Mr. Mayank S.

In general, depending on the model and method used.

Hi,

I slightly disagree with the answer of Saad Najeeb . The result of a model or its convergence/numerical robustness should not very much depend on the mesh type used, if we are talking about mesh independent solutions, i.e. if the meshes are not underresolved and are finely enough resolved to deliver a mesh independent solution.

The first question(s) to your comparison from my side would be:

1) How you measure precision / accuracy of your CFD solution? Do you have some quality or accuracy measure, which is independent from CFD, e.g. some comparison to data from experiments?

2) Have you compared only the results on 1 hexa mesh with an almost equally resolved tet/prism mesh or have you run a systematic and sufficiently fine mesh refinement with both mesh types and arriving with both mesh types on a mesh independent solution, which is than afterwards compared between the two mesh types?

Last bu not least it is important to know, how CFX as a node based solver is building control volumes. Building a control volume for a hex mesh around a mesh vertex again leads to a hex mesh with 6*4=24 integration points on the 6 faces of that control volume. For a tet/prism mesh CFX again builds the control volume around the mesh vertex. I.e. the control volume will have the shape of a general polyeder around that vertex. The number of integration points can gt 2-3 times as large in comparison to hexahedral meshes, depending on how much tets and prisms join in that mesh vertex for which the CV is build. Consequently the numerical effort for matrix assembly is usually around that factor larger for tet/prism meshes in comparison to hex meshes.

And further worth to mention - the accuracy of your results for general turbulent flows will not so much differ due to slight differences in skewness and aspect ratios, but it will quite crucially depend on the resolution of near-wall boundary layers and flow gradients wherever they occur at the wall or in the core of the flow.

Best regards,

Dr. Th. Frank.

Hello

Thanks for of the reply first of all

I have comparing the surface Nusselt Number and heat transfer coefficient for Re 30000 by dittus boelter for fully developed regime Nu should be 171 by terta I am getting 168 while with hexa with multizone o grid mesh the Nu value is 227 with sweep o grid mesh it is 237 while with multizone method and surface mesh set to uniform the Nu value is 297 if you want then I may share the screen shot of mesh too...

Thanks in advance Dr Thomas Frank and Mr. Saad Najeeb

Clear.. Dear, Mayank S.

Thankful.

Mayank Singh : Are your findings based on single mesh calculations or on an investigation with a hierarchy of more and more refined meshes showing some asymptotics towards your end results?

And btw Dittus-Bölter correlation is not an absolute truth. Correlations are developed from experiments under certain conditions, most of all under fully developed flow conditions. If conditions for your particular application differs from those in the experiments underlying the correlation development, than your results may differ and be nevertheless correct. And heat transfer is very, very sensitive...

Regards,

Dr. Th. Frank.