Friction Forged Blades : CATRA tests

[Cliff Stamp]

Here is the information:

http://www.cutleryscience.com/papers/DiamondBlade info.pdf

Now I am going to show you how this is an example of really misleading and biased statistical analysis.

Figure 1:

Look at the CHANGE in peformance. The S90V blade degrades by 0.1, the FF D2 blade changes by 0.2. Thus the FF blade degrades TWICE as much as the S90V blade.

Figure 2 and 3:

The entire superiority hinges on one point in the S90V blade which is lower. But the variation in the performance of the blades due to measurement uncertainty is WAY HIGHER and this small change is not significant.

In short, the information provided raises some interesting questions such as why was the FF D2 blade so much sharper initially (wrong abrasives used I would assume). But give no clear evidence of edge retention increase and in fact show it to be inferior in rate of degredation to S90V by 2:1.

This is why I have said many times that what you do has no bearing on if something is "scientific" or not, but how the data in interpreted and what conclusions are drawn. This is an example of a lot of precise numbers which are not utilized properly and the conclusions presented are not rigerously supported by the analysis.

-Cliff

 

[Jared Stenoien]

The part about differential hardening raised some major red flags for me.

 

[Cliff Stamp]

Essentially, it is like torch hardening the edge.

-Cliff

 

[Cobalt]

Mkes a lot of sense. Were is the toughnes data? They let that part of the equation out

 

[Larrin]

The only toughness information they give is bending it, which makes it look good because of the differential hardening, but that isn't in this article at all.

 

[Cobalt]

The only toughness information they give is bending it, which makes it look good because of the differential hardening, but that isn't in this article at all.

true and the only reason it looks good is because of it's differential hardening, but actual edge toughness is a whole different matter as you know. I wonder why they left that out?

 

[Larrin]

My inside information tells me it's not very tough at the edge. It might be passable, but calling it "tough" would be an overstatement, no matter what you're comparing it to. This is mostly my opinion, though I have heard a little bit about the edge toughness.

 

[tnelson]

Cliff,

Thanks for posting this information, but I think you need to hold on here. Your analysis and comments were based on a very selective data set which makes it a “statistically bias analysis.” I think it is great that DiamondBlade actually put out this much raw data for others to analyze. There is a lot of data there, so I feel we should look at all of it. I did some very simple analysis (for my simple mind). Here is my analysis and interpretation.

With regards Figure 1: If I do the math with data taken from the charts), FF drops from 0.45 to 0.30 (or 0.15) in the first 50 cuts. This is 33% difference. If I do the same for the S90V, it drops from 0.27 to 0.19 (or 0.08) in the first 50 cuts. This is a 32% difference. These are essentially equal in this range. However, FF blade is cutting about 50% more media per stroke compared to the other blades.

Now, I analyzed the data without (removed) the first 50 cuts for the FF blade (move the FF curve to the right so the fifth data point is on the vertical axis). I did this because, as Cliff said (and from Figure 1 it looks like), it appeared that the FF blade was initially much sharper. By eliminating the first four data points on the FF blade, it gives the FF blade an initial sharpness (or media cut per stroke) between S30V and S90V, two good steels. Now, doing the same analysis as above, the FF blade drops from 0.3 to 0.25 (or 0.05) in 150 inches of media cut. This represents a 17% change. The S90V blade drops from 0.27 to 0.15 (or 0.12) in 150 inches of media cut which represents a 48% change. That’s a difference of 2.8 times.

Now let’s consider the slope of the data for the S90V and FF blades in Figure 1. This essentially tells us which steel is losing sharpness the fastest. The slope of the FF curve (between 50 and 200 inches media cut) is 0.00033. The slope of the S90V curve over the same range (0 to 150 inches media cut) is 0.00073. This means the S90V is losing cutting ability a little more than two times faster than the FF.

To spare everyone an extremely long post (okay, even longer), I’ll summarize my analysis of Figures 2 and 3. If I look at the data beyond 300 strokes, the data looks fairly linear for all blades. If I take the slope of each data curve (blade), FF is nearly horizontal, i.e. it is NOT loosing sharpness very fast at all. The author defined shaving sharpness as a REST less than 3. Based on the slope of the FF blade, it will take a long time for this blade to exceed a REST of 3. All of the other blades have a much steeper slope and have already dropped below a REST of 3.

Also, it does not appear that the FF blade was sharper initially. All blades had the same initial REST value (0.5 from Figures 2 and 3) which is a strong indicator of sharpness.

