Accelerometer and resonance measurements of the motion system.

Discussion in 'Motion System' started by Ntesla66, Nov 22, 2019.

  1. Ntesla66

    Ntesla66 Active Member

    Joined:
    Sep 9, 2019
    Messages:
    34
    Likes Received:
    31
    Let me be absolutely clear from the start. I'm not making any blanket statements here concerning the performance or operational characteristics of the E3D Tool Changer nor Motion Platform.

    I cannot and will not assume that anyone else's machine has the characteristics that I've measured on my own machine.

    This post may easily fall into the tl;dr category but I don’t know where else to post it ,so here it is anyways.

    I undertook a series of tuning prints with a ringing test cube and was dismayed to find that there was nearly nothing that had an effect on ringing in the Y direction. I was very methodical starting slow and changing only one thing at a time. I then took the tool changer gear off and increased the spring tension on the compliant mechanism that holds the tools in place by using a stiffer spring. If the resonance I was seeing was caused by the tool rocking against the tool plate this would at least have changed its frequency... it did not. Finally I pushed the carriage all the way to the back and centered the tool holder and then tightened the belts quite a bit, plucking them on both sides, I measured the frequency with a microphone and o-scope . The high tone and low tone was 274Hz and 135Hz respectively(do not tune them this high it isn’t good for the machine). This got rid of quite a bit of ghosting on the Y axis. I then played around with M593 quite a bit and ended up with it set at 55Hz although I believed there was a lower fundamental in the carriage near 34Hz.

    I then obtained and attached several adxl337 accelerometers to the carriage and to the hot end of one of the tools. I’ve included pictures of how they were attached. Great care was taken to ensure that neither the accelerometer vibrated in its holder nor the holder itself vibrated against the attach point.

    The accelerometers power was supplied by a Siglent SPD3303x-e set to 3 volts dc and their XYZ outputs were connected to an HP35670A dynamic signal analyzer set to take an FFT spectrum power density measurement and average the samples. When taking a full XYZ compliment, I could only capture one accelerometer at a time as the signal analyzer is only four channels. To compare two accelerometers, I took only the XY components of each. A Rigol 1054 o-scope was monitored during measurements to ensure signal integrity and level of clipping if any in the output of the accelerometers during rapid traversals.

    The impulse I used to excite the system was 1mm step inputs starting with either the carriage or carriage and tool centered in the printing area and then stepping only in either the X or Y direction at one second intervals until the average sampling was completed. Most of the measurements were made with a twenty sample envelope.

    I did not attempt to gauge the accelerometers against an absolute reference, so all measurements are relative to the platform itself. However the test system was quite adequate to indicate the data I was looking for.

    The data was acquired from the analyzer over GPIB and parsed in MATLAB and then plotted.

    Results, I’ve no intention of turning this into a dissertation, so succinctly here’s my take on it. There’s no evidence that the tool holder is rocking or vibrating at the compliant point of the spring holding it in.
    However the longer lever arm that the tool head represents is magnifying the resonance present on the Y axis. The M593 command has an effect but many times only to move the resonance elsewhere (Bode's sensitivity integral). I’m not entirely sure but I’ve a suspicion that M593’s effect may shift resonances onto the other axis because of the inherent coupling in the corexy kinematics.

    There is a resonance on the Y Axis that is not controllable by M593. The resonant frequency that I’ve observed is at 45HZ with the belts tuned to a plucked frequency of 108Hz and 68Hz. There may be a belt tuning where the cross coupling effect of M593 and the natural Y resonance may cancel and ringing abated (my next endeavor). I suspect the long side belts as the cause but that’s pure conjecture and not proven.

    Here is the data and some of the photos of the setup. In some of the graphs I forgot to label the data in the Acceleration legend, it was always X Y Z top to bottom in the colors or if four then XY XY.

    AccelX_Tool_Mosq.jpg AccelXY_comp.jpg AccelXY_comp_stpx.jpg AccelXY_comp_stpy.jpg AccelY_Tool_Mosq.jpg XY30CubeComp44.jpg XY60CubeComp44.jpg XY60CubeComp50.jpg XY60CubeNoComp.jpg XYcomp41.jpg XYcomp41Ystp.jpg XYnoComNoToolYstp.jpg XYnoCompWtool.jpg XYnoComwToolYstp.jpg

    Setup

    174240scaled-min.jpg 182324scaled-min.jpg

    Cubes that go with the data
    cubesxscaled-min.jpg cubesyscaled-min.jpg
     
    #1 Ntesla66, Nov 22, 2019
    Last edited: Nov 22, 2019
  2. Amr

    Amr Well-Known Member

    Joined:
    Jun 2, 2019
    Messages:
    130
    Likes Received:
    31
    Great work, appreciate you sharing the results, I don't have access to such tools so I am grateful you went the extra mile, and looking forward for your conclusion.
     
    Ntesla66 likes this.
  3. Paul Arden

    Paul Arden Well-Known Member

    Joined:
    Aug 28, 2019
    Messages:
    162
    Likes Received:
    58
    Given the motion system part is pre-assembled you’d think that to some degree such research would be in some way transferable. There does seem to be a fair bit of variation in belt tension even in the pre-assembled units based on comments people have been making here. This is a really cool idea though, must less subjective.
     
    Ntesla66 likes this.
  4. Ntesla66

    Ntesla66 Active Member

    Joined:
    Sep 9, 2019
    Messages:
    34
    Likes Received:
    31
    I think in the long run it will, Paul. I just didn't want to be misconstrued as having said all these motion systems would have this response because I've changed the tool head and already mucked about with the belt tension.
     
  5. Greg Holloway

    Greg Holloway Administrator
    Staff Member

    Joined:
    Sep 4, 2015
    Messages:
    1,020
    Likes Received:
    631
    Fantastic diagnostics!

