Precision Load Cell (Wheatstone Bridge) Simulator

[Kleinstein]

The LTspice and Kicad plans don't really match. The Kicad one has the 10 ohm resistor upper end wrong.

The circuit does not really need supper stable resistors. As a simulator the circuit is in a way that the resistor drift has rather little effect - much less than the resistance in a real bridge.
Normally just 25 ppm/K resistors should be good enough in in most cases. With 10 ohm the layout part can also be relevant.
When the pot is used to trim the scale factor, there is no longer the need for low tolerance for most of the resistors (maybe still the 10 ohm). Trimmers also tend to have a rather poor TC, so ideally one would avoid them and maybe have 1 selected resistor for the trim instead.

[hozone]

Thanks!

Fixed the KiCad schematic. Added a Calibration resistor instead of the trimmer.
Because 0.1% does not costs a lot more, I think I'll go for 0.1% +25ppm, just to trim less All resistor will be 0805 SMD.
If all ok I'll move to layout and post there if this schematic seems ok to you.

[Kleinstein]

The circuit is still missing the output resistors (175 or 170 ohm). Ideally one would add separate resistors for the different settings (more with the 0 and less with 2 mV/V) to compensate for the extra 10 ohm resistors.  I would also expect to have a switch setting also for 0.
The scaling should not be fixed mV steps, but more in mV/V of excitation.  With the current values one may get it about rigth, though rather large effect for R_cal.

[hozone]

Thanks,

- added SIG+ as step (gain*VExc)
- added output resistor, it should match the 175 + 175 output

As for the switch settings for 0, what do you mean?

[hozone]

Find the PCB attached. Note that there is no ground plane. Don't know if it's better to put one.

[floxia]

How did it perform? I am considering to give it a go!

[hozone]

I'll test it the next week.
Anyway I've yet made a couple of changes. Like adding a fixed calib SMD resistor of 23R7 insted of the THT. And a toggle DPDT to invert the EXC without disconnecting the cables. This is usefull to simulate the "unload" of a load cell.

[tmd63]

I know this is a bit late in the day but have you tried to simulate it on https://www.falstad.com/circuit/ ?
I find the visual simulations work really well.

[hozone]

Thank you for suggestion.
I know falstad.com but I've simulated it in LTSpice, that is a little better than falstad.com
You can find the LTSpice simulation file above in my posts (it's the .asc file).

[floxia]

Please keep me updated because I'm really interested, I have been looking for such project and would be lovely to give it a go!

[hozone]

Hello,

it works.
I can not say more than this cause I've not precision instruments to test it against.
Latest version has 0.0mV output, and a dip setting to allow selection of calibration Resistor or calibration Trimmer or a parallel between those.
I'll post project files later on my blog and I will link here details.

[floxia]

Awesome job, please post your blog link when ready!!

[hozone]

Update.
I've notice that the output has a little offset, that depends on Vexc.
On 3 out of 5 device I've found an offset of 0.29mV with 10V excitation, and 0.14mV with 5V excitation. That seems related. I can read the offset even on 0.0mV range output.
Find attached the Excel with test on 3 devices.
LTSpice simulation of course does not have this offset.
Where do you think this offset comes from?

Measured using AN870. (Even on UT-61e there's this offset).

[hozone]

Update 2.
It may help.
The resistance between SIG- and SIG+ is almost 350ohm when no Excitation is provider. As soon as the Excitation is provided it change from range to range up to 470ohm almost on the 2.0mV/Vexc range.
Maybe something related to the PCB design?

[hozone]

I haven't find the reason why this happens, maybe @Kleinstein can help me?

[Kleinstein]

The layout looks not that great and could be the reason the the offset / not to perfect zero. The zero path it not proper from the right point.

[hozone]

Thank you.

Attached the KiCad project.
I'm not that sure that it depends on layout. Or better, It also depends on this but I've "big" difference.

So i think is something related to other, like the internal resistance of the excitation voltage supply. Note that even using a battery does not make changes.

Let's talk about resistance, below the measurements between SIG- and SIG+ in the different position of the switch
Without excitation voltage:
- all ranges 350ohm +- 2ohm
With excitation voltage 5V:
- 0mV/VExc: 350ohm +-1ohm
- 0.5mV/VExc: 350ohm +-29ohm
- 1.0mV/VExc: 350ohm +-59ohm
- 1.5mV/VExc: 350ohm +-88ohm
- 2.0mV/VExc: 350ohm +-116ohm
With excitation voltage 10V:
- 0mV/VExc: 350ohm +-2ohm
- 0.5mV/VExc: 350ohm +-61ohm
- 1.0mV/VExc: 350ohm +-120ohm
- 1.5mV/VExc: 350ohm +-179ohm
- 2.0mV/VExc: 350ohm +-238ohm

Considerations:
- As you see there's a LOT of difference in readings, we are speaking of almost +-30ohm per step on 5V and +-60ohm per step on 10V. So it's linear, and it should be +-6ohm per step at 1V.
- Note that there's a +-ohm on ranges. It means that the output resistance depends on the REV switch, so it's like inverting the meter probes. As example on 10V 2mV/VExc i read 588ohm or if i invert the probes i read 112ohm

I've try it in a simpler way using a breadboard. The behavior is similar.
I miss the knowledge to understand this phenomenon. Have you got any clue with this happens?

