Capacitive Z-height sensing
Posted by Radian
|
Capacitive Z-height sensing March 11, 2014 09:09AM |
In the continued quest to develop a better Z-height sensor I finally got around to testing a capacitive sensing probe. The principle is simple enough - create an oscillator that varies it's frequency with changes in capacitance in part of its circuit. By using the alluminium heat spreader as one plate of the capacitor and a square of tea-caddy tin under the X-carriage as the other, I'm generating a height dependent frequency using the onboard Capacitive Sensing Module in a PIC12F1822
The shiny tin part is soldered directly to the stripboard carrying the PIC. This board bolts on in place of original IR sensor board so the tin "foot" ends up sitting parallel to the bed, roughly at the same height as the tip of the nozzle. I also tried connecting the capacitive sensor input directly to the metal hot-end assembly but the frequency change wasn't anywhere near as pronounced. This is understandable when you look at the small surface area of the nozzle which is the only part that gets close to the bed. The purpose-made plate has a far greater area.
Initial tests look promising. So far I'm just reading back raw counter values from a gated timer using the CSM as clock source. To get a feel for the readings the counts are currently typically like this:
There's dancing in the lowest two digits, but the oscillator timebase in the PIC is only internal CR so I'm expecting better results using an crystal. The next step is to convert the counts into a PWM derived analogue signal that can be fed into the Duet as Z-height.
Unlike Ultrasound, the readings appear to be largely unaffected by temperature and of course ambient light! I expect Electrical noise will prove to be the main environmental hazard - although so far it seems pretty clean. As for humidity, further testing is needed to find out if this is an issue or not. Previous experience with this kind of sensing tells me that Long-term drift is going to need addressing.
RS Components Reprap Ormerod No. 481
The shiny tin part is soldered directly to the stripboard carrying the PIC. This board bolts on in place of original IR sensor board so the tin "foot" ends up sitting parallel to the bed, roughly at the same height as the tip of the nozzle. I also tried connecting the capacitive sensor input directly to the metal hot-end assembly but the frequency change wasn't anywhere near as pronounced. This is understandable when you look at the small surface area of the nozzle which is the only part that gets close to the bed. The purpose-made plate has a far greater area.
Initial tests look promising. So far I'm just reading back raw counter values from a gated timer using the CSM as clock source. To get a feel for the readings the counts are currently typically like this:
Z Counter 0.0 13500 0.1 13620 0.2 13750 0.3 13930 0.4 14200 0.5 14500 0.6 14750 0.7 14960 0.8 15190 0.9 15250 1.0 15400 1.5 15800 2.0 16100 3.0 16500 5.0 16900 10 17100
There's dancing in the lowest two digits, but the oscillator timebase in the PIC is only internal CR so I'm expecting better results using an crystal. The next step is to convert the counts into a PWM derived analogue signal that can be fed into the Duet as Z-height.
Unlike Ultrasound, the readings appear to be largely unaffected by temperature and of course ambient light! I expect Electrical noise will prove to be the main environmental hazard - although so far it seems pretty clean. As for humidity, further testing is needed to find out if this is an issue or not. Previous experience with this kind of sensing tells me that Long-term drift is going to need addressing.
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 11, 2014 10:15AM |
Hi dc42,
Input: A = 20mm x 20mm square of tin, D = 3mm above glass (K=5 ?)
C=K*Eo*A/D,
=5*8.854x10-12*0.0004/0.003
=6pF
Not much on paper, agreed, but all in a day's work for a half-decent CSM like the one in the PIC
I'm very keen on getting it to work reliably with the stock glass, but yes, coated glass would totally transform the situation. Mind you, in that case we could just dangle a wire contact to the surface!
RS Components Reprap Ormerod No. 481
Input: A = 20mm x 20mm square of tin, D = 3mm above glass (K=5 ?)
