Load cell mounting

[Jon86]

Not so much an electronic question here, but I've not thought of a better place to ask this.
I'm working on a set of ultra high-precision scales, with about 100g of resolution. I found this load cell:

http://www.ebay.co.uk/itm/Electronic-Balance-Four-Wire-Weighing-Load-Cell-Sensor-100g-/121137660908?pt=UK_BOI_Electrical_Test_Measurement_Equipment_ET&hash=item1c345e03ec

But I'm not sure how to go about mounting the plate on top of it. Is it just glued to the white patch on top? I can see that the base would be screwed on, but obviously it's one solid piece of metal so mounting both with screws wouldn't work.

[Andy Watson]

Think of it as a beam. You clamp one end - probably the end with wires - and put the load on the other end. The measurement would be due to the bending of the beam. If it were a beam you would have to pay attention to locating the load correctly, and ensuring there are no artefact forces acting to bias the measurement.

However, the inside of the beam has been milled away - this gives the device an almost parallel motion in the direction of the measurement. As well as making the device more sensitive in the direction of measurement, it also reduces sensitivity to off-axis loads and forces. All you need to do is clamp one end - using the bolt holes, and attach your pan/plate to the other set of bolt holes. Try to mount the plate such that it is balanced about the bolt holes - so as to avoid unnecessary twisting moments in the beam.

[turbo!]

Not so much an electronic question here, but I've not thought of a better place to ask this.
I'm working on a set of ultra high-precision scales, with about 100g of resolution. I found this load cell:

But I'm not sure how to go about mounting the plate on top of it. Is it just glued to the white patch on top? I can see that the base would be screwed on, but obviously it's one solid piece of metal so mounting both with screws wouldn't work.

Did you Google at all??? Set it up like a dive board, wire side on shore.

This specs are far from ultra precision. This is your typical eBay grade pocket scale.

This, can correct by A/D converter calibration programming
Non-linearity    ±0.03%

This one is harder.
Hysteresis Error    ±0.04%

These two you can't really correct it.
Repeatability    ±0.03%
Creep    ±0.05%

[retrolefty]

Keep in mind that the weight of the pan or plate will take up some of the dynamic range of the sensor. This can be 'zeroed out' electronically or by software but it still takes up some of the fundamental weight total range that you can measure. Also a mechanical stop under the moving end of the beam is needed to be sure you don't over-range the sensor mechanically beyond it's maximum rating.

[babysitter]

For 100g of resolution, a comparator will do, and you selected a bridge with a lot of headroom accuracy-wise 

You might be able to get a average-class scale out of it, anyway. Use the specs to calculate the error from % to weight.
Maybe you are lucky and get to a piece which is better than specified.

Do you have a comparison scale or a set of weights which are suitable for your requirements ?

[Jon86]

Thanks for all your help
Yes, I do have some precision weights which will be suitable for calibration, I didn't intend to do it without those.
What do you mean a comparator will do? A comparator where?

[turbo!]

I would just buy a cheap Chinese pocket scale. You could use some button combinations to get it to display the raw A/D count or if you choose not to use the circuitry, you can just pig tail off of the load cell wires and attach it to your choice of electronic circuitry stuff

[bronson]

I second the cheap Chinese scale recommendation.  Their quality has gotten disturbingly good in the past few years.  Except I would keep the pan and load cell, chuck the rest, and wire the Wheatstone bridge into an instrumentation amp feeding an A/D.  The cheap scale guts I've seen are too highly integrated to be of much use.

[babysitter]

Your scale has a capacity range of 100g, you wrote about 100g resolution. So a Comparator set to switch from low output meaning 0g to high meaning 100g would do the trick

I would recommend you to select a high-resolution ADC with *important* differential input and reference voltage input.
Connect the sig+ and sig- of the wheatstone bridge circuit to the ADC input, and supply a voltage which fits the ADCs reference input requirements to both bridge excitation connections and ADC reference inputs. This will only jiggle at the last bits of resolutions.

The alternative is connect the bridge signal connectors to a instrumentation amplifier and get the single ended output signal to a ADC. Again, monitor the excitation voltage, this time with a second ADC channel.

This device will exhibit a quite strong thermometer behaviour, the null value will drift all over the place then. Treat slowly moving measurement changes as temperature drift in software and supress, only "accept" step changes which are at least happening at the speed of the slowest weight rise you can foresee.

Consider using a temperature sensor to give your software the chance to compensate temperature effects, like "its five degree colder then on calibration, so i need to use different gain and offset values".

You will learn a lot about feeding and care of such sensors when beginning with a average piece like this. Maybe one day your electronics will see connection of a better one...

[David Hess]

The alternative is connect the bridge signal connectors to a instrumentation amplifier and get the single ended output signal to a ADC. Again, monitor the excitation voltage, this time with a second ADC channel.

