What are the key challenges in applying CFD to simulate and optimize multiphase processes in complex industrial systems?
The main challenge here is the ability to create a sufficiently accurate representation of the real-world situation for which you are trying to optimize, in order to have sensible results within the timeframe available for the analysis.
Often, you will need to make various assumptions to limit the computational extent or level of detail of the simulation domain, and choose among a variety of available models that capture the physics underlying the flow. The suitability of certain models depends very much on the process conditions (flow rates, pressures, temperatures, etc.) and to some extent on the geometry. Also, suitable boundary and initial conditions have to be deduced.
To a large part the modelling approach hinges on experience and engineering judgement. This is why at Dynaflow Research Group, the method of analysis for all CFD projects is always discussed internally with a team of experts prior to starting simulations. Also, we actively study relevant literature on simulation approaches for given flows or geometries.
The webinar was intended to illustrate that when using CFD in the early stages of process design, it is less about the exact simulation outcome, and more about the ability to signify trends in the flow on a qualitative basis, that are clear enough to already act on for optimizing your process. At a later stage, after tuning the process parameters, it is wise to follow up with a more detailed simulation for verification purposes.
How to deal with large time-to-solution in multiphase simulation?
Try to find a good trade-off between the required accuracy and the simulation time required. In the examples we’ve shown, our early assumptions often allow us to use single-phase flow results with a suitable model for the second phase, or split the problem into two stages. Another example: for gas flows at low velocities, the effects of compressibility are often negligible, and a (faster) incompressible solver can be used.
And sometimes, simulations just need to run for a long time. At Dynaflow Research Group, we make use of a computational cluster with 48 cores, and our longest simulations can run for multiple days too. That is not an issue, but it is important to test on a coarser level (less detailed mesh, larger timestep, simpler models) if the CFD simulation is set up correctly and the initial outcome more or less aligns with your engineering intuition.
How do you validate multiphase CFD simulations when experimental data is limited or unavailable?
This is a difficult matter, to which there is also professional disagreement. CFD software development employs benchmark cases that can be tested against experimental data, for instance, the results of wind tunnel tests. Good correspondence over various benchmarks is an indicator that can provide a sense of confidence in simulation outcomes for other geometries. This is not something we are actively focused on in our consultancy services at Dynaflow Research Group.
What we do always conduct is a mesh convergence study, to investigate that there are no simulation anomalies that cloud the results. For the same geometry, with an increased resolution mesh, the CFD simulation is re-run. The simulation outcome should be equal for both cases to within certain bounds. Larger differences in the results can then be attributed to numerical error and are thus not interpreted wrongly.
Also, varying one parameter to review its effects on the simulation outcome can provide you with an idea of the sensitivity of the outcome. We are always aiming to optimize clients’ process conditions to a stable state, that shows little spread if the real-world settings are slightly off from the CFD parameter settings. Optimization to such a state is then a relatively safe operation, and this allows us to provide our advice with confidence.
