Author Topic: Knot testing : Eliminating less reliable data  (Read 6112 times)

xarax

  • Sr. Member
  • *****
  • Posts: 2781
Knot testing : Eliminating less reliable data
« on: April 22, 2014, 02:20:27 PM »
   When we test the slippage or the strength of a knot, we often get results that differ quite a lot from each other. So, it seems that there are significant variations in the testing conditions and/or in the quality of the rope material itself, which we can not control. A reasonable thing to do, is to eliminate some of the data we collect, and base our evaluations / statistics on fewer, yet more reliable results.
   The relevant articles in Wikipedia are simple, sufficiently informative, and very useful for the amateur knot tester, IMHO.
   
    http://en.wikipedia.org/wiki/Outlier
    http://en.wikipedia.org/wiki/Anomaly_detection
   
   Personally, I would use a more naive approach - I would simply trim from the collected data a number of extreme values - at least 1, but preferably 2, or even 3. The "trimming estimator" technique is described at :
   
   http://en.wikipedia.org/wiki/Trimmed_estimator
« Last Edit: April 22, 2014, 03:57:50 PM by xarax »
This is not a knot.

TheTreeSpyder

  • Exp. Member
  • ****
  • Posts: 152
Re: Knot testing : Eliminating less reliable data
« Reply #1 on: April 23, 2014, 11:57:53 AM »
Middle Road:

i think that you should not waste anything; especially as sparse as data can be on some topics.
Perhaps more comprehensive to show all data and only certify some of it?
Each processor of data can make up mind at that point what data to include.

i think best would be certified results range; and then variance = un-certified range( or all results range) - certified range  ?

But, then whom is judge and executioner?



But testing is just a guide anyway. 
The best/extended position of a lacing, to capture with least deformation and least impact of change in real life;
might not be testable because lacing would compress before breaking.

But if not worked on that end of the powerband in actual use, might be best to adjust from test lacing to actual use lacing.
Especially, where some dampening of dynamic force is desired, but working line at low percentage of it's tensile strength (gives least elasticity) and/or exposing only a short length of line(or just static/inelastic line) to the impact/hit of dynamic loading.
In these cases, a lacing that would give some/slide/compress (to more where testing would take it), yet not release could work nicely to dampen shock, saving rest of system from hyper-loading (conceivably even averting anchor failing/slip on load impact).
Like, steam released from a pressure cooker by safety valve, it could lessen system pressure.
Unlike working line in it's elastic range, the slip would not be recoverable immediately(yet not hysteric); giving only so many 'dampenings' to dynamic loading.

xarax

  • Sr. Member
  • *****
  • Posts: 2781
Re: Knot testing : Eliminating less reliable data
« Reply #2 on: April 23, 2014, 01:41:38 PM »
Perhaps more comprehensive to show all data and only certify some of it ?
Each processor of data can make up mind at that point what data to include.

   Correct. However, you have to present some rounded numbers, even only in the title of the presentation - so, at some point, you have to decide which data to include to the calculations required for the statistics...
   Nowadays there are many software packages ( and many of them are free ) where one can insert all the data he has, push "Enter", and, abra cadabra, see:)
   Oftentimes, we need to specify the probability distribution we expect our data to fit in... I have read that, concerning tests of the MTB of rods, ropes and wires, the best practice is to use the Weibull distribution ( or "weak-link distribution" ):

http://en.wikipedia.org/wiki/Weibull_distribution

   Of course, with enough data, the pattern of distribution, if it exists, will manifest itself !  And it may be an atogether different one - which we will then name : "The Knot Strength distribution " !  :)

But testing is just a guide anyway. 
The best/extended position of a lacing, to capture with least deformation and least impact of change in real life;
might not be testable because lacing would compress before breaking.

   - or before slipping / sliding...
   True... However, regarding "tight hitches" at least, where the force of the pre-tightening is important, I think that we can "solve" this, by specifying the load by which the ends(s) are pulled, before the final test and measurement of the load under which the hitch will open up, or slide, on the pole.
   There are many things that we can not yet test and measure in practical knots - and even some that we can not even imagine how to measure ! ( like the "easiness to untie a knot", for example, discussed in another thread ). However, this should not mean that we should abandon any tests and measurements, and claim whatever crosses our mind !
   I believe that, in the ripe phase the study of knots has reached, it is paramount to have NUMBERS, collected by careful and repeatable tests. Otherwise, the commercialization of the knotting literature in the web, where anybody tries to gain some $ by selling knotting myths, would accelerate ! A certain mediocre knot seller comes and claims that a certain mediocre knot based on an overhand knot tied on the Standing Part is as "easy to untie" as the Zeppelin knot, or the bowline - how on earth are we going to confront this ?
« Last Edit: April 23, 2014, 01:48:49 PM by xarax »
This is not a knot.

Dan_Lehman

  • Sr. Member
  • *****
  • Posts: 4366
Re: Knot testing : Eliminating less reliable data
« Reply #3 on: April 23, 2014, 04:19:34 PM »
   I believe that, in the ripe phase the study of knots has reached,
it is paramount to have NUMBERS, collected by careful and repeatable tests.
Repeatability comes from knowing the test method
and pertinent conditions; this knowledge might come
from asking questions sufficient to supplement whatever
test report is initially available, or it might come more
easily if the test method is one that has been already
carefully provided for testers to follow!

Quote
Otherwise, the commercialization of the knotting literature in the web,
where anybody tries to gain some $ by selling knotting myths,
would accelerate !
((A certain mediocre knot seller comes and claims that a certain
mediocre knot based on an overhand knot tied on the Standing Part
is as "easy to untie" as the Zeppelin knot, or the bowline
--how on earth are we going to confront this ?))
Mon dieu, quelle horreur!!

Those darn mediocre knots sellers --the "1%"--
are getting away with all the "$$$" of knotting,
while we honest knotters fiddle our fingers into
frazzled fragments for futile fame, what a shame!

Well, how have you confronted knot myths all
through history --it is not a new phenomenon?


 ;)

DerekSmith

  • IGKT Member
  • Sr. Member
  • *****
  • Posts: 1573
  • Knot Botherer
    • ALbion Alliance
Re: Knot testing : Eliminating less reliable data
« Reply #4 on: June 01, 2014, 09:50:31 AM »
honest knotters fiddle our fingers into frazzled fragments for futile fame

Oh Dan, what perception.  If that doesn't get front page of the next KM, then there is no justice in this world...

But as far as "Eliminating less reliable data" is concerned.  From a career spent as an Analyst, I can tell you two things about 'outlier data'.

The first is that sometimes data values are just simply wrong - someone wrote down the wrong value, or misread the meter, or the 'tape measure' slipped or...  In these cases, if you can identify the error, then it is reasonable to remove the values from the results set (or hire a better analyst and do the tests all over).

The second is where there is no discernible fault in the making of the test nor in the taking of the measurement.  Those outliers I can tell you, if you eliminate them, they will have a nasty habit of coming back and biting you on the bum...  And of course, in the world of knot strength, that could well result in a fatality.

So, I must agree with Treespider.  If the tests were carried out competently with calibrated equipment, then they ALL have to be taken as valid members of a much wider results set.  That single very low result should not be a cause to delete it, it should be a red flag, warning us that a lot more understanding and testing needs to be done in order to ensure that this very real break does not happen again in a real use situation.

Derek

 

anything