Hi everyone,

I would greatly value any input on a load cell conditioning circuit that I designed for a livestock scale. I work for a small company as the sole electronic engineer, and being a recent graduate with no senior electronic engineers to learn from, am still inexperienced and cannot get technical input from superiors. I cannot divulge on the forum too many details of my design, at request of my employer, but will try to give sufficient information for informed advice.

A brief overview of the circuit:
Four load cells, excited with the same 5V supply rails used to supply the rest of the analogue circuitry, are each fed to the input of an INA332 instrumentation amplifier. The IA is configured with external resistors to have a gain of 905 V/V, which is within, but near the maximum of the rated gain range given in the datasheet (rated maximum is 1000 V/V). The REF input is fed from a MCP6022 opamp buffer, which in turn is fed by a resistive voltage divider and first-order low pass filter. The REF voltage is 50mV (as I am only interested positive voltages from the load cells). The IA is then fed to another passive, first-order low pass filter and then to an input of a CD4051 multiplexer. This is repeated for each of the four load cells, with four of the MUX inputs being used. The second opamp in the MCP6022 chip is then used to buffer the output of the MUX (improving the voltage loading factor between the MUX and the ADC). The buffer is then fed to the ADC of a STM32F4 microcontroller through a 2m cable. The STM32 sets the control pins of the MUX and then polls for an ADC reading, cycling through the four channels continuously and performing running averaging.
The load cells are 2T, shear-beam load cells, which are much more competitively priced than other load cells that I could source here in South Africa. I am not using their full range (hence the high IA gain) and have designed for a total capacity of 2T.

Some relevant given specs of the load cells:
- Sensitivity: 2.0000±0.002mV/V
- Total error: ±0.02%F.S
- Creep (30 minutes): ±0.02%F.S.
- Recommended voltage: 5V～12V
- Zero balance: ±1%F.S.
- Temperature effect on load: ±0.02%F.S/10℃
- Temperature effect on zero: ±0.02%F.S/10℃


My problem:
With four load cells connected and sitting on the bench with their casings grounded I have stable readings with little fluctuation, but with the the load cells installed in steel weigh beams or with the casings ungrounded, I experience fluctuations in the voltage read by the ADC. The voltages output for each channel seem, within a matter of seconds, to drift by 50 LSBs of the MCU's 12-bit ADC (~ 40.3 mV). The casings of the load cells seem to be isolated from the internal Wheatstone bridge and therefore are not grounded unless I ground them externally. When the load cells are installed in the weigh beams, they also seem to interfere with one another. For instance, if I disconnect or apply shear force to just one cell, the readings of all four channels vary (not so if they are not installed in the steel weigh beams).

My questions:
1. Should I be grounding the weigh beams (and thereby the load cell casings), and if so, how should I be doing so? Could this pose a risk of ground loops? I am uncertain, as I assume that the casings were intentionally not grounded by the manufacturer but grounding them seems to stabilize the readings. This may be a very trivial question, but I admit my ignorance, and am a little confused.
2. Am I too audacious to use a single amplifier stage with a gain of 905 V/V? I believe that my gain/phase margins are sufficient, given the very low frequency passband.
3. Is it generally unwise to use high-capacity load cells at only a quarter of their capacity with high-gain amplification as opposed to using lower-capacity load cells? Once again, my only reason for using the 2T load cells is cost, and if it is unwise then I should rather redesign for lower-capacity load cells.

Thank you in advance for your time and willingness to consider my problem and offer advice. Any input or pointers, however brief, will be appreciated immensely. I can share more details on the design privately if need be.

 

Hi Hartley,
Welcome to AAC.
Do you have a circuit diagram that you could post?

On your weigh platform Base plate, I assume the LC's are mounted at each corner?
If Yes, then applying weight anywhere on the Base plate will register on all 4 load cells.
With 4, LC's it is typical to parallel the load cells.

The casings of the load cells seem to be isolated from the internal Wheatstone bridge and therefore are not grounded unless I ground them externally.

The weigh frame and baseplate are connected back to 0V on the main electronic unit.

E

OT: My Son has a farm in PE, RSA, I have designed and built all his animal scales.

 

Hi Eric,

Thank you so much for your quick reply and willingness to help. I have roughly hand-drawn the schematic, showing just a single channel of the circuit for simplicity. The circuit repeats for all four load cells, except that the multiplexer and reference voltage circuit are shared by all four channels. I hope that the schematic is clear and comprehensive enough:


The load cells are mounted in two load beams (each with two load cells). The load beams are then fitted into channels underneath the weigh plate. I have not yet implemented any measure to ensure that the beams will consistently make proper electrical contact with the main frame to which the rest of the system is grounded (the frame will be painted), but it seems that will be necessary and I will ensure this.