C=K*Eo*A/D,
=5*8.854x10-12*0.0004/0.003
=6pF
Not much on paper, agreed, but all in a day's work for a half-decent CSM like the one in the PIC
I'm very keen on getting it to work reliably with the stock glass, but yes, coated glass would totally transform the situation. Mind you, in that case we could just dangle a wire contact to the surface!
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 11, 2014 11:31AM |
Excellent! Glad this looks promising. Someone also suggested using two plates side by side on the hot end. You 1/4 the capacitance but dont need a connection to the heater plate.
regards
Andy
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
Someone also suggested using two plates side by side on the hot end. You 1/4 the capacitance but dont need a connection to the heater plate.regards
Andy
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
|
Re: Capacitive Z-height sensing March 11, 2014 11:35AM |
Quote
kwikius
Excellent! Glad this looks promising.
I suggested that, but I was still intending that the heater plate be used as one side of the capacitor. One side of the bed thermistor is already connected to 0V, so I would just connect a resistor between this connection and the plate (a resistor rather than a wire, to give some protection for the Duet against short circuits to the heated bed connections).
Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod
Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].
|
Re: Capacitive Z-height sensing March 11, 2014 01:37PM |
My plan was to pass a sine wave of constant frequency and amplitude, amplify the result and rectify it. to give a signal inversely proportional to distance I believe there are various microcontrollers around that have all the required peripherals DAC, OpAmp, A2D and capacitive sensor too! Avoiding using the heat spreader as ground should eliminate it as a source of noise. ( I have a hunch that splitting the capacitance means any spikes cancel.. but not looked at it in detail) But I havent really done much about it its true!
regards
Andy
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
regards
Andy
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
|
Re: Capacitive Z-height sensing March 11, 2014 02:06PM |
kwikius, your plan makes use of a capacitive divider then. The Capacitive Sensing Module I'm using in the PIC chip uses a cute scheme where an internal constant current sink and source alternately charges and discharges the sensor plate generating a triangle wave that varies in frequency with the amount of charge taken by the plate.
In this case the aluminium heat spreader makes a great ground to drain the charge off the plate. If noise was a worry it could readily be grounded but I'm getting repeatable results with the bed floating. It's actually something like 1nF to ground so the series effect on the gap is utterly swamped. I'm in two minds about grounding the metalwork so I think I'll stay on the fence awhile longer.
RS Components Reprap Ormerod No. 481
In this case the aluminium heat spreader makes a great ground to drain the charge off the plate. If noise was a worry it could readily be grounded but I'm getting repeatable results with the bed floating. It's actually something like 1nF to ground so the series effect on the gap is utterly swamped. I'm in two minds about grounding the metalwork so I think I'll stay on the fence awhile longer.
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 11, 2014 02:40PM |
colsenfoto - thanks for your encouragment 
OK, so I've got filtered PWM giving a 0 to 3.3V Z-probe height form the CSM counter. To get sensible scaling I'm tracking the minimum/maximum counts found while moving Z and scaling the value sent to the 10-bit PWM generator accordingly. So here's the numbers from the web interface:
The readings dither more the higher the probe - within the first mm or two it looks most usable but there's a spread of values of around 10 to 20 which equates to less than 0.1mm
Sounds good, but I'm not happy with the way it seems to "lurch" - several readings in a row can be within a few counts then it drops by 10 and stays around there for a while... then... you get the drift. Still using the PIC RC timebase which is embarrassing but I think it's supposed to be good for 1% or so.
Where I see a problem is that the auto-scaling gets thrown right out if you touch the sensor. Without it, the sensitivity goes down considerably. The raw counter values have been ranging from 13000 to 17000 (out of 65535) so, while the upper value doesn't go much further (probe well isolated) the lower can fall all the way to zero if shorted or touched.
