The lathe came out of a fittings shop in Lima that closed in 2019 and then sat two winters with a hole in the roof over it. Thirteen inch South Bend, five foot bed, war gray paint under sixty years of chip oil and mouse nest. The serial dates it to 1943 and the man who sold it still had the original card, which almost never happens. I paid 900 dollars and I moved it myself, badly, and my left knee still has an opinion about that afternoon.
I am going to say tenths a lot below. A tenth means one ten thousandth of an inch, 0.0001. It does not mean a tenth of an inch. If somebody says a tenth and means 0.1, they do not work in metal.
The bed was worn the way every job shop lathe is worn. Hard in the first nine inches out from the headstock, where four generations of parts got chucked up and faced off, and hardly touched from 40 inches out to the tailstock end. The front flat way had a step you could catch with a fingernail, and a fingernail finds about half a thousandth, which is worth knowing about your own hands.
Everything came off, kerosene, then two days sitting on three jack screws so the casting rested on three points and was not being twisted by its own cabinet. A bed bolted hard to a cabinet bolted hard to a floor that is not flat reads as bent when it is not. Three points cannot twist anything. Start there or throw away your first week.
What I measured with:
The wear step got measured by bridging the parallel from worn ground to unworn and indicating the drop: worst case 0.0092 inch, three inches out from the headstock, which falls between my stations below. The long shape came from the level, walked along each way in six inch steps and converted to height. That second part is what beginners skip. A 24 inch straight edge cannot see the shape of a 54 inch bed. It knows about 24 inches at a time, and 24 inch truths add up into a lie.
Twist, front way against back, ran about a division and a half over the length. Call it 0.00075 inch per foot. Not surprising for a machine that spent its life on a wood floor.
| Station, inches from headstock | Before, tenths | After three weeks, tenths | Finished, tenths |
|---|---|---|---|
| 0 | 0 | 0 | 0 |
| 6 | -78 | -12 | +1 |
| 12 | -41 | -6 | +2 |
| 18 | -22 | -9 | +3 |
| 24 | -12 | -14 | +4 |
| 30 | -6 | -18 | +4 |
| 36 | -3 | -13 | +3 |
| 42 | -1 | -7 | +2 |
| 48 | 0 | -2 | +1 |
| 54 | 0 | 0 | 0 |
Heights are relative to a straight line through the two ends, and minus is hollow. The first inch and a half at the headstock end is unworn because the carriage cannot reach it, which is what makes that end usable as an anchor. So it arrived with 78 tenths of hollow near the head and left with 4 tenths of crown over the whole 54 inches, and 2 tenths over the middle 30 where the carriage lives.
My uncertainty over the full length is around 2 tenths, because a short straight edge stacked with a level is not an autocollimator. Four tenths means somewhere between two and six. Over the middle 30 I will defend the number, because I can check it two ways and I did.
Crown, never hollow. That got drilled into me by a man named Elmer Dockery at the shop in Bucyrus, who had been scraping ways since Eisenhower and who called me something I will not print when I handed him a bed 3 tenths hollow. A way sitting a hair high wears toward flat. A hollow way lets the carriage tip, and every year after that the machine lies to you a little more.
Around week four I had the headstock end knocked down and the bed reading beautifully everywhere I set the straight edge. Within 3 tenths anywhere, 24 inches at a time. Then I put the level on it and found 18 tenths of hollow through the middle.
That is what indexing a short reference does when you trust it too much. Every 24 inch bite was straight. The bites did not agree with each other, and I had spent three weeks cutting a very smooth curve.
Two more things had gone wrong and both were mine. The camelback had taken a nick from a dropped wrench, and the burr around it read as a high spot, so I was chasing metal that was not there. Ten seconds with a stone would have found it. And I had not proved the straight edge on the granite in six weeks. When I did, it had moved 3 tenths. Gray iron does that. It is not being difficult, it is letting go of stress it has held since it was poured.
