Chisel Hardness Study, the first step.

As I aspire to someday make chisels I've begun a study of the chisels made by others. Since hurricane Harvey has pretty much shut down all outside activity today I did a series of measurements on chisels made by Blue Spruce, Robert Sorby, Marples and Pfiel. Here are the results.Attachment 419138Attachment 419139Attachment 419137

Attachment 419134Attachment 419133Attachment 419132Attachment 419131
The HRC of the various manufacturers covers quite a range from 58 for Robert Sorby up to the very hard 64 of the Blue Spruce butt chisels. Interesting that the softest chisels of the bunch also have the lowest standard deviation. More later.

Rob,

Do you plan to include any kind of research on bevel angle performance through the range of hardnesses?

Cheers,
Luke

Quote:

---

Originally Posted by Luke Maddux

Rob,

Do you plan to include any kind of research on bevel angle performance through the range of hardnesses?

Cheers,
Luke

---

Hi Luke,

I may down the line but those subjects are pretty touchy in these parts.

At this point I'm looking to determine what the characteristics of commonly used lines of chisels are and from that try to infer the answers to questions such as:

1) How hard is hard enough?
2) How hard is too hard or soft?
3) Which lines of chisels have better and worse reputations?
4) What characteristics do consumers want in chisels?

It's apparent that megalothymia (https://en.wikipedia.org/wiki/Thumos) is prevalent in woodworking too i.e. everything must be the hardest, sharpest, most expensive etc. But is super-hardness really necessary or desirable?

Considering my studies of saw blades I can see that acceptable hardness encompasses about a 10 HRC point range, 20 points if Footprint saws are considered hard enough.

Cheers,
Rob

Rob,

Forgive my total ignorance of the subject!
The first table shows values tabulated into 3 columns. I assume that these values relate to your Nominal Block HRC. What do the columns represent?
Are all these values ultimately used to calculate a single "correction factor" applicable to future measurements?

The Measured HRC graph is headed "Measured vs Actual HRC Values...." - isn't a measured value an actual value or is the "measured value" an expected value to be compared with the actual measurement? Sorry if this sounds rather pedantic but I am trying to follow your "investigative thinking".

Yvan

Quote:

---

Originally Posted by rob streeper

Since hurricane Harvey has pretty much shut down all outside activity today

---

Saw a pic in our local news of a San Antonio sign blown down, and thought of you. Presumably you're (only just) far enough inland to miss the worst of it, but getting severely dumped upon?

Hi Rob

It's a very long time since I did stats, and even then I had trouble getting my head around standard deviations and similar.
Now I am possibly wrong, but I seem to recall that for the standard deviation to be meaningful, you need many more than 3 results contributing to the average.
Perhaps, it would be more meaningful to report the range of hardness values you measured
and perhaps your report could use Excel's high-low-close chart to display the results.

As to your conclusions,
I'd posit that the way butt and paring chisels are used is sufficiently different as make hardness comparisons between the two types mostly academic.
For example, a butt chisel is used to make shallow (<1/8" deep) mortices and to pare short (<1" wide) recesses.

Hi Yvan,

Those are the calibration data for the hardness tester. I routinely post the calibration data and curve to show viewers that the instrument I used for the measurements is calibrated and that all of the measurements have been corrected by reference to standard hardness test blocks.

Hardness testers typically have slope and Y intercept offsets that need to be accounted for in taking high accuracy measurements, particularly when they're being used to quantify the differences between two or more tested items.

I got a fair amount of flak on the subject when I started posting hardness studies several years ago, mostly not on this site however so I now routinely post calibration data to preempt such challengers who say "are you sure that thing's calibrated" or "that's too hard (soft), you don't know what you're doing" and such.

Cheers,
Rob

Tell me if this graphic makes it clearer.Attachment 419140

For each of the three test blocks, certified hardnesses of 27.2, 46.1 and 62.2 correspond to X axis values 1, 2 and 3 respectively in the graph above I've plotted the measured (red dotted line) and their actual (blue solid line) values. The gap between the two lines is due to the slope and offset errors I mentioned above. This tester has a very good correspondence between measured and actual hardness measurements.

I'm working hard to convert my data into forms and graphics that are straightforward so you comments would be much appreciated.

Hi Rob

Why is hardness relevant? Can you explain this?

Have you taken into account steel composition, and the manufacturers recommendation in this regard? Blue Spruce is A2, and some of the others variations of O1. I am not sure what Pfiel are?

