GSMT Assembly AURA New Initiatives Office

Workshop on Modeling of Telescope Wind Loading
Meeting Notes and Minutes


To send email to a participant, please replace "at" with "@" in your mail program.


Monday, November 26:

9:00Welcome and introductionLarry Stepp
9:20Gemini water tunnel testingPaul Gillett
9:40The Gemini South wind loading studyMyung Cho
10:10Analysis of the Gemini dataOleg Likhatchev
11:00Previous CFD modeling of Gemini telescope and enclosureDave DeYoung
11:30Enclosures for Extremely Large TelescopesDavid Halliday
1:00Wind modeling studies at TSUGuanpeng Xu
1:30What do we need to know for GSMT?George Angeli
2:00Discussion of what we have learned so far about telescope wind loading
3:20Discussion of next efforts

Tuesday, November 27:

9:00Planning session for additional studies
11:00Establish action items
12:00Adjourn Workshop and Lunch

Section 1. Presentations

Presentation 1: Welcome and introduction - Larry Stepp

Round-table introductions and agenda overview. Brief discussion of AURA and NIO, and the purpose and goals of the workshop.

Presentation 2: Gemini water tunnel testing - Paul Gillett

Presented summary of report by Wong & Forbes, July 1991. Study was performed for seeing results, not wind loading. Cylindrical enclosure model was tested in January but dropped in the later tests; it had lowest porosity so it might actually be another acceptable option if porosity were increased. The CTIO 4m enclosure model was added to the second test group but there is little explanation in the document; the assumption is that this was the initial test for seeing improvements that were indeed later done on the 4m Blanco. Conclusions stated in presentation: to avoid the ground boundary layer being elevated to telescope level, solutions are to make the pier tall (Gemini), open up the base of the facility to allow unimpeded ground flow (Magellan), or install guide plates to keep the ground air flow attached.

Presentation 3: The Gemini South wind loading study - Myung Cho

Used 24 pressure sensors to cover 8m mirror for first test, second test increased locations to 32. See presentation slides for discussion and results. Regarding scaling to larger sizes, no analysis was done to prove homogeneity of the structure function but it was not felt to be necessary at the time. Keep in mind this is only one idea of how to calculate these results, not an absolute.

Q: There seemed to be a lot of symmetry in the mirror deflections, what is the connection between the data and FEA model. A: Apparent symmetry was due to piston, tilt and focus being taken out of the modeling plots.

Q: Is 3m/s wind speed allowed at the mirror, or outside? A: At the mirror, inside the dome - goal is to control opening of the vent gates to allow no more than 3m/s at the surface at the mirror.

Q: What about the angle of the wind coming into the telescope when the dome orientation is changed? (not the elevation angle of telescope but a change in wind approach direction) A: Anemometer on top of the dome was indicative of the general wind flow to the telescope but it was only about a meter above the dome. If further testing is done then further data will be taken simultaneously from more locations around the site. There is also historical information on the wind flow above the Cerro Pachon site.

Q: Would the gaps between segmented mirrors provide an advantage because of the porosity that would allow air through the mirror surface? A: The goal is to minimize those gaps - current plans call for gaps to be less than 1% of the area.

Q: Are there likely to be more wind problems with a larger mirror? A: That is one of the fundamental questions for this workshop: what have we learned about wind loading that will predict the effects on larger telescopes.

Mike Sheehan gave a brief discussion about the vibration testing of the telescope that went on at the same time as the wind pressure test, with 50-60 accelerometers mounted on and around the structure at the time. This was somewhat successful, as a lot of data was obtained but some was corrupted. Some modal information could be used, which has led to a second modal test going on now; there are no conclusions at this time (in terms of wind shake). Plan is to combine the results of this test with the wind loading to get a better finite-element model of the telescope and mirror and a viable transfer function that can be extrapolated for use with GSMT.

Presentation 4: Analysis of the Gemini data - Oleg Likhatchev

This presentation is a continuation of the subject begun in Myung's presentation, from a different point of view (that of the physics behind the buffeting phenomenon). These are preliminary results only. In general a direct connection can be drawn between pressure and velocity, but our case is different because the dynamic portion of the wind velocity is as large as the constant portion, so first order approximations that relate pressure to velocity do not hold. Results from pressure measurements will be presented.

