Additive manufacturing: Terrible tolerances

[Generic Default]

As the post title suggests, I've come to believe that all additive manufacturing methods and machine types are inaccurate. I've noticed this over the last few years, but I was expecting it to improve and it hasn't.

Most of the tables I check say something like "+-0.008" for the first inch, and an extra 0.003" for each inch after that." This is more or less the same for SLA, SLS, DMLS, FFF, ect. PolyJet (the resin based one that is similar to an inkjet printer) has better tolerances, 0.005 for the first inch and only 0.0005 for each inch after that.

When I use a small nozzle on my Rostock Max with magnetic arms, I typically get 0.005 for the first inch and 0.0015 for each inch after that. The tolerances on flat sides are significantly better, it's circular features (especially holes) that are undersized or oversized.

Anyway, my point is that additive machines are extremely inaccurate compared to older methods like milling, turning, and grinding. The standard tolerances for post 2000 CNC lathes and CNC mills is a few ten-thousandths of an inch. CNC tool grinders operate within a few microns.

I don't believe that additive machines are inherently inaccurate; I think it's a lack of precision built into the machines that causes low tolerances.

Does anyone know of an extremely accurate additive manufacturing method that can make macro sized objects?

[Jimustanguitar]

I blame the materials. I feel that my machine is more accurate than the plastic that it prints. Things contract as they cool (as you know well), and the combination of irregular shapes and gridded infill make for a pretty wonky shrink that would take some serious computing to model and predict.

It doesn't bother me, though. I'm a decimal place or two away from what you're worried about, and for the kinds of jobs that I do, that's just fine.

This brings up an interesting thought that I've been throwing around lately, though... What if imperfect, printed parts were scanned with metrology equipment, and then somebody way smarter than myself wrote some code to compare the model to the print and then suggest changes to the model to make it accurate? Print a part, scan the print, calculate a new model, lather rinse repeat until machine learning could figure out the common tweaks and changes to apply to every model...

We'll probably see closed loop positioning before that, though.

[JFettig]

I would never expect any of these kit printers to hold even .005", you can get some dimensionals that hold that but overall, if you did a high accuracy 3d scan you'd be lucky to hold .010 over a large print. I'm including my custom build and yours as well as the TL max metal builds. Heck, we have a hard time getting these printers to print a flat first layer near the extremes, I can't imagine they print any better higher up.

One of my customers uses SLA machines almost exclusively and they hold pretty tight tolerances and have very high resolution. I don't have exact numbers to quote but I'm happy with the results that I get from them.

FFF printers probably won't ever get much better than the commercially available printers that are properly calibrated and built.

Its an inharently inaccurate process with lots of variables while machining and grinding is quite accurate. If we built printers like we build these machines, we would be one step closer to accuracy, then we'd have to deal with shrink and extrusion widths just like we currently do.

[bot]

Another large variable, especially for deltas, is the segmentation and other math tricks used to fudge the toolpaths. Until we have firmware for something like a Dynomotion KFLOP, the controller will be a limiting factor in accuracy as well as the others previously mentioned.

[Jimustanguitar]

Not to mention the polygons inherent to STL files and the faceting errors.

The problem has many partial causes, and the solution will need to as well. Hopefully new advances like the 3MF file format can help these faults and limitations in the open source end of things. Assuming that the firmware keeps pace and that control boards move up in processing power, many of the model and motion limitations that we have today will improve considerably in the next few years.

[briankb]

I know I'm gonna get beat for this -- but I do not see using a delta configuration for high tolerance work. As bot just said, the math is wicked. I've just about given up on using a Delta configuration for anything I expect to fit together with a high tolerance. Delta's are fun to watch and fast for a lot of objects that don't have to be measured with a caliper though.

My hope is we see a lot more progress with SLA/DLP and SLS in the DIY machines and especially the materials. In my mind SLS will find a home with printing metals and ceramics and SLA/DLP in synthetics or things than can somehow be dissolved to a solution.

Whatever the case, it's a fun ride to be on

[Captain Starfish]

Deltas and precision...

My belief is you've got to close the loop to get that precision. Using servos with encoders instead of steppers closes the loop at the servo but there's just too much mechanical scope for error on a delta downstream of the drive motor. And with a 3D printer you can't just chuck a linear encoder on the effector like you can on a cartesian.

