Hand scraping

[gfish]

In order to do precise machining, you need machine tools with precise surfaces. Which, unless you believe god hands down grade A surface plates as well as commandments, raises the question of how you get a precise surface to begin with. I spent the weekend at a class learning some of the basic techniques, thanks to a wonderful birthday present from corivax

The idea behind hand scraping is pretty simple.

Take a piece of metal:


Spread pigment across something that you already know is flat:


Rub the piece of metal against the inked surface plate. The highest points will end up marked:


Take a scraper, which is basically a long chisel, and scrape off the bits that are colored. With a nice sharp scraper and some skill, you can remove very thin layers of metal, easily down to 1/10000 of an inch, and 1/100000 if you're good.


Repeat...


...repeat...


...repeat


This piece took me at least 10 cycles to get reasonably flat. (Note that it is flat, not smooth. Hand scraping always leaves channels between the high points, but that's usually a very good thing. If you have two pieces of metal rubbing against each other, you want room for the oil.) It could have used several more cycles, but I wanted to continue with doing the other side, getting that parallel with the first, and then doing one of the edge at a precise right angle. By the end of the second day I was getting halfway decent at it, and the final surface I scraped is pretty darned nice.



If you've been paying attention, you're probably asking where the reference surface plate comes from in the first place. This is the really cool thing. Using these techniques, you can make a surface plate without any precision references at all. Take 3 rough castings, call them A, B and C. Ink up A, rub B against it and scrape off the high points. Now ink B up and rub/scrape C. Then C and A. Repeat a whole bunch. All 3 will converge on perfect planes, because that is the only shape that can nest perfectly across 3 objects. (If you only used two, they could become matching convex/concave spheres.) The same trick works for making right angles, once you have a surface plate to work with.

It takes a lot of effort and even more patience. This is not something to do if you're easily frustrated. One wrong scrape can add hours of work. The marking process is very exacting -- the differences you're looking at are so tiny that it's very easy to mess things up without realizing it. A speck of dust in the ink will lift up your workpiece and change the markings. Directly holding your workpiece as you walk over to the surface plate can warm it up enough to cause it to warp slightly. Everything has to be done deliberately and carefully for hours on end.

I loved it. It's all a contest against yourself, to see just how patient and exacting you can be. I love knowing that I can make precision out of precisionlessness. This is to machining what processor architecture and assembly language is to CS. You might never use it, but knowing how it works all the way down is still important. I think I just might get some (small!) unfinished cast iron surface plates and do the A-B-C trick, just to have done it. 3 of them at a square foot each would probably take me a day or two of actual labor, but it would be worth it.

Comments

[eleventoes.livejournal.com]

Neat!

[ivy]

That is pretty awesome; I too like knowing the roots of things.

[mdlbear]

The better technique for making flat surfaces, without scraping, is to lap the three plates against one another. Actually grind them, using a sequence of increasingly-finer abrasives until you end up with jewelers' rouge. This is the surface-plate version of mirror-grinding; you'll end up with three optically flat plates, smooth to within a fraction of a wavelength.

Or you can glue a piece of 1200-grit sandpaper to a piece of float-glass (which is made by floating molten glass on molten tin) and have something that's good enough for sharpening jointer knives and plane irons.

[gfish.livejournal.com]

Better for some purposes, but not always for machining. Flat pieces would stick to a surface plate that smooth, making it really hard to use it for measuring and layout purposes. And you can't lap complicated surfaces like lathe ways, which are a parallel V or dovetail.

[peteralway.livejournal.com]

This was reminding me of the art of telescope mirror grinding, and how you can get such amazingly precise spherical surfaces with 2 blanks, and optical flats with three blanks.

[mdlbear]

Good point, though you could always stop with whatever grit left you the right amount of roughness.

Actually I believe you probably could lap a V-groove (by flipping the matching block over occasionally) and end up with something flat on both sides, but not necessarily a precise right angle.

[stolen-tea.livejournal.com]

That's fascinating...

[randomdreams.livejournal.com]

I've read a paper that claims there are some weird multiple-maxima-and-minima shapes that can be formed by the three-plate method, but now I can't find a reference to it. Anyway, they said it was extremely unlikely it would happen in practice, and it wouldn't at all if you randomly rotate the plates with respect to each other every now and again.

