[mouli1991]

Hey guys!!How you doing?Myself sayee working in cfd department. I got a serious doubt coz i am new to CFD. As a matter of fact i always use to vlaidate lift coefficient since drag coefficient wont get predicted by my software. SO what i have to do to accurately predict the Drag of my object? Which one i have to refine
1. Mesh?---do i have to improve mesh in flow direction (For airfoil) to predict drag?
2. Boundary conditin?Do i have to give any specific important boundary condition to precit drag?

[t.teschner]

well, it seems obvious that if you increase your mesh that you should get automatically better results, however that is not "really" the case.
if you increase your mesh size, what you really get is a mesh-independent result, meaning that the mesh itself will not introduce any further errors to the simulation, however, cfd is an approximative science and hence there are so many other errors that you make which all contribute to the overall error.

there is a concept which lets you measure how much of an error you are making for using a specific grid resolution introduced by Roache which was termed GCI (Grid Convergence Index). It is basically a measure of how your error diminishes by going from coarser to finer grids.

Now lets say your mesh is not influencing your results anymore (by that i mean you can refine your mesh even more but the results won't change with increased mesh resolution), you will probably still feel that the drag is not accurately captured. the biggest source of "error" is the choice of the turbulence model. i assume that you are not running DNS (direct numerical simulations) or infact LES (large eddy simulations) but rather you are using a RANS turbulent model. in that case, each model will give you a different value for the drag, one being closer to the experimental data then the other. each model is trying to mimic the effect of turbulence(therefore it is a model) and we have so many different turbulence models so we are able to simulate different flow domains with acceptable accuracy. your job is now to find which turbulence model is best suited for the application you are having.

you mentioned that you are running simulations of an airfoil?! in that case, the spalart-allmaras model is usually used, at least for 2d flows as it is robust, fast and specifically designed for aerospace application in mind. it is however not really good for 3d flows and for predicting separation, so if you want to calculated stall behaviour you will most likely not get very good results.

then there are also other approximation in your model, like how to interpolate the cell variables to the cell faces. the commercial solvers will all give you at least first and second order approximation, fluent also has a third order one.

well, i could go on and tell you where else we introduce some sort of approximation that influences the result but for you it is probably enough and best to remember that everything is approximated and therefore errors introduced. your mesh should be fine enough to capture the flow field as good as possible while not being too fine to cause too long computational times.