Within the course of a hypothetical severe accident in a nuclear power plant, hydrogen can be generated in the primary circuit and released into the containment. Considering the possibility of a deflagration, the simulation of the hydrogen distribution in the containment by computer codes is of major importance. To create a database for code validation, several distribution experiments using helium and hydrogen have been performed in the German Thermal Hydraulics, Hydrogen, Aerosols, Iodine (THAI) facility. The experiments started with the TH13 test, which was the base of the International Standard Problem exercise (ISP-47). TH13 was followed by the Hydrogen-Helium Material Scaling (HM) test series conducted within the Organisation for Economic Co-operation and Development/Nuclear Energy Agency (OECD/NEA) THAI project. The objectives of the HM tests were (a) to confirm the transferability of existing helium distribution test data to hydrogen distribution problems and (b) to understand the processes that lead to the formation and dissolution of a light gas cloud stratification. The HM-2 test was chosen for a code benchmark.

During the first phase of the HM-2 test, a light gas cloud consisting of hydrogen and nitrogen was established in the upper half of the facility. In the second phase, steam was injected at a lower position inducing a rising steam-nitrogen plume. The plume did not break through the cloud because its density was higher than the density of the cloud. Therefore, the cloud was gradually dissolved from its bottom.

Eleven organizations performed blind calculations for the HM-2 experiment. The lumped parameter (LP) codes ASTEC, COCOSYS, and MELCOR and the computational fluid dynamics (CFD) codes FLUENT, GASFLOW, and GOTHIC were used. The main phenomena were natural convection, interaction between the rising plume and the light gas cloud, steam condensation on walls, fog behavior, and heat up of the walls. The experimental data of the first phase were published, and the atmospheric stratification was simulated reasonably well. The data from the second phase stayed concealed until the simulated results were submitted. The thermal-hydraulic phenomena were well predicted by several LP and CFD contributions, whereas the time intervals needed to dissolve the light gas cloud were either underpredicted or overpredicted. However, the other LP and CFD contributions showed larger deviations in the measured data. Reasons for deviations were identified, and model improvements were demonstrated in open posttest calculations. In this article, the experiment, the benchmark results, and the simulation features are described, and recommendations for code users are given.