[Kleinstein]

When measuring the resistance with the excitation still active, the extra voltage will in most cases interfere with the resistance measurement. How much depends on the meter. So different meters can read different things and nothing to worry about. Ideally one should do the resistance test with a 0 V excitation and thus the excitation terminals shorted. 

Another point can be leakage current between the excitation and meter. This would mainly be an issue with a bench meter and mains powered supply, but also the isolation of some of the cheap banana wires is surprisingly poor (can reach the low Gohm range).

P.s.:
The layout can cause a small voltage to be read when the switch is in the zero position. So the zero is not absolute accurate. This would be on the order of maybe 1 ppm of the excitation, so one may no notice this very much.

[Overspeed]

Hello

For 350 and 120 ohms Micro measurement sale high precision foil resistor as MR2-350-128 with +/- 0.005 % in half bridge configuration

I use them for calibration

wiring and connectors as ''banana'' shall be made with Hight quality component

Regards
OS

[hozone]

Thank you!

So, I will not worry about the resistance measurement, mostly because without excitation applied I read 350ohm +-2ohm between ranges, that is ok to me.
Also, thank you for making me notice the zero position drift due to layout.

What sounds strange to me is that:
- the absolute error is almost the same between rages, and it's twice the mV absolute error in range 0.0
- the absolute error is linear and depends on the excitation voltage

This seems something not fully related to component quality/accuracy, but to the design of the device. Am I wrong saying this?

Excitation 10.024V readings (expected, read and abserr in mV)         

 range   expected      read    abserr
   0.0      0.000     0.181     0.181   
   0.5      5.012     5.311     0.299
   1.0     10.024    10.316     0.292
   1.5     15.036    15.314     0.278
   2.0     20.048    20.320     0.272



Excitation 4.999V readings (expected, read and abserr in mV)

 range    expected       read    abserr
   0.0       0.000      0.087     0.087
   0.5       2.500      2.647     0.148
   1.0       4.999      5.146     0.147
   1.5       7.499      7.643     0.145
   2.0       9.998     10.142     0.144

[Overspeed]

Hello

I dont really understand the purpose of your setup I work with strain gage since 20 years

You want to generate a voltage value in mV to simulate a load on a full bridge = OK
You need a mV source as the output of your full bridge is a voltage = OK

SO

Why don t use a precision voltage source and a voltage divider set ?

Another strange point is the 0.1% resistor you use except if your balance the bridge / sort the resistor as even for a real strain gage bridge that mandatory to match the resistor value 0.1 % on a 350 Ohm resistor is huge value for a strain gage .

Regards
OS

[floobydust]

I think there is confusion about strain gages, they end up being highly imprecise. Their resistance can be quite close but the mechanical mounting and deflection is a mess. There is an offset error, a span error.
For this reason, I've always used a multi-turn potentiometer in a simulator (bridge) to dial-in the output voltage I want. It's really only used to check hardware amplifiers and firmware for calibration.
Otherwise, how much output you get, you have to measure from the actual load cell anyway instead of expecting a precise number. Temperature also causes huge shifts.

[Overspeed]

Hello

I agree regarding strain gage that very sensible to mounting / setup . a lot or even most strain bridge have '' zero pot '' system to set the output to zero / zero load

I work with 1/4 or 1/2 bridge and take also in consideration the linearity of the output voltage , in 1/4 bridge I use mostly 3 wire wiring to avoid wire resistance effect

to simulate the output of the bridge I use a voltage precision source with a resistor divider built with foil resistor

Regards
OS

[hozone]

Thank you all.

The purpose it to build a device as accurate as possible to simulate a load cell in the field, where there will not be a precision voltage source.
I should use a multi-turn potentiometer to simulate the bridge and a multimeter in mV range, but i will lose prefixed steps.
I know load cell are not that accurate, indeed they could come with a calibration sheet.
Of course I can still live with the accuracy of the device I'm using. I'm just trying to understand why it is acting like so.

Said this, still I don't understand why the circuit I'm using has an absolute error on readings when excitation is applied.
Simulating the internal voltage source resistance does not produce a constant between range absolute error.

Maybe I should just consider that this kind of circuit has a .003% VExc absolute error offset on all ranges. So for 10V it will be a 0.0003V, so 0.3mV offset. Like experimental results. And for 5V it will be 0.15mV.. and so on.

[Henrik_V]

Hello

I dont really understand the purpose of your setup I work with strain gage since 20 years

You want to generate a voltage value in mV to simulate a load on a full bridge = OK
You need a mV source as the output of your full bridge is a voltage = OK

SO

Whey don t use a precision voltage source and a voltage divider set ?

Another strange point is the 0.1% resistor you use except if your balance the bridge / sort the resistor as even for a real strain gage bridge that mandatory to match the resistor value 0.1 % on a 350 Ohm resistor is huge value for a strain gage .

Regards
OS

Well, the trick with a bridge simulator is the mV/V .. and it will run with AC bridge amplifiers too .
(And clever bridge amplifier ADC will not measure mV, it will use the bridge voltage as ADC reference to measure the ratio )
And a resistive simulator is powered by the amplifier.

An alternative simulator is the use of a DAC using the bridge exitation as a reference. 
See Fig. 5 of https://oar.ptb.de/files/download/310.20150208.pdf