Dispositivo de detecção de incêndios incipientes em secadores de grãos através de técnicas de Redes Neurais Artificiais (RNA)

Abstract

In recent decades, Brazil has consolidated as one of the largest grain producers in the world. To ensure product quality during storage and processing, grains undergo a drying process using grain dryers. This high-temperature step presents a significant fire risk. Due to their unique propagation characteristics, fires in grain dryers cause product waste and financial losses, posing a technological bottleneck. This study aimed to develop a fire detection system for grain dryers using Artificial Neural Networks (ANN). The methodology included: 1) Characterization of the study element; 2) Analysis of material behavior in the drying process regarding fire causes; 3) Exploration of conventional fire detection methods in grain dryers; 4) Development of an innovative fire detection device for grain dryers; 5) Laboratory-scale operationalization and testing of the device; 6) Industrial-scale operationalization and testing of the device; 7) Evaluation of intellectual protection requirements; 8) Systematization of results. The research resulted in the creation of a fire detection device for grain dryers that can identify early-stage fire situations by monitoring independent variables such as drying air temperature, atmospheric temperature and humidity, grain temperature, moisture and volume of grain mass, and drying speed. The device utilized a feedforward ANN model, MLP 5-16-10-1, with Tanh, ReLU, and Linear activation functions between layers, respectively. Simulations with the proposed model demonstrated the ANN's ability to predict normal operation conditions, identify fire risks, and detect early stage fires.