Poor Assessment of CFD for Rooftop Solar Arrays

In the press release announcing the report "Rooftop Solar Arrays and Wind Loading: A Primer on Using Wind Tunnel Testing as a Basis for Code Compliant Design per ASCE 7 (pdf)," there is the phrase:

"It also addresses the applicability of Computational Fluid Dynamics (CFD), which is sometimes proposed as an alternative to testing."

With eager anticipation I read the report and found a quite damning assessment of CFD.

In assessing CFD the report says:

  • "CFD is not currently in a state where it can be used reliably to estimate peak pressures on low-rise building roofs, let alone solar arrays on such buildings, and should not be used as a basis for design at this point in time."
  • "The primary challenge is that peak pressures and peak loads at the component scale are not well estimated by this method. This is unsurprising, given the fact that turbulence models (such as Reynolds-Averaged Navier Stokes, RANS, models) have been developed to estimate the mean flow field."

I was disappointed that in making these claims the authors cite no references or comparisons with their wind tunnel data.

I have to think there is a fundamental misunderstanding here on the nature of RANS modeling. The averaging is based on the turbulent fluctuations around a notional mean velocity. The time scales of the fluctuations are assumed to be orders of magnitude smaller than the time scale of the developing flow, whether the flow is assumed to be steady or unsteady. Typically in wind tunnel testing you employ the same averaging to report the localized quantities, such as velocity and pressure. RANS will be able to predict peak wind loads just as you are able to capture the data from a wind tunnel. For instance RANS has no problem predicting the peak low pressure (suction peak) on an airfoil.

Interestingly the paper cites a number of issues that arise in using a wind tunnel, such as problems relating to:

  • Blockage effects - restricted wind tunnel working area versus free air
  • Reynolds number effects - scaled wind tunnel model versus full size

As I discussed previously the great complementary strength of CFD is that you can perform simulations at full scale in open air. Alas the authors make no mention of this feature of CFD and did not use CFD to validate their wind tunnel assumptions.

I have to believe that if CFD is useful for simulating complicated flows around aircraft and automobiles then it would also be useful in assessing relatively simple flows around low-rise building roofs with solar arrays. Roofs and solar arrays more accurately resemble flat plates than curved wings and car bodies do, and therefore turbulence models should be in their sweet zone given that a basic test of any turbulence model is to accurately model the flow over a flat plate.


CFD for solar collector wind loads

There is no misunderstanding about the nature of RANS CFD in the UWO/Sunlink document. Boundary layer wind tunnel practitioners do not employ RANS-type averaging in measuring wind loads on solar collectors, rather we measure a time series of uplift of the panel and, based on the statistics of this time series, make a recommendation about the appropriate peak design load.

Roofs and solar panels are not simpler geometries than cars, as you have implied. The pressures on the building surface (and across the panels) is dictated by a flapping shear layer which forms at every corner, under which strong vortices can form, and which is rendered even less stable by significant inflow turbulence. The reason RANS-averaged simulations are of no value is that the mean flow never really exists. CFD is useful for aircraft and cars when it has been tuned for aircraft and cars. It is certainly not the case that separated flows are the sweet spot for RANS models.

As for "using CFD to validate the wind tunnel", while these methods can and should complement each other, it is CFD that needs validation. Wind tunnel testing is the basis for building codes worldwide because it has been validated against full scale testing.

Your comments on Professor Kopp's paper

Professor Kopp's paper is a fine discussion of the recent issues that have evolved in our knowledge of wind loadings on solar arrays. His discussion of the limitations of CFD (or the subset of CWE, Computational Wind Engineering, as it is known) in architectural aerodynamics (bluff bodies) is correct. CFD (using RANS, LES, or more fundamental turbulence schemes) has, to date, not been successfully validated against the full scale (prototype) or model scale (wind tunnel). I believe that someday it will, but it has not happened yet. With regard to peak design pressures on buildings, the comparative studies (CFD versus wind-tunnel/full-scale) in the literature generally show fair results on the peak positive pressures zones (front face), but an inability to replicate the negative peak pressure areas (roof, side walls and leeward wall) where peak pressures are generated by vortices and separated flows. The smaller solar arrays on buildings and in fields on the ground hold the same bluff body concerns as building do themselves. CWE does many things well, but currently it cannot produce reliable pressures and loads on building and/or components. In the field of architectrural aerodynamics CFD is currently a useful hybrid tool, but must not be used to create design loads. I suggest that you get more involved in the quadrennial International Symposia on Computational Wind Engineering (the last one was the Fifth in 2010, North Carolina). For a more detailed discussion of where CWE works and where it does not (and physical modeling limitations as well) I refer you to this recent paper:

Cochran, L.S. and Derickson, R.G., "A Physical Modeler's View of Computational Wind Engineering", Journal of Wind Engineering and Industrial Aerodynamics, Volume 99, Number 4, pages 139-153, 2011.

Lastly, regarding your comments on Reynolds Number constraints I suggest that you read more about bluff body aerodynamics and the Reynolds Number independence. This has been understood for about 50 years. A classic paper on the topic is below:

Cermak, J.E., "Application of Fluid Mechanics to Wind Engineering", A Freeman Scholar Lecture, Journal of Fluids Engineering, American Society of Mechanical Engineers, Volume 97, No. 1, March 1975.

CFD has a role

Thanks David and Leighton for your comments. I see a complete misunderstanding of my post though.

Let me be clear I am not advocating that CFD can replace the wind tunnel. I am merely suggesting that CFD can play a complimentary role with wind tunnel testing in the design of rooftop solar panels and buildings. For an example of such an approach see:

Robert N. Meroney and David E. Neff, "Wind effects on roof-mounted solar photovoltaic arrays: CFD and wind-tunnel evaluation," The Fifth International Symposium on Computational Wind Engineering (CWE2010), Chapel Hill, North Carolina, USA May 23-27, 2010

Meroney and Neff say "Computational Fluid Dynamics (CFD) can provide a flexible and cost-effective tool to select promising configurations from alternative design strategies, which can be subsequently tested in detail either in the laboratory or the field."

My problem with the SunLink report is that claims are made concerning CFD without citing references or showing comparisons. In my opinion the report leaves itself open to criticisms of bias with this glaring omission.

Clearly my comment about solar panels being similar to flat plates was wrong. However, to say that CFD can not simulate flow around bluff bodies or vortical flow on your part is also wrong. Take a look at a Formual 1 car - it is a mix of bluff bodies (wheels) and sharp edges that shed vortices all over the car - in fact the undertray relies on those vortices to generate downforce. There are no bigger advocates to the benefits of CFD than F1 designers. Interestingly those same designers are also big advocates of wind tunnel testing too.

Leighton, I'm well aware of Reynolds Number independence, I was merely citing the SunLink report in its reference of issues that need to be addressed when dealing with wind tunnel testing. Just like I might cite that the mesh resolution and turbulence model selection are factors that determine CFD accuracy.

If we include LES in our discussion, which increasingly is finding use in engineering, then I'm confident that CFD will become a first class member of the testing tool kit for future rooftop solar arrays and buildings - alongside wind tunnel testing.