I've come against a problem that I can't figure out, hopefully someone here may be able to help-

I'm modelling a cylinder in crossflow. The cylinder is in a rectangular channel and is attached to the channel walls at the base and top of the cylinder. The cylinder is heated from one end using a constant temperature condition. There inlet and outlet are set to pressure inlet and outlet boundary conditions.

The problem is that on the surface of the cylinder on the cell adjacent to the heated surface, the heat transfer coefficient is much higher than it should be. (y+ is in the log layer).The problem is localised to this cell only, and the heat transfer coefficient on the rest of the cylinder seems reasonable.

Any ideas why this happens and how I can prevent this from happening? I've tried reducing and increasing the grid, but to no avail. Any comments would be much appreciated.

Mat W,

If your turbulence model is incorrect the wrong heat transfer coefficeient will be calcualted. You might trying running the problem as laminar flow and see if it makes a difference in the coefficent. I have run into this problem with cylinders in crossflow before and have conqured them to within 10% or so. Give me a better idea of the range of Re and Pr you are running in and I can give you better insight.

Tighe Smith

Principle R&D Engineer

Delta M Corp

Tighe,

Thanks for taking an interest- here is some more info on the case. The Re (based on cylinder diameter) is 600 and the fluid is air at around ambient temperature so the Pr is about 0.7. I've since altered my model - instead of applying heat directly to the base of the cylinder and the prime surface, I have modelled the cylinder on a solid. The heat enters the model via the base of this solid and enters the base of the cylinder by conduction.

In addition, I have reduced the boundary layer grid further and turned on the enhanced wall function. The good news is that the excessively high heat transfer coefficients have since disappeared. The bad news is that I suspect I will have to embark on an unsteady investigation to fully determine the behaviour of the model.

Please do pass on any info you have on any of the problems I encountered, I'd still like to run models with smaller grids to give me a rough idea of performance before establishing grid independance on specific cases that show promise.