AI-Based Early Fire Detection & Alert System
Department of Artificial Intelligence and Data Science, Ajeenkya DY Patil School of Engineering, Pune, India · India
Abstract
In factories, bakeries and warehouses, as well as in homes, abnormal heating or smoke can trigger fires that are not caught by the fixed-threshold monitors until it is too late, especially if high temperatures are normal. The design of an AI-powered, IoT based early fire risk detection and alert system using an ESP32 microcontroller, a DHT22 temperature and humidity sensor, an MQ-2 smoke and gas sensor, an infrared flame sensor and a buzzer and LED for local alerts is presented. The system continuously takes readings from the environment, checks the conditions of the thresholds and temperature trends, determines the risk level, and uploads readings to the ThingSpeak platform for remote monitoring. To learn the normal behavior of the sensors, and detect anomalous patterns, an unsupervised algorithm called Isolation Forest is proposed, which uses time-dependent features like the speed of temperature change and the rolling statistics. The current prototype includes a pre-defined threshold-based decision logic, Wokwi simulation and data transmission to ThingSpeak, and the Isolation Forest component is still being developed and has not yet been trained and implemented. So, no accuracy, response time or false alarm rates are reported, accordingly. The paper describes the architecture, the decision logic, and the status of implementation and an evaluation plan with respect to a baseline defined only by thresholds. The system is a prototype for research purposes only and is not an alternative to certified fire-safety devices.
Keywords
How to Cite This Article
Rights
BODHIVRUKSHA JOURNAL OF DIVERSE DISCIPLINE