CFD and machine learning based hybrid model for passive dilution of helium in a top ventilated compartment

Abstract

Abstract For hydrogen management in the containment of Nuclear Power Plants (NPPs), besides the Passive autocatalytic Recombiners (PAR), the passive dilution of lighter gas plays an important role. This could be an attractive option to optimize the containment design and to estimate the extent of dilution. Passive dilution has many other applications in nuclear industry. The experimental studies of air entrainment in the upward rising helium plume and the resulting dilution of helium gas by the Canadians in terms of Volume Flow Magnification Factor (VFMF) have been utilized for Computational Fluid Dynamics (CFD) validation. The CFD based Fire Dynamics Simulator (FDS) predicted values of VFMF found to be in good agreement with the test data. After FDS code validation, parametric study has been carried out to generate a data base of VFMF for range of hydrogen injection, side opening area and opening height. In present study various Machine Learning (ML) models are evaluated based on two-parameter relationship i.e. non dimensional hydrogen injection and VFMF using the CFD code generated database. The trained ML models were used for the predictions of the mass flow rate of gas entrainment (through opening) in the rising buoyant plume in terms of VFMF. The ML predictions were in good agreement with the predictions against test data. Multivariate Adaptive Regression Splines (MARS) based ML model found to performed best and discussed in the paper. The paper highlights details of methodology of numerical simulation, results of the CFD studies and machine learning based predictions.