Oleg believes most of the wind buffeting on the mirror comes from vortex shedding from the telescope structure itself. We have a lot of data but it's difficult to analyze, since laboratory measurements suggest one thing but reality does not conform to those predicted results because of the vortex shedding. Because of the variation in wind velocity, the frequency of vortex shedding moves and creates the appearance of two strong peaks in the power spectral density (PSD) curves. There also may not be enough data points for high-enough data resolution. These issues will be dealt with more later; the hope of this initial analysis is to help us understand the physics behind the buffeting actions on the mirror for future use in learning how to design and control the dome and gates.

For case c00030oo where the dome slit was pointed towards the wind, the real-time pressure records show that the wind was coming slightly from the left (looking from the M2 towards the primary mirror) so the -X side was exposed to more buffeting. The autocorrelation function shows that there is almost pure random motion in the turbulence; it is difficult to judge any kind of information from this. But if the scale is increased, you can extract the time differentials and estimate the vortex shedding.

We have a mirror exposed to vortex shedding from the dome enclosure and from the mirror itself. Bernoulli's equation allows us to proceed with power spectral density analysis; will the static and dynamic sides of the equation resemble each other? Since we don't have measured upstream conditions only local conditions were used (which means this is all just estimation, not a precise analysis). For the +X sensor, the velocity was less so the peak frequency shifts to a lower value. We can compare different cases (oo and co) and evaluate the most energetic frequency in the PSD. From the Strouhal number we can estimate the characteristic length of the bluff body from which the vortices are shedding. We can conclude from these preliminary results that the phenomenon is due to vortex shedding from the enclosure and the mirror itself.

The second problem was buffeting forces on the secondary mirror. Information for vortex shedding from cylinders was presented. At Reynolds numbers below critical, uniform wind speed gives regular periodic vortex shedding at a frequency given by the Strouhal number; with turbulence from upstream obstructions, the wind produces irregular shedding. Pressure distribution changes behind the body give you a pulsed drag and lift effect due to the shedding. Low Reynolds numbers give regular lift but higher drag; increase the Reynolds number and both lift and drag become more irregular.

How do we estimate the buffeting forces on the secondary mirror? We can consider the secondary mirror assembly as a cylinder subject to transverse wind (assuming no velocity in the Z-direction). (These assumptions may or may not be appropriate for GSMT calculations.) There is a mean component of velocity; just take this component and project it on the cut of the cylinder and run the coordinate transformations. Due to the pulsating velocity of the wind component there will be varied drag on the cylinder -- the mean vector will change angles and give a component of lift as well. (This does not take the vortex shedding into account, however.) We cannot use a linear approach for this calculation; we must use the total expression to take all components into account.

It is hard to tell if the lift and drag forces are being provided by vortices shedding from the mirror itself or from other components of the telescope structure.

Q: Does the frequency of shedding change as the flow speed changes? A: Yes, it can be moved around but it is a nonlinear problem.

Q: Can you influence the flow over the secondary by the size and opening of the vents? Does the secondary see the same kind of wind flow as the primary? A: In general with vent gates open you see the same but if the vents are closed the flows are different, and if one vent is closed and the other is open it gets very different. This may be a real driver for GSMT. But this test program was designed for testing the wind load on the primary more than on the secondary. Also, the test program didn't address very many cases concerned with protecting the mirrors from high winds, such as with vent gates closed and the windscreens blocking part of the observing slit.

Q: It is interesting that on two different days the same spectrum was found for the same configuration; this implies that it isn't the outside wind spectrum but that of the observatory itself that mainly influences the results. A: Correct. We can take results from test cases on different days and use this scale to normalize the frequencies, to plot against a dimensionless number and see the results.

One of the main things we want to learn is how the features of the telescope and dome affect the wind environment.

Q: Are PSDs for the downward side of the mirror similar to those for the upwind side, which shows the oo and co cases being very similar? A: No, you'll have a very different story there, because the flow is being changed. You can't really compare in this case. We can try to compare cases with +Y and -Y and see what happens but we will need more time, as the data analysis is still ongoing.

Q: Has any research been done on the influence of type of vents on the flow (i.e. different masking, porosities, etc.)? A: We have not done this but it's something we'd like to understand to design a new enclosure; for example, are there characteristics of vents that we could use to minimize detrimental effects in the enclosure? We welcome recommendations for further tests.