That’s my two bits, but don’t use my analysis. We all have the same data. I would like to hear your analysis. I did this very quickly. I should have printed off the information and made sure all my numbers were correct. Let me know if I made mistakes.

TN

 

[Cliff Stamp]

... 33% difference.

Now the first thing you have to do is argue why it is meaningful to compare in percents and ignore the fact that the sharpness in absolute terms drops TWICE as much for the FF D2 blade? You also need to address why the initial sharpness was so high. As I noted the other two blades have high vanadium contents and it is likely a sharpening media problem. If you just translate that curve down, the FF D2 blade would be well behind the CPM blades.

Now let’s consider the slope of the data for the S90V and FF blades in Figure 1.

There is again a huge data bias here because the FF D2 blade is cut for longer and thus goes into the plateau that all blades reach. You can not do a slope analysis here as the curve is nonlinear, I have given the exact curve you need to fit in the past. If the CPM steels were extended they would also go flat in the end, all steels will because of several simple physical reasons. Thus the FF D2 blade will have a lower slope simply because it has more of the plateau.

My main point was that their analysis is very selective. For example look at figure three. If you stop at 400 then the S90V is clearly superior and the FF D2 blade is just infront of the S30V blade. As soon as the S90V blade randomally jumps up past the cut off line the cutting is terminated.

However it should be obvious from the size of the variation that this is not a meaningful difference. There is nothing in that data which would allow you to conclude that at the 1000 point that either of the CPM steels could also be right at the same level.

In order to do this sensibly, you would want to rerun the CATRA tests at least 3-5 times to smooth out the curves and see the real behavior from the noise and then apply a nonlinear curve and use the coefficients to discuss the differences.

-Cliff

 

[Broos]

Cliff,
Now, I analyzed the data without (removed) the first 50 cuts for the FF blade (move the FF curve to the right so the fifth data point is on the vertical axis). I did this because, as Cliff said (and from Figure 1 it looks like), it appeared that the FF blade was initially much sharper. By eliminating the first four data points on the FF blade, it gives the FF blade an initial sharpness (or media cut per stroke) between S30V and S90V, two good steels. Now, doing the same analysis as above, the FF blade drops from 0.3 to 0.25 (or 0.05) in 150 inches of media cut. This represents a 17% change. The S90V blade drops from 0.27 to 0.15 (or 0.12) in 150 inches of media cut which represents a 48% change. That’s a difference of 2.8 times.
TN

This is good way to look at it. I think some kudos should go to these folks for releasing this information as well. I am still looking forward to it getting some real world user testing.

 

[Cliff Stamp]

This is good way to look at it.

If you want to look at selective parts of the data you can support any conclusion you wish to propogate. That is one of the common ways statistics are used to distort reality.

-Cliff

 

[tnelson]

Good thoughts. Let me embellish on my analysis a little.

I am an engineer (engineering geek, according to my wife). It is standard practice to normalize data and evaluate relative differences (percentages) as I did in my previous post. This is especially useful if there are some differences (or some question) in the data (e.g. the initial sharpness?). This analysis did not ignore the difference in initial media cut per stroke (sharpness?), rather it was used as the normalizing factor. This is why I used this approach.

Regarding initial sharpness, I do not believe the FF blade was sharper. The initial REST (Razor Edge Sharpness Tester) values were the same for all blades in Figure 2 and 3. After looking at this information on the REST, I think this is a very accurate measure of edge radius, i.e. sharpness. This should not be ignored. As a result, I don’t think you can move the FF curve down in Figure 1.

Regarding the slope analysis: it was very simplistic, but it is what I could do with the data available. You are correct in that more data is needed in order to fit a good curve to each blade. Then we could evaluate R squared values to determine the accuracy of the fit, R squared adjusted values to evaluate the fraction of variance in the data, etc. Even if we had this, I do not believe it would create any dramatic change in the relative differences of these blades at which point (total media cut) they quit shaving. Once a blade is done shaving, it’s done. I’m ok with this criterion. I feel that once a knife is done shaving, it becomes a “sawing” comparison. I think S30V and S90V will win out in a sawing competition any day. FF was still shaving, the others were not.

It will be great to hear how the knifes perform in the field.

TN

 

[gunmike1]

In short, the information provided raises some interesting questions such as why was the FF D2 blade so much sharper initially (wrong abrasives used I would assume). But give no clear evidence of edge retention increase and in fact show it to be inferior in rate of degredation to S90V by 2:1.

