Compare Polars

Seems to work in Numbers on a Mac.

On Friday, December 22, 2017 at 5:33:36 PM UTC-7, Kiwi User wrote:
On Fri, 22 Dec 2017 16:06:55 -0800, stu857xx wrote:

Choices:
1) Buy a copy of MS Office / Excel
2) get a hooky copy of Office/Excel
(may be OK, may contain malware unless you know and trust the source)
2) Download and try Libre Office
This is a free/open source office suite that provides the same
functionality as MS Office. There are versions for Windows, OSX and
Linux

I tried it last night on the Libre Office (5.1.6.2) that come with Ubuntu 16.04. No joy, the graphs didn't get through.

It is nice to know that is works on Numbers on Mac.
That provides a second option.

I'm pretty sure there is no value in fitting a 6th order polynomial. The data isn't that good.

On Friday, December 22, 2017 at 12:22:05 PM UTC-8, wrote:
On Friday, December 22, 2017 at 11:28:11 AM UTC-5, Tony wrote:
There is a spreadsheet that allows comparison of quite a few polars he https://www.gliding.com.au/assets/docs/Polar10.xls

This is interesting data to me, but I open it up, and look at the "gliders" table of the raw data, and I cannot make any sense of what I am seeing.

What is "G/F" ... and then after that looking across the table, look at the first line, for an ADSH-25e ... yes, pretty high-performance sailplane .... but it says the units are "mps" (surely meters per second ?) ... but then the first entry is of 80? 80 what? 80 m/s ... yowsa, that's ≈ 160 kts!

and at "80" the value in the table is is 0.62 ... what is that? If that were the sink speed in m/s then the L/D would be 80/0.62 = 129 ... no way! So what are the numbers?

Can somebody explain WTF I am looking at here?

At this point I should mention that I really have been looking for these data. and I am an applied mathematician/scientist by trade and have been studying risk tradeoff optimization in speed-to-fly problems, this is also basically the same problem as optimal handicapping in disguise.

Is there interest in discussion of these issues? I could point out some issues where the current methods aren't entirely right or complete, that do have pretty straightforward solutions.

One other point I would make ... for all calculational purposes, you want to reduce the polef data to a fitted function of some sort, and boy ... are polynomials convenient for this purpose in this case. The standard "McCready Speed to Fly" just assumes the polar from best L/D on up is fitted with a quadratic, after trivial calculus and a little algebra the speed-to-fly is computed by solving the quadratic equation. This is easy to generalize to better fits.

And when you see this, and go through just a little math, what you see you want for the polar are stated minimum sink and best-L/D speeds & sinks, and then at least one, preferably 2 or 3 data-points at higher speeds ... and you don't need anything more than that.

The reason minimum sink and best-L/D are so important to defining the polar should be intuitively obvious, but there's a mathematical reason too ... these provide additional important equation constraints on the fitted function

minimum sink is of course s'(h) = 0

best L/D means that s(h) = h / s'(h)

where h is the horizontal speed (that what's really the independent variable in a polar, NOT airspeed, although at the very high L/Ds of gliders the differences between these two is negligible) s(h) is the sink rate at horizontal speed h, and ' means first derivative.

Forth these two "special" points you get two constraint equations, not one. Playing around with real polar data it takes a 5 to 6 order polynomial to really fit one well, and for manipulating all the ensuing calculations it is the coefficients of that polynomial that is wanted.

Also, I don't see listed gross weights for these test data? That is important ... to do ballast corrections etc.

I'd appreciate it if somebody would answer "what is G/F" ... is that the wing loading and if so in what units?

As to all of this dealing with ms-format xls spreadsheets ... if it's your tool of choice, fine ... but it's not mine, and I just wanted the data out, and other than the issue of the gross weight or wing-loading ... I've got that done.

I would make the following points however, as a working scientist/engineer:

* you'll find (if you look) that journal standards for "supplementary material" and funding-agency data retention requirements (NSF, NIH, DOE, NASA) ban, at least strongly discourage, proprietary or spreadsheet data formats, in favor of ascii-delimited, or NetCDF

* "can't help you there" ... is not accepted

On Saturday, December 23, 2017 at 11:44:41 AM UTC-5, jfitch wrote:
I'm pretty sure there is no value in fitting a 6th order polynomial. The data isn't that good.


Ah, now this is a more interesting issue, that has several aspects. Some things to think about:

* for almost every test, the minimum sink speed and the best L/D points are given, and in some cases this is all the data for which you get actual numbers without digging, or trying to digitize data from squinchy graphs etc.