Q: Why not try to actively control and/or organize the flow? A: One of the concerns is that we don't want to elevate the flows because it'll lift the ground layers and increase turbulence.

Q: What about recessing the telescope below ground level instead? A: that will still give the effect of lifting the turbulent ground layer; not a good thing.

Q: Are we convinced that this is a global and not local wind environment, i.e. is it coming across the mountain top or is it being created by the mirror cell and telescope itself or the wind hitting the edge of the enclosure and creating vortices from there? A: Another good question we'd like to see addressed, as we're not sure. Because of differences in pressure spectra at different points on the mirror, there appears to be evidence that much of the buffeting is not coming from vortices in the free air stream.

Presentation 5: Previous CFD modeling of Gemini telescope and enclosure - Dave DeYoung

Presentation on past numerical simulations for Gemini. Nominal wind speeds were 11m/s and 3m/s; wind direction was either perpendicular or parallel to vent gate openings. Mesh generation had to be handled very carefully; mesh size near the walls was approx 1cm, to accurately model the enclosure. David displayed and discussed the velocity vector plots and speed contour plots in the enclosure, for both wind orientations and at both wind speeds. A vorticity contour plot was presented and discussed. These results were presented to show what has been done and what can be done in the computer-modeling arena.

Q: Measured wind speed at the secondary is 60%-70% of the speed of the outside, and is relatively independent of the condition of vent openings; how does that correlate with the model? A: The model used a small slit opening that corresponds to the reduced size of the Gemini slit with windscreens deployed. Very few of the wind measurements were in this configuration, but we should compare those to the CFD modeling to see if they show similar levels of wind speed attenuation. (Dave DeYoung is willing to do this; he and Larry will coordinate.)

Q: Is there a difficulty in applying these 8m results to a 30m telescope model? A: There should be concern about the turbulent flow fields; it would also be good to do a time-dependent calculation. If you want to restrict yourself to a time-independent calculation, then you could scale it appropriately.

Presentation 6: Enclosures for Extremely Large Telescopes - David Halliday & Mike Gedig

In working on the XLT, AMEC wanted to develop the stiffest telescope possible in order to ensure it would work with a wide variety of optical configurations. David compared enclosure configurations for several large telescopes (Keck, Subaru, Gemini). He discussed the "Calotte" (French for "cap") enclosure configuration (see presentation; based on design developed for a smaller telescope on Pic du Midi). This is just one configuration being investigated at this time.

Next question was how to make the mirror cell as stiff as possible as well; proposed solution is a monocoque sandwich structure. Their solution keeps gravitational flexure below 2 mm at all zenith angles. Must look at an integrated solution; have to pay attention to how the cell and telescope structures interact with the enclosure so we can understand what the wind will do.

Presentation 7: Wind modeling studies at TSU - Guanpeng Xu

Brief introduction of Tennessee State University personnel and some of the projects they have undertaken. Overview of three methodologies that were studied; three wind turbine rotor configurations investigated; and results of the rotor studies. Discussed how this relates to the design of the 30m GSMT. 8m measurements could be scaled using non-dimensional and harmonic analysis. Also a full Navier-Stokes solver can be applied, with limited modifications. First the enclosure and mirror structure can be separately modeled, then after verification they can be combined for simultaneous modeling. The simultaneous modeling of the telescope and its enclosure will be very CPU intensive, but computing power and processors are becoming more easily available.

Presentation 8: What do we need to know for GSMT? - George Angeli

A summary of what we need to know about wind to design the GSMT. Discussion of the NIO approach, along with advantages and disadvantages. Expectations of the modeling efforts. The more we know about the wind, the less robustness we have to build into the control systems themselves. Discussion of the various items that we need to know and basic questions to be answered.

Q: Can the measured velocity field be used as a first approximation? A: Yes, it's being done right now, but it has to be characterized not just with the amplitude but also the direction; so the data at hand is not really enough. Have approached the project by "patching" together the results of the 8m tests to the 3m scale, basically working from an expanded 8m model. It is likely there are edge effects present in the Gemini data.

Can't assume we understand what the pressure patterns will be at the back side of a 30m mirror when all we know is what is happening at the front side of an 8m mirror.

