Millivolt generation for Load Cell Simulator

[vtx22]

Hello everyone,

I'm a student currrently working on a project for my university / for my work and I think I need some help and some ideas.

What is the project about?

I have to design a load cell simulator. Imagine you are creating software for an industrial weighing system (like a SIWAREX module for example). Instead of testing your ideas in the field you try to analyze and verify the behaviour in the lab. Currently we are using a manual setup for those tests: You turn a potentiometer knob to change the weight, you press buttons for digital outputs etc.
My task is it to automate those tests. The software part is not the problem, I am currently stuck on the hardware, more precisely on the load cell voltage generation. So the voltage that is proportional to the weight.

Hardware prequesits:

I have to use a Raspberry Pi for the base of the simulator
More importantly: I have to use a second add-on board that generates the analog voltages that have to be converted into the load cell voltage, so those are fix

The task im currently stuck on:

I have to create a circuit that uses the add-on board volatages (0 - 10V analog, 2 channels) to generate the differential load cell voltage. The load cell voltage is dependant on a lot of factors, so i cannot use a simple voltage divider type circuit (i guess at least).

Wheatstone bridge loadcell working priciple:


As you can see: I have to generate a differential voltage between the SIG+ and SIG- lines of the loadcell. The DC voltage of the cell is half the exitation voltage (here EXC=5V).  Those voltages are quiete small, and I have to mimic different types of load cells (1mV/V to 4mV/V). With possible EXC voltages from 5V to 10V the maximum differential voltage is 40mV.

What I have done so far:

I designed an opamp circuit to add 0 - 20mV on 5VDC based on the first 0 - 10V channel for the SIG+ line and to substract 0 - 20mV from 5VDC for the SIG- line to generate 0 - 40mV differential voltage between SIG+ and SIG-. You can see the circuit in the next picture.



VGT1 and VGT2 are the 5VDC, which is half of the (here for example 10V) EXC voltage. V1 and V5 are the two 0 - 10V analog channels. The upper part is an adder and an inverter, the lower part ia a subtractor.
The simulation looked promising, I made a PCB and assembled all the parts last week.
Immediately I noticed some problems with my circuit:

1. Noise: There is some noise on the output, however because there is a lot of filtering going on by the SIWAREX module I dont think that is a major problem as of right now

2. Offset: I noticed an extremly high offset of about 4mV when V1 and V5 are 0V. As i found out later, the circuit i designed is highly dependant on precise and stable resistor values. I used only 1% resistors, whis is quiete problematic i guess. However, I think i can get away with it by using a software based calibration and correction. I am working on that right now, but i do not have any results.

3. Temperature: Temperature changes have an extremly high influence on the differential voltage, because as i already mentioned the circuit is very dependant on stable reistors. The RaspberryPi emits a lot of heat causing temperature instability. After about 30min, when the whole device has warmed up, the output differential voltage fluctuates with about +-50uV, i think that is fine for my application

4. "Long" term drift. The output voltage drifts over the term of about 100uV / 30min. I dont think i can fix that problem without temperature controlling the device. Maybe a case would help, and i could do some software temperature correction, but i think i have to rely on the zeroing capabilities of the SIWAREX module.

My "simple" question:

Do you have any ideas on how i can further improve the circuit, to fix the problems i mentioned? Or should i consider redesigning the whole analog circuit?

Thanks in advance for any help / new circuit ideas (btw that was my first analog circuit design, so please be kind )

[Marco]

If I had to DIY it as discrete as possible I would use a railsplitter and a current source/sink to resistors connecting to the split rail.

If I had to quickly do it I'd google for single ended to differential converter :
https://www.analog.com/en/analog-dialogue/articles/versatile-precision-single-ended-to-differential-converter.html

[RoGeorge]

Currently we are using a manual setup for those tests: You turn a potentiometer knob to change the weight, you press buttons for digital outputs etc.
My task is it to automate those tests. The software part is not the problem

I will use the already existing setup, except replacing the potentiometer with digital ones.  Similar for buttons, instead of manually pressing them, use either relays' contacts, or conduction gates (analog switches).  See what fits better with the existing setup.  Make a new board with some microcontroller and the digitally controlled switches/potentiometers and connect it to a PC for automated testing.

This way all the mechanical fixtures, connectors and known working testing procedures remain unchanged.

There are digital potentiometers, where one can set the cursor by sending the desired value by I2C, other models have an SPI interface.  Often 256 position, some are more, some are less, some are dual, or logarithmic instead of linear, etc.

For example:
https://www.ti.com/data-converters/digital-potentiometer/products.html
https://www.ti.com/switches-multiplexers/analog/products.html
https://components.omron.com/us-en/products/relays/power-relays/dc-small-power-relay

[dzseki]

Maybe I like to overcomplicate things, but I'd take two finely matched transistors and put one to each side of a small peltier element. Transistors would have about 2mV/C heat dependence in their VBE by applying power to the peltier element. This would be quite slow to react though...

[RoGeorge]

Forgot to comment about the opamps issues 1-5 you asked.  I didn't look at the schematic, but regarding your concerns about offset and temperature drift, there are the so called auto-zero opamps (sometimes called, zero drift, switched opamps, chopper stabilized, etc. name varies).

It's an opamp that measures itself and auto-adjusts against offset and thermal deviations.  This is the idea:  Chopper Stabilized (Auto-Zero) Precision Op Amps - Analog Devices - MT-055 tutorial, and it is done transparently for the end user.  Such an opamp can be used just as any other normal opamp.

Example:
https://www.analog.com/en/product-category/zero-drift-op-amps.html

[Kleinstein]

In most cases the load cell is driven by the measuring instrument. So the test voltages should not be absolute mV, but fractions of the bridge voltage. This is especially true for high grade read-out instruments that may even work with AC excitation or reverse the polarity.

So the more logical setup is for the main part a dummy bridge with the usually 4 x 350 Ohms and than additional part to simulate an unblance coupled with relatively large resistors (e.g. 20-50 K). So using a simple voltage divider is very much an option and the obvious choice. For the unblance setting digital pots may be option. The tricky part could be the supply for the digital pots. A setting via a voltage source would be more like a less accurate way and would need an extra measurement of the bridge excitation.

[David Hess]

Real load cell simulators use the same arrangement that a load cell uses to adjust zero.  A bridge is made with precision resistors, and then a divider across the excitation applies a current through a high value of resistance to one of the outputs.

If I was designing an electrically controlled one now, I would do the same thing but apply the current using a multiplying DAC preserving ratiometric operation.  A true differential output is rarely needed so it should be sufficient to adjust the output of only one side.

A fully active simulator with a precision resistive divider providing a reference to a pair of operational amplifiers is possible but I do not think it has any advantages.  Note that the load that the bridge applies to the excitation should be simulated also.

[Terry Bites]

The ready made tester is a much better idea and you can have real cal done on it.