Tolerance stack and accuracy
I’m going to be doing some posts on how things are made, but before I do it’s important to make a few things clear.
One of the most critical issues in manufacturing is tolerance stack. Ordinarily this is defined as the possibility that a part can be non functional and still be within spec. I could link to Fritz Scholz, who pretty well has the science down, but it’s a classic TL: DR.
Instead, imagine a 2″ diameter shaft ten inches long. The engineer wanted grooves every inch, .100″ deep and .100″ wide. The width of the grooves isn’t critical, soo he toleranced them .100 +/- .005. Then he dimensioned the lands between the grooves .900″.
The Machinist looks at the print and determines he doesn’t have a grooving tool .100 wide, so he used one he had in his toolbox that was .104 wide. He also saw that the other dimension on the print, the .900″ one, had no stated tolerance, so he was very careful to make sure there was exactly .900″ between the grooves.
The top is “As drawn”. the bottom is “As made”
They look really close to the same, don’t they? Lets zoom in to the last groove on the right.
The addition of the extra .004″ per groove added to each segment until at the end, nine grooves later, the groove is .036″ off.
That’s tolerance stack. Proper tolerancing of engineering drawings is necesary to the manufacture of quality parts.
There’s another, less well understood kind of tolerance stack, and that is the tolerance stack of multiple fixturing. When something as simple as a bolt is made, it has one operation- it gets threaded. A shoulder bolt gets threaded, and then has the shoulder turned or ground, but even this is usually done without removing it from the machine. A pipe fitting, on the other hand, has threads on both ends, and is held in the middle to machine. So one end is cut, the fitting is turned around, and the other end is cut. The act of chucking and unchucking will cause the threads to be out of alignment, and it is the amount of “Slop” or innacuracy in the machine’s fixture that determines how far off this will be. Mostly a non issue with pipe fittings, this is a disaster in more complex manufacturing.
The upper receiver on an AR is machined in several steps, and the brilliance of the original Stoner design is that the critical dimensions are machined all in one operation. This means that the bore that houses the bolt carrier group, the mounting surface of the barrel, and the centerline of the threads are all in precise alignment, or as precise as the stated accuracy of the machine. In other words, if the bolt carrier bore is done in one op, and the part is removed from the machine and placed in a different fixture to do the threads, the liklihood of the threads not being concentric with the bore is accentuated. There are some manufacturers that do this, with varied results. This is the kind of “Tolerance stack” that is due not to bad design, or the misunderstanding of the design by the machinist, but to a failure to understand the process. The critical machining must be completed without removing the part from the machine
Here is an example of CNC machining which includes threading. This is simply a demo, but if you were to take the part out and very carefully measure it, you would find that the cylindrical surfaces are all parallel and concentric at a level that can tax the limit of the ability of the equipment to measure, because they were all cut in the same, new machine, without being removed or disrupted during manufacture. The thread roots are concentric with the shaft and cannot be otherwise.
When an AR upper is manufactured properly, the bore, the OD, the face, and the threads are cut in one operation, without removing the part from the machine, so there can be no fixture related tolerance stack. If the face is not perpendicular to the bore, it can only be because it was manufactured improperly. If the manufacturer didn’t go to the trouble to do that right, the whole thing is suspect, and if accuracy is important to you, you should purchase a properly manufactured part. If the face is not perpendicular to the bore the odds are good that the threads are off as well, and that cannot be corrected in a lathe or with some files.
Lots more on this later. Anyway, I just wanted to talk a little about tolerance stack before moving forward, it’s important to the discussion.
34 comments Og | Uncategorized
I’ve seen tolerance stack so mind achingly fucktarded you could actually skip several critical processes and technically still be in spec.
I’ve also seen tolerances applied in ways that make the parts impossible to make to spec, because different dimensions on the print utterly contradicted other dimensions on the print.
Very few things in life are sweeter than watching the 26 year old engineer with the masters degree who has been condescendingly assuring you that the print has to be right get the “Oh SHIT the machine jockey is RIGHT!” look on his face when confronted with this stuff.
Best one I have seen yet is a part that was 0.005″ think, and the tolerance was +/- 0.005″. . . . so air was in spec. . ..
Excellent intro. Now that you have described what ought to be the right way, the next logical question is: are the common parts manufacturers for AR’s doing it that way? But as you said: if they are not, you can’t fix it.
“Very few things in life are sweeter than watching the 26 year old engineer with the masters degree who has been condescendingly assuring you that the print has to be right get the “Oh SHIT the machine jockey is RIGHT!†look on his face when confronted with this stuff. ”
Seems the new ones get that look less frequently.
I’m looking forward to this thread of posts Og.
