Posts Tagged Arduino

3D Printer Design Conversation: Part 5

The final installment of musings about building a large-format 3D printer …

(Continued from yesterday)

Perhaps they saw your blog post?

The old-old (original) high-resistance Kysan motor costs something like $45 and, apart from minor cosmetic differences, looks /exactly/ the same as the old-new low-resistance motor. If you were picking motors and didn’t quite understand why you needed a low-resistance winding, which would you pick? Hence, my insistence on knowing the requirements before plunking down your money.

To be fair, I didn’t understand that problem until the Thing-O-Matic rubbed my nose in it. With all four motors. Vigorously.

So, yeah, I think I had a part in that.

comes back to the same numbers over and over

The new-new leadscrews have something like half the pitch of the old-new and old-old threads; I don’t recall the number offhand. In any event, that gives you twice the number of motor steps per millimeter of motion and roughly twice the lifting force. This is pretty much all good, even though it may reduce the maximum Z axis speed (depends on your settings & suchlike).

When it moves upward by, say, 5 mm and downward by 5 mm, you’re measuring position repeatability. That level of repeatability is pretty much a given (for the M2, anyhow), but it doesn’t involve stiction & suchlike.

Can you move the platform up by 0.01 mm, then down by 0.01 mm, and measure 0.01 mm change after each motion?

Do larger increments track equally well in both directions?

Move upward a few millimeters, then step downward by 0.01 mm per step. Does the measurement increase by 0.01 mm after each step?

Repeat that by moving downward, then upward in 0.01 mm increments.

If the platform moves without backlash & stiction in both directions with those increments, it’s a definite improvement.

I wish I knew more
everything you learned is burned into your head forever

The way to learn more is exactly what you’re doing.

Two things I learned a long time ago:

1. Whenever you have two numbers, divide them and ask whether the ratio makes sense.

2. Whenever you don’t understand a problem, do any part of it you do understand, then look at it again.

Also, write everything down. When you come back later, you won’t remember quite how you got those results.

Which is precisely why I have a blog. I search with Google ( microstepping) and /wham/ I get a quick refresher on what I was thinking. That’s why I keep link-whoring URLs: that’s my memory out there!

You’ll sometimes find scans of my scrawled notes & doodles. They won’t mean anything to you, but they remind me what I do to get the answers in that blog post.

modern controllers utilize much higher voltage and current bursts

More or less. Microstepping drivers apply a relatively high voltage, far in excess of what the winding can tolerate as a DC voltage, then regulate the current to a value that produces the appropriate waveform.

This may be helpful:

The mass of the bed APPEARS to be cancelling out any magnetic or mechanical stiction.

That can’t be true in both directions: the gravity vector points downward and the results aren’t symmetric. I think you’re reading noise. If the sequences of motions I described don’t produce the results I described, then you’re /definitely/ measuring noise.

From back in the Thing-O-Matic days:

E3D hot end setups vs MakerGear’s?

No opinion.

I’d want that groovemount post in an all-metal socket, though, rather than the traditional plastic, to get solid positioning and tolerance control. Makergear has the right idea with the aluminum V4 heater block mount.


, , , ,


3D Printer Design Conversation: Part 4

Continued musings about building a large-format 3D printer …

(Continued from yesterday)

taking your challenge and am starting by cloning the M2

That gives you an existence theorem: you know exactly what you want to end up with.

AFAICT, few of the M2’s parts bear standardized numbers you can simply order from a reputable seller. Makergear knows what it’s buying (obviously!), but they’re under no obligation to help out: you must reverse engineer the requirements, find a suitable part, find a supplier, then buy one item.

Let me know how that works out for cost & performance; “cost” should include a nonzero value for your time and “performance” should have numbers you can verify. I (obviously) think the build will be a dead loss on both counts (*), but good data will be interesting.

(*) Albeit useful for educational purposes, which I’ve used to justify many absurd projectst!

How the heck do you read out the current (estimated, obviously) X Y Z position absolute to the machine coordinates?

Perhaps M114 or M117?

My overall list may be helpful, although the RepRap Marlin reference has more detail on their command set:

The LinuxCNC (and, perhaps, Machinekit) G-Code languages give you access to built-in variables and extend G-Code into a true scripting language. Marlin evolved differently and doesn’t support that sort of thing.

