Identification of operator behavior through signal detection theory in a multitasking environment
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT REQUEST OF AUTHOR.] Nowadays, control rooms in oil and gas refineries consist of multiple complex human-machine systems. An operator in a control room observes numerous different control loops concurrently while performing other attention-demanding tasks. Although the systems generate lots of data, the amount of information transmitted to the operator is always smaller than the stimulus information. Hence, it is necessary to understand how operators reform their detectability under multitasking environment, to improve control conditions and avoid mistakes caused by missing the critical information. In this paper, we used Signal Detection Theory (SDT) as a tool to evaluate human performance in a multitasking environment. The primary objective of using SDT was to assess operator's sensitivity (d'), and bias (beta) in process monitoring environment. In addition, NASA-TLX was used to measure participant's workload under different complexity scenarios. NASA-TLX is a multi-dimensional rating task that derives an overall workload score based on a weighted average of scores on six subscales. These subscales comprise of mental demands, physical demands, temporal demands, own performance, effort, and frustration. During the experiment, participants were asked to detect abnormal and alarm signals on a gauge monitoring display as primary task. Meanwhile, they also needed to perform Multi-Attribute Task Battery (MATB) task at the same time. The gauge-monitoring screen contains the total of 52 gauges (flow, level, temperature, and pressure). The display design is based on Abnormal Situation Management(TM) (ASM(TM)) Consortium (www.asmconsortium.org). The MATB task consists of alarms detection system monitoring, target tracking, and dynamic resource management. Each task requires a different cognitive resource, such as visual searching, target tracking, and diagnostic control. Every scenario was developed based on the actual refinery operation. Participants experienced two levels of complexity (low and high) scenarios during the experiment. A total number of events in the high complexity scenario are twice larger than the low complexity scenario. There were 12 alarm events in low complexity and 24 alarm events in high complexity scenarios. The MATB tasks were used to keep the balance of complexity in each scenario. For example, when the number of alarm events was decreased, the number of events in MATB tasks was increased. The experiment was conducted as a two-factor experiment with repeated measures, within-subject factor: scenario complexity (low vs. high), between subject factor: display type (overview vs. schematic), and a multi-session study, which continued for five days. Day 1 was a training session, and Day 2 and 3 were practice sessions. The data from Day 4 and 5 was used to analyze operator's sensitivity and bias. A statistical model was developed to analyze the performance under different complexity scenarios. In overview display, the results support that participants showed various levels of sensitivity (d') in the gauge-monitoring task based on the degree of task complexity. When the scenario complexity was low, the sensitivity (d') went to high. However, there was no significant effect of the scenario complexity on the bias. One of the findings was the scenario complexity had a significant influence on the tracking task of MATB display. The tracking task performance was better under the low complexity scenarios. However, the results of the system monitoring and resource management were not significantly influenced by the complexity. On the contrary, in the schematic display, complexity influenced operator's bias but not sensitivity. Furthermore, MATB performance results showed that besides tracking task performance, complexity also influenced resource management task performance. Another finding was the negative correlation between sensitivity and mental workload. As the increasing of complexity level, the sensitivity went worse, and the mental workload was increased. From display design comparison, it is also shown that operator detected better signal when overview display was applied. Hence, from these results, we concluded that overview display might provide a better monitoring performance compared to schematic display. The outcomes of this study will help researchers to predict the sensitivity and bias of human operators by using the Signal Detection Theory (SDT) analysis method. Moreover, this study will also benefit industries in developing a proactive monitoring environment that can improve operator detectability.
Access to files is limited to the University of Missouri--Columbia.