Near Field Atmospheric Dispersion Modeling on an Industrial Site Using Neural Networks

Assessment of likely consequences of a potential accident is a major concern of loss prevention andsafety promotion in process industry. Loss of confinement on a storage tank, vessel or piping on industrialsites may imply atmospheric dispersion of toxic or flammable gases. Gas dispersion forecasting is adifficult task since turbulence modeling at large scale involves expensive calculations. Therefore simplermodels are used but remain inaccurate especially in near field of the gas source. The present work aims tostudy if Neural Networks and Cellular Automata could be relevant to overcome these gaps. These toolswere investigated on steady state and dynamic state. A database was designed from RANS k-İ CFD andGaussian plume models. Both methods were then applied. Their efficiencies are compared and discussedin terms of quality, real-time applicability and real-life plausibility