CFD Model Based Ann Prediction of Flammable Vapor Colour Formed by Liquid Hydrogen Spill

Unintended releases can occur during the production, storage, transportation and filling of liquid hydrogen, which may cause devastating consequences. In the present work, liquid hydrogen leak is modeled in ANSYS Fluent with the numerical model validated using the liquid hydrogen spill test data. A three-layer artificial neural network (ANN) model is built, in which the wind speed, ground temperature, leakage time and leakage rate are taken as the inputs, the horizontal diffusion distance and vertical diffusion distance of combustible gas as the outputs of the ANN. The representative sample data derived from the detailed calculation results of the numerical model are selected via the orthogonal experiment method to train and verify the back propagation (BP) neural network. Comparing the calculation results of the formula fitting with the sample data, the results show that the established ANN model can quickly and accurately predict the horizontal and vertical diffusion distance of flammable vapor cloud relatively. The influences of four parameters on the horizontal hazard distance as well as vertical hazard height are predicted and analyzed in the case of continuous overflow of liquid hydrogen using the ANN model.