Optimizing Sensor Placement for Hydrogen Leak Detection in Enclosed Infrastructure: A Comparative Study Using CFD-informed Genetic Algorithm and DeepSets Neural Surrogate

A multi-objective genetic algorithm optimized with a DeepSets neural surrogate identifies sensor locations to achieve a 96.1% hydrogen detection rate within 60 seconds. The framework uses a CFD database of 180 scenarios to reduce blind areas to 0.12% in 1,350 m3 enclosed spaces across variable leak rates and ventilation levels.