Or drive one side of the excitation to produce a virtual ground on one side of the bridge producing a single ended output directly.  This is shown on the last example of figure 4 on page 4:

http://www.linear.com/docs/4134

Delta-sigma instrumentation converters are so good now though that they can interface with a bridge circuit ratiometrically without any additional active signal conditioning so there is little reason to amplify or convert the output of the bridge from differential to single ended.

[Jon86]

I don't know why the hell I wrote that, I meant 0.001g resolution 
Anyway I think I know what I'm doing now, time to order some parts 

[wraper]

I don't know why the hell I wrote that, I meant 0.001g resolution 
Anyway I think I know what I'm doing now, time to order some parts 

You are crazy if expect any usable resolution out of it, even if order of magnitude worse.

[Jon86]

Why would it be worse? Surely I could make a circuit with at least slightly higher precision than the one I bought?

[SeanB]

It is a cheap loadcell. At best you can get 0.05g resolution that is repeatable on it, you will find that anything higher resolution will show up the repeatability, nonlinearity and thermal performance of the loadcell quite well. 10000 counts is a 16 bit converter at a minimum, which is pretty hard to get to work stably at what is effectively DC for the lower bits. You basically need a 6 digit DMM to read it, and then you will have to keep it at a stable temperature to get any usable results out of the lower digits.

Loadcells make pretty good thermometers if you only use the one leg. The higher cost ones have a lot more care taken during assembly to get the temperature compensation side right. Right materials, right processing and assembly, and you get a loadcell that is an industry standard 150R 4 wire unit.

[Jon86]

Ah I see, so how do the $20 scales achieve 1mg resolution then?

[wraper]

Just for comparison 5 1/2 digit scales insides photos. Just were remains of the scales lying nearby (with dust layer on them  ). If you notice there is LM399AH heated reference on the board. These scales goes up to 4 kg, but for smaller weight ones are very similar.

[SeanB]

Take 500 samples and average them, and treat any variation of the zero reading under a certain rate as being effectively zero. Also they do not stay on long enough for the loadcell to heat up appreciably from self heating, so you do not see the drift with a load. Turn on, it does readings till it settles around a value, decides that will be zero and displays it, then only regards a step change bigger than a certain amount as being a valid mass. That hides a lot of the errors and drift. At 100g or so full scale that will allow you to show 10mg resolution reasonably easy, even if the last digit will drift around a little. A lot also will use the stable load reading and if the sampled value is changing at a rate below the trigger level it will silently just track the change without changing the displayed value. Easy to check by placing a load near full scale and leaving it there for a few minutes with a stable reading then take it off, you will see a zero offset for a few seconds till it gets autozeroed again. Software compensating for the limitations of the hardware.

[SeanB]

Just for comparison 5 1/2 digit scales insides photos. Just were remains of the scales lying nearby (with dust layer on it  ). If you notice there is LM399AH heated reference on the board. These scales goes up to 4 kg, but for weight ones are very similar.

Mettler Toledo, nice units. See the big capacitors used in the dual slope ADC there, with the microcontroller to drive it all and do the corrections. Decent loadcell as well, will not give much error with off centre loads, though it will absolutely need to be mounted level. Might only be 2 counts off at full scale at any corner compared to the centre of thrust. This probably only uses 3-4 samples per reading with the dual slope ADC, or just increases the counter stages to do the same with a single reading. I see it is over 25 years old.

[wraper]

Made in 94-95 according to the date codes on the parts.

[Rufus]

As others said it is a beam so you fix one end and put load on the other.

The more difficult problem is the rated overload of 200% which means if you place more than 300g on it it will take a permanent set and be damaged.

You need mechanical stops which take the load when it exceeds maybe 200g. The load cell deflection will be tiny, probably a fraction of a mm so the stops have to be robust and precise. Usually some kind of spring mechanism is required so the spring collapses between maximum and overload giving a much bigger deflection which makes the stops much less critical.

[David Hess]

Ah I see, so how do the $20 scales achieve 1mg resolution then?

Resolution or accuracy? What full scale capacity with 1mg resolution?

Resolution depends on the electronics but the later is limited by the mechanical design and implementation.  Linearity, hysteresis, creep, temperature effects on offset, and temperature effects on span all impact accuracy and compensating for them in the electronics is not trivial.

Transducer accuracy in the range of 0.01% to 0.05% is typical for strain gauge load cells yielding from 2,000 to 10,000 counts of accuracy.

[wraper]

Old Japanese Vibra PF-3000CD scales. Max 3 kg, min 5g, d=0.01g, e=0.1g

[bronson]

wraper, those are pics of the scale's underwear.  Keep disassembling!

[SeanB]

I have a Saeco unit that will be going to bits soon.