Loade Cell and sensitivity in terms of actual weight

[Gibson486]

Is there a way to determine the physical sensitivity of a load cell? I have one rated for 50 grams, but I wanted to see if I can detect 1 mg. So far, results have been pretty bad. I used a typical differential amplifier circuit (one optimized for low noise...money) and and a pretty high resolution adc (32 bits and more money). As expected, noise is a huge issue, but I cannot even get a shift in the noise. I know people will say, "oh, it's analog, you have infinite resolution", but there has to be a cut off as to when light is just too light to detect. Oh, and this load cell is 1mV/V and has a hysteresis of 0.03% the rated output. Also, I am a little terrified that the hysteresis is almost ten times the sensitivity I need.  Wouldn't this suggest that anything in the hysteresis band would be suspect?

[timpattinson]

Calculate the cross section of the strain gage and then calculate the strain for a given force. Use the gage factor to convert that to a change in resistance. Then calculate the output voltage from the bridge configuration.

Suggestion: use a lever/pivot or another assembly to amplify the force. Don't use bearings as they will have too much stiction.

Sent from my SM-G930F using Tapatalk

[CatalinaWOW]

You are  pushing things pretty hard, but should be able to at least detect something.  Have you tried larger loads, 1 gram for example?  Once you are convinced your amplifier and bridge are connected properly you can attempt to work your way down to the goal loads.  I agree with tim on the mechanical gain suggestion, although that has a very good chance of making system hysteresis worse.

You are right to be wary of the hysteresis spec, but might be able to work around it if it is repeatable enough.  Which will depend on your device (serial number), mounting, range of loads, and possibly other factors such as temperature.  Only experiment will tell.

Just remember, if this was easy, everyone would do it.

[Gibson486]

I can see something when I put a bigger weight (like 1g), and the results look fine. I can go crazy with running it with AC, but I do not want to go through that hassle only to find out that I need a different sensor that is more sensitive than 1mV/V. The best I can get is that I can detect something (although it is noisy), but the next day i cannot detect it at all. I am guessing that I running against its limit, especially since I am trying to get results that go beyond the hysteresis. 

[wraper]

Besides problems with load cell itself, doing something like this without very stable base and shielding against air movement is plain impossible.

[b_force]

I don't know who says that "analog has infinite resolution", but these people don't really seem to understand the concept of tolerance, noise and accuracy.

To get 1mg resolution from a 50 gram loadcell (assuming that is full scale) you need to have an 16bit or 24 bit ADC.
The big  question is why you're getting so much noise?

I was even able to detect a small screw on a 1kg load cell.
Another way to get rid of a lot of noise it to average your measurements

[mc172]

Well, if the hysteresis is that large relative to your minimum, you might not be able to detect 1 mg, but you might be able to detect the difference between say 20 g and 20.001 g?

[b_force]

Well, if the hysteresis is that large relative to your minimum, you might not be able to detect 1 mg, but you might be able to detect the difference between say 20 g and 20.001 g?

The problem with hysteresis is not that you're can't detect anything at all, but that your values are shifted when going back and forth.
So you might get shifted readings, but at least you get readings.
Hysteresis is something that can be calibrated out.

[mc172]

That's not what I'm suggesting to do. When at the zero point or near it, you could be in that hysteresis band. So by moving out of where the load cell is around its natural zero point, you are into a region where you are only moving in one direction. i.e. Put 20 g on the load cell, then add 1 or 2 mg without taking the 20 g off.

[b_force]

That's not what I'm suggesting to do. When at the zero point or near it, you could be in that hysteresis band. So by moving out of where the load cell is around its natural zero point, you are into a region where you are only moving in one direction. i.e. Put 20 g on the load cell, then add 1 or 2 mg without taking the 20 g off.

Yes, I have read that, but there is clearly a misconception here about hysteresis.
It seems that you mix it up it accuracy or threshold value.
Hysteresis only means that your value is shifted (to an unknown value), not that it's not detectable.

The only think that could be in play, is that there is a minimum threshold value.
Underneath that threshold the load cell doesn't measure anything.

[Gibson486]

I don't know who says that "analog has infinite resolution", but these people don't really seem to understand the concept of tolerance, noise and accuracy.

To get 1mg resolution from a 50 gram loadcell (assuming that is full scale) you need to have an 16bit or 24 bit ADC.
The big  question is why you're getting so much noise?

I was even able to detect a small screw on a 1kg load cell.
Another way to get rid of a lot of noise it to average your measurements

It is actually a 32 bit ADC. I have a feeling it is just lost in the noise.

[b_force]

I don't know who says that "analog has infinite resolution", but these people don't really seem to understand the concept of tolerance, noise and accuracy.

To get 1mg resolution from a 50 gram loadcell (assuming that is full scale) you need to have an 16bit or 24 bit ADC.
The big  question is why you're getting so much noise?

I was even able to detect a small screw on a 1kg load cell.
Another way to get rid of a lot of noise it to average your measurements

My advice is to plot the output of your ADC to see what really is going on.
32 bits here is rather overkill btw,
It is actually a 32 bit ADC. I have a feeling it is just lost in the noise.

[Gibson486]

I don't know who says that "analog has infinite resolution", but these people don't really seem to understand the concept of tolerance, noise and accuracy.

To get 1mg resolution from a 50 gram loadcell (assuming that is full scale) you need to have an 16bit or 24 bit ADC.
The big  question is why you're getting so much noise?

I was even able to detect a small screw on a 1kg load cell.
Another way to get rid of a lot of noise it to average your measurements

My advice is to plot the output of your ADC to see what really is going on.
32 bits here is rather overkill btw,
It is actually a 32 bit ADC. I have a feeling it is just lost in the noise.