STL format's approximation errors on curves, look-ahead algorithms rounding corners, mechanical error all come into play. But the very nature of the materials and the process - let's make something gooey, layer it up and let it set as it cools - means I'm surprised we get something as accurate as we do.

I think the FFM process is a great way of creating a blank shape without huge amounts of hogging and scrap metal wasted, much like a casting of molten metal. And, just like casting, it's often good enough to be getting on with. But, just like casting, if you need precision: you need to oversize the part then use subtractive machining to achieve that precision.

And I think that might be what we start seeing more of, five or more axis machines with toolchangers, one of the tools being a laser sintering unit for deposition, then the rest being milling tools of various types to get the part into spec.

[lightninjay]

I know I'm gonna get beat for this -- but I do not see using a delta configuration for high tolerance work.

Just felt like I should put my 2 cents in about what can be achieved with the accuracy and size of our particular delta machines.

I just finished my 2nd iteration of my cello today. This one is electric, includes a small class D amplifier and a speaker with a pre-amped, piezo-electric pickup.

Quite a few parts that require decently tight tolerances, such as the tuning pegs up top, and the battery compartment just above the fine tuner box.

Sounds pretty good in my opinion! I'll probably make a more comprehensive post about it in the "What are YOU making?" forum later.

[briankb]

That is amazing! Obviously, I just haven't found my groove with a delta yet.

[Captain Starfish]

Beautiful work.

And you demonstrate an important point: how accurate is accurate? How precise is precise?

A peg in a hole for an instrument string needs to be accurate to within about 0.2mm if tapered, maybe 0.1mm if it's straight through. The plastic deforms a little to take up the slack and so on. Inserting a sliding fit metal piston or dowel in a metal block needs something closer to 0.02mm to be right.

Horses for courses, and for an awful lot of courses (just like cast metal in my earlier analogy) the 3d printer horse is just fine.

But I agree with Generic Default - I don't think the additive manufacturing techniques we use now are ever going to be refined to the point where they are as accurate as good quality subtractive equipment.

[Generic Default]

I'm actually not as pessimistic as you guys about the potential; I think filament type machines will max out at tolerances of around 1 or 2 thousandths. I think a lot of the current inaccuracy comes from the machines being designed and built on price points more than performance points.

We're already close to common injection molding tolerances with FFF printers. The surface finishes aren't as good but we can print varying wall thicknesses without warping and drooping. I want to build a machine with an integrated 5 axis CNC mill for light machining, basically just shaving the outside surfaces and holes so that they're more precise and smooth.

The real accomplishment of reprap and DIY stuff like what we do is that we/they managed to reduce the price of CNC machines by 90%. The electronics are often the biggest cost, and we can get a working 4 or 5 axis machine with less than 500 dollars worth of electronics now. On 3d printers the extra axes are used for extruders.

[Captain Starfish]

Fair enough.

If you're looking at a build, I reckon you're better off taking an existing 5 axis machine and adding an FFM head and extruder to it as a tool. I've got a 3 axis machine going reasonably well, 4th axis is there but needs a little work to get it going. 5 axis is a whole new world of pain and I'll be another year or two getting my head and hardware around all that. Then (in 2021?) I expect to be adding a hot-end and extruder to it somehow.

[briankb]

I agree about the price over quality but the price had to go down to spread as fast and wide as it did. Now we just get to complain about it as the honeymoon is fading into normalcy and we are pushing those machines to higher and higher standards.

We're already close to common injection molding tolerances with FFF printers. The surface finishes aren't as good but we can print varying wall thicknesses without warping and drooping.

What are the tolerances expected for modern injection molded parts?

I want to build a machine with an integrated 5 axis CNC mill for light machining, basically just shaving the outside surfaces and holes so that they're more precise and smooth.

I think that is an ideal mashup. Are you talking about milling the parts from a FDM printer like PLA,ABS,ect..? If so I think the parts to make the 5-axis can be reduced considerably in power and especially size. I remember seeing a blog post by Bart Dring on Inventables using an rc motor as a spindle for shapeoko2.

