The increasing need for more accuracy of the codes and more detailed descriptions of new cooling concepts leads to a tendency to use at least for local detailed analyses more and more 3D and time-dependent computational fluid dynamic (CFD) codes. There are several motivations for using CFD codes in the nuclear field: Some problems can only be treated in 2D or 3D, like the flow-regime determination, thermal stratification, phase separation, and phase distributions in large pools. CFD codes are required if dynamical 2D or 3D phenomena have to be investigated, like the oscillations of velocities, pressures, temperatures, or phase distributions. The relevance of such investigations may come e. g. from ensuring safe heat transfer from all sub-assemblies under natural convection, from the thermal striping induced thermal fatigue, or flow instabilities due to geysering or sloshing, and from the phase and boron distribution dynamics coupled to neutronics.

Finally CFD codes have to be used for scaling up detailed results from model experiments to full-scale reactor conditions. They are also more and more used even for designing experiments and their instrumentation arrangement. They can also reliably be used to study in a more qualitative manner the relevance of certain phenomena in flow and heat transfer problems, so far as the governing physics is included in the equations or in the numerical modelling. Thus, they are also used to determine the dominant physical phenomena of nuclear flow problems, so that finally a reliable efficient numerical model configuration (input) can be formulated for a system code.

Most CFD codes for two-phase or multi-phase multi-component flows are based on the multi­field approach, assuming that all fluids and phases are defined everywhere in the flow field; this means, the fluids are interpenetrating. The spatial distributions of the different fluids or phases are determined by their relative local volume fractions. As a consequence of this method, one has not only a tremendous increase of the number of equations to be solved, but there are also no explicit phase boundaries existing. This kind of modelling is the basis for the typical working tools available. As such flows are physically and topologically complex, such codes give a valuable support to understand in more detail the macroscopic physical phenomena occurring in multi-phase flows.

The physical models in such codes, e. g. to model interfacial phenomena, are rather simple. Therefore, a quantitative use of such CFD codes for two-phase flows is currently limited to mainly homogeneous flows. Modelling improvements are related to bubble or particle diameters which have to be specified. New developments are going to include multi-group concepts of bubble diameter classes or to develop interfacial area concentration equation models which allow up to now for simple flow regimes to calculate the dynamics of the surface of the interface, which is an important parameter in determining momentum, heat, and mass transfer between the phases. This could also help to improve the modelling of bubble coalescence and fragmentation. For other flow regimes than bubbly or droplet flows, the modelling of interfacial phenomena needs serious improvements.

In going from 1D discretisation to 2D or 3D, one has to use turbulence models when turbulent flows have to be investigated. Some of the two — or multi-phase codes have turbulence models for two-phase flows, but our current knowledge and experimental data base on turbulence is very weak: we have only a few turbulence data for the liquid phase in bubbly flow. This basis is not sufficient to adapt, calibrate, and validate these models. Therefore, none of the existing models can reproduce all those data. For other flow regimes, data as well as models are missing; problematic flow regimes will be bubbly, churn, slug, and droplet flows. The turbulence in non-adiabatic flows is characterized by a rapid change of bubble diameters which will influence the turbulence characteristics. Another related problem occurs in the missing wall functions and in the boundary conditions for the interface phenomena at free surfaces. Thus, the turbulence modelling in two-phase flows can only be accepted as a first step. The models can be used strictly only in an interpolative manner, this means in a parameter range, in which extensive validation was done before. R&D is necessary to improve the models and to extend them for all flow regimes, so that more reliable prediction capabilities can be achieved.

Despite these model deficiencies such codes were rather successful, because many multi­phase problems are governed by large-scale interfacial phenomena. Therefore, most development in the past was based on developing adequate numerical methods to solve the set of equations of multi-field models. The challenge here comes from the weak compressibility of the fluids or phases engaged and from the strong density variations from liquids to gases across interfaces. Therefore, the development of more efficient and more robust solution schemes and solvers is always a target for further improvements.

For such cases, in which separated phases have to be considered, special numerical tools were developed to keep the interface sharp, like interface capturing, interface tracking, or some approximate numerical methods like surface sharpening. In case of dispersed flows special highly accurate numerical schemes would be required to avoid or at least minimize numerical diffusion, but most codes use standard lower order schemes; thus one still has to live with more or less strong numerical diffusion.

In the past, most codes for nuclear applications with multi-phase flows were developed mainly in the large national research centres or in the industry, like AFDM (LANL/FZK), ATHLET (GRS), CATHARE (CEA), COBRA-TF (Pacific NW Lab), ESTET-ASTRID (EDF), IVA (Siemens), MATTINA (FZK), MC-3D (CEA), RELAP5-3D (INEEL, DOE), ERHRAC (NPIC), TRAC (SNL, USNRC)… They have the advantage that the required nuclear specific modelling is included, and the models and numerics were selected according to the special requirements of nuclear applications. Now, powerful, and comfortable commercial CFD codes are becoming available, which are increasingly suitable for nuclear applications, like CFX-4,

COMET, FIDAP, FLOW-3D, FLUENT, PHOENICS, and others. There are major differences between the principal modelling capabilities available in the commercial codes. None of those has all the modelling required for certain flow phenomena in innovative reactor systems; thus it has to be decided from case to case which code should be used. This brings some concerns regarding the certification effort when one is progressing towards a licensing procedure of a reactor.

With increasing requirements on accuracy and details of codes, also requirements regarding the accuracy and details to be provided by experiments are growing. Whereas for design purposes and for system code improvement and validation mainly realistic large scale experiments are of interest, for the development and validation of CFD codes for two-phase flows one needs primarily single effect experiments with very detailed instrumentation, preferably providing not only local data, but also field data, like PIV and tomography. Topics should be the interfacial phenomena and turbulence near bubbles with rapid diameter change, near free surfaces, and near walls with boiling, the bubble formation, coalescence and fragmentation, data on the interfacial area, and the turbulence in the continuous phases; similar investigations are also needed for all other flow regimes, and the flow regime transitions. Nevertheless, some large scale more complex experiments with a detailed instrumentation are also required to validate the correct interrelation between the different models engaged in the CFD codes.