PDA

View Full Version : Digitizing Performance Graphs


AC103
25th Oct 2018, 23:22
Hi folks,

Has anyone had any luck digitizing performance graphs for take-off, landing, time to climb. Looking to fill out my W&B spread sheet to include all aspects of performance.

Much kudos in advance!

VinRouge
26th Oct 2018, 09:35
Hi folks,

Has anyone had any luck digitizing performance graphs for take-off, landing, time to climb. Looking to fill out my W&B spread sheet to include all aspects of performance.

Much kudos in advance!
This is a service provided by Jeppesen and the likes, they take tab/graphical data and turn it into algorithmic/equation based data. The output format is known as SCAP, or standardised Coputerised Aircraft Performance. There are standards set for the assurance of this, so its not a case of getting on excel and just doing a best fit to a curve.

Its also expensive - however, manufacturers can often have this data and will resell on as a service. i suspect though that you will be shipocked at the price- aviation isnt cheap!

john_tullamarine
26th Oct 2018, 11:06
A conceptually straight forward exercise the technical difficulty of which is around mid-secondary school level maths.

Problem is that you MUST model the AFM lines, not the usual performance equations. The test of a good setup is that the delta between the accurately read line data and the computer output is both conservative and minimal. You cannot do a simple regression analysis as the errors will be too great. I did a lot of this stuff in the deep past - all good fun if excessively tedious.

Some thoughts -

(a) read off the points at a reasonable density consistent with the line shapes. Blow the AFM charts up. If you want to do it the hard way, rescale as photostats and then pore over the graphs with a magnifying glass reading off data points and setting up tables for the lines. Generally easier if you scan to high res and then do the reading in a vector package. Keep in mind the aim is to minimise the delta between the line and what you read while being conservative. Either way you end up with tabular data replacing the lines. If the AFM data was tabular to start with ... you can avoid this bit.

(b) set up a look up model. Either use the tabular data as a lookup table and interpolate or run regressions on sensible lengths of the lines. Don't even waste your time trying to determine regression coefficients for the usual squiggly lines .. won't work. Simple linear interpolation doesn't generally cut the mustard but there are standard techniques which help out

(c) test the model output against the read off tabular data. When you reckon you have an adequate balance between accuracy and conservatism move to the next activity

(d) set up the carpet models. That's just a computer analogy for reading the charts manually. Including enough relevant tests to capture errors and the like is where the time is spent.

(e) when you reckon you have it under control, run an every point comparison between the final model and the original read off data to check for errors.

Can you set this up in Excel ? Of course you can. Alternatively, use any of the appropriate programming languages.

AC103
26th Oct 2018, 18:07
Thanks John, just great mate! Not proficient in a programming language as yet, python is on the list though. For creating tabulised data I was looking at software like this http://plotdigitizer.sourceforge.net/.

Digitizing plots (http://dankelley.github.io//r/2014/03/12/digitizing-plots.html)
Digitizing jpeg graphs (http://www.rscriptsandtips.********.com.au/2014/02/digitizing-jpeg-graphs-in-r-i-have-been.html) http://rscriptsandtips.********.com/2014/02/digitizing-jpeg-graphs-in-r-i-have-been.html
Converting plots to data (i) (http://wiekvoet.********.com.au/2014/01/convering-plots-to-data.html) http://wiekvoet.********.com/2014/01/convering-plots-to-data.html
Converting plots to data (ii) (http://wiekvoet.********.com.au/2014/01/converting-plots-to-data-ii.html) http://wiekvoet.********.com/2014/01/converting-plots-to-data-ii.html
Getting data from figures in published papers (http://www.r-bloggers.com/getting-data-from-figures-in-published-papers/)
Digitize graphs with multiple point sets (http://www.r-bloggers.com/digitize-linear-and-semi-log-scale-graphs-with-multiple-point-sets/)
Digitizing data from old plots (http://lukemiller.org/index.php/2011/06/digitizing-data-from-old-plots-using-digitize/)
How to digitize plots (http://lamages.********.com.au/2012/02/show-me-data-or-how-to-digitize-plots.html) https://magesblog.com/post/2012-02-27-show-me-data-or-how-to-digitize-plots/
Wikipedia: Converting scanned graphs to data (http://en.wikipedia.org/wiki/Converting_scanned_graphs_to_data)

Anyone had any success in this area that would care to share their results??