* It is important for general application that the fit get the minimum sink and best-L/D right. If you don't believe me on this point, we could discuss further. But you'll find that you cannot get a polynomial that does these right AND gets the higher-speed data right ... without 5 or 6 terms. Go away and give it a try..

The fundamental reason for all of this is that the drag is reasonably described as the sum of drag terms, with the dominant terms are the induced drag and the "profile drag."

The induced drag is proportional to the lift-coefficient squared, and when you grind that out it falls as 1/v ... and polynomials don't fit that well with few terms ... that's the real reason you need more.

This leads to the following point/idea ... nothing magical a about a polynomial, and if you want something with few terms that fits a polar pretty well and has a "reasonable" basis" consider (Ax + 1) ( B/x + C * x^2) ... easy to derive the standard best-glide and speed-to-fly equations with this. (and A is generally small)

Related to the above issues is "yes, the data aren't that good" in many cases, but OK, we know physics (aerodynamics) and our job isn't just to get "some line" through the data, our job is to get the best physically plausible estimate of the polar, from the data we have. Another reason for this is that often the data don't extend to higher speed ranges, etc.

Now this leads to one of the things that I am playing with to deal with these data, that consists of doing a drag fit to a model that includes terms for induced drag (including the epsilon term), a wing washout term vs AOA, using the airfoil drag-bucket curve if known (or a generic one if not), and a simple fuselage profile drag vs AOA function.

This yields a physically-realistic polar, given the physics we know (gets to be a much bigger nuisance for flapped sailplanes though, without getting handbook data and applying those) , and in principle allows us to "fix" not so good data to a a degree. It's also the most physically-plausible way to extrapolate the data to higher airspeeds if you need to, though doing is is always a reach.

The resulting system is something of a mess that one doesn't want to use as the function for speed-to-fly etc (it's messy and derivatives are messier), so it is easier to fit the result with a 5 or 6 term polynomial for subsequent use ... and this is just another way of saying what I have said above -- takes a polynomial with a fair number of terms to approximate a real drag model, because the physics has that pesky induced drag term, and some sort of drag-bucket approximation.

With today's computers, 6th order polynomials are easy to deal with, so have at it. But - getting a fit to within 1% of data that has errors of 5 or 10% is a fools errand. Idealized drag equations, even complex ones, still do not account for many large unknowns: how accurate is the wing profile and how does it change with temp/humidity, how good are the seals, what are the disturbances around surface intersections and how does that change with speed, how dusty or buggy, how much of a slip is the pilot flying in. Any or all of these can easily amount to a few percent, and may vary by individual glider, not just design. Go for it - but be prepared for reality to strike..

If you could just solve the problem of correctly handicapping for wing loading in known conditions, that would be a much greater advance than fitting the polar curve better.

On Saturday, December 23, 2017 at 10:31:46 AM UTC-8, wrote:
I'd appreciate it if somebody would answer "what is G/F" ... is that the wing loading and if so in what units?

As to all of this dealing with ms-format xls spreadsheets ... if it's your tool of choice, fine ... but it's not mine, and I just wanted the data out, and other than the issue of the gross weight or wing-loading ... I've got that done.

I would make the following points however, as a working scientist/engineer:

* you'll find (if you look) that journal standards for "supplementary material" and funding-agency data retention requirements (NSF, NIH, DOE, NASA) ban, at least strongly discourage, proprietary or spreadsheet data formats, in favor of ascii-delimited, or NetCDF

* "can't help you there" ... is not accepted

On Saturday, December 23, 2017 at 11:44:41 AM UTC-5, jfitch wrote:
I'm pretty sure there is no value in fitting a 6th order polynomial. The data isn't that good.


Ah, now this is a more interesting issue, that has several aspects. Some things to think about:

* for almost every test, the minimum sink speed and the best L/D points are given, and in some cases this is all the data for which you get actual numbers without digging, or trying to digitize data from squinchy graphs etc..

* It is important for general application that the fit get the minimum sink and best-L/D right. If you don't believe me on this point, we could discuss further. But you'll find that you cannot get a polynomial that does these right AND gets the higher-speed data right ... without 5 or 6 terms. Go away and give it a try..

The fundamental reason for all of this is that the drag is reasonably described as the sum of drag terms, with the dominant terms are the induced drag and the "profile drag."

The induced drag is proportional to the lift-coefficient squared, and when you grind that out it falls as 1/v ... and polynomials don't fit that well with few terms ... that's the real reason you need more.

