Analysis and performance of IoT-Based framework for fire incidence applications

Abstract

Natural disasters are one of the leading causes of death worldwide. Statistics show that many people die in various catastrophic events. The damage to property and infrastructure from these types of disasters is worth millions of dollars, and many more dollars are spent on disaster recovery. In many countries, including the East African region, indoor fire outbreaks are the most common natural disasters with fatalities. Least developed or developing countries do not have efficient mechanisms to detect, count or predict human occupants in buildings in the event of a fire. In Rwanda, common strategies to reduce the risk of fire outbreaks include emphasizing the use of firefighting equipment and following fire codes to enable firefighters to intervene and rescue a reasonable number of casualties. These methods do not guarantee that trapped people have the skills to use the device or know these guides in case of panic. The internet of things and machine learning are now available. The former will play a key role in developing data collection prototypes to collect and monitor environmental parameters and report them to the cloud in real-time. The latter helps predict occupancy using sensor-captured data from indoor environments. This research aimed to reduce the risks of fire outbreaks in public buildings by identifying and analyzing the environmental factors to be considered as indicators for fire existence, and the recent technologies adopted to fight against the matter. After identification of parameters and devices capable to support on fire detection and human counting, the development of a scalable spruce fire IoT-based system aiming to warn and predict humans in case of fire is done. The system uses wireless sensors to collect real-time environmental data, report estimated occupancy, and then predict the human presence. By using the developed IoT system, the building environmental parameters were collected, and machine learning techniques have been used to predict human presence. The IoT framework was containerized to support system scale-up and performance metrices were assessed. The research was conducted in three systematic phases: The data acquisition IoT prototype was developed, calibrated, and tested to gather real-time data to be used in the prediction. In this phase, the environmental parameters’ thresholds were experimentally captured. In the second phase, the IoT framework has been improved with machine learning model integration. The data

xxvi generated was used to build the insight between people’s presence and the environmental parameters in a room using machine learning prediction techniques. Different machine learning models were tested, and the long and short-term memory model was selected with accuracy of 96%. In the third phase, the framework was designed using the microservices paradigm, to analyze the performance during the increase of the distributed sensor nodes over the same managerial process. The performance was measured experimentally on 3 nodes by varying different requests 0, 100, 10.000, 100.000, and 1.000,000 respectively. Using Grafana and Prometheus, the system resources considered are CPU utilization, memory usage, as well as data transmission (receiving and transmitting bytes). The results showed that due to the containers, the CPU used did not cost high by increasing the number of requests. For memory resources, the increases do not cost any attention. The transmission increases considerably to the bytes received starting from 10.000 requests. The system prototype was successfully tested at the selected site and shows the better performance in case of scaling up. The study area of this research is the University of Rwanda, Kigali City, and Nyarugenge districts in Rwanda.