-Cliff

The initial sharpness seems to be the where all of the edge retention "advantage" is gained, as you said. Seeing the actual sharpening process (what machine, what abrasives?) would be very informative, as even if it is finer grained like they state I wouldn't expect to see that large of a difference in initial sharpness if each knife was sharpened to it's true potential. They state a lot about using the same geometries (what are the angles?) and everything being "best processed" for each steel as far as heat treat, it would stand to reason that they should use ideal sharpening techniques or abrasives for each steel so that they reach their own individual peak sharpness instead of a "one size fits all" sharpening. Using the "best processed" ideal you can then add in the variable of the ideal edge geometry for each steel. Since FF D2 is suposed to be so fine grained that you should be able to go more acute with the geometry than the other 2 steels before microchipping rears it's ugly head, that may help to swing the advantage back into it's favor, even if the other 2 blades were able to get much sharper than they were tested at. Of course varying the geometry to find the ideal for each steel would probably take forever and a day, but that would probably make for a more meaningful and useful test, even if it isn't the most practical. It would be interesting to see what kind of initial sharpness they were able to get with the simpler carbon steels, also. Either way they do provide a lot of raw data to look at and sort through.

Mike

 

[IronWolf]

One thing that hasn't been pointed out yet, and will make a HUGE difference in the testing.....

"the D2, CPMs90, and CPM s30 all had HRc values of 59-61, and all had deep freeze cryogenic treatments"

at the END of the article, not side by side with the other test knife information

"as the Friction forged zone is a narrow 1/2" wide zone along the knife edge with HRc values between 65 and 67."

Now that is completely unfair. If I were to use two pieces of O1, one with a properly hardened and tempered balde, and one that was edge quenched, and left untempered, of course I would see a difference in cutting ability.

Ken

 

[STR]

I have not read all the data yet but based on what I've read so far it seems a bit early to call all of the data biased, or even part of it biased particularly when the interpretation is biased to begin with. I think its a bit premature before further testing is done to elaborate on things that may be interpreted from the data in other words. Now that I think about it I could say interpretations made so far are biased to a large extent in that it is selectively picked showing each interpretation as accurate while leaving out other parts of the data that may say the opposite or at least negate the intepretation(s) presented.

Its good that they released it though for it to be dicussed.


STR

 

[Broos]

If you want to look at selective parts of the data you can support any conclusion you wish to propogate. That is one of the common ways statistics are used to distort reality.

-Cliff

I believe you are the one misinterpreting the data here.

You complained that the D2FF knife was "so much sharper" than the other, but the paper states the exact sharpening procedure and same edge geometry was used for all blades, and the tests provide initial sharpness as TN pointed out.

Regardless, TN looked at the retention test and started at the point where the blades cut the same amount of material per cut (not the same sharpness levels). This also showed that the D2FF blade cut more media per cut after 100 or 150 cuts.

I'm not vouching for this steel or these tests, but just looking at the data like everyone else. I would hope that the knife geeks here are happy that a mfg is providing this kind of information, because it is pretty rare, and this thread is a good illustration why most mfg's do not release this kind of info.

 

[Cliff Stamp]

It is standard practice to normalize data and evaluate relative differences (percentages) as I did in my previous post.

The underlying physical behavior has to able to of a suitable modeled. I have shown in the past the exact model needed for knives and the underlying physics of the actual mechanisms for blunting and how you can determine the edge retention advantage. An example is given below.

Regarding initial sharpness, I do not believe the FF blade was sharper. The initial REST (Razor Edge Sharpness Tester) values were the same for all blades in Figure 2 and 3.

This actually tests a different kind of sharpness than what is required for high performance in the CATRA testing. This is another problem with the analysis. If the blades are of identical geometry and sharpness then the initial cutting ability has to be the same, since the latter is false so is the former.

... it would stand to reason that they should use ideal sharpening techniques or abrasives for each steel so that they reach their own individual peak sharpness instead of a "one size fits all" sharpening.

Again, another problem with the comparison

If I were to use two pieces of O1, one with a properly hardened and tempered balde, and one that was edge quenched, and left untempered, of course I would see a difference in cutting ability.

Can you expand on this a little.

I have not read all the data yet but based on what I've read so far it seems a bit early to call all of the data biased...