This leads to the following point/idea ... nothing magical a about a polynomial, and if you want something with few terms that fits a polar pretty well and has a "reasonable" basis" consider (Ax + 1) ( B/x + C * x^2) ... easy to derive the standard best-glide and speed-to-fly equations with this. (and A is generally small)

Related to the above issues is "yes, the data aren't that good" in many cases, but OK, we know physics (aerodynamics) and our job isn't just to get "some line" through the data, our job is to get the best physically plausible estimate of the polar, from the data we have. Another reason for this is that often the data don't extend to higher speed ranges, etc.

Now this leads to one of the things that I am playing with to deal with these data, that consists of doing a drag fit to a model that includes terms for induced drag (including the epsilon term), a wing washout term vs AOA, using the airfoil drag-bucket curve if known (or a generic one if not), and a simple fuselage profile drag vs AOA function.

This yields a physically-realistic polar, given the physics we know (gets to be a much bigger nuisance for flapped sailplanes though, without getting handbook data and applying those) , and in principle allows us to "fix" not so good data to a a degree. It's also the most physically-plausible way to extrapolate the data to higher airspeeds if you need to, though doing is is always a reach.

The resulting system is something of a mess that one doesn't want to use as the function for speed-to-fly etc (it's messy and derivatives are messier), so it is easier to fit the result with a 5 or 6 term polynomial for subsequent use ... and this is just another way of saying what I have said above -- takes a polynomial with a fair number of terms to approximate a real drag model, because the physics has that pesky induced drag term, and some sort of drag-bucket approximation.

I'd appreciate it if somebody would answer "what is G/F" ... is that the wing loading and if so in what units?

The help sheet says G/F is wing loading in kg/sqMeter

10% error means a 10% change in L/D ... and that would sure be embarrassing as test data.

Your comments about individual A/C performance are "owner's responsibility" ... it's actually pretty easy to do basic testing on your own sailplane -- the real issue is a good static.

If you could just solve the problem of correctly handicapping for wing loading in known conditions, that would be a much greater advance than fitting the polar curve better.

Hey, it's as if I fed you this question! There is in fact a pretty straightforward solution to this BUT no possible solution ends up with a single "handicap number" independent of the soaring conditions ... and in fact that's obviously true just about sailplane handicapping in general.

In very weak soaring conditions low-performance sailplanes fly "much better than their handicap" ... it's obvious, just look at the speed-to-fly optimization. Any "one-size-fits-all" handicap factor implies a known headwind/tailwind and mean rate-of-climb, even from simple McCready speed-to-fly.

Handicap schemes that can deal with changing gross weights are really the same issue as dealing with changed assumptions about the conditions enroute.

Beyond that, one of the biggest real problems is risk optimization, and "transitions."

One of the biggest problems of handicapped contests is that the lower-performance gliders may be simply incapable of a transition that the better ones make, and DNF. A great deal of this lies on the CD and forecaster, to pick tasks that don't have impossible transitions for the weaker sailplanes. Obviously ... sometimes this screws up.

The broader problem is more subtle ... when you are gliding there is always some risk you won't find the next thermal ... how does that risk depend on the gliding range you will have? Pretty obviously the lower-performance gliders face a higher risk of not finding a next thermal ... how do you handicap that? What can be done about it?

Stay tuned (but don't hold your breath) ... that is the problem I am working on. The math of this is pretty straight forward but the problem of the risk model isn't. We don't actually know that much (statistically) about this issue, particularly how to estimate it from met forecasts.

Just to mention two snarky ideas that I am sure would cause a riot if actually implemented

* risks could be equalized by height-limit handicapping: if you know the predicted lift ceiling then the lowest-performance gliders get to climb that high, and all the higher-performance gliders are given lesser altitude limits that equalize the gliding distance (at predicted speed to fly for the conditions) down to 1000' AGL (and like some contests a safety DNF is enforced at just a little below that, from your flight recorder)

Betcha all the pilots who show up with the low-handicap gliders would really love that one ... right?

* and then for a really wacky one ... it would be technically really easy to build a little "auto-spoiler" device that measures your airspeed and degrades the performance of your glider to some pre-programned polar ... the polar of the worst glider in the contest! Yowza! Now everybody in the contest is flying an equal glider! Dontcha think people would just love that one? NOT!

A couple percent error in static/pitot is considered to be pretty good. A couple percent error in ASI instrument is pretty good. A couple percent error in sink rate however measured is pretty good. Now we have 6% on an unlucky day, even with good data. Dick Johnson's test data is probably as accurate as can be practically done. It has a lot of noise and scatter, like +/- 5%. Not many people have that equipment, or patience.

You aren't going to get anywhere with handicapping the gliders by degrading performance directly or indirectly, as you know. In this day of turned in IGC log files though, one could analyze them post race to determine conditions, and apply corrections to handicaps. For example average climbs each hour for the top 5 pilots could define conditions. According to pilot poles, rules and scoring are already too complex, so acceptance of such a large step in complexity might be an uphill battle.