All of it came from believing the tool I enjoyed using over the tool I found tedious. @grep/explain-analyze-has-never-lied-to-me is the same failure in a trade I know nothing about, and it made me feel slightly less stupid, which is about all you can ask.
The mistake almost every beginner makes, and I made it at 27 with a witness: chasing points before the surface is straight. Blue looks like progress. You can build an even, gorgeous, 25 point per square inch bearing surface that is bent 15 tenths, and it will feel wonderful under the straight edge and turn tapered parts forever. Get the shape with the long instrument first. Points are a bearing specification and they say nothing about geometry.
Ink the straight edge, not the work, with a film thin enough to read newsprint through, somewhere around twenty millionths of an inch. Thick ink makes everything look like contact and teaches you nothing. Roll it out on plate glass first, then onto the edge. Set the edge down on the way, do not slide it on, two or three strokes over a third of its length, no rocking and no leaning on the far end. Lift straight off and read it wet. Pencil around the darkest spots.
Cut with a carbide blade, one inch wide, in a pull scraper with a handle I made out of a broken shovel. Cross the way at 30 to 45 degrees and change direction every pass, because scraping twice the same direction builds a waviness you will spend a weekend chasing out. A roughing pass with the power scraper takes 3 to 5 tenths. A pass by hand takes half a tenth if you are honest with yourself and two tenths if you are tired.
Count points with an index card that has a one inch square cut out of it. Count the spots inside the window, five places, average them. Crude, and exactly as accurate as it needs to be.
passes tool goal PPI worst deviation
1-6 power rough break the wear step 2-3 78 tenths
7-14 power get the long shape right 5-8 21 tenths
15-22 hand straighten it again 10-14 9 tenths
23-31 hand shape and build points 16-20 6 tenths
32-38 hand fine final points, oil pockets 22-25 4 tenths
Ninety hours on the bed across eleven weekends, plus the saddle, cross slide and tailstock base after, which is another 40 and another story. Five months elapsed, including the wasted three weeks and a February when I did not open the garage. @dust/two-hundred-and-eighteen-hours-on-one-galaxy is a different sort of patience, better written than anything I will manage, but the arithmetic felt familiar.
Oil pockets last. Half moon flakes, light touch, cut so they hold way oil and give it back. They cost a point and a half of bearing and they earn it. A way with no pockets runs dry at the edges and polishes itself into a mirror, and a mirror is a bearing that has quit.
One more thing, which cost me a Saturday to learn. Gray cast iron moves about 5.8 millionths of an inch per inch per degree Fahrenheit, and the length change over 54 inches bothers nobody. Uneven heating is the problem. A shop light two feet from one end of a bed will bow it enough to see on the level, so I worked six to nine in the morning with the door shut and took my hands off the casting for five full minutes before any reading I meant to believe.
Two inch 6061 in the four jaw, eight inches out, no tailstock, 0.015 depth of cut at 0.006 per revolution, fresh insert. Chuck end 1.9698, free end 1.9703. Five tenths of taper, and most of that is the bar bending away from the tool, not the bed.
I put the last pass on it on a Sunday morning in March with the radio off, because I wanted to hear the blade. It is a pile of gray iron and I know what it is. But somebody poured it in 1943 to a piece rate, it spent seventy years in a shop where nobody thought about it at all, and it holds two tenths across the middle of itself after 83 years. I am not going to apologize for saying good morning to it when I turn the lights on.
The old federal specification GGG-P-463c set granite plate flatness as laboratory grade AA equal to (40 plus the diagonal squared divided by 25) millionths of an inch, with inspection grade A at twice that and toolroom grade B at four times. An 18 by 24 plate has a 30 inch diagonal, so AA lands near 0.000076 inch, A near 0.00015 and B near 0.0003. The spec went inactive in 2013 when ASME B89.3.7 replaced it, and the grades and numbers carried straight across. ↩