I want to make a 1 kg x 0.01 g digital scale with an Arduino (to control other things).
That's 1,000,000 mg with a resolution of 10 mg. That's an accuracy (non-linearity) of 0.001% FS or RO (full scale or rated output). Accuracy and resolution are two different things, but I want the resolution to be as good as the linearity.
I'm willing to pay a few $100 for the parts if necessary. I'm having trouble finding parts that I have confidence will give me this accuracy, starting with the load cell. A load cell is not the only thing that affects the accuracy of a scale. But, a scale can't be any more accurate than the accuracy of the load cell. Finding a suitable load cell is only one part I need.
In other threads someone often comments that this is unrealistic unattainable accuracy. But, I see lots of scales for sale that are 1 kg x 0.01 g. They are reasonably priced with many under $150. https://amazon.com/s?k=1000g+x+0.01g
I'm sure many of those are exaggerating, and the resolution of the display does not mean it's that accurate. Some people have calibration weights so it's pretty easy to verify if these were junk, yet they have pretty good reviews.
The load cells I'm finding usually have an accuracy of 1% or 0.1% or if I'm lucky 0.02% or 0.01%. Hardly any are 0.001%. I found a few on obscure sites for high prices. Digikey list 344 force sensors, but none are more accurate than 0.02%.
Ebay has tons of load cells. Most don't list any accuracy. Why would anyone buy a load cell without knowing anything about its accuracy?
What am I missing? How can there be so many 1 kg x 0.01 g scales for sale, but almost no load cells for sale that are that accurate?
Am I looking in the wrong place or for the wrong thing?
It sounds crazy, but do I need to buy a scale and remove the load cell in order to get one for under $500?
Rather than DIY and end up with something of uncertain combined quality (no offense ), buy a commercial scale that meets your requirements and that has an Arduino-friendly interface, such as RS232 (common), I2C, SPI, RS485...
There are 2 companies that i can recommend from experrience for professional load cells, amplifiers and inducators. Omega Engineering and Interface. I used products from those companies to test components for the US military and NASA so they do meet rather strict specifications for accuracy and precision. Not cheap, though.
It would be safer to use a commercial scale. I wanted the scale integrated as part of a device with its own enclosure and other parts. I really just want a good load cell. I could redesign everything around a commercial scale, but it's just not what I'm looking for. Either way, I would still like to understand why I can't seem to find what I'm looking for.
It doesn't look like Omega Engineering has anything accurate enough. This one is the closest I found and it's 0.028% linearity. So, at best it could have 280mg resolution. Right?
I don't object to the price. It's $164 which would be fine if it was accurate enough.
The fact that Omega and Digikey don't have anything accurate enough, makes me think there's something I'm missing.
You will find it almost impossible to measure with that accuracy . I’d suggest googling for weight calibration to find what can be reasonably done
Why do you want to do this ? What is the project ?
Let's first start with a scale. Even if I end up going with a different design, I would like to understand these challenges I'm having with load cells either way.
The scale you posted is very interesting and not at all typical! It's not showing linearity as a percent FS, but as an absolute 0.3 mg and readability of 0.1 mg. That's 300 micrograms.
It's a 510 g scale. So, 0.3 mg linearity would be 0.3 mg / 510,000 mg = 0.00006% FS. That is crazy accurate. I have never seen such an accurate scale. Wow. In fact, I wonder if that's a typo.
Given the scale you posted and the many scales at Amazon that have the resolution I want, I don't know why you say it will be impossible to measure with that accuracy.
I suspect that many of the less expensive 1000g x 0.01g scales are not certified in any way, so the "0.01g" may be essentially meaningless. Expensive / certified ones are likely achieve their performance through calibration. One way to verify that hypothesis is to talk with a helpful engineer at a precision weigh scale manufacturer. Make some calls/emails. It would be interesting to know what you find out.
I think you must be right. And that's a good idea. I'll ask Omega.
So, what you're saying is that a scale can be more accurate than it's linearity, within limits, because poor linearity can be calibrated and corrected for. Then, repeatability is actually the critical spec. If nonrepeatability % is very low, then theoretically you can correct for any error in linearity as long as it's monotonic (always increasing with increasing load) and you take enough calibration samples.
I asked Omega if their load cell would be capable of the accuracy I want, and they didn't answer me. They only said their load cell has a good linearity.
Thanks for the link to the Analog article Dave. Great info.
Hey te09, did you end up finding any more information? I have been searching for load cells and have been having the same difficulty. I'm looking for something in the range of 0.002% for a private project...
However, you also need to think about temperature effects, creep, and as previously mentioned, repeatability. I'm guessing that variances and quality of the bridge resistors on these cheaper load cells add to the inaccuracy.
I'm doing my Masters right now, and developing a cost-effective load cell with this kind of accuracy is something that I have put into my list of possible topics..
No. I bought some cheap load cells and I still plan to see how well I can correct for their non-linearity. I haven't gotten around to it yet.
I agree, temperature effects would be a major issue. You could try to correct for that by including a temperature sensor and calibrating at a number of different temperatures and compensating. But, that sounds difficult and tedious to take all those measurements. It might be easier to insist the environment is set at a particular temperature and calibrate only at that temperature.
Creep seems like a particularly nasty quality. I can't see how you could practically correct for creep.
I wonder how much you can correct for poor repeatability by taking many samples over a period of time.
), buy a commercial scale that meets your requirements and that has an Arduino-friendly interface, such as RS232 (common), I2C, SPI, RS485...
