Quantitative Risk-Informed Sensor Placement Optimization for Chemical Leak Detection in a Complex Laboratory Layout Using the CFD-ANN Approach

The rapid and accurate localization of hazardous chemical leak sources is critical for mitigating environmental damage, protecting public health, and ensuring an effective emergency response. However, the conventional source localization method can be slow and inefficient. To address this challenge, this research proposes and develops a novel integrated system leveraging computational fluid dynamics (CFD), sensor network optimization, and artificial neural networks (ANN) for the precise localization of chemical sources, focusing on ethanol leaks within a laboratory environment. The dispersion of ethanol gas was simulated using CFD, and the simulation data were used for sensor optimization. The sensor system, consisting of 10 sensors at vital locations, is the foundation for the development of the ANN-based source localization method. The environmental factors related to gas dispersion considered were wind speed, wind direction, and temperature, as well as the presence of exhaust and air supply systems. A total of 4 leakage points were studied. The concentration data measured by the sensor system were used to train the ANN to identify the most probable location of the ethanol leak. The model achieved a validation percentage of 87.5% and an average error of 0.0001% in determining the ethanol leak location of four release points in the study area. The findings demonstrate that this combined CFD-ML approach offers a powerful and efficient tool for improving emergency response protocols, enhancing safety measures, and mitigating potential financial losses during chemical incidents.