How to test sharpness in digits.

[Cougar Allen]

If you use a statistical calculator you can just type in your results and it calculates standard deviation for you. That's the easy way....

 

[Peter La]

That's a great idea.

 

[Cliff Stamp]

1/4

90.7 +/- 5.9

1/2

92.0 +/- 4.8

3/4

75.3 +/- 3.2

This last part is significantly sharper than the first two.

The standard deviation of the mean is a mess to calculate by hand it is :

square root of (the difference between the sum of the elements squared and the square of the sum of the elements divided by the number of elements) divided by the number of elements minus one, now divided all of that by square root of the number of elements

or ((Sx^2-(Sx)^2/n)/(n-1))^.5/n^.5

x is the element, S is a sum, n is the number of elements

as an example,

90,80,70

answer is

(((90^2+80^2+70^2)-((90+80+70)^2)/3)/(3-1))^.5)/3^.5 = 5.8

Doing this by hand is an invitation to insanity, especially if you make a mistake and type in the wrong number in the middle.

On the Ti-83+ you can see the numbers you type in and make sure visually you are calculating the right thing.

Can I put the numbers for the Sharpmaker rods in the review I wrote, it already had similar work for the medium/fine, but I don't have the diamond and ultra fine.

-Cliff

 

[nozh2002]

Thanks, Cliff you are right it is too complicated to do by hands.

I found that StarOffice spreadsheet has standard devastation. I bet free OpenOffice has it also - it is much more convinient to use spreadsheet then calculator.

Again my point is not to do all testing myself but encurage everydody to do this. Like fulloflead came up with idea to test sharpness on different angles etc. Anybody who has idea can now easyly execute it. And it is easy, free and not very hard to do. And no need for scientific calculator - any computer suitable (if you on the web you already have one I assume?).

It takes may be 5 min. to do testing - not too much to be honorably named "mad scientist".

Thanks, Vassili.

 

[fulloflead]

...be honorably named "mad scientist"

You're welcome.

 

[Cliff Stamp]

I found that StarOffice spreadsheet has standard devastation.

Most spread sheets do, you are looking for the standard deviation in the mean here. So take the standard deviation and divide it by the square root of the number of trials.

-Cliff

 

[Ed Schempp]

Gentlemen, good thread, good information, fair evaluation of data. I appreciate information; specifically the work that Cliff and Vassili have presented in this thread.

American Sieptman had a CATRA sharpness machine at the SHOT Show. CATRA had previously developed a machine to test edge retention. The edge retention machine made repeated CNC cuts into a special quartz impregnated stack of fixtured paper. The machine records the cutting progress and provides a quantitative record of the test. The sharpness machine is a table top affair that sells for about $18,000. The edge retention machine is over $100,000.

The sharpness machine measures the pressure it takes to cut a rib on a radial fixtured piece of rubber. The rubber medium that is cut can be described as: Parallel flat surfaces on the sides of the rubber strip about .25 inches wide. A rounded side is opposite a flat side with a raised rib gives a total width of about .3125 thick. The radiused side is held against the a round fixture. The rib on the exposed or test side is about .100 wide and about .7 protuding from the remainder of that side, two flats that are separated by the test rib. The blade is clamped to the test table. The test table pushes the knife straight into the fixtured rib. The machine senses the bottoming out of the blade against the supporting flats. The rubber medium was fixtured over a radius so the rubber material cut would open up and not drag on the blade. The machine measures the pressure to cut the rib. The baseline pressure on a mach III razor blade was .6- .8. Surgical scapels were rated next as .8-1.8. Kitchen Cutlery was next 1.8-2.8 General use utility hunting and pocket knives were next @2.8-4.0., etc....

These folks also had a machine that was used to measure the inclusive geometry of the edge, an important variable in steel comparison tests.

These folks had a contest to determine the sharpest knife at the shot show. The Spyderco Persian with VG 10 Steel that I've carried for more than a year was quickly tuned with a Spyderco Sharpmaker; and proved to be the sharpest knife measured at the show, 1.18 edging out a Kershaw @ 1.28, and a Serbenza at 1.34. A H-1 Steel Spyderco Salt off the shelf tested 1.34.

Thanks...Ed

 

[Cliff Stamp]

The sharpness machine is a table top affair that sells for about $18,000.

I got an email about that recently. You can make a very similar setup yourself with a Ti-83+ and some force probes from Vernier for a few hundred dollars, probably a lot less if you got it all on ebay. This doesn't actually gain you any more information that the above setup except that it require less readings as it reduces some random influences. You would still want multiple readings along the blade as it is still a spot check. The CATRA setup has possibilities for a lot more than edge retention and sharpness though, you can examine cutting ability, wedging and so force as it actually generates a full force curve as the knife is pushed through a media.

-Cliff

 

[nozh2002]

I know about CATRA, it was one of my first thread here. Only problem with it - $18000 coas + continiues payment for paper. And nobody testing knives with it (at least for us)! Spyderco and Buck have it but knife community - general public hasn't.

