A Worker's Fitness-for-Duty Status Identification Based on Biosignals to Reduce Human Error in Nuclear Power Plants

摘要

Human error has been highlighted as main cause of industrial and nuclear accidents. One of the key issues related to human error is a worker's fitness for duty (FFD). FFD refers to the mental and physical ability of employees to safely perform their job. The objective of this study is to investigate the feasibility of identifying a worker's FFD status using biosignals. The FFD statuses examined were with respect to alcohol use, depression, stress, anxiety, and sleep deprivation. Biosignals examined in the study include electrical activity in the brain measured by electroencephalogram and referred to as EEG, electrical activity of the heartbeat measured by electrocardiogram and referred to as ECG, galvanic skin response (GSR), blood volume pulse (BVP), dynamic changes in blood pressure and referred to as BPHEG, and respiration. A total of 114 volunteers participated in the study as experimental subjects from whom biodata were collected during their resting states (eyes closed and eyes open). The steps followed in the study include signal preprocessing, power spectrum feature analysis, important feature selection, and support vector machine (SVM) classification using 5-fold cross validation to identify a worker's FFD status. Among the 70 biosignal indicators, important features were selected by Multivariate Analysis of Variance (MANOVA). The best model developed with the SVM used 64 biosignal indicators and showed a binary (fit or unfit) classification accuracy of 99.4% and a multi-classification accuracy of 97.7%. While limitations of the current work remain, the study indicates the possibility of implementing an effective FFD management program to reduce human error in plant operations. A thumbnail sketch of the study is as follows: 1. To reduce human error in nuclear operations, use of biosignals was investigated to identify FFD status of workers. 2. EEG, ECG, GSR, BVP, BPHEG, and respiration signals were used to identify a worker's FFD status. 3. The SVM-based model was successfully implemented for multi-class and binary-class FFD classification.