India Four Two
27th Oct 2018, 02:52
I've done a bit of digitizing for oil exploration projects. I've had great success using Plot Digitizer, free Java software that runs on a Mac, Windows or Linux. It's not perfect, but it gets the job done and is much quicker than picking numbers off plots by hand.

http://plotdigitizer.sourceforge.net

john_tullamarine
27th Oct 2018, 10:28
Ooh, that's nice. My, things have progressed since I last played with this stuff .. back when it was do it yourself or buy very costly specialist software. l will have a looksee and be amazed.

Le Flaneur
27th Oct 2018, 15:16
I've successfully digitized graphs using Engauge Digitizer, an open source digitizer available for download on sourceforge. I will then curve fit using either Matlab or GNU Octave. I wouldn't use Excel curve fitting for anything but the simple linear and quadratic equations. You must be extremely careful to check the registration, as scanned of prints (like you would get from an old AFM downloaded from MyBoeiingfleet that has been scanned) usually have distortion.

Building a SCAP module is involved and requires access to an AFM or other primary source.

AC103
28th Oct 2018, 20:57
I've successfully digitized graphs using Engauge Digitizer, an open source digitizer available for download on sourceforge. I will then curve fit using either Matlab or GNU Octave. I wouldn't use Excel curve fitting for anything but the simple linear and quadratic equations. You must be extremely careful to check the registration, as scanned of prints (like you would get from an old AFM downloaded from MyBoeiingfleet that has been scanned) usually have distortion.

Building a SCAP module is involved and requires access to an AFM or other primary source.

That is just spot on sir! Ok to share an example?
FYI https://www.youtube.com/watch?v=P7GbGdMvopU&t=903s is not a bad tutorial for a web based option. The question is how to do this with all of the split axis performance graphs used in aviation?

john_tullamarine
29th Oct 2018, 10:28
When you refer to "split axis", I presume you are looking at the usual parameterised carpet presentations. They can be treated just as sets of xy data.

AC103
29th Oct 2018, 11:05
Thanks for the language John! I am checking with www.OriginLab.com to see if their Digitizer supports carpet plots.

Anyone actually done this for a typical takeoff or landing performance chart?

john_tullamarine
29th Oct 2018, 23:10
Nothing fancy about carpet plots other than that they look fancy. Just a drafting sleight of hand to get multiple data sets onto the one bit of paper. Confused me for an hour or so when I first had to work out how it was done (50-odd years ago, now). Once it fell into place, it's a dead easy technique to apply. My approach when putting line segments into the PC is to run regressions on each of the segments and then play with the results.

Main thing is to break the segments up (as necessary) into smaller lengths so that they can be fitted with no more than a cubic. Higher orders can give you headaches with unwanted inflexions .. especially if you want to play with short extrapolations. One has to keep in mind the goal is that we are trying to replicate the AFM pictures, not the physics behind the lines.

My comment about Excel, elsewhere, was not to suggest you use it to run the regressions. However, it is a fine vehicle to structure the final data output engine. I don't even bother using Fortran any more in preference .. does anyone remember Fortran ?

VinRouge
30th Oct 2018, 16:02
Yep. Used it extensively with Abaqus FEA to do complex structural analysis work. Think you can still use to this day, it's not all drag n drop unigraphics/catia if you need to do the complex stuff.

Certainly high strain stuff, complex pressure fields and complicated boundary conditions between adjacent parts (for example, a turbine for tree root).

AC103
30th Oct 2018, 23:35
https://cimg9.ibsrv.net/gimg/pprune.org-vbulletin/1119x757/take_off_performance_flaps_25_854b8c4aedc4a63887479f4572036b c06f7f1855.png
Still waiting on Origin to come back on whether their software can do carpet plots. Anyone want to have a crack at this in the mean time?

john_tullamarine
31st Oct 2018, 09:57
Don't waste your time .. a multivariate regression won't work to a sensible accuracy in most all cases. Run regressions on each of the lines separately (and that might require the lines to be segmented), use that to look up some values for the independent variable of interest ... and then run a sensible interpolation or regression to get the final answer ... works real well.

The end requirement is negligible to no detectable delta betwixt the line data as she is read .. and the calculation as she is run .. and any delta must be conservative.

AC103
31st Oct 2018, 23:42
Run regressions on each of the lines separately (and that might require the lines to be segmented), use that to look up some values for the independent variable of interest ... and then run a sensible interpolation or regression to get the final answer ... works real well.

Thanks John, do you think you would be able to demonstrate that in an SS for a portion of the Press. Alt / Temp (Density Alt) carpet. Understood you don't recommend reverting to the DA theoretical model nor using XL for regression.