Kb58's granite CNC router build

[kb58]

I have never dealt with a project requiring so much precision on so many levels. There are so many surfaces that need to be within a fraction of a millimeter for the router to work well. Any misalignment will not only produce objects that are "off a bit", but can also cause the servos to bind up if they have to fight misalignment. That last bit has really been beating me up. What doesn't help is not having $10,000 worth of metrology equipment to do it right, never mind much faster and more easily. On top of this is the concern that because I bought the ball screws and linear guides used... it begs the question: why it was removed from equipment? Was the equipment being recycled, or was it because there was a problem with the bearings or drive screws? Indeed, things have made strange noises from time to time, then it all goes back to working fine - a very mixed blessing.

On top of this is buying a CAM package that's designed for home hobbyists, and while that's helpful up front, it suggests that it may be limited once venturing out of the children's pool. Already it's doing some strange things, mostly seemingly unnecessary moves across the work and then back again. I've been in contact with the support staff about its difficulty at machine a simple slot. There is virtually no documentation, so I pester Support to explain everything - that's what they get for having no manual. Anyway, I'm considering returning it and buying a serious CAM package, which is very much at the deep end of the adult's pool, where instead of it setting most variables for you, it lets you do Everything, which means the user has to know what they're doing...

Lastly is the annoying little voice that's been saying, "if you made the dust shoe how you normally would, you'd be done weeks ago." Yes, and I wouldn't have a working router or know how to run the software...

 

[kb58]

A new issue... or, revelation:

When the servos and control board are powered up, I noticed that the servos shift position slightly. This is not a surprise given that the controller doesn't know where anything is when powered on because it doesn't have absolute position encoders. Being magnetic, servos and steppers jump to the nearest magnetic "slot", be that electronic or mechanical. So, a given servo may or may not shift position slightly, and may or may not move in the same direction as another servo.

This normally doesn't matter since each axis is typically controlled by one servo, and once homed (zeroed out), everything gets back in sync. The exception is an axis powered by two servos, where both servos must move in lock-step. Powering up in different positions is unacceptable on two counts:
- If the gantry is flexible, it'll offset one end from the other, causing a misalignment that can persist if the homing operation has no way of moving each servo independently. Worse, the offset may increase over time if additional power-ups add more offsets.
- If the gantry is infinitely stiff (as some owner mistakenly believe), the gantry will stay aligned, but the servos will be trying to move to their differing power-up positions. They can't move, but they will be drawing some amount of current which causes unnecessary heating. Worst case, every time the machine is powered on, the gantry moves slightly further out of alignment, causing higher and higher holding current, and may eventually overheat the servos.

How to fix this:
Up until now, the servos have been sharing the same commands on my router, so there was no way to correct any offset in one servo. Controller manufacturers are aware of this and developed software that separates control of paired servos, allowing dynamic correction of offsets. Even though the controller doesn't know where anything is at power-up, it can use the limit switches if they're offset slightly by a known distance. This ensures that one end of the gantry always reaches the end of travel first. They then "unpair" the servos and slowly drive the lagging end forward until it hits its limit switch, then backs it off by the offset amount, ensuring it ends up square. It also makes it easier to align because it can be checked with an accurate square, and whatever error is found can be added or subtracted to the offset number to improve accuracy.

 

[kb58]

I don't know who originally said it, but Carroll Smith, race crew chief, quoted it in one of his books:
"The racing car is an inanimate object; therefore it must, eventually, respond to reason."

Substitute "CNC router" for "racing car" and that's where I am right now. The journey to getting it aligned and stable in operation has been, hmm, somewhat trying. Gantry alignment has been a veritable kettle of monkeys and if doing it over, I'd consider using one servo instead of two. Anyway, that mole has finally been whacked, now that dynamic auto-squaring is enabled, calibrated, and working well. So I'm almost done. Almost.

There is still a pesky suspicion that there may be an issue with one of the gantry servos. Even moving slowly, it reports torque overload errors, even though the torque that it's producing is about 5% of maximum. Turning the shafts manually shows little difference in effort, yet the other servo never ever complains. I may switch their positions to see if the problem travels with the servo, or not. I'm in contact with the manufacturer's excellent Support department and expect some answers tomorrow.

 

[kb58]

Talked to the ClearPath manufacturer and all is clear now. We made several setting changes and that's what the servo was complaining about. No harm done. Tomorrow I plan to finally mill the first part of the dust shoe. What's proving more of a challenge at the moment is heat-forming a strip of polycarbonate (Lexan) to curve around the front. While more effort, it's much more elegant than square corners. So far though, it's proving difficult to get a smooth bend, while at the same time not overheating the material and causing it to craze.