The statement of bias was in regard to the analysis, not the data itself. You could argue for data bias for the reasons Mike noted, but it isn't the critical issue I wanted to point out which was the improper interpretation of the nonlinear data. I have shown detailed examples in the past of proper analysis of those curves :

The above is one such example. Note that the cutting advantage is dependent on the sharpness end point. This is a very critical note which was strongly promoted by Landes. This ratio can also invert during the comparison which is why the same data can show A superior to B, and B superior to A. This is why proper analysis isn't so trivial.

You complained that the D2FF knife was "so much sharper" than the other, but the paper states the exact sharpening procedure and same edge geometry was used for all blades, and the tests provide initial sharpness as TN pointed out.

This is flawed for reasons noted in the above.

Regardless, TN looked at the retention test and started at the point where the blades cut the same amount of material per cut (not the same sharpness levels). This also showed that the D2FF blade cut more media per cut after 100 or 150 cuts.

You can not look at subsections of data in that manner. I showed in the above if you want to do that you can show the S30V blade is just as good as the FF D2 blade for example.

... this thread is a good illustration why most mfg's do not release this kind of info.

Nonsense, the ones who are confident in their results release the information like Glesser. All such analysis is SUPPOSED to be subject to critical analysis. Science doesn't proceed on faith and blind worship.

-Cliff

 

[Cobalt]

I am an ME (which means nothing at all) as well and I read the initial sharpness as being much greater. I also read that edge curve of the FFD2 being worse than the S90V. FFd2 seems to be about as good as S30V, but no where near as good as S90V in edge holding. if you throw out initial sharpness and bring all three curves up to the same starting point.

But I also have some other concerns here:

"WHAT IS FRICTION FORGING® A KNIFE BLADE?
“Friction” Forging® is a localized forging process using high temperatures
and high loads against a blade to perform the rapid quench. The Friction
Forging® is performed on the knife blade in the area that will eventually
become the edge. Friction Forging® uses a specially designed tool made
from Polycrystalline Cubic Boron Nitride (PCBN), a material second
only to diamond in hardness. During forging, the PCBN tool penetrates
the blade while rotating, which creates frictional heating. When the
tool is fully engaged, it moves along the eventual blade edge, creating
dynamic microstructure shearing and the high forging pressures that
produce excellent blade microstructures. The speeds, heat and down
forces are monitored and controlled by the process computer to ensure"

The above denotes and explains friction forging very well. But it also explains that the performance is only skin deep. It seems to me that friction forging of only the edge is basically like case hardening the edge or hammer forging or rolling the edge only. The remainder of the steel is then differentially heat treated soft to make up for the very hard edge. So you have a severe transition of hardness and thus stresses within the steel. The HT is what will decide how good this knife is. Now that I know more about how this blade is being made the less I am likely to want to use it. I will reserve judgement when others have wasted their money and tested it heartily. But I predict edge fractures along the transition from FF section to non-FF section.

 

[IronWolf]

yes, looking at that statement again, it was a little short.

If I were to take two blades made from the same bar of O1 steel, and make test knives from them to show my position I would do it in this manner.
Both blades would be heated (in a salt pot to eliminate decarb, and to make surt the heat is even) One blade would then be quenched in an oil fast enough to get a full conversion to martensite, and then tempered three times for a final HRc of 59-60, Hard enough to hold an edge, tough enough not to break easily. The other blade would be edge quenched also fast enough to insure a full conversion 1/2" up from the edge, the spine simply air cooling, so it would be softer. at this point the edge should be HRc 65-66. Then finish both knives with their procedure for testing to keep the edge geometry the same.

The edge quenched blade will out cut the tempered blade, it is much harder, it will take much more to dull it. As thin as those edges were, it will even flex over a stainless rod. However, drop it on a concrete floor, or a rock, and see how much of a tip you have left. It will also be a little weaker overall. Yes, it will bend before it will break due to the soft spine, but the fully hardened and tempered blade will resist bending. At the level of torque that the edge quench will stop flexing, and start bending the hardened and tempered blade will still be flexing. It will take more force to bend the second blade, and there is a very small window between bending and breaking, not the wide window of the edge quenched blade.

Hope I made a little more sense this time
Ken

 

[Cliff Stamp]

IronWolf, I understand now, perfectly clear, basically full hard edge/soft spine vs uniform 60 HRC blade.

Cobalt, I would be very interested in the actual alloy distribution after the hardening. They imply it is VERY different from standard D2.

-Cliff