Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning

[carcamerarig]

@Argo said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

Since yesterday I'm testing Klipper at one of our printers. In particular the input shapers and their auto tuning with a ADXL345 accelerometer woke my interest.

It turns out my bed slinger printer (Bear Frame with Bondtech Extruder) has different resonance frequencies on each axis.

After the auto calibration /measurement it recommended:
Recommended shaper_type_x = 3hump_ei, shaper_freq_x = 61.4 Hz
Recommended shaper_type_y = 2hump_ei, shaper_freq_y = 57.4 Hz

I see there are input shapers on the way for RRF but I couldn't find exact infos what shaper will be used and if it will be possible to enter a shaper frequency for each axis separately.

To sum it up, here are my wishes for RRF 3.4+:

• support for a accelerometer (preferable ADXL345 as it's easy to find and buy) to measure the needed frequency. The input shapers 2HUMP_EI & 3HUMP_EI may need this as it can't be easily measured
• Input shapers: 2HUMP_EI & 3HUMP_EI
• separate frequencies and dampening for each axis (at the moment RRFs DAA uses the same frequency for both axis)

Gets a +1 Vote from me.

[Argo]

@Egon-Net

it's still a fast 32 bit board but I don't know if the available connections are fast enough for a accelerometer.
But you could run Klipper with your Duet 2

[dc42]

The accelerometer we are testing with is the LIS3DH. The Adafruit breakout board for it is widely available.

Please note, the use of EI input shaping in 3D printers is patent protected. That's why we intend to support ZVDD as well.

[zapta]

@Argo does Klipper support also backlash compensation? I wonder if the accelerometer can be used to calibrate it as well.

[Egon.Net]

@Argo said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

@Egon-Net

it's still a fast 32 bit board but I don't know if the available connections are fast enough for a accelerometer.
But you could run Klipper with your Duet 2

At leats be able to measure externally and input the results... (With and arduino, pc, whatever)

[MikeS]

@dc42 please consider on supporting also ADXL345. It's the most common and cheapest one. I had one laying around that i used on my previous delta to do probing of the bed. I understand that, as a business, you have to come up with your own implementation and maybe better solution for the resonance problem, but klipper one for now would be great. We have time to improve it and have a better one!

[dc42]

LIS3DH boards from well-known suppliers such as Adafruit are half the price (or less) of ADXL345 boards.

[Argo]

The ADXL345 cost me 6 EUR and a LIS3DH would cost me 10 EUR.
But honestly I don't really care if I had to buy another one for my Duet / RRF.

@zapta said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

@Argo does Klipper support also backlash compensation? I wonder if the accelerometer can be used to calibrate it as well.

It's measurable, so yes.

[zapta]

The Adafruit LIS3DH is available in my country on Amazon, $7 Prime shipped.

https://www.amazon.com/Adafruit-LIS3DH-Triple-Axis-Accelerometer-ADA2809/dp/B01BU70B64

@dc42, will input shaping be supported on stand alone Duet2 Wifi?

[dc42]

@zapta said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

The Adafruit LIS3DH is available in my country on Amazon, $7 Prime shipped.

https://www.amazon.com/Adafruit-LIS3DH-Triple-Axis-Accelerometer-ADA2809/dp/B01BU70B64

@dc42, will input shaping be supported on stand alone Duet2 Wifi?

Yes, but perhaps not on the Duet Maestro because it doesn't have floating point hardware.

[MikeS]

@dc42 no problem at all at buying another accelerometer, but just for reference, amazon IT has ADXL345 at €5.99 with prime, LIS3DH €15.46 with prime. Most important is: are there differences on precision of the readings between these 2? 10€ difference is not a problem compared to accuracy of readings

[printernoodle]

I am looking forward to this feature set...

[dc42]

More at https://duet3d.dozuki.com/Wiki/Accelerometers, and sample results at https://forum.e3d-online.com/threads/accelerometer-and-resonance-measurements-of-the-motion-system.3445/#post-38177.

[zapta]

@dc42 said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

sample results at https://forum.e3d-online.com/threads/accelerometer-and-resonance-measurements-of-the-motion-system.3445/#post-38177.

@dc42, Can you post the graphs here? E3D requires registration to view them.

[dc42]

The first two are without input shaping. The second two are with ZVD input shaping at 42Hz. ZVDD didn't give any further improvement, and both EI2 and EI3 were worse.

[Argo]

Very nice seeing progress there!
Also as usual it looks very tidy and well thought out so far.

just ordered a Adafruit LIS3DH and plan to experiment with the new feature.
I see (experimental?) input shapers are also already implemented in the 3.3-input-shaper branch.

Is it already worth to try and experiment with the feature also during printing conditions?
Would like to see how the input shapers work with a typical i3 bed slinger as these printers tend to have different ringing frequencies at once. And I also could provide some results/data.

[dc42]

@argo said in Input Shapers: 2HUMP_EI & 3HUMP_EI + Auto Tuning:

I see (experimental?) input shapers are also already implemented in the 3.3-input-shaper branch.
Is it already worth to try and experiment with the feature also during printing conditions?

Experiment if you like; but the code is by no means complete, so be very careful. Pressure advance is disabled - I haven't even tried extruding. Delta kinematics is not implemented.

Input shaping will only work on moves for which the acceleration time is long enough to apply it, i.e. at least 2 * the shaping time. So you may need to reduce acceleration to get it applied. I believe there is some scope to apply it to shorter acceleration times, down to 1 * the shaping time + one half period, at least when using ZVD or ZVDD, but that isn't implemented yet. Also I will probably implement reducing acceleration (but not below a minimum) where necessary to allow input shaping to be applied.

I don't expect to include the input shaping code in the 3.3 release.

[CNCModeller]

Hi All,
Can anyone recommend a link to a good article on what input shaping is and what print features it addresses? I'm assuming it's about removing excitation of the printers structural modes from the demand, but I guess there are more subtle nuances too.

Edit: I assume this is the accelerometer mentioned.

https://www.ebay.co.uk/itm/232427148679

Many thanks
Barry M

[Argo]

@cncmodeller

Yes that's the sensor.

Input shaper ZVD(D) paper:
https://www.researchgate.net/publication/316556412_INPUT_SHAPING_CONTROL_TO_REDUCE_RESIDUAL_VIBRATION_OF_A_FLEXIBLE_BEAM

EI2 input shaper:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.465.1337&rep=rep1&type=pdf

Most simple explanation: It basically cancels out ringing frequencies.

@dc42
Defining a minimum and maximum acceleration range might be good approach for input shaping.
With Klipper's (automatic) calibration measurement you can set a "max smoothing target" as too much acceleration may lead to losing fine details. It then recommends to stay below a specific acceleration, depending on the used shaper the value differs.
What Klipper does not offer though is to set a specific acceleration range. Meaning, if printing small details, the acceleration slows down as needed, so the shaper works properly.

Speaking of pressure advance. Is some sort of pressure advance smoothing planned?

[Dougal1957]

I have just order two of these https://shop.pimoroni.com/products/adafruit-lis3dh-triple-axis-accelerometer-2g-4g-8g-16g in anticipation.

Doug