What we need to understand are the characteristics of the turbulence introduced by the telescope and enclosure. Is there a way to dissipate enclosure-introduced vortices before they get to the telescope? What kind of measurements do we need from more tests and modeling? Two directions we'd like to take: (1) understand more and more about what makes Gemini work, and make it better; (2) how do we take that and predict what a larger telescope can do?

Q: How effective are the "great walls" being used by Subaru to isolate the telescope environment? A: At present, we don't know.

Section 2. What have we learned so far about telescope wind loading; next efforts.

What do we already know? Paraphrase of discussion:

(LS) We believe there are 3 effects we are concerned about:

1.	The spectrum of velocities in the free wind stream on the summit;
2.	The effect created by the enclosure itself as wind passes through it;
3.	The effect created by the telescope itself regardless of enclosure concept.

What's the consensus - is the turbulence in the airflow approaching the mountain pretty
well understood or do we need further research?

(GA) If the telescope is elevated a bit above ground layers, then probably we know enough.

(DH) Is the data taken at Gemini South in '91-92 still available?

(LS) Do we need to take more real measurements at this time to understand airflow over the

(OL) It's not crucial at this time; I don't think it's necessary.

(LS) What do we need then?

(OL) Simultaneous measurements, at high resolution, just upstream in the flow.

(LS) What key questions would we answer with this?

(OL) Velocity spectrum, vertical components, etc.

(LS) Do we need to take these measurements at the height of the telescope, assuming that
being at the same height would be most important?

(MW) How many data points can we take simultaneously?

(MC) Currently 30 points of measurement (100 sensors), with 6 anemometers.  Can probably
get a few more.  Can take up to 30 samples/second, and they can be repositioned.

(MW) It would be nice to get a solid batch of data in one particular volume at one location,
move it around the telescope to get three dimensions at different locations.

(LS) DD modeled the summit of Pachon 6-7 years ago with a coarse and fine grid; measurements
taken were very close to his models.  We could ask him to revise his model to take the
current, modified topography into account and run the models again.

(OL) We need to know what portion of wind buffeting contributes to the total turbulence?
We need hard data.

(LS) Can we do that with, say, a single anemometer if we have a model of the site?

(MW) It would be a good idea to do a topological model and do a CFD of that; enclosure concept
has a lot to do with the area around it.

(OL) What about two, one on the upwind enclosure side and one in a stagnant spot, to compare

(LS) Would we learn enough in general from these kinds of measurements to apply to a new
telescope, or would we be better off applying our time and efforts to other measurements?

(GA) I'm really interested in seeing the spectral densities for this.

(DH) We're serving two masters here, looking at the future and trying to improve Gemini.
Which comes first?

(LS)  It's more a matter of understanding the differences and there may be some efforts that
benefit both.  There is some segregation of funding, though, which is a consideration.  We're
interested in both, but the focus of this meeting really should be what have we learned to
help us develop new telescopes.

(MW)  Cannot get telescope data absent the enclosure, or vice versa; it seems we'll have to
get a high-resolution model of both the Gemini telescope and enclosure, and see if it correlates
to the model; then we can scale it for GSMT.

(LS) Ok, let's go in that direction for a bit.  In DD's model, it was very simple geometries,
no telescope inside.  How much more complicated can we make a model nowadays and still get
practical results?

(MW) In my opinion we need a telescope model as complex as a "cartoon" of the telescope to
be effective.

(LS) Is that level of detail practical?

(GX) If something like this is to be done, it has to be started very simply; very simple
numerical geometries in the computer.  We can't take the structure details of the supporting
structures into account.

(MW) Can we insert other simple geometric bodies into the simulations?

(GX) Yes, that's possible.

(GA) Is it possible to come up with a transfer function between the outside wind and the
interior flow boundaries?  Then verify it with this CFD process?

(MW) That transfer function would be different -

(GA) Yes, but it would give us some kind of theoretical consideration, and we could use it
for other estimations.  In that case we are chasing something and not just running the test.
We want to verify it.

(LS) When you say transfer function, are you talking about a parameter or a group of

(YT) Can you identify the functional relationship?  Can that be understood in functionality,
not the detailed mathematical conformity but the independent variables?  If we do that and
go for an experimental approach to measure components, then could we measure the data and
come up with this function?

(GA) We know what it looks like from the models we've already done.