    If there is anything we can do to help please drop me an email.
     
    Ntesla66 likes this.
  6. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
    Bear in mind that for M593 to reduce ringing effectively, you need to set the acceleration limits of the machine high and the jerk limits reasonably low. Unfortunately, some slicers include M201 and/or M204 commands in the GCode file they generate, and this is likely to prevent RRF using a high enough acceleration to cancel the ringing.

    What speeds did you configure for your 1mm step input, and what acceleration limit did you set?

    Ideally, the jerk limit would be set to zero or near-zero allow M593 to work best, but this is impractical because it causes stuttering when printing curves. Still, it would be interesting to see the effects of different jerk values on the ringing that you measure.
     
  7. Ntesla66

    Ntesla66 Active Member

    Joined:
    Sep 9, 2019
    Messages:
    34
    Likes Received:
    31
    Max Velocity was 400mm/sec both coordinates , Acceleration was 3000mm/sec^2 both coordinates, Jerk was 6.667mm/sec both coordinates. The weight of the tool was 423 grams as built. An interesting effect was to take a full measurement with M593 off and then changing nothing other than turning it on , take another average. You could see the sidebands forming on one of the other axis.

    I've been playing around with non-symmetrical settings on the acceleration and jerk while setting M593 to the second harmonic and achieved somewhat acceptable results but nothing earth shattering. When I set jerk too low the machine sounds like it's grunting.

    Since you're familiar with this motion system is there a particular analytical series of measurements you would like to see? I would gratefully run any tests you'd like.
     
    #7 Ntesla66, Nov 30, 2019
    Last edited: Nov 30, 2019
  8. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
    I've been using an accelerometer on a Hemera tool recently. These are the results I got for a Y move:

    upload_2021-4-4_21-34-39.jpeg
    There is significant ringing in Y and Z. I think this is because the centre of mass of the Hemera tool is well below the crossbar, so Y motion induces a torsional oscillation of the crossbar and/or the mounting system.

    I am using this as a testbed to experiment with input shaping.
     
    garethky and Joe Pomo like this.
  9. Joe Pomo

    Joe Pomo Well-Known Member

    Joined:
    Sep 21, 2019
    Messages:
    107
    Likes Received:
    41
    Thanks for posting your results. Also great to see you doing this in DWC? Awesome!
     
  10. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
    Here are some more plots. showing the vibrations from a 100mm Y move, speed 12000, acceleration 4000, before and after enabling experimental input shaping (ZVD @ 42Hz).
    upload_2021-4-5_22-6-47.png
    upload_2021-4-5_22-8-8.png
    upload_2021-4-5_22-7-29.png
    upload_2021-4-5_22-8-41.png
     
    W1EBR.Gene and Greg Holloway like this.
  11. W1EBR.Gene

    W1EBR.Gene Well-Known Member

    Joined:
    Apr 10, 2016
    Messages:
    294
    Likes Received:
    39
    How are you collecting data from the accelerometer? What is the sample rate? (I have one and want to set it up)
     
  12. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
  13. W1EBR.Gene

    W1EBR.Gene Well-Known Member

    Joined:
    Apr 10, 2016
    Messages:
    294
    Likes Received:
    39
  14. W1EBR.Gene

    W1EBR.Gene Well-Known Member

    Joined:
    Apr 10, 2016
    Messages:
    294
    Likes Received:
    39
  15. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
    Nothing yet. If the vibration is severe then I would expect the layer thickness to vary as it vibrates. The peak amplitude is about 0.5g and the frequency 42Hz, and from those figures you could work out the peak amplitude.
     
  16. W1EBR.Gene

    W1EBR.Gene Well-Known Member

    Joined:
    Apr 10, 2016
    Messages:
    294
    Likes Received:
    39
    @Greg Holloway why is every other bolt hole on the x axis liner rail used? With gentle pressure on the bottom rear of the tool holder cover I can make the tool holder move slightly backward and forward. It flexes more when the tool holder is positioned over a spot on the linear rail that is not bolted down.
     
    #16 W1EBR.Gene, Apr 9, 2021
    Last edited: Apr 9, 2021
  17. Greg Holloway

    Greg Holloway Administrator
    Staff Member

    Joined:
    Sep 4, 2015
    Messages:
    1,020
    Likes Received:
    631
    We assumed it would be fine with fewer fixings due to the considerably lower loadings than the rated capacity of the rail and carriage.

    How did you measure the flex and what is the displacement delta?
     
  18. W1EBR.Gene

    W1EBR.Gene Well-Known Member

    Joined:
    Apr 10, 2016
    Messages:
    294
    Likes Received:
    39
    I measured the flex by touch. The flex is rotational; I gently rotated the tool holder carriage and it moved. I also found that I had a few bolts that were loose. I read that 90 N-cm is the max force when they are screwed into aluminum so I never tried to use a hex key in "high-torque" orientation. Tightening the few loose bolts helped some with the flex (I think)

    I don't think there is a load issue at all. I am wondering whether the low frequency ringing in the y and z axis that @dc42 is seeing is related.

    I just got an accelerometer and hopefully can start collecting data soon. One test will be to see if re-loosening a few bolts changes the ringing :)
     
    Greg Holloway and Joe Pomo like this.
  19. dc42

    dc42 Well-Known Member

    Joined:
    Aug 16, 2016
    Messages:
    484
    Likes Received:
    216
    Here are my latest RRF 3.4beta + accelerometer + input shaping plugin results, for one of my Hemera tools. The red trace shows the ringing of the tool on the Z axis with no input shaping applied. The blue trace shows the same movement with ZVDD input shaping applied atv42.8Hz with a damping factror of 0.05. The first plot is the time domain, the second is the frequency domain.
    image2.png image.png
     

Share This Page