Data-Driven Fire Safety Management at Building Construction Sites: Leveraging CNN

Abstract

Fire safety management on site is important during the implementation of construction projects. However, many factors have caused fires at construction sites, where workers are in close proximity and large amounts of materials and machinery are stored. Traditional smoke- and temperature-based sensors cannot be used because of the open-environment conditions and environmental complexities of construction sites. Moreover, traditional fire management on site mainly relies on artificial patrol mode, which is inefficient. Most previous studies focused on traditional real-time fire monitoring of constructed buildings. Therefore, a new, intelligent, and effective method should be developed for real-time fire monitoring of construction sites. The current study proposed a data-driven approach based on convolutional neural network (CNN), which is suitable for various construction environments and can recognize real-time fires on site. This research built a fire-recognition model and developed a real-time construction fire detection (RCFD) system. Experiments were conducted to verify the applicability of the proposed system in different environmental conditions. Experimental results showed that the fire detection model based on the CNN algorithm can be applied to various field construction environments, and the recognition accuracy is above 90%. This study used a data-driven method to solve the problem of construction fire safety management. Results indicate that the RCFD system can guide project teams in the timely detection of fires on construction sites, improvement of safety management efficiency, and reduction of fire-related losses.