3d Z-axis errors

[erniebro]

My first project on my recently completed Joe's 4 x 4 mill is to cut a plug for a hand launch R/C glider fuselage. I designed the shape in Rhinoceros and generated a .stl file. I then imported the .stl file into Cut3D and generated a roughing and finish path cut file.



I then setup a piece of MDF and cut the part. "http://www.youtube.com/watch?v=pXjkQ3paNe4"]YouTube - Joe's 4x4 cutting

The results look like this:


I was pretty excited to be able to get this far but then I started to look closer. The finished part has several areas where the z-depth appears off. Being new to this game I thought I must have some real problems with the mechanics. I spent the next several days tuning up my mill, looking for sources of error. I replaced LoveJoy connectors, I tightened down the v-bearings on the carriages, I realigned the ACME screws and end bearings and a bunch of other stuff I won't list here. Anyway, the machine is now running very smoothly with very little to no backlash. I can cut perfect circles all day with no flats or gouges. I cut another plug and there was some noticeable improvement in the quality of the rasters, but there were still seemingly random z depth banding problems.

Then I shifted my focus to Mach3. I know it is doing a lot of work in an interrupt driver with limited time slices, so my theory was to give Mach more time to compute. I'm running Mach on a 1.8 GHz Intel dual core processor with 1 gb memory so there should be plenty of CPU cycles, but you never know what Windoz is doing. The computer is running WindowsXP and has been stripped down to the bare essentials. It's not connected to any network, I stopped all the extraneous background tasks, turned off the screen saver, virus checker, disk defragmenter, power management, etc. When I look at the task manager it looks pretty clean.

I tried changing a bunch of parameters and cut more plugs.


I started out by slowing down the Z axis. I tuned the motor acceleration from 15 down to 5 and reduced the velocity to 80 ipm. That made some very visible improvements. The edge where the fuselage meets the bottom plane smoothed out a lot and the cuts were much more regular. However, the banding was still appearing.

Then I tried upping the frequency of the interrupt task. I went into the Config Ports and Pins menu and changed the kernal speed up to 60000hz. I think that means the kernal is running 60,000 times a minute, pretty fast, but not much of a load when the CPU is running at 1.8 billion cycles per minute. All that did was just put the bands at different places.

Then I did a kind of counter intuitive thing. I changed the z-axis jumper on my controller board to generate 1/2 stepping. It was set at 1/8th stepping. That means I went from 6400 steps per inch down to 1600 steps per inch. Mach now only has to compute 1/4th as many steps on the z-axis. Surely this would solve the computational problem? No dice. Just moved the bands to another location. Surprisingly, I couldn't see any differences in the accuracy of the z depth of the cuts. When cutting wood and MDF going from 1/6400" to 1/1600" resolution just isn't going to be visible. The stepper motor sounds very raspy at the lower stepping mode, but it works just fine.

Which brings us to constant velocity mode. I think there may be some fundamental errors inherent with the way Mach computes constant velocity. My next step was to bite the bullet and change to exact stop mode. Execution times were much longer and the mill was definitely vibrating lot. Oh, well.

Here's the results -


The cut starts at the nose and moves towards the tail. It got to the middle of the tail section with out a hiccup and then a band appeared. I didn't hear the motor stall and I can't detect any dings in the rails or ACME screws in that area. So close....

This is getting very frustrating. I am excited because I now think my mill is capable of generating pretty high quality surface finishes, but I don't know how to solve the software problem.

What should I try next? Is it possible to cut 3d parts using Mach3? Should I try Quantum? Is there other software out there that works better than Mach? Does the new SmoothStepper board solve these types of problems? Should I break up my finish pass into several smaller files and run them separately? Does anyone have experience with EMC on Linux?

Or is this some other problem like random electrical noise? I'm using a HobbyCNC pro board hooked to a parallel port in my Dell pc.

I guess I can always just cut the plug oversize and sand out the imperfections...

Thanks for any input-

Ernie

[ger21]

Run the mach3 driver test, do you get smooth results?
I'd say it's one of three things.

1)Resonance
2)Something in the PC interfering with Mach3.
3) still might be an acceleration issue. Drop it in half again and you can eliminate this one if it does the same thing.

Since it's only happening on the Z axis, you could also try swapping the X and Z wires on your drives and see if the problem moves to the other axis.

What drives, motors, and what size power supply are you using?

