A MACHINE LEARNING-BASED PREDICTIVE MODEL FOR HAZARDOUS AREA CLASSIFICATION GAS LEAK SCENARIOS

This study proposes a predictive model capable of estimating Hazardous areas in hydrogen leaks by integrating Computational Fluid Dynamics (CFD), computer vision, and Artificial Intelligence. The methodology employed image processing in the HSV color space to determine the bodies of the simulated gas clouds and identify their spatial distribution, generating a dataset for training an Artificial Neural Network (ANN). The model, trained using 36 distinct areas across 72 leakage scenarios, yields the identified Hazardous zones as its output. The results indicated high performance in distinguishing between hazardous and safe areas, demonstrating the technique as a valid alternative for gas dispersion monitoring.