RS Components Reprap Ormerod No. 481
OK, so I've got filtered PWM giving a 0 to 3.3V Z-probe height form the CSM counter. To get sensible scaling I'm tracking the minimum/maximum counts found while moving Z and scaling the value sent to the 10-bit PWM generator accordingly. So here's the numbers from the web interface:
Z Zprobe 0.0 1010 0.1 990 0.2 908 0.3 844 0.4 760 0.5 701 0.6 667 0.7 626 0.8 597 0.9 558 1.0 542 1.5 420 2.0 340 3.0 260 4.0 228 5.0 178 10 100
The readings dither more the higher the probe - within the first mm or two it looks most usable but there's a spread of values of around 10 to 20 which equates to less than 0.1mm
Sounds good, but I'm not happy with the way it seems to "lurch" - several readings in a row can be within a few counts then it drops by 10 and stays around there for a while... then... you get the drift. Still using the PIC RC timebase which is embarrassing but I think it's supposed to be good for 1% or so.
Where I see a problem is that the auto-scaling gets thrown right out if you touch the sensor. Without it, the sensitivity goes down considerably. The raw counter values have been ranging from 13000 to 17000 (out of 65535) so, while the upper value doesn't go much further (probe well isolated) the lower can fall all the way to zero if shorted or touched.
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 11, 2014 06:01PM |
Well, this isn't helping:
Dumping the values I was putting into the PWM generator I could see they were much less noisy than the Zprobe readings after being A/D converted by the Duet. So I hooked up a couple of 1.5V batteries in series with a 100R (100uF/10nF in parallel with output) to the input of the A/D and measured 200 seconds worth of readings. This will come as no surprise to dc42 who went to the trouble of fixing a new header to get a better 0V connection for the Z-probe board. I had no idea it was this bad though.
dc42: is there a possibility of accepting a serial data input into a spare UART anywhere on the ATSAM3X8E? We're throwing away Zprobe repeatability mainly because of an awkward grounding arrangement. Standard 115200 or 57600 baud 8bit data received as a five char. ASCII string e.g. #nnnn every 16 or 32 ms for example.
RS Components Reprap Ormerod No. 481
Dumping the values I was putting into the PWM generator I could see they were much less noisy than the Zprobe readings after being A/D converted by the Duet. So I hooked up a couple of 1.5V batteries in series with a 100R (100uF/10nF in parallel with output) to the input of the A/D and measured 200 seconds worth of readings. This will come as no surprise to dc42 who went to the trouble of fixing a new header to get a better 0V connection for the Z-probe board. I had no idea it was this bad though.
dc42: is there a possibility of accepting a serial data input into a spare UART anywhere on the ATSAM3X8E? We're throwing away Zprobe repeatability mainly because of an awkward grounding arrangement. Standard 115200 or 57600 baud 8bit data received as a five char. ASCII string e.g. #nnnn every 16 or 32 ms for example.
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 12, 2014 05:49AM |
Quote
dmould
One thing I could think might affect the readings is the presence of people or other objects around the printer - did anything nearby move when you saw the reading jump? The other thing is whether the readings are the same for a given height at all places over the bed - if not it might require a different parameter to be set for each probe position.
Hi Dave, fortunately the capacitive coupling effects from items outside the bed area are negligible thanks both to the relative permeability (K) of air being 4 to 8 times greater than glass and then all divided by distance (D) (where C=K*Eo*A/D)
But I have now grounded the heat spreader because despite being itself a few thousand pF to ground (due to its large area), the coupling into it could be even greater under some conditions. With it grounded, any coupling from the periphery is shunted away from coupling to the sensor foot.
dc42 - So basically then, MIDI format
Funny you should suggest that - I always seem to get it in the neck for doing the top-bit-set control/top-bit-clear data by customers who'd rather see something more stylish - but you can't beat simple and concise! I suppose receipt of a two-byte Async. packet could be enough in itself to flag to the firmware that Z readings are available through that route?colsenfoto - the analogue signal is mostly suffering on the Duet board where it passes by the 5V switching regulator (unshielded magnetics, dubious grounding etc.) It's also referenced to the digital ground whereas the A/D convertor has a HF decoupled ground (used by all the other analogue inputs). A digital communication would be a welcome relief from all this!