Yeah, I did the calcs and I only need 24 bits, but I figure start high and downgrade later...and plus, this adc already had a arduino library made for it. 

[mc172]

Yes, I have read that, but there is clearly a misconception here about hysteresis.
It seems that you mix it up it accuracy or threshold value.
Hysteresis only means that your value is shifted (to an unknown value), not that it's not detectable.

There can be no hysteresis if you only travel in one direction.

[David Hess]

So 1 millivolt/volt and you want to detect 1 milligram out of 50 grams full scale.  1 milligram out of 50 grams is one part in 50,000 which is at the high end (or beyond) for accuracy (1) and 1 millivolt/volt is on the low end of sensitivity but detecting a change of 1 part in 50,000 is still feasible.  What you left unanswered was:

Strain gauge resistance?
Is it a full bridge or just part of a bridge?
What is your excitation voltage (or current)?  This is the important one.

If your excitation voltage or current is low which is required in some applications like 4 to 20 milliamp interfaces and portable instruments, then the signal to noise level will also be low.

Bridge resistance has some effect on signal to noise but is usually not a factor unless excitation is very low.

(1) As you might guess from your hysteresis of 0.03% or about 1 part in 3,000, strain gauge load cell accuracy is not even up to 20,000 let alone 50,000 counts but that has little to do with resolution.

There can be no hysteresis if you only travel in one direction.

In one direction you have creep instead.

[EmmanuelFaure]

Mechanical engineer here.

1mg over 50g means you need 50.000 counts at least. Achieving that performance is not a piece of cake, but it's perfectly doable.

"Is there a way to determine the physical sensitivity of a load cell?" : Yes, it's even written on the load cell... 2 to 4mV/V at full scale usually. That means for an excitation voltage of 5 Volts, you'll get 10 to 20mV at 50g. And 0.2µV to 0.4µV for 1mg.

The ADS1231 from TI is a nice chip : Integrated amplifier, low noise, 50/60Hz rejection, blabla... 0.2µV peak to peak noise at 10 samples per second.

BTW, beware of the temperature coefficient of your load cell, most low cost ones are rated at 0.03%/°C FS drift : For each °C the output of a 50g model drifts of 15mg... And your 1mg is gone! Beware of air streams too.

[David Hess]

BTW, beware of the temperature coefficient of your load cell, most low cost ones are rated at 0.03%/°C FS drift : For each °C the output of a 50g model drifts of 15mg... And your 1mg is gone! Beware of air streams too.

I remember a load cell at a trade show intended intended for weighing diamonds and such.  Someone pushed on it and it did not return to zero, "Aha!  It has a zero problem."  "Nope,"  as the salesperson wipes off the platform, "it was weighing your fingerprint."

[Kleinstein]

At 50000 counts resolution one starts to see mechanical relaxation: so a cheap spring may not longer instantly react to a load change. Rather similar to dielectric relaxation it takes time to settle. Part of the hysteresis might actually be relaxation.

The temperature can be another problem. The heating of the DMS might change the load cell properties. There can also be thermal EMF. So finding the correct level of drive can be important.

With a load cell the ADC is usually the smaller problem. It is more about thermal stability of the load cell and drift and noise of the amplifier. The important point for the ADC is more that is gives good line frequency suppression and not too much noise BW.

[b_force]

At 50000 counts resolution one starts to see mechanical relaxation: so a cheap spring may not longer instantly react to a load change. Rather similar to dielectric relaxation it takes time to settle. Part of the hysteresis might actually be relaxation.

The temperature can be another problem. The heating of the DMS might change the load cell properties. There can also be thermal EMF. So finding the correct level of drive can be important.

With a load cell the ADC is usually the smaller problem. It is more about thermal stability of the load cell and drift and noise of the amplifier. The important point for the ADC is more that is gives good line frequency suppression and not too much noise BW.

Temperature shifts only change the offset, not the fact if you actually can detect such a change (that was the issue the TS had)

To summarize; things like hysteresis, temperature changes or any other stability issues only give you a relative change in the measured value.
Unless they move the system to some kind of threshold.
In practice it means you need to adjust your calibration values or compensate for it somehow.

Relatively speaking, you should still be able to detect something at least.

[MosherIV]

Hi

I used a typical differential amplifier circuit

Then calculate the output voltage from the bridge configuration.

Stupid question :
Are you using the load cell/strain guage in a bridge configuration (Or does the load cell contain the bridge circuit) ?

https://en.wikipedia.org/wiki/Bridge_circuit

Besides problems with load cell itself, doing something like this without very stable base and shielding against air movement is plain impossible.

I agree.
From what I have seen of commercial weight scales of that kind of resolution (sub 1gram) they have a box around the weighing platform.
Just breathing on the scale will measuring something!

possibly other factors such as temperature.

Yes, I forgot about temperature. I beleive the bridge circuit compensates for temperature because the resistors should be thermally coupled and matched to each other and the strain guage BUT the mechanical part should be checked as well.

Good Luck

[David Hess]

Yes, I forgot about temperature. I beleive the bridge circuit compensates for temperature because the resistors should be thermally coupled and matched to each other and the strain guage BUT the mechanical part should be checked as well.

The bridge usually includes a temperature compensation resistor in series with one of the bridge elements for temperature compensation of the zero and another in series with the whole bridge at one excitation input for temperature compensation of the modulus of elasticity.

[Doctorandus_P]

How are you powering the load cell?
Any ripple on the load cells power wil directly couple into your ADC.
ADC and loadcell should share the same reference.

How are you measuring?
Taking a bunch of averages can drive the noise floor down.

How high is your power supply voltage?
Using a higher power supply voltage increases the signal output of th load cell.
You can also use PWM with a low duty cycle if synchronized with the ADC to keep the average power (heating) down.