When I first became involved with commercial CFD programs, I was favorably impressed by the companies that showed how well their code performed compared to data from well conducted experiments. [Likewise, I was not impressed by companies that did not provide validation results.] These data were provided primarily by university fluid research programs. The test cases were all for single phase fluids. When simulation of multiphase systems came on the scene, there were almost no test data for validation.
Recently, I was looking on the website of a major CFD program supplier and I found that the situation hasn't improved greatly. When I talked to a senior manager in the company, he agreed that validation data were scarce. He cited the difficulty in measurement as the reason.
After looking at some research papers on fluidized beds, I see that much of the observations and measurements are indeed expensive and complex. The reason is that the experiments are designed to further develop the closure relationships needed for the CFD models. Thus, bubble sizes and shapes are being measured by X-ray or other non-intrusive means. However, for validation of a model, there are more simple, macroscopic measurements that can be used.
In this post, I would like to outline some validation data that should be easy to obtain in a laboratory setting. I'm assuming that cold flow studies can be made in transparent equipment or with sight glasses in metal pressure vessels. The following data should be obtained for all four types of particles (Geldardt classification), but types A, B and D are the most important for reactors. CFD programs should be able to correctly model all particle types, not just smooth spheres.
Minimum fluidization velocity, obtained by observation of the point where the pressure reaches a maximum as flow is increased.
Dense bed height as a function of velocity above the minimum fluidization velocity. The bed height increases for type A particles up to a maximum.
Superficial velocity at onset of a bubble phase. For B and D particles, this velocity will be equal or slightly above the minimum fluidization velocity. The onset of a bubble phase may be determined by visual means, or by the tracer method discussed next.
The main objective for a good fluidized bed model is to adequately represent both the gas and solid mixing and local residence time behavior. This information can be obtained via gas (helium) and catalyst tracer studies (radioactive tagged particles). Although tracer studies have been used in a few research projects, they haven't become a common model validation tool.
I would like to see one or more laboratories develop a good set of tracer studies for model validation. The CFD companies can then model the experiment including the tracer study. Comparison of the tracer output data from the experiment with the CFD result will show how well the CFD program performed.
Tracer curves obtained at different gas flow rates could be used to determine the onset of gas bypassing, i.e. the presences of a bubble phase.
Judging the significance of curve differences
The tracer curve comparison can be conducted without using another model. However, by assuming a simple model as I have shown in previous posts, the tracer information can be converted to parameters such as the fraction of gas to the bubble phase, the gas exchange coefficient, and emulsion phase void fraction. The values for these parameters from the experiment and the CFD simulation can be compared. Such a comparison provides an engineering perspective of the results as opposed to a visual comparison of two curves. It must be recognized that this simple model may not represent the flow well. If the CFD simulation suggests another model might be more appropriate, the parameter estimates for that model, using the tracer curves from experiment and CFD might be compared.
Beware of validation with reactions
There are examples of validation cases involving reactions. Frequently, it is not clear whether or not the reaction parameters were adjusted to get a fit to the data. Adjustments to a reaction model can easily mask deficiencies of the flow model. The validation data I have suggested above deals only with the flow model.
Do your own validation?
Yes, we should all make an attempt to validate our own models for each application. However, I don't want to spend a lot of time and money investing in a CFD program to find out it doesn't provide good results. The generic validation case data proposed above would make program selection easier and increase confidence in the the program results. CFD simulations can provide very valuable design and diagnostic information. However, can we believe the results without some validation?