(LS) You're overstating our knowledge somewhat.

(OL) It's not an exact answer - reality is quite different.

(GA) I don't need an exact answer; we need an estimated wind velocity that's in the ballpark.

(LS) We need so much more - it's a very complex situation to try to characterize.

(MW) We can decouple a system - it may be possible to come up with an enclosure design such
that the telescope structure doesn't aerodynamically couple with the enclosure structure.
Then they could be separately modeled.

(LS) Although it's interesting and possible, it seems that if we need to come up with a new
enclosure design in order to move forward, that's putting the cart before the horse.  We
need a fundamental understanding first.

(DH) We need tools that will give us comparative results.  How do we fix the current issues
with a new design?  Several solutions - we need the computational tools that will tell us if
we're on the right track, not necessarily a hard solution.  This is what we need to move

(LS) A lot of this already exists in the current literature, and we need to understand what's
already been done first.

(DH) There are also the thermal effects of the wind itself to keep in mind.

(LS) Let's turn it around a bit - one of the tools we have available is CFD; if that's not
adequate to give us a full detailed model, what is the most productive thing we can do with
CFD to improve the situation?  Look at mountaintops, details, components, somehow bound
some other modeling, what's the best thing we can do with CFD to gain insight on these issues?

(OL) We can ask what contributes the most in this problem; what is the major effect we are
dealing with?  We can learn about the vortex shedding.  At the first step we can disregard
turbulence from the atmospheric boundary layer, and look at just how does the vortex shedding
affect the pressure distributions on the surface.  Disregard upstream and enclosure flow,
and look just at the vortex shedding.   Later on, upstream wind turbulence could be included,
see how it affects again.  It's impossible to calculate the entire wind load, so let's take
it a piece at a time.  We can also look at the mirror itself.  There is definitely some useful
information, in regard to physics, in this information.  It would help to get some measurements
from a tower separate from the enclosure as well…we need free air flow information.

(DD) The most profitable but most expensive way would be to do a model of the interior with
a model of the telescope there, in a time-dependent way.  The model can be built to match as
closely as possible the existing telescope structure, but it depends on how much you want to
spend.  It would help answer some of the vorticity questions inside the dome and on the
telescope structure itself.  And modeling the independent structures is a great idea.

(LS) Would it be better to model Gemini and then correlate it with other measurements, or
just model the 30m and learn from there?

(DD) It would be better to get some comparison with real data first, for validation of the
numerical scheme.

(LS) Can you do it with separate measurements from inside and outside the enclosure?

(DD) Probably not, since it won't be a free flow inside the enclosure.

(OL) But can you at least reproduce the pressure distributions?

(DD) I guess, depending on how instrumented the primary is.  With the CP testing, that should
be sufficient.

(LS & GA) What about using a PSD?

(OL) No, it's not the same -

(MW) What about aliasing affects, filtering and corruption of the data?

(MC) We looked into that, and checked with the vendor - there's no corruption, the company
said that there wouldn't be effects at 10Hz frequencies.

(MW) which means they must be filtering the results… plots should be rolling off between 2
and 4 Hz.

(GA) It's not there; we're not seeing that… if you calculate the PSD without integrating,
you don't have useful data.

(MW) So you're saying they did not filter it, they just sampled it?  There's no way once the
data is corrupted with aliasing to determine what's good and what's bad.  Be sure to find out
if they put analog filtering in there or not.  (Further discussion on this topic between GA,
MW and OL.)

(LS) let's get back to proposed studies.  It seems that we might be misled by the GS data, as
much of it might be unusable for a 30m telescope.  Should we be concentrating first on the
closed cases instead of the open cases?  Open cases show flushing; could we design a closed
enclosure well enough to decouple the enclosure and telescope?  Should we focus on those
studies instead?  We can propose other tests to be done, including seeing tests, within limits
(we can't replace the primary mirror anymore but we can place anemometers just about anywhere
we want to).  Can we think of anything particularly useful to measure?

(DD) Studies with the vents closed would be very interesting to evaluate.

(MW) Can we snake a pressure tap into the gap around the mirror?

(LS) We could put pressure taps on the aperture plate, barely outside the surface of the mirror
but still part of that continuous surface.  We might be able to get some kind of correlation
with the current data set, in a bulk way, so that we can infer what's going on across the
mirror by measuring edge effects.

