operators using eye-gaze behavior

Highlights

• Loss of operator’s situation assessment is a typical precursor to process safety incidents and accidents.

• We seek to develop proactive strategies to assess operators’ situation assessment.

• Operator’s eye gaze behavior is a direct indicator of operator’s cognitive processes, especially during abnormal situations.

• A entropy based measure is proposed to quantify eye gaze dynamics.

• Experimental studies are conducted which validate the proposed entropy measure.

Abstract

In an attempt to improve process safety, today’s plants deploy sophisticated automation and control strategies. Despite these, accidents continue to occur. Statistics indicate that human error is the predominant contributor to accidents today. Traditionally, human error is only considered during process hazard analysis. However, this discounts the role of operators in abnormal situation management. Recently, with the goal to develop proactive strategies to prevent human error, we utilized eye tracking to understand the situation awareness of control room operators. Our previous studies reveal the existence of specific eye gaze patterns that reveal operators’ cognitive processes. This paper further develops this cognitive engineering based approach and proposes novel quantitative measures of operators’ situation awareness. The proposed measures are based on eye gaze dynamics and have been evaluated using experimental studies. Results demonstrate that the proposed measures reliably identify the situation awareness of the participants during various phases of abnormal situation management.

Introduction

Process safety is a major concern in the chemical and allied industries. Accidents large and small torment plants regularly; their annual cost is estimated to be in the millions, even for medium-sized facilities (Mannan, 2004). Over the last three decades, numerous interventions have been made by governments and industries around the world to improve process safety. Despite these, there is no significant abatement in accident occurrence – a recent survey (Marsh, 2014) reported that 25% of the accidents that led to largest losses in the hydrocarbon industry over a period of 40 years happened in the last 5 years from 2009. There has however been a notable change over time in the key contributory causes of accidents. In the early days, inadequate system reliability and insufficient understanding of process phenomena were the key reasons leading to accidents. More recently, process plants widely use highly reliable systems with sophisticated automation and control strategies. Statistics show that the predominant root cause of accidents in the process industry now is human error (Mannan, 2004). An analysis (Sepeda, 2006) of over 80 incidents revealed that human-related factors contribute significantly to incidents. Symptoms of human factors related deficits in a plant include stress (Rasmussen and Laumman, 2014) on operators and shift supervisors, which in turn translates to slips, lapses, mistakes, or violations and cause equipment outage, plant shutdown and various production accidents (Kidam et al., 2010). This highlights the need to develop a deeper understanding of human error in the process industry and new techniques to prevent them (Gordon and Rachael, 1998).

Traditionally, human errors have been accounted for during risk assessment of the process design. Various types of human failures and their expected probabilities of occurrence were incorporated in Process Hazard Analysis (Munger et al., 1962, Swain, 1990) i.e., human error was viewed as the initiating event of incidents using likelihood approaches, similar to the way that a piece of hardware is expected to fail at some frequency. The role of the human in any complex system such as a process plant has evolved over the last three decades from being predominantly manual (physically ‘doing’ a task) to being predominantly cognitive (requiring ‘thinking’). The nature of human errors that affect the system’s safety has therefore evolved, as exemplified by the Three Mile Island Nuclear Plant accident (Le Bot, 2004). Rasmussen suggested a model of behavioral shaping mechanisms to model failures in human performance using cognitive science concepts (Rasmussen, 1997). Motivated by these, effort has been focused on designing ergonomic control rooms or user-centered design of human-machine interfaces (Cochran and Bullemer (1996)). A number of risk assessment techniques have also been developed (Chang and Mosleh, 2007, Pate-Cornell et al., 1996). There has however been little effort to understand the cognitive state of operators in real-time during process operations and its impact on process safety (Kodappully et al., 2016, Sharma et al., 2016).

Recent advancements in the biomedical arena has led to various tools such as Electroencephalography (EEG), functional Near Infrared imaging (fNIR), Galvanic Skin Response (GSR) and eye tracking which could be used to understand the cognitive behavior of humans. Among these, our research has focused on eye tracking since it is non-invasive and can be relatively easily deployed in a control room (Kodappully et al., 2016, Sharma et al., 2016). Our previous studies demonstrated qualitatively that during process operations, the control room operator’s gaze on the Human-Machine Interface can reveal the extent of his situation awareness and thus his ability to handle any disturbance successfully before it escalates into an accident. The current work aims to develop quantitative measures of eye gaze behavior that can supplement the qualitative understanding developed in our prior work. The rest of this article is organized as follows. Section 2 presents an overview of eye tracking and reviews the literature on various attempts to quantify the gaze behavior of operators in various domains. The proposed entropy measures used in this work for quantification of eye gaze behavior are discussed in Section 3. Section 4 reports three potential applications of the proposed entropies as well as detailed results from large-scale studies.