"Fitness-For-Duty" evaluation using bio-signals

Abstract

In high-reliability systems such as Nuclear Power Plants and the civilian aircraft industry, Fitness-For-Duty (FFD) is often the cause of human-error related accidents/incidents. Although the U.S. Nuclear Regulatory Commission (NRC) highlighted the importance of NPP worker’s FFD to ensure personnel reliability, current FFD programs only address drug and alcohol testing and fatigue management. However, today’s healthcare bio-signals technology makes it possible to monitor the physical and mental state of humans. Thus, the objective of this thesis is to develop an FFD evaluation method using Electroencephalogram (EEG), Electrocardiogram (ECG) and Galvanic Skin Response (GSR) signal indicators to identify potentially-at-risk workers, especially those with unstable psychological distress, drug abuse and a sleep-deprived worker. Bio-signal data was collected from 6 different categories: Normal healthy group, Alcohol-use group, Sleep-deprived group, Stress/heavy workload group, moderate depression and anxiety group. Three tests were conducted, on a total of 124 subjects (not all participating in each test). The EEG, ECG and GSR signals were recorded during eyes closed and eyes open resting status. These subjects performed experimental tasks that included multitasking, working memory, attention, speed, problem-solving and cognitive flexibility. The cognition tasks were directly related to an NPP operators’ duties. The first experiment identified significant bio-signal indicators to classify worker’s psycho-/physiological/alcohol--use status. A Support Vector Machine (SVM) classification model, specific to this research, was developed using data from first experiment. The database from the second verification test was used to evaluate the SVM classification model’s prediction performance. After determining the fitness classification, a Bayesian Network Model (BNM) was developed to assess each subjects FFD for each job duty by calculating work efficiency. In addition, another BNM was developed to reveal the relationships between job stressors and fitness status. This can be used to find the cause of Psycho-/physiological impairment. As a result, the bio-signal indicators showed a statistically significant difference between at-risk workers and healthy workers. The performance of the newly developed SVM and BNM models were also reliable when identifying worker’s fitness status and evaluating their work efficiency. These results can be applied directly to FFD monitoring systems of nuclear power plants as well as other high reliability fields, such as aerospace, military and transportation.