For a long time I've been trying to make a DIY milligram-accurate scale, and milligram-accurate strain load cells are expensive. Does anyone know if the resolution of this is high enough?

My dad built one with an electromagnetic coil many years ago, can't find the design, capacity was limited but had 100? microgram resolution, maybe similar to this applied science video https://www.youtube.com/watch?v=ta7nlkI5K5g but I think simpler.

maybe something like this: https://www.erowid.org/archive/rhodium/chemistry/equipment/s... https://www.erowid.org/archive/rhodium/chemistry/equipment/s...

I found some old emails about the scale from 2005/6 but can't find the link and it would probably be gone anyways.

Thanks for the reference! If you do find the link maybe it will be in the WABAC machine.

That's the most interesting misspelling of "wayback" I've seen so far.

TIL it may not necessarily be a misspelling: https://en.m.wikipedia.org/wiki/Wayback_Machine_(Peabody%27s...

My uninformed opinion based on no experience is that this will creep so you should use a Kibble balance instead. (See sibling comment by s0rce.) Or maybe use a lever arm to amplify milligrams into tens of milligrams of force. Or just a regular pan balance. How big a weight do you want milligram precision on?

If you have some way to cut a precise shape out of some kind of metal sheet of well-controlled thickness, could you cut out a milligram precision bismar balance or steelyard?

Probably not, PCBs are also terrible as load cells because fibers break.

In general if you want a precise and accurate strain gauge you’ll be paying a lot for it, especially for one that doesn’t need to be recalibrated before every use and after nearly every measurement.

In the video, the author tries it with a small component that I think is a SOT-323-5 or similar. Based on [1], that weighs about 5-8 mg.

[1] https://www.mccsemi.com/pdf/ComponentWeightInformation.pdf

Milligram accuracy isn’t a really a direct property of the load cell. A lot of it comes down to creep and hysteresis behavior.

You can just buy strain gauges which are specially cut copper foils on a thin plastic substrate that you glue to something, like a metal strip. The resistance changes very slightly as the metal bends.

You measure the change in resistance with a wheatstone bridge tuned correctly.

You basically just need a strain gague (a few dollars), 4 resistors, an op-amp, and a microcontroller with an ADC.

Calibration is important and you'll run into things like the metal bar creeping, permanently bending as a result of weight being put on and off.

But also, milligram accurate scales are $20 on Amazon https://www.amazon.com/GRAM-PRES-Precision-Milligram-Reloadi...

Not every scale with milligram repeatability has milligram accuracy, and not every scale with milligram resolution readouts has milligram repeatability. You probably know that, but not everyone reading this does.

(This one might be fine? It does claim to have a 50.000g calibration weight, which is a good sign, but it doesn't say anything about metrological traceability, which is a bad sign.)

Precision and accuracy is very expensive there is a reason why high end measurement equipment costs as much as it does.

At some level it becomes expensive, but it's far from clear that five significant figures of mass is that level.

A US$2 quartz watch measures time to 5½ significant figures, US$10 multimeters routinely measure voltage to 4½, and US$5 GPS receivers can provide you with time measurements accurate to 40ns that inherit the drift of world metrology standards, a precision of 16 significant figures if you are measuring a long enough time interval (over 10 years).

As a quick and dirty rule of thumb measuring parts per million in anything except time or frequency will get expensive. Temperature drifts will cause expansions and contractions on that order if you’re measuring lengths.

Yes, but in this case we're measuring weights, not lengths, and we're looking for 5 sig figs, not 6.

Where are you finding $10 4.5 digit multimeters?

Hardware store, usually. 3.5 digits is common, 4.5 digits less so.

3.5 digit ones sell for far less than that (especially the infamous 830-series based on the ICL7106 and its clones). I haven't found any 4.5 digit ones at that price.

>metrological traceability

It's a $20 scale, if you had need of metrology you wouldn't be buying a $20 scale. Most of us do not need metrology. I want to make small amounts of pickles with perhaps unreasonably precise measurements of salt at scales where half a gram or maybe a tenth of a gram is significant.

Shahriar has a video on repairing a milligram resolution scale: https://www.youtube.com/watch?v=Lz9mBc6FGzE

They don't tend to use strain gauges I think.

Can anyone explain why the BRIDGE_SUPPLY voltage is connected to the voltage regulator output and PWM signal at the same time (through the FSA5157L6 analog switch)?

I don't think it is. PWM_BIAS is used as the select for the analog switch. High connects COM to B1 (GND) and low connects COM to B0 (BRIDGE_SUPPLY).

Ah yes, you're right, my bad.

Piezoelectric or Strain Gauge Based Force Transducers?

https://www.hbkworld.com/en/knowledge/resource-center/articl...

Piezo vs. strain gauge https://www.kistler.com/US/en/piezo-vs.-strain-gauge/C000001...

The challenge is that while you can make a strain gauge out of just about anything, making them repeatable over temperature, humidity (in the case of hygroscopic materials, like PCB FR4) and repeated flexing is where it gets difficult.

For this, while I'm sure it works, if the humidity and/or temperature changes, the same deflection will result in different readings.

If you can calibrate it immediately before each use, or you don't care about absolute values, this is a completely valid option.

"Real" strain gauges generally use a constantan resistive element to deal with the temperature variability, deposited on a plastic carrier film (typically polyimide). The film elements then get glued to the stress sensing member. They're fairly inexpensive too.

All sensors are thermometers, some measure other things too.

I hadn't thought about the hygroscopic and expansion questions; I think FR4 is, like wood, almost immune to longitudinal variation with temperature and humidity due to its anisotropy. (But I haven't tried to measure that.) And strain gauges are low enough impedance that I'd expect the capacitive effects to matter.

The temperature coefficient of resistance of the strain element seems like a concern, though, and so do thermal EMFs. My kitchen scale zeroes when I turn it on, a procedure that should be able to cancel one of these two but not both. Maybe you could have a diode thermal sensor, as an MCXO does, to measure the temperature so you can cancel both?

This project seems to do the first-order temperature correction thing:

> The included sample firmware will wait until a serial console is opened, perform a 5 second offset calibration, then sample continuously at the lowest gain setting. The graph.py script can be used to display the output.

> For high sensitivity measurements, it's important to let the board reach a stable operating temperature for at least 5 minutes before calibrating.

The thing I intuitively worry about here is creep. Does FR4 creep enough to worry about? Normally you make strain gauge bodies out of steel because it doesn't.

I'm sure there's a general rule of thumb where this approach works best as an approximation (such as a simple on/off switch)... which may give us opportunities to simplify the BOM list further with cheaper parts at cost to accuracy which wouldn't apply to PCB anyway.

Where I mostly seen this approach is with 3D printers where we just want to know if the nozzle is touching the print base.

But if we can quantify the general worse case variation between most PCBs then maybe we can create a recommended strain sensor element with a semi-quantified level of accuracy so it's not just an on/off sense.

I figured that's why they put it in a sealed chassis in the demo.

> Thinner boards will result in a smaller output voltage swing.

For the same weight? I would expect the opposite.

For the same deflection, I assume.

The device effectively measures mechanical strain at the surface of the PCB. The surfaces of a thicker board will experience more strain from bending because the radius of the curve (at the surface) is greater.