G-Code is pretty much a write-only language, but you can do some interesting things:

I use the gcmc compiler whenever I can for actual CNC machining:

Works for me, anyhow, although I don’t do much CNC these days.

move my nozzle up .01 at a time

Stiction / microstep errors / command resolution prevent that:

The only way to measure the nozzle position is to measure a finished part with a known height, because any variation comes from the first layer offset. That’s if you have Z=0 at the platform, of course, rather than whatever offset you get by defining Z=0 at some random height based on jamming business cards / feeler gages / special Japanese rolling papers under the snout. [ptui & similar remarks]

For example:

You need numbers. Lots of numbers. [grin]

strip basic tools out of the control interface

Yet another reason I don’t use S3D: that “Simplify” thing gets in the way of my obsessive need for control.

(Continues tomorrow)

, , , ,


3D Printer Design Conversation: Part 3

More musings in response to questions about building a large-format 3D printer.

(Continued from yesterday)

make a direct clone of the M2. No thinking required.

The present-day M2 has survived four years of rather fierce Darwininan winnowing, so it’s a much better thought-out product than, ahem, you may think just by looking at it.

To build a one-off duplicate, you’ll spend as much money collecting the parts as you would to just buy another M2 and start printing.

Should you buy cheap parts to save money, without considering the requirements, you’ll get, say, the same Z-axis motor Makergear used on the original M2, the complete faceplant of Thing-O-Matic electronics, or crap from eBay described as being kinda-sorta what you want.

Sometimes crap from eBay can be educational, of course:

I encourage thinking, particularly with numbers, because it leads to understanding, rather than being surprised by the results.

increase the rigidity of the X and Y axis

In round numbers, deflection varies as the fourth power of length: enlarge a frame member by 50% and it becomes five times bendier. If your design simply scales up the frame, it won’t hold the tolerances required to produce a good object.

If you add more mass (“stiffening”) to the Y axis, then the Z axis motor (probably) can’t accelerate the new load upward with the original firmware settings and the Y axis motor may have trouble, too. Perhaps you should measure the as-built torque to support your design:

Reduce the acceleration and lower the print speed? Use bigger motors (if you can find a Z motor with the correct leadscrew) and lose vertical space? Make the frame taller and lose stiffness? Use two Z motors (like the RepRap Mendels) and get overconstrained vertical guides? Try building a kinematic slide and lose positioning accuracy? Your choice!

If your intent is to print more parts at once, buy more M2 printers, which will not only be cheaper, but also give you more throughput, lower the cost of inevitable failures, good redundancy, and generally produce better results. Some of the folks on the forum run a dozen M2s building production parts; they’re not looking for bigger print volumes to wreck more parts at once.

Conversely, if your intent is to learn how to build a printer, then, by all means, think about the design, run the numbers, collect the parts, then proceed. It sounds like a great project with plenty of opportunity for learning; don’t let me discourage you from proceeding!

However, I’ll be singularly unhelpful with specific advice, because I’m not the guy building the printer. You must think carefully about what you want to achieve, figure out how to get there, and make it happen.

To a large extent, searching my blog with appropriate keywords will tell you exactly what I think about 3D printing, generally with numbers to back up the conclusions. Get out your calculator, fire up your pencil, and get started!

(Continues tomorrow)

, , , ,


3D Printer Design Conversation: Part 2

Wherein I continue dumping my responses to a large-format 3D printer project …

(Continued from yesterday)

What do you mean by 12 hour mean time to failure

In round numbers, the cries of anguish on the M2 forum seem to increase as parts require more than a dozen hours from start to finish; while you can print things that require 48 hours, that’s not the way to bet. There are more ways for things to go wrong than for them to go right, given the rather rickety collection of software & firmware making everything happen, plus the gummy nature of squeezing hot plastic into precise heaps.

Most of the time, it works fine.

much cheaper hardened polished rod system that the taz 6 uses?

Unless they’re doing something non-obvious to make a kinematic assembly, two rods on four hard mounts with four one-degree-of-freedom slides will be severely overconstrained and, I expect, a continuing hunk o’ trouble:

FWIW, linear slides don’t eliminate the need for a rigid and well-aligned frame. Even the slab atop an M2 can deform by more than 0.1 mm under belt tension, which is enough to wreck the nozzle-to-platform alignment across the length of the X axis.

“Arduino-class firmware (Marlin, et. al.) is a dead end” Why is that?