[randomdreams.livejournal.com]

Buh. Wholly missed the point: that's really cool. I've always wanted to sit down and do that. I've done it with aluminum but that's different because it's not really useful. The chisel you were using: what was it like? Single or double bevel? Roughly what angle was the tip? I've read about people using glass to do this, or brazed carbide toolbits, but I didn't actually get a chance to talk to them about it.

[gfish.livejournal.com]

These were brazed carbide, though you can use HSS if you want to touch up the edge every minute. It had a profile radius of about 3 inches and a cutting bevel of about 100 degrees. The tool itself was made for the purpose, not just any old chisel:

[samildanach.livejournal.com]

What does a precision-flat surface allow you to do, in terms of further machining?

[gfish.livejournal.com]

The primary direct use of a surface plate is to be a reference datum for measuring and layup. If you want to measure how high something is in thousandths of an inch, you need a very flat surface to measure from.

More importantly, if not something most people actually do themselves anymore, all the precision equipment (mills, lathes, etc) needs precision surfaces. The same basic techniques that you use to make a surface plate are used to make the ways so that the bed or carriage moves exactly parallel or normal to the axis of rotation of the quill or head. These are Vs or dovetails instead of simple planes, of course, but the same ideas hold. At the core, they are made by reference against a surface plate.

[cian-moriarty.livejournal.com]

How did you use the flat you had created on the front of each plate to make the edges 90° to the front and to each other? And how did you get the back of each plate parallel to the front? Please give us some details I am dying to know!

I think I have worked out something that might work...

Ink the edge of plate A, and place it and plate B on plate C with the newly flattened fronts of the plates facing. Gently butt the inked edge of A to the edge of B to be flattened, keeping the already flattened surfaces of all 3 plates in contact. Then reink A if necessary and mark a similar edge of C in the same manner, this time resting A and C on B. Clean the ink off A. Scrape the inked high points on the edges of B and C. Then similarly use B to mark C using A as the rest. Clean B, scrape C. Swap the names of plates A and C and start again.

Is this correct? I have convinced myself that it would work but that doesn't mean I'm right.

Presumably the next step would either be to:

1. Work an edge that meets the newly flattened one. Work would proceed as before but during marking the newly flattened edges of the plate instead of it's front would rest on the reference flat. Maybe after this was complete it would be checked for right-angles to the front of the plate?
2. Work the back of the plate.

The back of the plate could be worked using this technique from the first edge worked to make it theoretically parallel to the front, but wouldn't that introduce error? Or is this error averaged out by checking against all of the other worked faces and shaving if necessary?

Also did the instructors on the course say which edge to right-angle first for greatest accuracy? My guess would be to do one of the longest first as a smaller edge would be more prone to errors from it tipping over slightly.

[gfish.livejournal.com]

The technique we were taught to get sides parallel was to measure the thickness of each corner with a micrometer. Mark whichever one is the thinnest, then scrape that entire face except that corner. Repeat many times, and the new face will converge on parallel with the old one. Since you don't actually care how thick it is, you don't even need to use something as advanced as a micrometer. Something like an uncalibrated height gage would work just as well.

In the class, we kind of cheated to get the right angles, by using a right angle block on the surface plates. We did that last, on the thin face, so we could easily hold the wide sections flat against it.

If you were going to do it completely from scratch, you'd need a surface plate plus 3 blocks (A, B, C). Using the normal techniques, flatten one side of each of the blocks (sides A1, B1, B2). Next, place A on the surface plate, A1 down. Ink the unfinished face, A2. Put B1 down on the surface plate as well. Keeping A1 and B1 flush with the surface plate, rub A2 against B2. Scrape the highpoint, then repeat with B2 and C2, etc. It would be fun to do, but I haven't even found time to start with the little surface plates I got. =\

[ww-kayak.livejournal.com]

Technically you wouldn't have the surface plate either ;)

[phil robinson (from livejournal.com)]

Lapping will only produce two surfaces which match closely, but will not make them either truly flat planes. The same is true of the emery paper approach. for a discussion, read "Engineering Reminscences" by Charles T Porter