And if you leave the task manager open while cutting, does the CPU usage ever hit 100%?

Also, Art always recommended a 2Ghz or faster CPU for any kernel speeds over 25Khz.

[protman16]

If your LookAhead is still at the default (20) increase it to about 200. Reset your accel and steps/unit values back to what started with.

This should help.

[fatal-exception]

That banding looks like missed steps, coupler slipping on the motor shaft or something else mechnical to me.

If you write a program to do a few hundred z ups and z downs, does it return to the same position as you started at?

Paul

[erniebro]

First, thank you everyone for the input.

I tried out your suggestions and I am still getting the banding problem. The z axis looses or gains a step for multiple rasters and then it gains or looses it back to where it should be. This results in either a slightly gouged band or slightly raised band. (See pictures above)

I replaced the LoveJoy couplers between my motors and the ACME screws with some couplers from Dumpster ( http://www.dumpstercnc.com/acme_couplers.html ) and now there is absolutely no slippage and no backlash in the coupling. Great product. The ends of the ACME screws are held by very sturdy flanged bearings and the set screws are all tight. The ACME screws are very securely anchored. No slop there. I'm using Dumpster anti backlash nuts on the screws and they have very little to no backlash. The mechanics are very tight at this point. Here's a video of the z-axis while cutting the file:
"http://www.youtube.com/watch?v=yg4NJZPtJCU"]YouTube - Dumpster Couplers on Joe's 4 x 4 CNC mill

I don't hear any resonance while running the machine. The audio on the video gives you a pretty good feel for how smoothly the machine runs (except when it is in exact stop mode...)

I lowered the kernal speed back to 25000, cut the z acceleration in half again, changed the look ahead to 200, turned off constant velocity, turned it back on, changed the constant velocity angle to 5 degrees, etc. There are a lot of possible combinations of a lot of Mach variables. More than I can test in my lifetime...

I've been running the task manager all the time now and it registers between 3 and 9 percent cpu usage while cutting. I haven't seen any spikes above 9 percent. There appears to be lots of head room. Over 700 mbytes of memory are in the free pool and I don't see any paging going on while the job is running.

I tried moving the wiring around. I separated the power wiring to the router from the cable run for the stepper wiring. I swapped cables. Nothing seems to help so far.

I think I'm going to have to live with this bug...

[protman16]

What are your motor tunning srceen settings for all axis? Post everything you see on the screen.
Also you might want to check if your drives require an active HI or an active LOW for step pulses.

[fatal-exception]

If you think it's the software/hardware, post your code for the file, and I will try cutting it out...

Paul

[ger21]

Increase the pulse width to 2 or 3.

[erniebro]

Here are the cut files. I cut the hlgRough first using a 1/4" upspiral square endmill. The hlgFinish file is cut using a 1/8" ball endmill. The part is about 2.5" x 15". You may have to cut a shoulder around the top after the first cut depending on the reach of your 1/8" ball mill.

[Khalid]

Please see if ur controllers are far away from each other (if u are using modular type)...check again for loose connections, any high power equipment near ur controller....(Speakers etc)..try to run MAch3 alone no other program running while cutting...i hope this may solve...

[keithorr]

I think I'm going to have to live with this bug...

Is that the name of your machine? "Bug"?
I looked at the toolpath in simulation and there isn't anything in the text file. You mentioned loosing and gaining steps. You can't gain steps, you can only loose steps. The program doesn't store pulses somewhere and inject them into the program at unexpected intervals. Your machine axis is is loosing steps in two directions which is a mechanical fault, not software.

[erniebro]

I haven't named my machine yet and it won't be "Bug" but thanks for the suggestion. LOL

Here's another picture.

The machine is not randomly gaining and losing z steps. There is a definite pattern. The z axis gets off at the beginning of a raster and keeps the same error for some number of rasters and then restores to the correct z depth at the beginning of another raster. I have seen no examples of the z depth changing incorrectly in the middle of a raster. Also, the depth errors always happen in pairs- first up and then some number of rasters later back to where it should be.

It would be a very clever mechanical problem to exhibit that behavior. Anything is possible, but I don't know what to look for mechanically. There is no slippage in my z-axis motor/shaft coupler, the motor is very securely fastened to the mill, the ACME screw is supported in a very beefy flanged bearing with tight set screw. There is no unwanted mechanical movement of the magnitude showing up in the cut error.