Treth - As you point out, humidity may well be a significant variable although I've had working touch panels that were mounted on bikes and rain wasn't a problem. What difference it will make to the fine-grain readings remains to be seen.
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 13, 2014 10:30AM |
Re: Capacitive Z-height sensing
OK, that passes the "clever test" with flying colours
I've already improved the max/min detection by passing the values through an exponential moving average IIR filter with long time constant. The values themselves were already the mean of 16 samples but the slightest tickle when at Z=0 could drop the min. to below the correct value. This was all too easy to do when using paper to check the nozzle gap. Also, I turn on a LED for a fixed duration whenever min or max change so there's a visual indication that calibration is taking place. All I have to do is ditch the switch and use the touch to initiate the calibration process. A timeout after no changes have been detected could terminate the calibration process. Sounds like a plan.
As for the rest of the process, by tweaking the power levels in the CSM and the reference voltages to drop the triangle wave frequency from ~1MHz to ~250kHz the readings have become significantly more stable. The integration period is now 32ms but I think that gives an acceptable update rate given dc42's revised Z-homing speed. The resolution is now so fine, I can "see" the displacment of the bed due to the weight of a one-pound coin on the opposite side!
RS Components Reprap Ormerod No. 481
Quote
dmould
...you could enable "calibration mode" by detecting when the capacitive sensor is touched with a finger 3 times quickly (say within 1.5 seconds).
OK, that passes the "clever test" with flying colours
I've already improved the max/min detection by passing the values through an exponential moving average IIR filter with long time constant. The values themselves were already the mean of 16 samples but the slightest tickle when at Z=0 could drop the min. to below the correct value. This was all too easy to do when using paper to check the nozzle gap. Also, I turn on a LED for a fixed duration whenever min or max change so there's a visual indication that calibration is taking place. All I have to do is ditch the switch and use the touch to initiate the calibration process. A timeout after no changes have been detected could terminate the calibration process. Sounds like a plan.
As for the rest of the process, by tweaking the power levels in the CSM and the reference voltages to drop the triangle wave frequency from ~1MHz to ~250kHz the readings have become significantly more stable. The integration period is now 32ms but I think that gives an acceptable update rate given dc42's revised Z-homing speed. The resolution is now so fine, I can "see" the displacment of the bed due to the weight of a one-pound coin on the opposite side!
RS Components Reprap Ormerod No. 481
|
Re: Capacitive Z-height sensing March 14, 2014 04:13AM |
Sounds like its getting better all the time
The only bit I'm not getting is why the values change dependent on the machine are unknowns.
The PIC charging current can be measured by attaching a known cpacitance
The plate area is known
The physical properties of the glass (thickness, dielextric constant ect) are known or can be looked up.
Any other parameters (Air gap, stray capacitance, leakage) can be added to the model
So for a particular machine (EDIT : many of) these parameters can be entered and unknowns eliminated.
Re the offset. Isnt there some op amp arrangement that gives a negative capacitance? Or is that getting to complicated?
regards
Andy
Edited 1 time(s). Last edit at 03/14/2014 04:20AM by kwikius.
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
The only bit I'm not getting is why the values change dependent on the machine are unknowns.
The PIC charging current can be measured by attaching a known cpacitance
The plate area is known
The physical properties of the glass (thickness, dielextric constant ect) are known or can be looked up.
Any other parameters (Air gap, stray capacitance, leakage) can be added to the model
So for a particular machine (EDIT : many of) these parameters can be entered and unknowns eliminated.
Re the offset. Isnt there some op amp arrangement that gives a negative capacitance?
Or is that getting to complicated?regards
Andy
Edited 1 time(s). Last edit at 03/14/2014 04:20AM by kwikius.
Ormerod #318
www.zoomworks.org - Free and Open Source Stuff
Sorry, only registered users may post in this forum.