Abstract on automatic leveling and calibration

[Generic Default]

I don't know if anyone else has brought this up yet, but it recently occurred to me that automatic bed probing can actually eliminate the need to use precision linear rods.

For example, our rostocks use T slot aluminum, which is straight to several thousands over its length of a meter or so. Precision linear rod can be much more accurate and straight over the same length.

But if the bed is probed at multiple points, like a 7x7 grid, the correcting algorithm unintentionally accounts for error in the linear motion (such as mismatched arm lengths, bent rails, ect). It won't work with 4 point leveling because that only checks the radius. The more math intensive algorithm assumes that the bed is lumpy or curved when it is actually the rails that aren't straight, but the end result is the same for both. I guess the moral of the story here is that as we move towards more automated calibration, we can use cheaper build materials and get the same precision as with more expensive, precise parts.

Hopefully this made sense.

[teoman]

I have been pondering about the same thing.

Attach a probe and let the printer go crazy tuning every parameter. But there are a lot of parameters. Like 30 or more.

[dtgriscom]

The problem is that you can correct for any set of errors (bed, arms, towers, etc) at Z0, but once you start increasing Z you must assume that everything moves in parallel or you will start distorting. For instance, if the towers lean out, then as you increase Z then the actual Z height at (X0, Y0) will get higher and higher (as the towers pull out the arms, decreasing their angles). However, points that are near a tower will be pulled in towards the center of the printer (as the opposing towers pull out their arms, swinging the adjacent tower's arm out towards the center).


Dan

[teoman]

If you have a digital dial indicator directly hooked up to the system the you can also measure that.

You can get readings at several different levels to determine the differences related to height.

[dtgriscom]

If you have a digital dial indicator directly hooked up to the system the you can also measure that.

You can get readings at several different levels to determine the differences related to height.

Yes, but you wouldn't get the horizontal errors. It's possible that knowing the vertical errors would let you infer the horizontal errors, but with the number of independent variables in play I'm betting not.

Dan

[redoverred]

If you have a digital dial indicator directly hooked up to the system the you can also measure that.

You can get readings at several different levels to determine the differences related to height.

Yes, but you wouldn't get the horizontal errors. It's possible that knowing the vertical errors would let you infer the horizontal errors, but with the number of independent variables in play I'm betting not.

Dan

What if you put a high definition camera directly in the effector? You could place some sort of calibration grid on the bed and measure the horizontal error.You could have a super narrow angle lens to ensure the best dimensional accuracy.

[dtgriscom]

What if you put a high definition camera directly in the effector? You could place some sort of calibration grid on the bed and measure the horizontal error.You could have a super narrow angle lens to ensure the best dimensional accuracy.

Interesting. Actually, use a standard-resolution head [edit: I meant "lens"], and you can measure the distance to the bed as well, which should give you three dimensions of measurement (if the effector stays perfectly level, which actually doesn't seem that likely...)

I'd rather have three fixed cameras, one at each base, watching the nozzle.

(Actually, I'd rather have a hardware system that was inherently precise and didn't need calibration. And a pony.)


Dan

[Generic Default]

We know the tolerances on standard hardware. The vertical aluminum extrusions are the biggest source of error, since they are all curved at least a little bit. As long as the base is accurate so that three towers are very, very close to 120 degrees and the linear slides are perpendicular to the base plane, delta printers can be very accurate.

With 400 rev steppers and good microstepping, positioning resolution can get down to within a few microns. For that level of accuracy, we also need precise, concentric, smooth, low friction arms. We need excellent repeatability for Z height and bed leveling.

I guess my point is that we should all think about ways to make our machines more accurate.

[dtgriscom]

At this point in my 3D printing career (i.e. pretty much a noob) I don't need absolute X and Y positioning better than a millimeter, but Z=0 positioning close enough to ensure good bed adhesion (within 0.1mm?) and delta X, Y and Z close enough for consistent layer thickness (within 0.05mm?).

[teoman]

At the lab we had used a laser on the effector and a small ccd on the platform to calibrate our delta.

Industrially usually a calibration grid used with a camera. The end effector is swapped for kind of a needle and the operator moves the robot with a teach pendant to desired locations on the grid to synchronize the camera and robot plain.

So, the problem or process is reversed. Instead of the robot moving and detecting where it is, the robot is moved first to a specific location and then the calculations are performed.

[redoverred]

I saw this method that someone is using to calibrate bed level: https://www.kickstarter.com/projects/de ... sts/555370

They are using an accelerometer and a probe to gently tap the bed and determine where it is. Seems like an interesting way to do it. I'm sold on a narrow-angle lens with an HD camera and a special grid on the Onyx plate. Then you can calibrate the XY positioning and use some other method for Z like the one above.