(MW) There are too many modes that are indistinguishable on the mirror surface by the edge
effects - not really a useful measurement.

(DH) It appears that we need to run two simultaneous paths - one to collect as much information
as possible from Gemini, while modeling what we believe will be a better solution for the future.
AMEC might be able to do that at their end with our modeling group.  But we're looking at a more
active ventilation system, instead of passive flushing.

(MW) I agree, let's put more effort into investigating controlling the airflow within the

(MS) We can perform that kind of control right now on Gemini with the controlled vent gates; it
might not be more effective to actively control it.

(GA) I have a bad feeling, we are talking about launching a huge effort and I'm not convinced
we don't understand what's happening on the telescope.  I don't know if what we see on the
primary is just a stochastic atmosphere or actually an effect from the telescope.

(LS) I don't quite agree with you; procedurally I want us to brainstorm a list of possible things
to do, then we can discuss the brainstormed list.  We've also seen in some of the talks today that
we're getting turbulence from the enclosure itself; while the slope of the PSD curve may be similar,
there's much more in effect.

(DH) Is it possible to control the vortex shedding with controlled flow?

(OL) It's possible….  Using vortex generators, blow air out, suck air in, etc - but it's very
expensive solution.

(GX) Is it possible to measure the flow field halfway between the enclosure opening and the
mirror?  If we can do that then we can simulate the telescope structure with the enclosure - it's
very fancy for a CFD simulation - but we can use those measurements as numerical input for the
simulation of the telescope.

(LS) We could put anemometers on the telescope structure itself.

(GX) The numerical input of the incoming wind in the enclosure but a significant distance away
from the telescope would be needed.

(LS) Here's another question: we're using ultrasonic anemometers.  Are these as good as anything
for the measurements we're taking?  Is there anything better we should be considering?  We need
three components of velocity.  Would differential pressure anemometers be better?

(OL) Is further testing really necessary, or just a waste of time and money?

(MC) It would help verify the CFD models.

(OL) I still don't think it's worth to do it - better to redistribute the pressure sensors around
the edge of the mirror and take more data to learn about vortex shedding of the primary mirror
itself, which I feel is the largest contributor here.

(YT) We really want to use this CFD modeling to identify the free stream environmental turbulence
intensity, the shedding caused by the opening, and the shedding caused by the mirror itself, we
can construct 3 separate CFD models but we need the input.  Would help answer which of the three
is more important.  Could be done with the limited resources at hand.

(MW) I'm not sure that we really need to correlate data 1-to-1; no matter how much real data
you've got you'll never match it to the model.  The point is we need a design tool to help us
identify trends and issues.  Our best effort would be spent to develop these to get the insight
we need.  Let's not get more data that won't give us more confidence; let's develop the tools
to gain more insights into GSMT.

(GX) yes, we need to develop the tools but we also need to develop backups as well, and validate
the tools with the measurements to find the most accurate one.  We must correlate at an existing
scale before we can increase the scale with any confidence.

(MW) It would help to give us confidence but the structures are so different I don't think we'd
get a lot of information out of it.

(OL) Can you model unsteady vortex shedding?

(GX) Yes, it's very similar to a fluctuating airfoil - it can be calculated and simulated.  That's
why it would be good to have measurements between the enclosure and the primary, to get the turbulent
wind speed and frequency to input.  But we already have enough data for preliminary inputs; more
measurements aren't urgent but an upwind measurement within the enclosure would be useful.

(YT) I was wondering, if we directly model the structure of GSMT, how feasible would it be to have
a model for validation experiments (for water tunnels, for example).

(LS) Not really economical to do at this time.

(YT & OL) But for that kind of measurements we don't need a detailed model, just use standard
wind/water tunnel models like rectangles and spheres for validation.

(LS) What about in the field itself?  Attach a known shape with a strain gauge or load cell to
an existing structural member and measure direct forces and wind effects right there?  Would
that be useful?

(OL) we have the same problem - we don't know the flow field.  Better to verify within a wind
tunnel with known structures, to validate the CFD simple structure models.

