DIY Ball Bar for CNC Machine: Laser, Webcam, and Flexure Design

[@ianloy1854]

Love your work and the way you share all of the development - its amazing what you achieve with a web cam and a laser.

[@stezz90]

For the buckling of the flexure you can make the blades thicker in the center, leaving only the "hinges" thin and flexible. Also increasing the overall thickness of the assembly should help with stiffness in unwanted motion directions.

[@PatrickHoodDaniel]

Excellent work! I love it's ability to nicely flex and your experiments with the 3D printed version is nicely done. Love these types of experiments. It would be nice to see this as a standalone tool where maybe it has its own display, logging the values and displaying it on the display in a circular fashion like on the desktop.

[@chiparooo]

Very cool! I’m familiar with the commercial ball bars. This is next level for diy CNC calibration. Very interesting data and iterative work. Thanks for sharing!

[@michaeldean4362]

This is a super cool project. I love how approachable this solution is. I’m currently getting my DIY CNC mill up and running again after moving and will definitely be looking into characterizing its error with a DIY ball bar like your’s. Thanks for the great videos!

[@zyzzyva303]

Now I know what a ballbar is. Nice approach and I especially like the flexure with the laser. Since you're using lasers, throw in a mirror or two and get down to the nm range.

[@robertschoemer1250]

All this reminds me of my CNC machine tool career in the 80's. We used a HP laser to check the ball screws and Hamar lasers to check alignment. Now if someone could come up with a DIY laser like the Hamar it would a real help for DIY CNC builders.

[@ianloy1854]

I was using CNC's back in the late 70's and nothing like this was ever seen. I can understand the need if you are working below 10 microns. but we were working above that level and in a total uncontrolled work environment 0-35C on large pieces 600mm up to 7000mm so expansion and contraction mattered more, but was never allowed for ..... The CNC machines were WAY better than the manual machines, ball screws vs ACME and digital readouts vs dial and dial indicators. For those that don't know - all of the Gcode was written by hand and was transferred to the machine by "paper" tapes that had holes punched in it. Archaic by todays standards - but the results are the same - just involves WAY less coding and walking. :) 70's was the peak time for manual machines - ball screws started to enter and the machines themselves were well developed.

[@BryanHoward]

People now have it too easy with all the advanced CAD and CAM software. There's parallels with software where nobody writes low level assembly which I guess you could say is similar to writing gcode by hand. Way too easy for someone to buy linear motion components and a spindle from China and through a CNC machine together in there garage too.

[@hrmny_]

​@@BryanHowardtoo easy makes it sound like it's a bad thing 🤔

[@bernhard_derProtoTyp]

Thats super cool and interesting! Thanks for sharing your progress!

[@AbdulazizAl-Khater]

Awesome project! Very nicely done.

[@maggieMakesMechs]

Another awesome video man thank you

[@F4bb3rs]

Awesome work! A video on simulating the compliant mechanism would also be interesting!

[@ferriswhitehouse1476]

another banger idea!

[@jaedinthompson1845]

I just wanted to say that the high frequency wobble in the measurement could come from 2 sources in my mind , 1) a c5 grade ballscrew has a +/-8 micron of axial error per revolution and +/- 18 microns of error per 300mm of lead, 2) I've been looking into bearings for my own project. If you're not specifically ordering P6 or lower grades of bearings for the BK/BF housings your axial runout can easily be in the double digits of microns. Otherwise it may just be vibration as you're making a realtime measurement

[@compucar03]

Neat! I'm working on making a ballbar adaptor for my 3d printer right now! Using a QC10 so different method

[@lukealchinsmith]

What's a ballbar for?

[@compucar03]

@@lukealchinsmith makin babies Or in the case of a cnc machine, exactly what this video talks about. Determining the error in things like squareness, circularity, backlash and some other dimensions of a motion system.

[@chip-load]

@@lukealchinsmith ​ @compucar03 When creating a circle one axis REVERSES at the start of a new quadrant. The test shows the quality of both your mechanical and control electronics. It also shows the orthogonality of your mechanical structure.

[@ewildgoose]

I think you can just measure runout and then correct for it? Rotate the spindle precisely by 90 degrees and measure the change. Half the 180 degree error gives you runout and as long as you know the spindle orientation it's just a linear XY offset. Essentially just add it to the initial position and treat it as a constant

[@BryanHoward]

The main issue you'll have with trying to do it that way is you have a few places where the runout can be on. It could be on the spindle, collet, 3d printed joint and the lapped washer. It's actually pretty quick to get the runout removed with the screws. An improvement to the design would be too use smaller fine pitched bolts with brass heat set inserts. I opted not to do it because I was worried about the brass inserts getting pulled out but I think it would work out now.

[@ewildgoose]

@@BryanHoward true. But as long as your bearings/joints are very close to being spherical, then it really doesn't matter (I think)? Essentially the runout reduces to being a case of the centre of rotation is shifted in XY versus the centre of the spindle. So as long as you rotate the centre of the runout around the actual centre point, then all should be well? Renishaw is coming around to this idea with their probing macros for spindle probes. Blum has always worked this way (as far as I know). Quite cool 😎

[@tobeproduced]

Using the washer seems like alot of work for very little gain, consider using a tri-bar kinematic copping, it reduces the uncertainty of the system

[@dmitryplatonov]

If the distance is constant, you can use LED with lens to get smaller dot.