Marlin is a dead end: they’re trying to jam hard real time motor control, soft real time command parsing, and non real time UI control into an 8 bit microcontroller teleported from the mid 90s. AVR microncontrollers worked really well up through the Cupcake and have held back printer design & performance ever since.

Which inexpensive all in one board would you go with

Machinekit on a Beaglebone seems to be the least awful of the current alternatives, but I haven’t examined the field recently enough to have a valid opinion. You’ll find plenty of proprietary “solutions” out there, none of which I’d be interested in.

Am I wrong?

I think so, but, then, I may be wrong, too. [grin]

It’s incredibly easy to slap together a bunch of parts that look like they should become a 3D printer. It’s remarkably difficult to engineer a reliable, stable, accurate device that actually produces dependable results.

Mooching design cues and parts from here & there doesn’t get you to the goal; if it did, Kickstarter wouldn’t be a graveyard of cheap 3D printer projects.

design a very rigid system for cheap

If it’s for your personal satisfaction, have at it, but a one-off large-format printer won’t be any cheaper than, say, a Taz 6. Some diligent searching will uncover any number of homebrew printer projects along the lines of what you’re considering; learning from their mistakes will certainly be edifying.

Anything is possible, but if you want to end up with a state of the art machine, you must begin with numbers showing how & why it actually meets the requirements. 3D printing now operates at accuracies, speeds, and controls comparable to CNC machines, with corresponding structural demands. There’s a reason high-end CNC machines aren’t made of sheet metal and don’t use 8 bit microcontrollers.

You might want to start at the beginning of my blog and read through my adventures with the Thing-O-Matic, which will explain why I’m such a curmudgeon …

(Continues tomorrow)

, , , ,


3D Printer Design Conversation: Part 1

I recently engaged in a wide-ranging email exchange with a guy planning to scratch-build a large-format 3D printer. He figured it would be a straightforward exercise and asked for some advice; I may be more cynical that he expected.

Over the next few days, I’ll dump my side of the conversation so I can refer to it in other contexts. I’ve left his side of the conversation as the short quotes that prompted my replies, but you can probably infer what he was thinking.

He’s well-acquainted with CNC machining and recently added a Makergear M2 to his collection …

I’m hooked.

All of sudden, you realize what you’ve been missing!

In round numbers, I’ve been designing & printing one “thing” every week for the last five years. Granted, my “things” look a lot like brackets, because they go into other shop projects, but 3D printing is how I make nearly all the shapes I formerly bashed from metal.

I loves me my 3D printer!

an open source design with AFFORDABLE, EASILY ACCESSIBLE parts with a build platform of at least 150% X/Y volume of the MakerGear

Some years ago, I had the same general idea. Then I bought an M2 (replacing my Thing-O-Matic), considered LinuxCNC / Machinekit for motion control, and realized there wasn’t much point; I didn’t want to devote far too much time & effort to solving an already solved problem.

A larger build volume doesn’t buy you as much as you think, while imposing far too many hard constraints. Basically, good-resolution extruders run at 2 to 10 mm³/s, so large objects require print times beyond the 12-hour MTTF of the “printing system”: something will go wrong often enough to drive you mad.

Bonus: plastic’s thermal coefficient guarantees bed adhesion problems. Using high-traction materials (PEI / hairspray / whatever) introduces problems in the other direction. There’s a limit to how big you can make things before they either don’t stick or stick too hard.

Some the fundamental design problems that nobody recognizes until far too late in their design:

  • nozzle-to-platform accuracy < ±0.05 mm
  • XY axis speeds 30 mm/s to 500 mm/s
  • Z axis stiction & backlash < 0.1 mm
  • filament drive with excellent retraction control / speed
  • bed adhesion vs. part removal vs. Z accuracy
  • Arduino-class firmware (Marlin, et. al.) is a dead end
  • Windows is crap in any part of a machine-control problem

Those are hard requirements. At a minimum, your design must satisfy all of them: miss any one and you’re not in the game. It’s easy to build a cheap and crappy fused-filament 3D printer (see Kickstarter), but exceedingly difficult to build one at the state of the art (see patent litigation).

The M2 descends from the original RepRap design, with the Y axis slinging far too much mass back & forth. That kills nozzle-to-platform accuracy, introduces temperature instability, and soaks up bench space. On the other paw, look at the problems Makerbot (not Makergear) had with their direct-drive extruder on an XY platform; getting that right requires nontrivial engineering

Bowden filament drives have improved, but really can’t provide enough retraction control / speed. Delta printers always use Bowden drives, because they can’t sling a direct-drive extruder with enough XYZ speed & accuracy. Bowden on an XY platform has the worst of both worlds: bad retraction and difficult mechanical design.