(GA) My problem is we don't have a hypothesis to work towards.  We want to model, we want to
measure, but we don't have a design that we're working towards.  These are things to be done
when we have a design, not before.  What I'd like to see is the development of a model - we're
talking about verifying a model that we don't have yet.  Is there any way to generate an
approximate analytical expression for what's happening, and then validate that analytical model?
Even if the validation is just in the ballpark, we can use that model to design the enclosure
and telescope.

(OL) The major problem is vortex shedding from the mirror itself; if we simulate that and learn
how the shedding affects the pressure distribution on the mirror, and if I can see if it
resembles what you measured, will that satisfy the need?

(GA) Yes indeed!  We can also fill in the missing data with that as well.  This would give us a
theory to work towards.

(LS) There are two main things we need to do, (1) model the performance of a 30m telescope,
and (2) have some idea of how to design the facility, enclosure and maybe even the telescope,
so that it will respond well within the environment it's put in.

(MW) Then I agree with George, we're putting the cart before the horse, and we need to come up
with a good baseline design model and do testing on that.  Until we have a model of a point
design we don't really have enough insight into what problems we have to solve.

(DH) we're learning through the modeling.

(OL) I want to explain a bit more my talk from yesterday and how it pertains to CFD; and I'd
like to propose a different design of flushing of the mirror.  First, let's talk about helicopter
blades and variation of lift with adjustable airfoils.  And what is our primary mirror but a
large airfoil?  Airflow moving across the mirror will create vortices of airflow separation,
proportional to upstream velocity.  Every time we change the airflow across our mirror, we'll
create pulsating lift forces across the mirror surface.  This is a major cause for wind buffeting
on the primary mirror, under all conditions (doesn't matter about vent gates, slit opening, etc -
even in steady case with no turbulence - we would still have buffeting.)  We not only could have
vortex shedding downstream from the object, but most definitely on the leading edge of the object
as well (like the edge of the mirror, the mirror cell, etc.)

(LS) Will these vortices move in the same direction the wind is moving?

(OL) yes, that's right.

(LS) Can we do a correlation of time and distance with pressure across the mirror?  Should you
be able to see some kind of pattern across the mirror?

(OL) Yes, you should, but your speed of propagation reduces - it's difficult without knowing local
scales to know the time delays.

(LS) By finding out the time delay we can learn about the local scales.

(OL) What I propose is the CFD people should calculate without the enclosure just some kind of
disk-like mirror, placed on the ground (or even just in an infinite environment) exposed to flow
but at different angles.  Let's see how vortices would shed, and how the pressure distributions
and lifts act - will it resemble what we have?  This would prove the concept.  Start with steady
flow and then adjust the velocity variation upwind, see how things change.

(LS) hasn't something like this already been done in wind tunnels?

(OL) No, not really - they didn't use a thick disk, and they didn't care about pressure
distributions across the disk.

(LS) How do we proceed based on this understanding?  What do you want the CFD people to do?

(GX) We can address this.  It's about how much time and money you want to spend.  Based on
the current capability we have and limited resources and time frame, let's look at dynamic
modeling of the GSMT primary mirror.  With the increase of the M1 diameter to 30m but wanting
to reduce the distance from the focus to the primary, so the cost can be reduced - so the
curvature of the PM had to be reduced to drop this distance.  But by increasing the curvature
we'll see more flow separation on the edges, both leading and trailing, leading to vortices
both above and below the mirror surface.   Makes flushing even more important.  Based on
this assumption, the question has been asked how do we use CFD to study the turbulence of
the incoming wind, whether this incoming turbulence contributes more to the turbulent flow
across the primary mirror.  Turbulence is good to flush and smooth the temperature field
across the mirror, but bad for dynamic wind loading, especially at higher wind speeds.  We
also want to study the harmonics of the wind loads.  What I propose is the CFD modeling
without the cylinder enclosing the primary and secondary, but just the mirror.  Put the
mirror in ambient air, and using unsteady compressible CFD and imposing harmonics at the
inflow, with the vector of the wind being equal to the mean velocity of the wind plus V'1 + V'2
+ ....  If we manually put oscillations in the incoming wind flow we can see the effects across
the dynamic vortex shedding across the mirror, and if such effects exist at what frequency do
the most important effects exist.  What we can output is the pressure at both the upper and
lower surfaces of the mirror, as a function of spatial coordinates and time.  And further using
harmonic analysis tools we can study the dynamic effects.  We can integrate the pressure to get
the lift, which is also a function of time.  So I think we can also realize unsteady flow fields
to see if there is wind buffeting on the mirror, and what this frequency and amplitude of the
wind buffeting are.  Such CFD modeling requires limited effort from current CFD methodologies,
it would be possible to finish it very soon by the Tennessee State CFD team if needed.

