3 wire load cell - no signal

[gildasd]

I have been fighting with 3 wire load cells all week and I am in a rut...
I am trying to get a measurable signal to amplify out of a load cells found in bathroom scales.

There is a wealth of data  about 4 wire cells available, but these would cost about 8 times more to employ.
Very little is said about 3 wire set up; people usually go from "here is a Wallmart scale ~blurry lines~ here is the finished project"...
My project needs to measure weight on each cells individually, not wire 4 together.

I have measured the cells: unpowered, 1.5V, 3V, 4.5V and 6V. All have no visible movement on a scope in the 5mV scale.
Using a bench meter I get about 0.13mV per kilo, but I am starting to see RF signals down there... Boosting it 100x or 1000x with a 741 is getting me nowhere - nearly no shift.
To the point I thinking the 0.13mV is my body conducting stray charges...

The best result I have to date is measuring resistance and getting 1.002 kOhm unloaded and 1.004kOhm with 15 kilo load between the black and white wire - with no current.
As soon as I give 1.5V, the resistance goes into the MOhm range and I can't see variation any-more (not enough digits on a Fluke 175).

Am I missing something obvious or should I spend cash on the big alloy cells (these bring a whole host of problems due to their size and heat sensitivity).
Here is some data, but the numbers are at least 10x off from what I am getting.
https://learn.sparkfun.com/tutorials/load-cell-amplifier-hx711-breakout-hookup-guide?_ga=1.265664391.925322740.1443502696

[retrolefty]

I believe a '3-wire load cell' acts like a simple 3 terminal variable pot. If you wire say regulated +5vdc to the red and ground the black you should see a change in voltage measured between the white wire and the ground wire. You should see some change in millivolts output as you very the pressure on the cell. Just maybe 10s of millivolts change so don't expect large voltage changes. That is why they almost always require amplification before wiring to something interesting.

[Paul Moir]

Just had a quick look on aliexpress and similar ones are quoted as being half-bridges, with a sensitivity of 1mv/V.  So if you complete the bridge with a pair of 1K ohms resistors, excite it with 5v and load it to it's max, it'll develop 5mV across the bridge.
You should be able to identify the pinout with a multimeter.  You would expect to get 2k between the two excitation wires and 1k between either excitation and the signal terminal.

http://www.ni.com/white-paper/4172/en/

[Dago]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

[gildasd]

Just had a quick look on aliexpress and similar ones are quoted as being half-bridges, with a sensitivity of 1mv/V.  So if you complete the bridge with a pair of 1K ohms resistors, excite it with 5v and load it to it's max, it'll develop 5mV across the bridge.
You should be able to identify the pinout with a multimeter.  You would expect to get 2k between the two excitation wires and 1k between either excitation and the signal terminal.

http://www.ni.com/white-paper/4172/en/

Ok, i get it. I was barking up the wrong tree... Simple now!

[gildasd]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

If complete the bridge as Paul Moir suggests, is a simple Op amp amplification valid to get a good signal before the ADC?

[retrolefty]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

If complete the bridge as Paul Moir suggests, is a simple Op amp amplification valid to get a good signal before the ADC?

 Most all load cells are configured in a Wheatstone bridge configuration and utilize an instrumentation op-amp giving the desired gain to match the range of the ADC processing the measurement value. Lots of examples on the web.

[gildasd]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

If complete the bridge as Paul Moir suggests, is a simple Op amp amplification valid to get a good signal before the ADC?

 Most all load cells are configured in a Wheatstone bridge configuration and utilize an instrumentation op-amp giving the desired gain to match the range of the ADC processing the measurement value. Lots of examples on the web.

The three wire examples in my possession do not (I wish they did!).

[retrolefty]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

If complete the bridge as Paul Moir suggests, is a simple Op amp amplification valid to get a good signal before the ADC?

 Most all load cells are configured in a Wheatstone bridge configuration and utilize an instrumentation op-amp giving the desired gain to match the range of the ADC processing the measurement value. Lots of examples on the web.

The three wire examples in my possession do not (I wish they did!).

But are sure that two were not wired in opposite polarity in a bridge configuration?

[gildasd]

The signal levels from load cells like that are extremely small and it is no wonder you don't see anything with an oscilloscope

If complete the bridge as Paul Moir suggests, is a simple Op amp amplification valid to get a good signal before the ADC?

