Learning Forum

[ibrahimallam087]

I am simulating a Ceiling Radiant Cooling Panel (CRCP) system in ANSYS Fluent as part of a validation study against a published paper. The paper follows the same numerical setup, but I am facing convergence issues that prevent me from achieving the expected results.

• Solver: Pressure-based, steady-state

• Turbulence Model: Realizable k-ε with enhanced wall treatment

• Radiation Model: Surface-to-Surface (S2S)

• Pressure-Velocity Coupling: SIMPLE (tried Coupled as well)

• Spatial Discretization:

  • Pressure: body force weighted

  • Momentum, Energy, k, ε: Second-order upwind

• Boundary Conditions: there are no inlets and outlets

  • Panel Surface: Isothermal (15.83°C)

  • Walls: Adiabatic

  • Dummy Heat Sources: Constant heat flux (113 W/m²)

  • Radiation Settings: Emissivity = 0.92 (panel), 0.82 (walls)

  • Gravity: Enabled (Boussinesq Approximation applied)

  • Mesh:

    • Tetrahedral with inflation layers (First layer height = 0.001m)

    • Y+ < 10 (Enhanced Wall Treatment used)

    • 1.26 million elements

      Issue:

      • Continuity residual fluctuates between 4e-2 and 1e-1, does not drop further.

      • Energy residual is low (~1e-5) but fluctuates with velocity.

      • k and ε residuals remain high (~8e-3).

      • Average temperature in the domain initially increases, then fluctuates up and down.

      • Ran 5000 iterations, but no stable convergence . i tried to refine the mesh i reached five million cells which is far bigger than my validation paper (1.2 millionc)but it the same problem .

[Rob]

That sounds about normal. How do the point monitors look? 

[ibrahimallam087]

I refined my mesh more making it 3 M cells with min orthogonal quality around 0.4 and make 10 boundary layers instead of 5 with a first layer thickness half of that mentioned in the paper and this is an image for temperature value it flucuates between (25.2 and 25.6 C) after 7500 iterations 

 

[Rob]

That's pretty normal.  So, monitor is fluctuating around 0.05 in 299K, is that precise enough given equivalent measurement stability in a real room? 

[ibrahimallam087]

 

 

 

I understand that in an experimental setting, these fluctuations are reasonable and within expected precision. However, since I am validating against a published paper where the authors reported residuals reaching 1e-5, I would like to know if there is a way to further improve convergence and bring my residuals below 1e-3.

Would refining the mesh, adjusting solver settings, or trying a different numerical approach (such as pseudo-transient or a different turbulence treatment) help in achieving this level of convergence?

 

[Rob]

I'd question how the authors hit that level - is there a plot to prove it?

I generally use RNG k-e for HVAC but Realisable should be fine. 

[ibrahimallam087]

 

I checked three different papers validating the same experiment, and all of them mentioned that their convergence criteria was 1e-5, but none of them included a residual plot to confirm it. That’s why I wanted to see if there’s a way to reach that level in my case or if it’s just a reported value without further validation.

 

Would you say that for natural convection cases like this, reaching convergence is typically necessary, or is it more important to focus on physical stability in point monitors?

 

Also, regarding turbulence models, you mentioned using RNG k-ε for HVAC cases. Would switching to RNG k-ε instead of Realizable k-ε help in improving convergence, or would it mainly affect the flow physics?

 

 

 

[Rob]

I always focus on the monitors. As long as the residuals more or less flatline that's usually sufficient. If you review the Learning materials you'll see we focus on residuals, monitors and fluxes for convergence: it's common for one method to less useful. 

[ibrahimallam087]

That makes sense. I'll focus on the stability of monitors rather than forcing the residuals lower. I'll also review the learning materials to understand how different methods impact convergence.

I appreciate your insights on this!

[Rob]

You're welcome.  Each type of model is a little different. Hence HVAC I may check residuals but focus on monitors: for a pipe flow system residuals and fluxes may be more useful (with monitors coming in if the residuals don't behave). It's just a case of a little practice.....