[bot]

I'd like that same machine!

[geneb]

5 axis is awesome, right up until you find out what the software costs to create a 5 axis toolpath. Then the crying starts.

g.

[Captain Starfish]

It's insane.

Looking at PTC for work a while back, $5k for the base CAD, add $5k for basic 3d CAM, add $5k for 4 and 5 axis indexing/positioning, add $5-10k for full 5 axis machining. Per seat. On a half million dollar machine, that $25k looks reasonable. But sitting next to a hobby machine that cost maybe $5k? No, it's kind of silly.

There's only one solution, I'm going to have to save up for a Haas machine so the software doesn't seem so ridiculous by comparison...

Actually - a hell of a lot of "5 axis" work is really 3 axis work with two extra axes that just save multiple discrete fixtures and orientations for the workpiece. Time saving, convenient, more accurate (not having to re-align the part every few operations), but not essential. And, if we can pocket and profile stuff layer by layer, 3 axes would be fine for everything but overhangs.

[briankb]

I believe Fusion360 includes up to 3-axis (vs 2.5), I wonder if they will ever add 4 or 5 axis CAM to Fusion360.

What is Pocket NC http://www.pocketnc.com/ using for their 5-axis machine?

We are really excited about the offerings of Fusion 360 and believe that it will make a huge difference for our users as well as a ton of makers, educators, and start-ups the world over. Autodesk has committed to offering the software for free (including 3+2 machining capability!) for hobbyists, educators, and businesses making less than $100,000 per year. See this great blog post by CEO Carl Bass for more info on their vision.

- blog post mentioned http://forums.autodesk.com/t5/design-di ... -p/5496355

Answered my own question! Autodesk is making their 5 axis CAM available to Pocket NC customers. AND it seems it may be available to anyone, I have not found a definitive answer on this yet.

[geneb]

It's not 5 axis. It's 3+2. A 5 axis tool path can result in all five axes moving simultaneously. What that "cloud" crapware does is NOT 5 axis.

g.

[gchristopher]

...I do not see using a delta configuration for high tolerance work. As bot just said, the math is wicked. I've just about given up on using a Delta configuration for anything I expect to fit together with a high tolerance. Delta's are fun to watch and fast for a lot of objects that don't have to be measured with a caliper though.

I'm just a programmer, so can someone explain what is "high tolerance work" in terms of pieces of plastic?

I use a run-of-the-mill digital caliper to measure prints and the delta printer is definitely more accurate than my ability to use the calipers. (e.g., if I squeeze the plastic between the calipers, I can move the measurement by 0.05 - 0.10 mm.) Exterior dimensions reliably come so close to the model values that I can't measure the difference. Am I just really bad at caliper-ing? I bet I am.

Interior holes, especially vertical ones, seem to always be smaller than modeled, but they're repeatable, so I can tune a good radius by testing for a particular material/temp/extruder/print speed combination and get good enough results to be self-threading with a machine screw, or wiggle-free with a smooth rod. That seems like about all I could expect from a plastic piece, whether FFF or injection-molded. (Though it's definitely a lot more work to tune than exterior measurements.)

Can plastic pieces really be significantly more high-tolerance than that? What are they used for? (That sounds like it must be something really cool.) Is my lack of ability to measure things making me miss out on a whole world of awesome precision stuff?

[Generic Default]

Plastic part tolerances are worse than metal part tolerances because plastics are typically 50-200 times less stiff than steel. And they change with humidity and temperature a lot more than metals. And most plastics creep over time.


Real 5 axis CAM software is typically expensive, but the majority of parts don't need full 5 axis profiling. The 3+2 axis stuff can make most things. You can program it by hand in some cases, like if you just need to drill holes or mill stuff at an angle other than 0 or 90 degrees.

On a 3d printer, an integrated 5 axis machine would not have to be built to the same tolerances and stiffness as a half a million dollar metalworking machine. Getting accuracy to around a thousandth of an inch isn't that hard to do, and the cutting forces would always have to be less than the adhesion holding the printed part to the build plate. Using a small end mill to flatten the top and sides, as well as interpolate holes to more accurate sizes would not be that hard for mechanical parts.

