Improvement of large-scale hydrogen combustion models using detailed measurements of flame front properties

International audienceDuring the course of a severe accident in a nuclear power plant, combustible gas, i.e. hydrogen, can be released into the containment atmosphere initially filled with air. Explosion mitigation measures in large dry containment can reduce the size and the duration of occurrence of combustible gas cloud, but none of them can avoid the possibility of having an accelerated flame during the combustion process. The Three Mile Island accident in US and, more recently, the Fukushima accident in Japan have recalled the devastating effect of large scale combustion events.As a result, the study of large scale combustion processes remains an important issue for nuclear safety demonstration for present and future Pressurized Water Reactors (PWR). Several large scale models have been developed worldwide with the capability to model hydrogen flame acceleration phenomena. Despite a huge validation effort most of the models are not able to predict the combustion behavior in a blind exercise. The main reason is the difficulty to capture the detailed flame acceleration processes and the underlying phenomena i.e. laminar flame properties, flame instabilities, interaction between combustion and turbulence, flame shock interaction, deflagration-to-detonation (DDT) transition and detonation itself.In the open literature flame acceleration experiments usually provide pressure transient and flame arrival time at a limited number of points in the experimental facility. The latter is used to compute the local flame visible velocity (VF) corresponding to a mean velocity between the measurement points. In the past, the main objective of the numerical studies was to predict the maximum pressure occurring during the combustion process. In general, a rough estimation of this maximum can be computed by pure thermodynamic analysis such as Adiabatic Isochoric Complete Combustion (AICC) or Chapman-Jouguet (CJ) detonation characteristics. Detailed structural dynamic analysis needs not only the maximum pressure but also the whole pressure transient in order to estimate the pressure impulse transferred to the structure. As a consequence, advanced numerical models must be able to accurately predict this local pressure time history.Generally, the knowledge of VF versus time allows the code modeler to recover the pressure transient. Unfortunately, VF is not a parameter of the combustion model and trial an error tests on the combustion model parameters are necessary to obtain the correct flame visible velocity. The main reasons for this are as follows. If one considers a 1D flame propagation process, the flame visible velocity is the sum of a fundamental flame velocity ST and the gas velocity in front of the flame U. The former usually contains the combustion model parameters. Acting on ST will change U and, consequently, VF in a nonlinear way. In a general 3D analysis another variable arises, such as the flame surface AF. Thus, the prediction of the pressure transient in an advanced combustion model requires an accurate computation of AF, ST and U in each cell of the numerical mesh. None of the existing combustion experiments can provide simultaneously these three variables locally. Therefore, a correct VF can be obtained with different combinations of these three parameters without any possibility of the selection of the relevant ones.Based on the previous remarks, new experiments were conducted by CEA and NPIC in the SSEXHY facility to measure these variables. In parallel, computational analyses were performed to check the predictive capabilities of the combustion models and also to allow the code developers to improve their numerical models. The present article reports these two issues