 Most all load cells are configured in a Wheatstone bridge configuration and utilize an instrumentation op-amp giving the desired gain to match the range of the ADC processing the measurement value. Lots of examples on the web.

The three wire examples in my possession do not (I wish they did!).

But are sure that two were not wired in opposite polarity in a bridge configuration?

All four cells made a single sensor, I'll post the skematics when I'll get home, but I was hoping that the white wire was the output of a bridge.
It's not.
For my use I need the sum of the sensors, not the average, so I cannot wire them as you say.

[gildasd]

It's working!
It was a half Wheatstone bridge!
I get nice repeatable results:
- 3.66mV no weight
- 3.32mV with 2kg
This is very usable once amplified a 100 or a 1000 times - success!
I'll post the full instrumentation amplifier schematic tomorrow (based on a Tl084 4xOp amp): I need to redo the values as I was not expecting the variable resistance values to go down with weight...
I'll see what kind of value range I can get before choosing an ADC (I2C flavor).
Bed time, I'll sleep well!

[lapm]

It's working!
It was a half Wheatstone bridge!
I get nice repeatable results:
- 3.66mV no weight
- 3.32mV with 2kg
This is very usable once amplified a 100 or a 1000 times - success!
I'll post the full instrumentation amplifier schematic tomorrow (based on a Tl084 4xOp amp): I need to redo the values as I was not expecting the variable resistance values to go down with weight...
I'll see what kind of value range I can get before choosing an ADC (I2C flavor).
Bed time, I'll sleep well!

All you need to do now is balance the bridge so it gives 0.00 mV at no load and its even more usable. 

[SeanB]

Going down with mass added simply swap the power and ground leads of the half bridge, problem solved. Those half bridges only have the 3 colours so the assembler can tell them apart, I have seen a few with the red lead connected to the battery negative.

[gildasd]

Conclusion: 3 wire load cells suck for my project.

I made a full bridge with a normal and a Trim resistor.
Next the trim was adjusted to as close to 0mV as possible with no load (I got to -0.2mV).
The resulting signal is fed basic Op Amp instrument amplifier (with no gain to start with) to see what amplitude is available to feed to the ADC.
However I could not get a logical result out of the amplifier... There was a base signal that should not have been there.
Frustrated I went to bed.
I have now added a load and ceramic cap on the 3v line and still I got weird results.
Thusly I went back to load cell without the Opamp and found this:
(Series of measurements)
No load  _  2Kg load
-1.09mV -1.00mV
-1.10mV -1.01mV
-1.12mV -1.03mV
-1.13mV -1.04mV
-1.14mV -1.05mV
-1.16mV -1.06mV
-1.18mV -1.07mV
etc, the load cell is drifting like hell...
I tried adding a load to the 3V line, a ceramic cap etc, it changes nothing.
This explains why 4 cells are wired together and the scale calibrated at each use.
The rig I'm building has to measure wind force non stop for 1 month without calibration, apparently this cannot be done with this sensor type.

Edit, going to test if it's not the trim pot that's the issue - but I'm pretty sure it ain't.

[Andy Watson]

but I was hoping that the white wire was the output of a bridge.
It's not.

I might have missed this in the thread, but, have you actually identified which wire IS the output? Other, similar devices reported on the web suggest that the excitation voltage is applied between white and black - leaving red as the output (contrary to what any normal person would assume!) Measure the resistance between the three possible pairs of wires - this will tell you how the half-bridge is wired.

[Paul Moir]

All load cells drift a lot so there is more to a scale than a load cell and adc.  You have to do zero tracking, weight change detection, maybe even temperature compensation. 

[gildasd]

but I was hoping that the white wire was the output of a bridge.
It's not.

I might have missed this in the thread, but, have you actually identified which wire IS the output? Other, similar devices reported on the web suggest that the excitation voltage is applied between white and black - leaving red as the output (contrary to what any normal person would assume!) Measure the resistance between the three possible pairs of wires - this will tell you how the half-bridge is wired.

It is the white. I was measuring voltage not resistance as I thought it was the output of a full bridge.

All load cells drift a lot so there is more to a scale than a load cell and adc.  You have to do zero tracking, weight change detection, maybe even temperature compensation.

