I’m a student at NTNU in Norway, and I’m currently working on my masters thesis about aeroelastic software and Turbsim.
A part of my task is to perform a parameter study on Turbsim. I’m now inspecting the .sum files of simulations with different amount of gridpoints e.g. 10x10, 20x20 and so on. I’m only using IEC kaimal parameters.
I have a few questions regarding the statistics in the .sum file:
What would be the most important statistic for mechanical engineers (I’m studying computer science primarily)
As far as i know, the friction velocity is used to describe shear-related motion in moving fluids. Different amount of gridpoints gave me very different output for friction velocity (or u_star) ranging from 0.1 to 0.6. What does this mean? Shouldn’t this statistic be the same irregardless of the different grid resolutions?
Do you know any literature on the statistical approach to turbulence apart from Burton, Hau and Hansen’s books on the subject?
I’m having some trouble understanding the turbulence models like Kaimal and Mann.
In case some guys want to take a look at my runtime analysis, I’m adding a plot which shows runtime as function of gridpoints (square grid) with IEC Kaimal standard input.
As far as I recall, Dr Jonkman’s analysis only covered lower grid resolutions.
Thanks for the questions and plot. It’s interesting to see how the performance scales. Often times the number of gridpoints is not something we can choose: typically you want to have your grid fine enough that the grid resolution is similar to the turbine’s chord length, so that you correctly capture the turbine response.
Here goes with some answers to your questions:
That’s very difficult to answer! I am an engineer and am interested in different things for different projects. Sometimes all I want is the mean hub-height wind speed. Other times I want the turbulence intensity (standard deviation of wind speed / mean wind speed), and other times I want the wind speed at 10 different heights. Maybe we can give a more useful answer ig you tell us why you asked this?
That’s not something I would have expected. Could you share a plot that shows the friction velocity against resolution? Do other statistics, like mean wind speed, shear or turbulence intensity vary? I’ll also ask colleagues to look at this.
Kaimal and Finnegan wrote something on this that may help. It’s more about the observations and their analysis than the coding, but it may help. See: Kaimal, J. and Finnigan, J. (1994). Atmospheric Boundary Layer Flows their Structure and Measurement. Oxford University Press Inc. New York.
The Mann model is from Jakob Mann. There’s a lot out there if you use a search engine (e.g. google, bing) to look for “mann turbulence model”. I think the a good starting paper is “Wind Field Simulation”, sciencedirect.com/science/ar … 2097000362.
FYI, Francesco Perrone posted on these forums about creating his own version of the Mann model in Matlab. It’s now available through mathworks.com/matlabcentral/ … mann-model. That might help you in digging in to the code.
You’re welcome! I can’t take any credit for the coding, though - that’s mainly Bonnie Jonkman and Levi Kilcher, who might have some more answers for you.
The friction velocity is based on the u and w components of the flow. In TurbSim (the Veers/Sandia method), the u and w components of the wind speed are generated independently, which means this value is basically random–it varies based on the random phases (which will vary if you vary the number of grid points). In the non-IEC models, you can specify a u-w cross-correlation term, which will result in a more consistant u* value at hub-height for given boundary conditions.