I came out of my ME degree wishing I had a chance to work with guys who actually did it for a living. And hoping I didn’t screw up too badly. What kind of person comes out of college thinking they know everything? That’s what high school is for.
I find inconsistencies in drawings all the time…Usually from the cut and paste feature in the CAD…and punk assed children/engineers who use it…I was going to say lazy…but workload is a factor too…unrealistic expectations…Just Damn!
From the historical trivia file:
The tolerance stack for the internals of a WWI torpedo could have resulted in machinery too long to fit in the housing.
Something I learned from touring the Old Torpedo Factory in Alexandria. It opened for production 11 November 1918. Timing, like tolerances, is everything.
20 years ago, I was working for (Shall Be Nameless) Electronics Company up here in N.E. Ohio. We had a Contract from DOD to make portable Satellite Antennas for Soccom. They look like small Christmas Trees. The Main Shaft had a Diameter of .750. Bushings would be slid down the shaft, locked into place, and Antenna Stubs would be screwed into the Bushings. We got the Bushings from another Supplier. They came in, and Half of them wouldn’t fit. There was much Gnashing and Grumbling, the Military Inspector from Crane wasn’t happy, the Supplier wasn’t happy, everyone was blaming everyone else. During a Lunch Break, I walked over to Inspection Table, took a handful of the Bushings, and started to mic the I.D. Then I looked at the Print. The Gooberment Spec said the I.D. was .750 +/- .005. All the Bushings were in Spec TO THE PRINT. I asked our Machinist if it would be possible to Chuck them into the Lathe, Open them up to .755, and move on. Not a Problem. When the Lunch Break was over, I walked over to the Cluster Fuck and told them the Emperor had No Clothes, but here’s the Fix.
You could have heard a Pin Drop from all the College Educated Engineers.
We had the Order Completed and Shipped 3 days later, and to my knowledge, none have ever come back for Repair/Replacement.
OOH! Anecdote time.
I once saw a junior officer open a box of chemlights. Activate one. Nothing happened. Takes out another. Same result. Goes through half the box before I told him to put his nods on and look down. The Chemlights were IR visible only. He was ecpecting the regular kind.
2. I once established a Unix-based combat models lab. The new workstations came with optical mice, back wheneveryone else was still using a mouse with a ball in it. Upon installation, we discovered that half of them didn’t work. We sent those back and half the replacements also didn’t work. At that point, someone (maybe me) decided to read the instructions. The metal mousepad they came with was polarized. When oriented correctly, the mouse would work. When turned 90 degrees, no input.
3. Back when zip-ties were new, we were using them to bind detainees. We were experienceing a 50% failure rate. half the time the locking lug woulnd’t engage. Again, failure to read the directions was the problem. Zip ties are directional.
Not as good as your anecdotes, but still fun.
Ah, I remember the days when optical mice had those shiny special pads.
Good times. To be long past.
We (when I worked there in QA) would have the occasional stack-up CF when fitting a multistage steam turbine rotor that’s 15 feet long, with location dimensions at +/-.005.
I think it was just luck that most times things worked out OK; running clearances on those machines are pretty tight.
I call it ‘stacking errors’ when I teach my students precision measuring. Well…. SEMI-precision measuring. We only go to 0.001″ with them.
I have them measure a sheet of paper. Then fold it, and measure two thicknesses. And again…..
By then, the sharp ones are saying something like “Hey! This can’t be right! One sheet is .003″ so four sheets CAN’T be .015!”
That lets me point out that each sheet is actually .003″… plus a smidge…. and the smidges ADD UP.
As a young engineering co-op, I got a design assignment to build a lab scale chemical reactor vessel. It was interesting, but the real lessons were with the lead machinist. I’ve never forgotten the lesson that the power is in the combination of concept and experience.
While my experience with tolerances were no where near the precision of those being discussed here (I built ocean-going cargo barges and freightcars for rail), I can honestly say I never did get a straight answer from the engineers as why a particular sidesheet for the boxcar was specced at XXX-23/32 +/-1/8″ when general ‘shop tolerance’ (if not otherwise called out on the print) was +/-1/16″. It would have been much less of a headache to just call it for XXX-3/4 and you’d wind up with a tighter overall tolerance in the finished product too!
Outstanding post – of deep interest to all of us with interests in metalworking, CNC and/or shooting!
Looking forward to more in the series.
Very interesting, and the moral of the story seems to be, “Talk to your machinist(s).”
Og, I must be missing something obvious. Prom a process perspective, it seems that unchuck/move to new machine/rechuck adds time (i.e. cost) to the manufacturing process. It seems that doing all the machining at once would not only be the high-precision process, it would be the low-cost one as well.
If this is so (and I very well may have missed something), then it makes the quality suspicion even more urgent – if the manufacturer can’t even figure out how to save $$$ (i.e. increase profit) then there’s all sorts of things to wonder about.