My idea again is to came up with methodology to measure sharpness which anybody can run without any investment and without significant effort - it take few minutes to do several measurements and then modern math statistic will takes care of random influenses etc. (what is MathStatistic for?). And Even it may be sound scare - it is very easy to put number in spreedsheet and got result instanteniusly.

Thanks, Vassili.

 

[nozh2002]

So take the standard deviation and divide it by the square root of the number of trials.

I found that this way it soon degradate to zero. If you simply triple same experiment the number goes down, but probably should not.

Thanks, Vassili.

 

[GRMike]

Anybody interested in generating some data of this kind and applying the appropriate test of statistical significance might want to know about this site, which makes it painless. No need to calculate anything or even crank up a spreadsheet. It's all online. Works as fast as you can cut and paste (or hand enter) the data.

To compare two things (two knives, two sharpening techniques with the same knife) select Student's t-test and follow whichever link (enter or cut and paste) you prefer. The output is enough to please a stats prof. Look for the probability, given in the text below the various statistics. Values below .05 are generally (though arbitrarily) regarded as showing that you have a statistically significant difference.

If you're comparing three or more things, such as Vassili's data on the four Spyderco rods, the preferred test is an ANOVA. Unfortunately the application lacks post hoc tests, so all you will be able to determine in this case is that at least one thing is significantly different from at least one other thing--not so helpful, but it will create a nice box plot that lets you visually compare the means and confidence intervals.

Here's a sample of the output using Vassili's data from the 5/10 post. Note the probability is .0001, so at least one sharpening rod produces statistically different results from one other:

ANOVA: Results
The results of a ANOVA statistical test performed at 15:03 on 12-MAY-2005


Source of Sum of d.f. Mean F
Variation Squares Squares

between 1.1309E+05 3 3.7696E+04 98.98
error 3.1230E+04 82 380.9
total 1.4432E+05 85
The probability of this result, assuming the null hypothesis, is less than .0001
--------------------------------------------------------------------------------
Group A: Number of items= 18
150. 160. 170. 180. 180. 180. 180. 196. 200. 200. 200. 200. 200. 204. 220. 220. 220. 220.
Mean = 193.
95% confidence interval for Mean: 184.2 thru 202.5
Standard Deviation = 20.9
High = 220. Low = 150.
Median = 200.
Average Absolute Deviation from Median = 16.0


--------------------------------------------------------------------------------
Group B: Number of items= 18
100. 120. 120. 120. 120. 126. 128. 130. 130. 130. 130. 130. 132. 140. 140. 140. 140. 140.
Mean = 129.
95% confidence interval for Mean: 119.5 thru 137.8
Standard Deviation = 10.3
High = 140. Low = 100.
Median = 130.
Average Absolute Deviation from Median = 7.11


--------------------------------------------------------------------------------
Group C: Number of items= 27
70.0 70.0 80.0 80.0 90.0 90.0 90.0 90.0 100. 100. 100. 100. 100. 100. 105. 105. 110. 110. 110. 115. 120. 120. 130. 140. 140. 140. 140.
Mean = 105.
95% confidence interval for Mean: 97.90 thru 112.8
Standard Deviation = 20.5
High = 140. Low = 70.0
Median = 100.
Average Absolute Deviation from Median = 15.7


--------------------------------------------------------------------------------
Group D: Number of items= 23
70.0 70.0 70.0 70.0 80.0 80.0 80.0 90.0 90.0 90.0 90.0 90.0 90.0 100. 100. 100. 100. 105. 107. 130. 130. 140. 150.
Mean = 96.6
95% confidence interval for Mean: 88.51 thru 104.7
Standard Deviation = 22.5
High = 150. Low = 70.0
Median = 90.0
Average Absolute Deviation from Median = 16.2



Hope this helps,
Mike

 

[Cliff Stamp]

That is a very nice site thanks for the link, I use a fortran program to do similar analysis which includes tests for deviates as well. If you are going to apply statistics you also want to check for normality in the sample, have a look at the data and see how it is distrubuted. If it is spiking non-normal you may want to have a look at the method. The thread testing is normal when I checked it, several times.

-Cliff

 

[nozh2002]

Thank you very much GRMike!

I added ANOVA results to the data I gained:

http://playground.sun.com/~vasya/Testing-Res.html

Thanks, Vassili.

 

[nozh2002]

Only today I realize that I shoud test method itself.

So I did 6 runs for same knife - 30 checks each (10 at the begining, 10 in the middle and 10 at the end of the edge).

Results are from 98 to 121 for runs and 111 for all. Now weight scale I use has scale by 10, so I should round it to 10 - one unit on scale. So it is actualy 110 +/-10 and in scale units it is 11+/1 which is pretty good result!

I think everybody who try this should do same - runs several times on the same knife better just sharpened to werify that results will be representative.

Thanks, Vassili.

 

[Cliff Stamp]

Now weight scale I use has scale by 10, so I should round it to 10 - one unit on scale.

When taking readings from analog devices you estimate the fraction inbetween the calibrated lines, so if it calibrated to 10's then you estimate to 1's. After taking your sample data you round according to the standard deviation in the mean.

-Cliff