(LS) So what you're saying is let's do this on the geometry we'd expect on a 30m telescope -
this is not calibrating effects on Gemini but exploring the design for GSMT.

(GX) I don't know the current configuration -I think we should use the current measurements
and known configurations of the 8m before we start exploring a new structure.  I think this
proposed study will greatly reduce the effort needed for the results.

(LS) Would you do this with the mirror tilted at a number of different angles?

(GX) Yes, of course.  Once the geometry of the mirror is set, the angle problem is simply
changing the angle of the incoming wind.

(LS) The mirror we're looking at for GSMT isn't a cylinder but an irregular edged curved
surface, and that irregular edge would influence the effects.  I think we need to model
the irregularity to get more accurate solutions.

(MW) I think it wouldn't be hard to model, just a circle with triangles around the edge of
the circle to simulate the irregular edges.  It should be straightforward.  A question I
have on the 8 vs. 30 m - as long as we keep the winds at 3 m/s on Gemini you can live with
the turbulence - the question I have is what is the situation with 30m?  Will it be the
same?  One of the simple things we can do is compare an 8m to a 30m at the same wind speeds -
will we see the same wind flow?  Can we get a comparison like this?  See how many times
worse is the dynamic loading?

(LS) My concern is that if we model the 8m but don't model it accurately will we get an
accurate correlation and would it be useful?

(MW) No, I'm saying take a representative 8m disk and run these tests, then extrapolate up
to a 30m disk and run the same tests, see what we find out.  Simple procedure to give us
some basic insight to the effect of increasing in size.

(DH) The telescope structure itself could have more effect on the primary wind flow than
the enclosure, anyway, so we can model without the enclosure and still get useful data.

Discussion of the structure of the Gemini mirror cell, the mirror covers, vent gates, etc.,
and display of pictures.  Then discussion and pictures of proposed GSMT mirror structure,
and discussion of modeling rough-edged disk.  Discussion of flows across both top and bottom
surfaces of GSMT mirror segments, and the complex understructures, and how they will influence
other things.

(LS) To summarize: let's model the 30m mirror, just the glass, supported in the airstream,
and bringing the air flow in at different angles, and adding different harmonics into the
airstream, and see what we can learn from this model.  Then see what we have learned, what
else we can learn, and get insight into the next round of questions to ask.  And by
adjusting the Reynolds number we can look at 8m and 30m sizes and see what happens then.

(OL) I want to propose a new design to organize flow inside the enclosure.

(LS) what's the goal of the new concept?

(OL) Reduce velocity while still keeping flushing.

Section 3. Planning session for additional studies

Have we reached an agreement for the next planned phase of studies? What else could be done?

CFD studies: GX to write up a short description of what has been proposed; will send to LS. Study is mainly to identify the characteristics of the vortex shedding across and around the primary mirror.

Enclosure studies: we should look at proposed designs and do preliminary analysis of options. Come up with a representative model that can be used down the line as well. AMEC will do this; for simplicity, it should be done separately but in parallel to the TSU work, and we must keep communications going. We should also model different vent sizes; that might show us if one or the other is better for what we need. Our opinion is that forced ventilation is not as effective as a well-designed natural ventilation scheme but it can be kept as an option to investigate.

Measurement programs: do we need more measurements at Gemini South? Yes, of the upstream conditions, to determine the inflow properties. We agree that it would be useful to get simultaneous measurements of wind velocity inside and outside the dome. We can take measurements from the anemometer on the outside tower and from the existing anemometers on the telescope structure, at 30Hz. This should be for about an hour each time, and should be repeated over several days. Second stage of this would be to rig up a differential pressure sensor in a low-pass filter box next to each anemometer.

Data analysis: Is there any additional information or are there references on scaling information and/or functions? (This may be more a CFD question than a physics analysis one.)

It would be productive to get together occasionally, either in real time or via videoconferencing, to keep in touch and compare notes.

Section 4. Action items

Ruth A. Kneale / / December 3, 2001