On Saturday, December 23, 2017 at 1:13:37 PM UTC-8, wrote:
10% error means a 10% change in L/D ... and that would sure be embarrassing as test data.

Your comments about individual A/C performance are "owner's responsibility" ... it's actually pretty easy to do basic testing on your own sailplane -- the real issue is a good static.

If you could just solve the problem of correctly handicapping for wing loading in known conditions, that would be a much greater advance than fitting the polar curve better.

Hey, it's as if I fed you this question! There is in fact a pretty straightforward solution to this BUT no possible solution ends up with a single "handicap number" independent of the soaring conditions ... and in fact that's obviously true just about sailplane handicapping in general.

In very weak soaring conditions low-performance sailplanes fly "much better than their handicap" ... it's obvious, just look at the speed-to-fly optimization. Any "one-size-fits-all" handicap factor implies a known headwind/tailwind and mean rate-of-climb, even from simple McCready speed-to-fly.

Handicap schemes that can deal with changing gross weights are really the same issue as dealing with changed assumptions about the conditions enroute.

Beyond that, one of the biggest real problems is risk optimization, and "transitions."

One of the biggest problems of handicapped contests is that the lower-performance gliders may be simply incapable of a transition that the better ones make, and DNF. A great deal of this lies on the CD and forecaster, to pick tasks that don't have impossible transitions for the weaker sailplanes.. Obviously ... sometimes this screws up.

The broader problem is more subtle ... when you are gliding there is always some risk you won't find the next thermal ... how does that risk depend on the gliding range you will have? Pretty obviously the lower-performance gliders face a higher risk of not finding a next thermal ... how do you handicap that? What can be done about it?

Stay tuned (but don't hold your breath) ... that is the problem I am working on. The math of this is pretty straight forward but the problem of the risk model isn't. We don't actually know that much (statistically) about this issue, particularly how to estimate it from met forecasts.

Just to mention two snarky ideas that I am sure would cause a riot if actually implemented

* risks could be equalized by height-limit handicapping: if you know the predicted lift ceiling then the lowest-performance gliders get to climb that high, and all the higher-performance gliders are given lesser altitude limits that equalize the gliding distance (at predicted speed to fly for the conditions) down to 1000' AGL (and like some contests a safety DNF is enforced at just a little below that, from your flight recorder)

Betcha all the pilots who show up with the low-handicap gliders would really love that one ... right?

* and then for a really wacky one ... it would be technically really easy to build a little "auto-spoiler" device that measures your airspeed and degrades the performance of your glider to some pre-programned polar ... the polar of the worst glider in the contest! Yowza! Now everybody in the contest is flying an equal glider! Dontcha think people would just love that one? NOT!

On Saturday, December 23, 2017 at 1:31:46 PM UTC-5, wrote:
I'd appreciate it if somebody would answer "what is G/F" ... is that the wing loading and if so in what units?

I haven't really looked deeply into this spread-sheet but most likely G/F stands for 'Gewicht / Flaeche' or weight(mass)/surface area. The typical UoM would be kg/m^2

Uli
'AS'

Jfitch -- I think you don't understand how decent measurements are made. They are not made simply from ASI and VSI. Read any of the various documents on the topic.

The classic method involves calibrating the aircraft's pitot/static versus a trailing reference cone and then having done that making long straight glides in still air and using altitude differences ... with a lab-calibrated altimeter and altimetric corrections made from the hypsometric equation for the particular lapse rate.

There are other/augmented ways that are coming into vogue that allow good accuracy with "amateur" gear ... one of the better methods is to to use GPS and either circling or quadrants to establish tthe wind-aloft heading (at a good altitude where it will be stable over a good delta), and then fly out and back legs on the pure cross-wind heading. The GPS along-track data will give TAS very accurately if the legs are 2 km or longer e.g., small errors in the true cross-wind angle cancel by summing the out and back leg data due to sin(theta) ≈ theta for small angles.

And another way, used to some aircraft testing and apposite to sailplanes I believe, is towing with a tensionometer.


The biggest problems occur for the very highest performance sailplanes -- one tends to make these measurements in the still air of a high-pressure stagnation, and there is subsidence that may be as much as 10 cm/sec, and there are arguments about how to diagnose that, whether such corrections are right or not. Dick Johnson and the Germans battled over that one, I think that argument is still in play.

This all being said, it's easy to make mistakes or have bad data one way or another. If you want to see data that clearly looks like something is pretty wrong, look at the Discus A (L) data; that's clearly aphysical (for a ship with no flaps).