Like I said, I might have missed something. Closest tolerances I ever dealt with was building a coffee table in Shop class.
it can add cost, but it’s about equipment. The older and cruder a machine is, the more likely you are to have to do multiple operations to a part. Modern machines have a lot of ability they didn’t have even ten years ago.
Excellent post, I look forward to more. The machinist has to actually make the part, the drawing (or CAD file) can be or say anything. I find many misapplications of GD&T as they sometimes don’t know what they are asking for, or if it is right.
Notice that the lathe cuts with the spindle going forwards and backwards —
I had a six month contract job as a drafter to draw several parts at their extreme tolerance ranges.
They were trying to demonstrate to someone that if all of the dimension went on way, the part would not fit anymore.
The really fun thing was they didn’t believe my CAD drawings and had their model shop make all the parts to my drawings. Sure enough, they didn’t fit.
What really amazed me was the machinist nailed the dimensions first time, every time. It didn’t even take him very long. I am in awe of genuine skill and craftsmanship.
Interesting post. I had never considered just how the process could have such an impact on final quality, living under the happy assumption that, given good tools and conscientious workers, one should be able to make “good” parts with any number of steps.
I believe the last segment is .032 off, not .036 when you think of it.
Its .036 on the outside edge. I didnt calculate it, i measured it.
Nicely put, Ed, thanks for chiming in. You wouldn’t be by chance talking about the company named after a male cervid, would you? Buncha nice guys, those. We looked at some automated finishing projects there some time back. If so, I’ve seen the honing process you describe and like it just fine. The fixtures you’re talking about- do they still use those big twin spindle Stamas? You guys gotta buy some new machines to keep us distributors in business!! Actually, like you said, if it ain’t broke, don’t fix it.
We have four or five “boutique” shops here that are using processes that eliminate the true position issues that you talk about, but then they don’t do anything like the sheer volume you do- most make in a year what the male deer company does in a month.
I must say I’ve always been impressed by the “flip the dust cover over and use it on the other side” thing, that was a stroke of genius.
BTW I assume you mean ‘Half in the barrel extension and half in the receiver”
I guess I am a little slow, but if each land is exactly .900 and there are 10 of them, that comes out to 9 inches. If there are 9 grooves and each is .104 wide, that comes out to .936 inches. But it was stated that the part was 10 inches long. I seem to be missing something, I guess.
What you are missing is that the last land can never be 900 wide. I drew a picture, already! That is the whole problem with tolerance stack.
Sorry boss, I misunderstood my verbal directions – the last segment would be .964. It would have been an inch if we had the .100 tool. Good thing I checked before we ran the parts. A little embarrassed here.
[…] No, not the hippie tree-hugging kind. The useful kind. […]
Liberty: hey, you’re the only one to catch it. The original point is that you can’t have a 10″ bar with 10 .900 lands and nine .100 grooves, the design itself is flawed, and the execution just makes it worse. Pat yourself on the back and put a gold star in your datebook.
You can get stacking problems even in cases other than precision machining. My brother is an architect and once some years ago he says he found one of the other architects making a bunch of calculations down to 1/8 of an inch for a floorplan of a multi-unit condo complex the firm was designing. He asked her what she was doing, and she said the measurements had to be done accurately or the floorplan wouldn’t lay out correctly. He told her that the contractors can’t work that finely.
As I recall the story went, when the contractor was framing the building they got a call about how there wasn’t enough room left at one end of the foundation for the stairs.
Exactly the situation you describe.
Good practical explanation.
It’s also good to remember that CAD can draw and dimension something that cannot be machined on ordinary machines (LMD being excepted).
Working in Silicon Valley in the 80’s/90’s, I found out it was common for companies to put engineering prototype designs into production.
Tolerance stack, and clearances in general, became big problems due to short circuiting the proper design process. And when the product is pushing the boundaries of what is even possible to do, it can snowball into a machine that never gets out the door due to failing Final QA.
Discovered the lens housing (of bolted together 1/2″ plate steel) had a stack that equaled +0.080″, for a lens that was designed to replicate one-to-one a chip design image with 1.25 micron line width geometry. (photo lithography projection stepper)
My experience in the transportation industry is that any assembly with three or more parts often will not work at minimum material condition and will not assemble at maximum material condition. One answer to this reality is to use a process called continuous process improvement to reliably make parts to nominal dimensions at a small fraction or allowable tolerances.
“will not work at minimum material condition and will not assemble at maximum material condition” that’s probably because they were toleranced by idiots. Properly toleranced parts will always, always, always assemble and work fine at both minimum and maximum material conditions, and in any combination therof. That is the entire point of geometric tolerancing. Tolerance stack between parts is just as important as in a single part.