Hello everyone. Well, the publication aims to resolve doubts about the results obtained by FLUENT when compared to experimental results, I share with everyone for general understanding.

The flow in question is the flow through the divergent channel, backward facing step geometry. The simulation is based on the experimental study of Jovic and Driver (1994), an experimental study carried out to serve as a basis for numerical analysis. From this, we have the computational domain schematized in the image below. (Geometry values in h function)

The Reynolds number is based in step height h= 0,98 cm, U velocity = 7,72 m/s and std air 20° C conditions is adopted like in JD experimental study. At inlet velocity magnitude is prescribed normal at boundary with U= 7,72 m/s and viscosity ratio of 10 (for k-epsilon and k-omega I set the turbulence intensity to 5%). At wall no slip condition and adiabatic surface is imposed and at symmetry I set as symmetry line.

A quadrilateral structured mesh is generated with y+ < 1 in adjacent wall cells. We can see in the image below a zoomed view of mesh.

So,Then I will demonstrate the graphs of Cf vs x / h and Cp vs x / h displaying experimental and numerical results from the Sparlat-Allmaras model.

As expected we can see a double change of sine on the chart Cf vs x / h which indicates the presence of a recirculation zone composed of two bubbles, one smaller at the base of the step and one larger a little forward. This is shown by the image below with the colored vectors for velocity, results for model S-A.

The Xr indicates the reattach length and for this case we get a Xr/h value equal Xr/h = 6.5. According with the experimental data from JD the mean value of rettach length is Xr/h = 6+- 0.15. An acceptable difference of 8.33%.

It is visible that, by plotting Cf vs x / h, the result obtained by FLUENT overpredicts the coefficient of friction experimental curve and has values closer after the reconnection zone.

But what intrigues me is because the graph of the pressure coefficient is so below the experimental one that the graph Cf vs x / h has an acceptable agreement between the curves? Is it because of this approach being 2D and the real approach being 3D? Together with the experimental material, it contains results for a 3D simulation by DNS that the values coincide for both the pressure coefficient and the skin friction coefficient. I tested severe differences between models (I only showed the result S-A because they all exhibited similar behavior), I changed pressure-velocity coupling, etc. but they all exhibited satisfactory agreement for Cf, but for Cp no? What could be behind it? This is the big question ...

Enjoy the conversation!

Mantovani.