Regards from Perth

Derek

Hi Derek,

Why? Because it's there of course. :)

The purpose of this effort is to determine the hardness of the chisels that have gone before is relevant in that the results of my analyses will hopefully guide how I will make my chisels. My perception is that hardness is important, but how hard is best is unclear to me. Other processes will also be considered in due time.

Steel composition influences hardenability, toughness, and wear resistance, subjects I hope to address in the future insofar as they are relevant to woodworking tools. The openly available literature in this area is scant. I'm sure that the companies that make industrial tooling have done a lot of work to understand these factors but none of it is available to me.

Regards,
Rob

Quote:

---

Originally Posted by FenceFurniture

Saw a pic in our local news of a San Antonio sign blown down, and thought of you. Presumably you're (only just) far enough inland to miss the worst of it, but getting severely dumped upon?

---

Brett,

No, we're cool, thanks for asking. A little rain and some moderate wind. We're situated on the back side of a hill relative to the storm.

Regards,
Rob

Quote:

---

Originally Posted by ian

Hi Rob

It's a very long time since I did stats, and even then I had trouble getting my head around standard deviations and similar.
Now I am possibly wrong, but I seem to recall that for the standard deviation to be meaningful, you need many more than 3 results contributing to the average.
Perhaps, it would be more meaningful to report the range of hardness values you measured
and perhaps your report could use Excel's high-low-close chart to display the results.

As to your conclusions,
I'd posit that the way butt and paring chisels are used is sufficiently different as make hardness comparisons between the two types mostly academic.
For example, a butt chisel is used to make shallow (<1/8" deep) mortices and to pare short (<1" wide) recesses.

---

Hi Ian,

Yes, the numerical analysis could stronger from a sample size standpoint, that's why I didn't present calculations comparing manufacturers or comparing individual tools within the manufacturer sets. In this dataset I've got 15 measurements on the tools of each manufacturer, 5 per chisel, which is in conformity with ASTM suggestions for work of this type. I'll progressively increase the dataset size as I go along, just getting started here.

It's taken me a couple of years but I now have several thousand individual measurements of saws in my database so it will likely take me a while to build this database up to comparable size.

As in the saw study another objective of this effort is to gain some insight into how consistent the products are from tool to tool.

As to conclusions I honestly have no idea what the optimum physical characteristics of chisel steel are, I only know what I read on the net and what I can measure. I assume that those companies that make chisels know more than I do so I'm hoping to study their examples and draw some insights.

Regards,
Rob

Added data for more manufacturers today. Fixed a math error present in the first post above too.

Attachment 419181Attachment 419184Attachment 419179Attachment 419183Attachment 419182

The Lie Nielsen bench chisel data shows a surprisingly high standard deviation because one of the chisels is softer toward the cutting edge, apparently it has been overheated at some point.

N.B. Group Std. Dev. = Standard deviation of all of the measurements taken on the chisels of a particular manufacturer, Avg. Std. Dev. = Average of the standard deviations of each chisel for a given manufacturer.

Just how many measurements must you make? How big does the data set need to be?
There's a formula in statistics textbooks which is used to decide how big the data set of measurements needs to be.
I've had the luxury of forgetting most of these stats things that we used in the lab.
n=3 is really on the low side, but I'm guessing. 's' is the symbol for standard deviation.

Remember, this is just a first look, designed to give me an idea of what the hardnesses and variabilities within groups are.

For a power of 80%, alpha of 0.05 and a standard deviation of 0.9, (Overall average of the average standard deviations of this dataset.) and an effect size (difference in HRC between two groups) of 1, N is 25. For an effect size of 1.5, N is 11 and for an effect size of 2, N is 6.

To detect a statistically significant difference between the Blue Spruce paring chisels and the Lie Nielsen bench chisels where E is 0.16 I would need to measure 993 chisels. To detect a statistically significant difference between the Japanese fishtails and the Sorby bench chisels where the E is 7.1 N is 1.

Translated this means that if the difference between two groups of chisels is at least 2 HRC units my sampling of 3 chisels from one manufacturer and 3 from a second is sufficient. If I want to measure differences of 1 HRC unit I need 12.5 sets of measurements per group. N gets bigger of course for groups with higher standard deviations.

It's also important to bear in mind that, as in many applications of statistics, these calculations are estimates. Preliminary power analyses of my first data sets on saws seemed to suggest that N should be 24 per group. The N's got smaller however as I added more data.