need 0.1 g resolution scale with 4x 50 kg load cells

I build insect class combat robots for fun, they are ending up oddly shaped and hard to measure on my cheap small kitchen scales. Plus, the one I have seem to have a randomly occurring non-linear error when I test them with a 500 gram standard weight.

I want to make a scale made of a 3D printer glass bed and 4 “pucks” strain gauges, and have more control over calibration than just a simple tare button. There are off the shelf options but they all suck (10 g resolution is not enough for me, and also I want to control calibration).

The key specifications I have gathered so far:

HX711 uses a 5V supply, is basically setup to use the same supply for AVDD, and has its own band gap reference

HX711 has a PGA capable of 128 gain, ADC resolution is 24 bits, input should be ±20mV differential

Now the confusion comes from the load cell’s spec sheets. Sensitivity is listed as 1.0±0.1 mv/v, what does this mean? I am expecting a specification that relates force to resistance (or voltage) and I don’t see it, I don’t see force mentioned at all except for the total capacity being somewhere between 40kg and 50kg.

Question 1:

How do I calculate the relationship between grams and the LSB read from the HX711? If the load cell’s spec sheet is missing an item, or if I’ve misinterpreted it, please tell me what it is.

Question 2:

If I have a calibration routine, am I allowed to simply decouple the excitation voltage from AVDD and then feed 10V (from a boost converter) into the load cell, in order to achieve a higher resolution? (of course this might lower maximum capacity)

Thanks!

As you may have suspected during your research of OTS instruments, 0.1 grams is distantly unrealistic for 200 kilograms of load cell capacity and less-distant-but-still unrealistic for one 50 kg cell.

It’s useful to think of measuring instruments in terms of number (n) of intervals/graduations/divisions the device can resolve (d) over its rated range (c) and comparing those figures to what’s commercially available. The scale on the deli counter to weight the potato salad and feta is something like 30 x 0.01 pounds, 3000 divisions. The one at the airline baggage counter to weigh your suitcase is 500 x 0.5 pounds, 1000 divisions. A freight scale for pallets might be 5000 x 1 pound, 5000 divisions.

A capacity 50000 gram load cell taken to 0.1 gram resolution is 500,000 divisions. One gram resolution (d = 50k) is really pushing it and I’d expect stability, linearity, and repeatability problems, even under ideal conditions and with a really good load cell. Try to combine those anticipated performance shortfalls across four load cells into one device and it’s certain to be a mess before the first weight is even added.

The 1mV/V spec (“One millivolt per volt”) is the load cell’s signal output when loaded to its rated capacity and a primary figure of merit. For 5 volt excitation, this is 5 millivolts. All things equal, more signal is better. 1mV/V is sort of low, 2 is more common but 3 is easy to find. [3mV/V example

You can try what’s called external excitation but it’s not going to grant any miracles and I doubt you’d see any benefit at all.

If (IF!) you actually need 0.0001 at 50 kg+, you’re looking at magnetic restoration force measuring instruments. Using this technique, an electromagnet lifts the load by way of a series of levers and correlates the electrical current to the magnet with the force applied. If that sounds expensive, it is. Here’s one that’s 70kg by 0.1 grams that starts at $15k so see if you can borrow one from a friend.

[High cap High res](Sartorius Cubis II Balance | MCE70201S-2S00-0 | Data Weighing Systems)](MB Miniature Beam Load Cell Low Capacity)

Thanks, ok I understand now. And while my biggest problem is just what is in-stock at SparkFun at the moment, I just realized that since I am unhappy with my current scale, I can just take the strain gauges out of it and achieve what I want to do since it’s a much lower capacity scale. I just don’t like it since the platform won’t fit my robot and also it’s now reporting slightly wrong numbers.