 

[kb58]

Like Whack-A-Mole, the issues keeping success at bay are slowly falling. Somewhat ironically, the seemingly-simplest part of the work flow - CAM - is proving to be a pain, providing quirks both large and small. One is that when a project is saved, it forgets about half the settings. So, you set up a job, save it, close the app, then later come back, reopen the app, change one thing, and run the job again, and it fails worse than the first time. I eventually realized that a bunch of settings had reverted to default values - but not all of them. Huh?

Then there's how it managed to correctly route the work piece to a depth of 0.25", yet machine a step to 0.087" instead of 0.125". Both are dictated by the imported CAD model so I don't understand how it can "partially screw up". Sigh, software...

 

[kb58]

Well this is kind of a funny story...

I contacted the Support department regarding the issues I'm having with their CAM package. I fully expected to be told that I had misconfigured things, set up stuff wrong, or left out various steps. They replied the next day with a modified version of what I'd sent them, which was pretty sweet. Loaded the job, put in a piece of polycarbonate, and watched it work. It ran more efficiently than what I had configured, with less jumping around across the work piece. When it was done - it was clear it was still messed up. Holes that were supposed to be 0.625" were 0.615". Worse, the step around the edge of the panel that's supposed to be 0.125" wide is 0.198" wide. Huh?

If the company selling this can't get it to work, what does that say about the product? I initially felt slightly better, that the problems weren't entirely my doing, but now there's trying to convince them to give me a refund, and then starting the search all over again for a good CAM package that works with Alibre CAD.

 

[kb58]

The search for another CAM package came down to four demo versions. One was immediately eliminated due to being subscription based. Two others are probably off the list because the user interface was clunky. The last was a bit frustrating in a funny way - it's the most expensive - but the salesperson offered to make a short video of running my file through the app, and he did, and it's pretty nice. The issue is that it's $2,200 - I can't justify it since I'm not in business. After going through the four, curiosity got the better of me and I tried the CAM app that caused me much grief, just to approach the issue with fresh eyes (this is the app that the seller couldn't get to work either). Tried doing things slightly different and it worked. While I'm happy, it could still be a problem with future jobs if each job requires randomly trying things different ways until it works. Shown below are most of the failed attempts and test pieces. I don't want to be repeating this again with every job, or with expensive stock.

So anyway, the top of the dust shoe and the vacuum adapter are done. Three more pieces need to be designed: the bottom plate; a magnetically attached ring, and a magnetically attached brush. I managed to heat-bend a strip of polycarbonate to use around the sides once the bottom plate is done. Fun times.

 

[kb58]

With fabrication seeming to have been solved, I of course changed things to accommodate thinner polycarbonate side material. The 1/8" material was tricky to heat-bend, with surprisingly strong spring-back, and ended up "not resulting in the desired level of workmanship I wish to present" - bending too much in areas and not enough in others. The concern was that even if forced into a fixed radius, it wouldn't stay there while the glue set up, plus attaching the second plate would prove difficult. The new panels now have a teardrop-shaped groove for 2mm-thick polycarbonate which will become the wall connecting the upper and lower panels.

Because the CAM app (MeshCAM) worked okay - yesterday - I thought about contacting the seller and offering to pay for it again (they're refunding my money). That thought went away -again - because while routing one of the panels, the cutter deviated off the CAD model's edge, then returned... why!? It wasn't enough to ruin the part but was yet another quirk that's proving it very untrustworthy. The CAD model is good, yet this CAM app just decides sometimes to do its own thing. It's a shame because it's so easy to use, but its unpredictability is a real problem. I may try and help the manufacturer (the actual creator, not who I bought it from) to make it a better product. Will be interesting to see where this goes.

 

[kb58]

First, the good news:

After way too many weeks, the dust shoe is finally done! Making a dust shoe seems to be a rite of passage after building a router, and for good reason. After watching chips fly everywhere at the first cut, and getting tired of holding the vacuum hose on it the entire time, the shoe seemed more like a requirement than a nice-to-have. Connecting the vacuum and lowering the spindle into it for the first time resulted in a great deal of wind noise, so ear protection is no longer an option.

The bad news - well, not so much for me, but for MeshCAM - is that pointing out issues seems to be falling on deaf ears. I sent an email with a pic of their simulated router's path showing a strange and unnecessary cut. I also said that the physical part did in fact have that issue. Their response: "That's just an artifact of the rendering software and is not an issue." Sigh, yeah well the part says otherwise, but okay... As a really smart mechanical engineer once said: "No one knows anything, the results speak for themselves." Indeed.

 

[kb58]

I've been in touch with the spindle manufacturer concerning two issues:
1. Every time the coolant pump is switched on, bubbles are seen emerging from the spindle. I can't figure out where all that air is coming from because there are no leaks, and both the source and return lines are submerged. The manufacturer surprised me by saying that the coolant lines should be cleared after each use - which doesn't address the mystery. I asked how that's supposed to be done, compressed air?

2. Spindle speed has always been unstable, traced to using insufficient shielding. Switching to double-shielded cable in addition to a small capacitor across the input solved the problem.