Conclusions on the conditions in the reactor sump can in principal be drawn from the experi­mental results by two means, by applying scaling laws, and by applying CFD methods. Scaling laws were applied by Knebel & Muller (1997) to determine from the 2d experiments that in the first days the temperatures in the water above the core melt will be large enough to achieve boiling. As a consequence, the investigations in SUCOS are only applicable for the long term cooling after about ten days. A condition for applying scaling laws is, that the physical trans­port phenomena are of similar relevance in the model experiment and in the reactor sump. The analyses of SUCOS-2D in Carteciano et al. (1999) have shown that specific 3d effects, like the
feed water pipes to the coolers, the bolts, and screws going through the fluid domain strongly affect the experimental results. In the small cross section of the SUCOS-2D slab geometry these structures become more important for the heat transfer and for the flow resistance than in the huge reactor sump. Thus, the results of this experiment cannot be used for scaling up to predict accurately the temperatures in the reactor sump. In contrast, the first numerical cases, which did not record those structures, are more feasible as a basis for this extrapolation. The experimental results of SUCOS-3D could be a better basis because there the cross sections oc­cupied by those structures are of relatively less importance, but there the above mentioned strange flow distribution was found which is specific for this experiment and not for the reactor sump.

Scaling up by CFD-tools is also not free of serious problems in this context. On one hand the SUCOS tests could successfully be interpreted with the FLUTAN code. Such CFD codes have the flexibility to tackle all the experiment or reactor specific structures. On the other hand the model experiments showed laminar flows, whereas the reactor sump will have turbulent flow conditions (Knebel & Muller 1997). Hence additional investigations are necessary by experi­ments of similar flow types to validate the turbulence models and boundary conditions used with the turbulence models. Purely buoyant flows are currently a challenge for any turbulence model, see e. g. Hanjalic (1994, 1999). Also the TMBF, which is explicitly developed for buoy­ant flows, has up to now only been validated for two-dimensional forced, mixed, and natural convection (Carteciano et al. 1997, 1999b). An additional feature, which is inherent to most purely buoyant flows, is its local time dependence. The cold plumes plunging down through the chimney are a low frequent phenomenon which was also not completely filtered out in the experiment by time-averaging over two minutes. This causes the wall heat flux on the copper plate to change in time, Fig. 4. And also the hot plumes rising from the surface of the copper plate are no stationary phenomenon. The surface temperature on the copper plate, Fig. 10, is very straggly as it is typically found in Rayleigh-Benard convection in which the hot plumes rise mainly from the knots of those straggly structures (Worner & Grotzbach 1997).

As a consequence of the 3d and time-dependent nature of buoyant flows and of the current status of development of the standard turbulence models, there is currently no way to achieve reliable results by standard models on the temperature fields in the reactor sump. The only way which is nowadays often considered to give a better solution for 3d time-dependent flows is to apply Large Eddy Simulation methods (LES). Indeed, there exist already several applications of LES to reactor typical flows; for an overview see Grotzbach & Worner (1999). These show the tremendous potential of the LES method and that its possibilities are going far beyond those of standard Reynolds averaged turbulence models. The main problems which need to be solved for LES methods in this context are e. g. the development of more universal subgrid scale models, of boundary conditions for buoyant flows, and of numerical methods in commer­cial codes that fulfill LES requirements.

5. CONCLUSIONS

The former numerical interpretation of SUCOS-2D experiments with the FLUTAN computer code showed that good agreement between experiment and calculation can be achieved when local thermal disturbances like the feed water pipes to the coolers are recorded in the simula­tion. Here, the single phase natural convection experiment SUCOS-3D is interpreted. The nu­merical results confirm, natural convection in large pools, in which pressure drops are negligi­ble, is very sensitive against small disturbances. Here, some other discrepancies are found by analyzing the experimental results from SUCOS-3D: In these experiments the maximum tem­perature did not occur far above the horizontal coolers as in SUCOS-2D, but below the tilted roof, and the heat fluxes over the horizontal and vertical coolers were not homogeneously dis­tributed. Comparable pool temperatures could only be achieved numerically by using the measured heat fluxes at the coolers instead of temperature boundary conditions, but the calcu­lated flow pattern was still different. One explanation for this strange result is supported by the experimental data, this is to postulate that one of the horizontal coolers was slightly tilted against the main flow direction. Additional numerical investigations indeed show that a slope of only one percent would explain the experimental flow field. From this problem of the ex­periment one can learn how to improve this sump cooling concept: Foreseeing a small slope of the horizontal coolers downwards in the expected flow direction would stabilize the flow and would drastically increase the efficiency of the horizontal coolers.

Based on the performed numerical investigations it can be concluded that a transformation of the experimental results from SUCOS-2D and -3D to reactor sump conditions by means of scaling laws is questionable. Such transformation will only be possible by applying well vali­dated CFD-Codes and experienced code users with a sound physical and engineering back­ground. Current standard turbulence models form the working basis for engineers, but they can only be used for approximate predictions, because the statistical models fail for buoyant flows which are locally time-dependent. The only more reliable solution could come from adequate Large Eddy Simulation methods and LES-suitable codes for complex geometries.

[1] .