[@paul3555]

Like The Project, could it be that your 100 micron error comes from the fact that both points arent on the same plane and if you move out a certain distance along the axis the hypothenuse is maybe longer ? and second idea maybe try an z yoint for your complient mechanism or mirror one side of the joints because at the moment you laser is not moving at a straight line over your sensor. would like to maybe one day build along, love it wonderfull project for diy enthusiasts

[@BryanHoward]

The 100 micron offset is just the scaling factor. I tried determining the sensor size (to get the scaling factor) by taking a closeup photo of it with the known length of the PCB as reference dimension. A better way is to use gauge blocks with the sensor configured in one of my earlier laser/webcam projects measuring relative height changes. I won't be doing it with this webcam though. I’m going to be changing the webcam to a camera module that is compatible with the camera bus on the Pi Zero. There is a number of benefits of going that route. But if one had to find the scaling factor of an unknown sensor without a datasheet like this one in the video, gauge blocks would be the best way to do it. You’d need to be working in a temperature controlled room as you can easily see fluctuations of 30um due to thermal expansion.

[@stefanguiton]

Excellent

[@blauesKopftuch]

19:53 I'd guess the scale is off because of the smoothing. You can make the value drift further to the edge by moving the smoothing slider further to the right. I'd guess it is due to sensor width being devided by the reduced pixel count (e.g. 1920 - 2x smoothing (once per side)) instead of using the native µm~pixel ratio. Thus the µm per pixel increases when increasing the smoothing.

[@BryanHoward]

The scale is off because I haven’t calibrated the scale factor as good as it could be. The smoothing slider will have a negative impact if the gaussian hill is clipped too much off the edge of the sensor. I find it should only be used sparingly to remove some of the high frequency noise but no further. I need to properly calibrate the scaling factor using some gauge blocks. It can easily be done by stacking them and offsetting the webcam sensor or the laser (see the previous videos I have on the subject). I’m going to change the camera sensor and webcam in the future. Buying a proper camera module I’ll get the datasheet that’ll tell me the pixel size.

[@AlexGontmakher]

I wonder if you could use components of a (high resolution) optical mouse as a sensor

[@dmitryplatonov]

I would switch lapped washers with 3 hardened rolers/pins for true 3 point contact.

[@samuelschuur7044]

Are you worried that with your system where you clamp with the Z- axis to place the base you might be deflecting your gantry and offsetting your mesurements slightly as a result?

[@BNETT21]

Would a 18650 on the bottom of the bar along with a pi zero w work as a counterbalance to keep it level enough and remove the cable headache?

[@BryanHoward]

The whole setup is quite rigid that I don’t think it’d need a counterbalance. Also a counterbalance would limit you to only working in the XY plane and couldn’t measure in the XZ/YZ planes. I am looking into a Pi Zero with a camera module over the camera bus. Make it all wireless and work with a raw camera feed instead of mjpg over usb.

[@BuildItAnyway]

What is your max error during 1 rotation. Curious what it is.

[@BryanHoward]

I'll check it tonight. I think it's quite large because I have not squared and trammed the machine after putting on the be ball screws. I think it's upward of 400 microns. I might make a video on trying to better interpret the results and a before/after on trying to remove error in the machine. Might look at how mapping ball screws work in linuxcnc too.

[@BuildItAnyway]

@@BryanHoward any specific reason you did not go for lvdt?

[@MrJTJINX]

I was thinking of using an LVDT (linear variable displacement transducer) en.wikipedia.org/wiki/Linear_variable_differential_transformer as the transducer for a ball bar tester. I figured it would be variably adjustable pretty much in the same way you made the flex coupling. would it be possible to adapt your design to incorporate such a device instead of the laser/camera idea. My concern would be the dynamics of the plastic in your flex joint or refection's on the camera lens from the laser. I would be very interested in your thoughts.

[@BryanHoward]

Yeah I looked at those as well. They can get quite costly on Mouser. I don’t see why they couldn’t be incorporated into a flexure design. Especially one that has mirrored flex arms. I went with using the laser and webcam sensor because I’ve already built the software around it and know it can give reliable results. The webcam doesn’t have a lens on it. That was taken off and I’m shining the laser directly onto the silicon. A lens would have all kinds of distortions that would skew the results.

[@kesor6]

From the video, it doesn't look like the two balls are on the same Z height. Would it be advantageous to have them at the same plane while doing the calibration? or is having them just on the same ball circumference good enough? Did you have to consider polar coordinates when drawing the results?

[@BryanHoward]

Yeah it's probably best to get the balls on the same plane for best results. But I think as long as the Z axis is not moving, it won't hurt things too much. I see the main problem being if you're working with absolute distance on the ball bar you don't get that. I'm mapping the data in polar coordinates.

[@kesor6]

@@BryanHoward the gradient on the camera sensor might be an oval instead of a circle as well.

[@GuyFawkes911]

you are awesome

[@leonordin3052]

🤩

[@AdamMaszynotwor]

🦾🦾🦾🦾