Classification of construction hazard-related perceptions using: Wearable electroencephalogram and virtual reality
Introduction
The construction industry has recorded a large number of fatalities every year. According to surveys conducted by the U.S. Bureau of Labor Statistics (BLS) during the last five years (2015–2019), the number of annual fatalities in the construction industry has increased by 13.23%, from 937 in 2015 to 1061 in 2019 [1]. Also, the construction industry accounted for 20% of all fatalities in the private sector in 2019.
To improve construction workers' safety, hazard identification—a process to identify potential hazard factors that can pose adverse effects (e.g., accidents) on workers—has become a critical component in safety management [2,3]. In the current practice, safety managers usually perform manual inspections on a construction site based on the procedure, which is known as job hazard analysis [4]. However, previous studies revealed that a significant amount of construction hazards still remain unidentified or not well-assessed at the jobsite, continuously exposing workers to unsafe working conditions [2,5,6]. Such poor hazard identification performance is due to the subjectivity and inconsistency in the inspector's hazard identification abilities [7], a limited number of resources (e.g., inspectors) to inspect the large area [8], and the dynamic nature of construction environments which decrease the inspector's ability to recognize hazards [9].
Extensive research efforts have been made to improve hazard identification performance by analyzing the physical and physiological responses of humans measured via wearable sensors. The underlying rationale is that workers show unusual or abnormal responses when they encounter a hazard [[10], [11], [12]]. Physical responses (e.g., gait and stride) have been studied by employing wearable sensors (e.g., inertial measurement units), and the relationship between the existence of hazards and changes in physical responses has been demonstrated [[13], [14], [15]]. Similarly, physiological responses (e.g., eye movement, brain activity, and electrodermal activity) have been investigated to infer and locate the hazards in the surrounding environment [[16], [17], [18]].
Among them, analyzing electroencephalogram (EEG)—a method that can detect and evaluate brain activities—offers unique opportunities since it contains direct and critical information associated with human perception, cognition, and intention [19,20]. In addition, the recent advancement of wearable sensing technology has enabled the real-time collection of EEG signals and the application of wearable EEG in many real-world scenarios (e.g., construction jobsite). Within the construction safety domain, EEG has been increasingly used to measure, assess, and monitor varying mental aspects of construction workers in terms of vigilance, distraction, emotional state, stress, and mental workload [[21], [22], [23], [24], [25], [26]]. Several researchers have attempted to identify construction hazards by adopting wearable EEG technology [16,27].
Despite the contribution of previous research efforts, there remain two limitations in the current knowledge base. First, how EEG signals collected from the wearable device can be utilized to locate and identify construction hazards remains unclear. Most previous EEG studies focused on assessing the changes in human mental status (e.g., emotional states, fatigue, and stress) induced by the construction hazards. Although changes in mental status have the potential to be used to identify hazards, to what extent of such changes can be used to classify EEG signals associated with a hazard or no-hazard has not been revealed. Also, considering that EEG signal is the first response of human hazard perception, its direct application to classify hazards needs to be studied. Second, most previous studies are based on the experimental setting that uses planar stimulus; many researchers used static images to simulate construction hazards and displayed them to subjects to analyze their influence on the human risk perception. However, such an approach often leads to information distortion, resulting in a significant gap between the indoor experiment and the real-world construction site [28]. Considering that wearable EEG holds great potential to be applied to the real-world jobsite, there is a critical need to conduct an experiment in a more realistic setting (e.g., immersive virtual environment) to understand the impacts of hazards on workers' perception.
In this context, the paper investigates the feasibility of identifying construction hazards by developing a wearable EEG classifier that can classify EEG signals associated with perceived hazards in the virtual reality (VR) environment via the methodology, which comprises the following five steps. First, various types of construction hazards (e.g., fall, struck-by, etc.) are simulated in the VR environment. Second, EEG data is collected in VR-based laboratory experiments. Third, collected data is preprocessed to remove signal noise and artifacts. Fourth, EEG features are extracted from the preprocessed data. Fifth, classifiers are developed based on each of the fifteen most widely used machine learning (ML) algorithms. Then the performance of different classifiers are compared, and important observations are discussed. The results of this paper will contribute to the ubiquitous hazard identification based on wearable EEG devices and thereby improve the current hazard identification capability.
Section snippets
Current construction hazard identification practice
Hazard identification is one of the most effective approaches for safety management in the construction industry [2,29]. In the current practice, hazard identification relies on human inspection; safety managers perform manual inspections on construction sites [30]. To improve the inspector's hazard identification ability, job hazard analysis and safety programs have been developed and proposed by the Occupational Safety and Health Administration (OSHA) [31,32]. However, a large number of
Simulation of construction hazards in VR
Five categories of construction hazards—which were identified as the leading cause of injuries and fatalities in the construction industry by the Occupational Safety and Health Administration (OSHA) [45,46]—were simulated in the VR environment. Table 2 presents the five hazard categories (fall, slip and trip, struck-by, chemical and electrical, etc.), each of which contains a varying number of scenarios.
For example, five different scenarios from No. 1 to 5 in Table 2 were developed to
Comparison of 15 EEG classifiers' performance
To compare and evaluate the performance of fifteen EEG classifiers, four metrics (accuracy, precision, recall, and F1-score) were adopted. Classification accuracy, the percentage of correctly classified observations (classified vs. actual class) with respect to the total number of observations, is the primary measure of classification performance (Eq. (8)). Precision is defined as the fraction of EEG signals correctly classified as signals associated with the perceived hazard (Eq. (9)). Recall
Conclusion
This paper investigated the feasibility of identifying construction hazards by developing a wearable EEG classifier based on the experiments conducted in the immersive VR environment. Five types of hazards (fall, slip and trip, struck-by, chemical and electrical, etc.), identified by OSHA as majors construction hazards, were simulated in thirty VR scenes by consulting OSHA Safety and Health Regulations for Construction. EEG signals were collected from twenty-eight subjects who wore both
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (69)
- et al.
Construction job safety analysis
Saf. Sci.
(2010) Workplace hazard identification and management: the case of an underground mining operation
Saf. Sci.
(2013)- et al.
Occupational risk assessment in construction industry – overview and reflection
Saf. Sci.
(2011) Velocity versus safety: impact of goal conflict and task difficulty on drivers’ behaviour, feelings of anxiety, and electrodermal responses
Transp. Res. Part F Traffic Psychol. Behav.
(2012)- et al.
Inferring workplace safety hazards from the spatial patterns of workers’ wearable data
Adv. Eng. Inform.
(2019) - et al.
Validating ambulatory gait assessment technique for hazard sensing in construction environments
Autom. Constr.
(2019) - et al.
Feasibility analysis of electrodermal activity (EDA) acquired from wearable sensors to assess construction workers’ perceived risk
Saf. Sci.
(2019) - et al.
New insights into rhythmic brain activity from TMS–EEG studies
Trends Cogn. Sci.
(2009) Where does EEG come from and what does it mean?
Trends Neurosci.
(2017)- et al.
Monitoring distraction of construction workers caused by noise using a wearable electroencephalography (EEG) device
Autom. Constr.
(2021)