You could generate your gcode with a slicer that already exists, like Cura, then run a C program to inject code in at specific layers or before / after the printing.




If you need full 5 axis interpolation where every axis is moving or rotating at the same time, you would have to have full 5 axis software. I don't see why it would be that difficult to program. If you have a triangle (polygon) with three coordinate points in 3d space, you can get a center position and a surface normal easily. STL files have this information. Once you have the position and the normal, you convert it to an angle and offset the position by the normal multiplied by the length of the cutting tool from the 5th axis. That's about all you need to calculate the (X, Y, Z, P, Y) gcode that the 5 axis machine reads. You just make a bunch of these coordinate sets from the STL file then have the machine interpolate between them.

[626Pilot]

Another large variable, especially for deltas, is the segmentation and other math tricks used to fudge the toolpaths. Until we have firmware for something like a Dynomotion KFLOP, the controller will be a limiting factor in accuracy as well as the others previously mentioned.

Smoothieboard 2 is going to include an FPGA, and according to Arthur Wolf, the missed-step detect pins on their drivers will be wired up to the controller.

Filament width sensors would be a nice start. I measure all my filament in ten places about a meter apart, but that doesn't compare to letting the machine do it in real time.

I know I'm gonna get beat for this -- but I do not see using a delta configuration for high tolerance work. As bot just said, the math is wicked. I've just about given up on using a Delta configuration for anything I expect to fit together with a high tolerance. Delta's are fun to watch and fast for a lot of objects that don't have to be measured with a caliper though.

If I could get a math nerd to help me figure out how to implement tower lean correction, that might be made better. I'm not that good at analytical trig.

Of course, a lot of people are printing delta printer structural bracing at 150mm/sec on delta printers that are also using structural bracing that was printed at 150mm/sec. They use Z-only correction to level the bed, and they run the probe so fast that it's probably noising up their results. Great if all you ever do is print statues, I guess.

I think that is an ideal mashup. Are you talking about milling the parts from a FDM printer like PLA,ABS,ect..?

You can tap and sand PLA, but machining it would require absurd cooling equipment. It will soften the instant you hit it with any kind of rotary tool.

[briankb]

You can tap and sand PLA, but machining it would require absurd cooling equipment. It will soften the instant you hit it with any kind of rotary tool.

I didn't even think about possible friction heat on PLA, I ASSumed it wouldn't be much of an issue as it such a soft material. I bet there is a sweet spot with feeds and speeds that would work with PLA, without cooling. It will be interesting to try though. I have a shapeoko2 with the variable speed spindle, hopefully maybe this weekend I can try.

[Captain Starfish]

Might be more a case of low spindle speed, high feed rate and really, really sharp tools with a lot of rake to minimise contact/rubbing surface area, too.

[626Pilot]

When I was designing 3D printed fixturing for my big Probotix CNC, I had an idea at one point to run the tool along the periphery of the fixturing to make sure it was compliant with the CNC's coordinate system (similar to a surfacing operation to make sure your spoilboard is orthogonal with the router's coordinate system). It failed. The PLA was ripped out in globs, and what was left behind was certainly not flat. It was a mess!

I don't think it's worthwhile to think of milling PLA unless you have some crazy liquid-cooled tool. (Not the motor, the tool itself.) I put some PLA pieces in the dishwasher and they came out warped like a piece of cardboard that's been outside too long. I don't think you can expect sub-dishwasher temperatures milling PLA with a drill spinning at tens of thousands of RPMs. Perhaps ABS would be a better target material.

[Generic Default]

I think a printer with subtractive capabilities would have to be able to cut any plastic or any material you print. The main problem I think would come up is the maximum holding force.

Normally, parts being milled or turned are clamped in a chuck or a vise. On a printer, you would have to rely on the adhesion between the part and the bed, unless you popped it off and re-clamped it. But then you have to worry about alignment.

Plastics are known for being hard to machine because of thermal expansion, burrs, lack of stiffness, ect. I found a video on youtube that shows a nylon block being milled. Notice that the mill only turns at like 100 RPM.

https://www.youtube.com/watch?v=fQVe9x0VwSE

A smaller endmill with longer flutes might be better, and you could definitely run at at least a few hundred RPM without melting problems.