I think the M2 occupies a sweet spot in 3D printer design: excellent results without excessive complexity or expense. It’s not perfect, but good enough.

But, then, I’m a known curmudgeon …

(Continues tomorrow)

, , , ,


Cheap WS2812 LEDs: Test Fixture Failure 2

A second WS2812 RGB LED in the test fixture failed:

WS2812 LED - test fixture failure 2

WS2812 LED – test fixture failure 2

The red pixel in the second row from the top sends pinball panic to the six downstream LEDs (left and upward). Of course, it’s not consistently bad and sometimes behaves perfectly. The dark row below it contains perfectly good LEDs: they’re in a dark-blue part of the cycle.

The first WS2812 failed after about a week. This one lasted 7 weeks = 50-ish days.

The encapsulation seal went bad on this one and, for whatever it’s worth, the remainder still pass the Sharpie test. Perhaps the LEDs fail only after heat (or time-at-temperature) breaks the seal. Assuming, equally of course, the seal left the factory in good order, which seems a completely unwarranted assumption.


DDS Musings: AD9850 and AD9851

The general idea is to build a specialized sine-wave source as part of a test fixture to measure quartz crystal / tuning fork resonators in the 10 kHz to 100 kHz band. AD9850 / AD9851 DDS modules are cheap & readily available, albeit with crappy two-layer PCB layouts and no attention to signal integrity:

AD8950 DDS module

AD8950 DDS module

Documentation seems scanty, at best. The Elecfreaks page may be as good as it gets, with a summary:

Serial mode just need connect GND,D7 WCLK, FQUP, REST, VCC, I-R

More doodling will be required.

Because the AD9850 has an upper clock limit of 125 MHz, of course those boards sport a 125 MHz oscillator-in-a-can. The AD9851 has a 180 MHz limit and an internal 6× multiplier, so it gets a more reasonable 30 MHz oscillator and can run at either 30 MHz or 180 MHz.

The included 70 MHz (?) reconstruction filter won’t do much to improve a 60 kHz signal. A much much lower outboard lowpass filter will be in order.

As far as the AD9850 goes, the output frequency comes from a 32 bit value determining the phase increment:

f = Δφ · osc / 232


  • f = output frequency
  • Δφ = phase increment
  • osc = oscillator frequency (30, 125, or 180 MHz)
  • 232 = DDS phase counter width

The smallest frequency difference between successive phase increments is thus osc/232, which is 0.029 103 83 Hz at 125 MHz. Obviously, you can’t get nice round frequencies like 60.000 kHz, except by accident, but you can get close.

Alas, the resolution conflicts with tuning fork characterization, where you (well, I) want to step the frequency in nice round 0.1 Hz units across a 2 Hz range around 60.000 kHz. Because the crystal characterization requires closely spaced frequencies, where the difference between the test frequencies matters, you shouldn’t work from the nicely rounded frequencies on a display.

For example, successive values of Δφ produce these frequencies near 60 kHz:

  • 2 061 580 → 59.999 874 793
  • 2 061 581 → 59.999 903 897
  • 2 061 582 → 59.999 933 006
  • 2 061 583 → 59.999 962 104
  • 2 061 584 → 59.999 991 208
  • 2 061 585 → 60.000 020 312
  • 2 061 586 → 60.000 049 416
  • 2 061 587 → 60.000 078 520

In an ideal world, you could pick an oscillator frequency to produce nice increments. For, oh, say, 0.025 Hz increments, all you need is osc = 0.025 × 232 = 107.374 182 MHz. Riiiight.

The computations require more numeric resolution than built-in Arduino data types and math operations can provide. Floating point numbers have 6-ish significant digits (double is the same as float), which cannot represent the Δφ values or frequencies. Unsigned integers top out at 32 bits (unsigned long long int is not a thing), enough for 9 significant digits that can hold the Δφ values, but integer multiplication and division do not produce 64 bit results and overflow / underflow without warning.

Other than that, an Arduino would be just about ideal: the generator needs a small display, a knob, and a few buttons.

Perhaps storing precomputed Δφ values for specific frequencies in a table, then computing nearby frequencies as offsets from that value would suffice. This will require doodling some absurd significant figures.