Yeah but 100% variation (the loaded value drifts to the no load initial value) after 10 cycles is awful. Simply crap. Even membrane cells do better.
In gusty conditions, 10 cycles is a few seconds. I can't put a box over the sensors to reset the null every 5 seconds. It's not feasible.
I can live with heat drift because all the cells will drift the same (shade and moving air).
The weight will be a constant during the whole campaign.
I think I will have to go to 4 wire alloy cells. More complicated but I would only have to do a drift compensation measurement once a week (if I read the specs right).

[gildasd]

I've been fooling about all day:
From what I've tested, up to 5kg, the cells are imprecise and drift.
The 0 moves as much as the loaded value.
Beyond 5kg, this goes down. At 10kg, the absolute value repeats quite nicely... And the 0 drift does not matter as much (equivalent to hundreds of grams).
The drift actually goes up and down slowly over a period of minutes... Apparently this can be partially cured by using two cells to form a single bridge. This would cost twice a much obviously.
However, at this price, I can use alloy bar cells that are for better:
https://www.sparkfun.com/products/13329
And it's dedicated amplifier:
https://www.sparkfun.com/products/13230

Edit1; The Amp Hour's latest guest has managed it!
I think he avoided most of my pitfalls by using a better scale (thus sensors) and "preloading" them with the heavy wooden hive.
https://hackaday.io/project/4648-analogio-a-full-stack-iot-platform/log/23345-measuring-the-weight-of-beehives

Edit2; The 0 seems to stabilise after about an hour...

mV   time 0.5   21:14:00 2.4   21:23:00 3.0   21:34:00 4.3   21:44:00 32.6   21:54:00 29.1   22:04:00 28.8   22:14:00 28.8   22:24:00 28.8   22:34:00

[gildasd]

I've changed the way I generate voltage and set the reference:

0kg  mV   2kg  mV   Diff   0kg  mV   2kg  mV   Diff   time -1.51   -1.39   -0.12   -2.03   -1.9   -0.13   10:00:00 -1.5   -1.38   -0.12   -2.02   -1.88   -0.14   10:05:00 -1.52   -1.41   -0.11   -2.04   -1.92   -0.12   10:10:00 -1.51   -1.4   -0.11   -2.05   -1.93   -0.12   10:15:00 -1.51   -1.4   -0.11   -2.05   -1.93   -0.12   10:20:00 -1.5   -1.39   -0.11   -2.05   -1.92   -0.13   10:25:00 -1.5   -1.39   -0.11   -2.05   -1.93   -0.12   10:30:00 -1.5   -1.38   -0.12   -2.05   -1.93   -0.12   10:35:00 -1.5   -1.38   -0.12   -2.05   -1.92   -0.13   10:40:00 -1.49   -1.38   -0.11   -2.05   -1.92   -0.13   10:45:00 -1.5   -1.38   -0.12   -2.05   -1.93   -0.12   10:50:00 -1.5   -1.36   -0.14   -2.05   -1.91   -0.14   10:55:00 -1.49   -1.37   -0.12   -2.05   -1.92   -0.13   11:00:00 -1.5   -1.38   -0.12   -2.05   -1.92   -0.13   11:05:00 -1.5   -1.38   -0.12   -2.05   -1.92   -0.13   11:10:00

Much better, the no load voltage is still too high, but I have ideas to solve this.

[gildasd]

I tried this with PT10 trimmers of 2.5 kOhms, then 100 Ohms, but I had problems getting it to 0.01mV the measured voltage would go straight from +0.1mV to -0.1mV.
This was solved by using  100 Ohms Vishay Cermet trimmers.
The drift due to the supply, mains, whatever etc was cured by using a 3.3V Zener on the -Vcc.

I have 0.02mV drift in a few hours, I think linked to heat changes in the room.
In any case, it works but I have to start soldering because the thin pins on the Cermets slowly slip out of the bredboard and change the measurement.

[gildasd]

Well, that did not f#####g work...
The soldered circuit became a bloody RF nightmare...

It was picking up the earth's magnetic field (tested with a compass), the power supply of my computer and was building up charge (static?)...
Not to mention high sensitivity to heat, pressure and noise (FFS).
After working on it to remove all the unused traces and putting ceramic caps here and there... It was beginning to be way too complex, no amount of make-up is going to make up for that, and even if it did, it would crack at the slightest provocation.

So a bit of potential precision was sacrificed to make the simplest circuit possible:

The results a much better, in fact, if I can buffer and sum 4 sensors, I've solved this.


Edit: how do I limit the size of the images in the post, but leave them as big in the link.