It is fascinating to hear that your son farms near Port Elizabeth. While I have only visited PE as a young boy, I remember it as a beautiful area of the country. I am sure that it is a great region for livestock farming.

Edit: I see that I initially looked over the two application notes that you shared. They both look very helpful and I will eagerly read through them.

Thank you again,
John (Hartley)

 

hi H,
Please check that this LTSpice simulation circuit is correct?

Do you have a circuit like for each load cell, if so IMO it is over complex.

The Ref voltage source requires a voltage reference, the 5V may drift.

E
The INA is untidy on my sim I will redraw later.
What is the LC specification?

 

Hi H,
Slightly modified, in order to show the response, to a dummy load.
E

 

Hi Eric,

Wow, thank you for putting in so much effort and even simulating the circuit in LTspice. That is well over and above the help that I expected someone would be willing to offer.

That simulation looks correct and matches my simulation. Yes, I do indeed have a circuit like that for each load cell. I see that it is rather complex.

The LC specifications given by the supplier are:
- Sensitivity: 2.0000±0.002mV/V
- Total error: ±0.02%F.S
- Creep (30 minutes): ±0.02%F.S.
- Recommended voltage: 5V～12V
- Zero balance: ±1%F.S.
- Temperature effect on load: ±0.02%F.S/10℃
- Temperature effect on zero: ±0.02%F.S/10℃
- Input impedance 380±10Ω
- Output impedance 350±5Ω
- Insulation impedance ≥5000MΩ
- Safe overload: 150%F.S
- Ultimate overload 200%F.S

The circuit is supplied by an L7805CV voltage regulator.

This circuit is mounted in an aluminium junction box with a grounded chassis, and the indicator's chassis is also currently grounded. As recommended in the "Load Cell Cabling" AN that you shared, I will try disconnecting the internal chassis ground connection in the junction box. I will give feedback once I have done that.

I will also try adjust the REF voltage as you recommended. I had it set to only 50mV as I wanted to maximise the positive voltage range, but perhaps that is too near the rails?

Once again, thank you for the extensive help.

 

hi H,
This plot shows the effect of the very low Vref, any signal less than 3mV is lost.

What are the dimensions of the scale Base plate?
Also, what animal types are you planning to weigh and the accuracy required?

I guess you know most animals will not stand still in the cage while being weighed.

The 7805 Vreg is not stable enough for a Vref signal source, use a TL431 for input to the Vref buffer OPA, then a trim pot to set the actual Vref. Do you need a circuit example ?
E

 

hi H,
This shows a TL431 Vref.
And some additional info.
E

 

hi H,
This plot shows the effect of the very low Vref, any signal less than 3mV is lost.

What are the dimensions of the scale Base plate?
Also, what animal types are you planning to weigh and the accuracy required?

I guess you know most animals will not stand still in the cage while being weighed.

The 7805 Vreg is not stable enough for a Vref signal source, use a TL431 for input to the Vref buffer OPA, then a trim pot to set the actual Vref. Do you need a circuit example ?
E

Hi Eric,

Thank you for all your input and for the time that you have given to helping me. I understand the points that you have brought up regarding the reference voltage. I will amend my design and have a new PCB prototyped.

In the meantime, I have tested the existing prototype with grounding as was recommended in the application notes that you sent me and it had significantly improved the fluctuating mass readings. As mentioned yesterday, I had a grounding connection to the indicator chassis as well as the junction box chassis. It seems that that may have been causing a ground loop. I still have some fluctuation.

Regarding the fact that the animals won't stay still, I am currently performing a running average on the STM32 over a period of about 5 seconds (I may adjust this when testing with an animal on the scale). I perform an average of 250 samples and store the value in a FIFO buffer, on which I perform a running average of 30 samples. This seems to smooth out the reading decently, apart from the load cell drifting with which I have been struggling.

The load cells are mounted in a rectangular arrangement of about 1m by 2m.

I notice that in this simulation, the wire between R2 and the RG input of the IA is missing, but your points are understood and remain valid. I notice with the prototype that the zero output of each IA is a little different - possibly due to component variances.

hi H,
This shows a TL431 Vref.
And some additional info.
E

Thank you for this Eric, I will replace my resistive voltage divider with a TL431 Vref as you recommend. I can see how it would be more stable. This is useful information for future designs as well as this one.

I think, based on your advice and the ANs that you shared, that I should connect the load cells in parallel and then amplifying the parallel combination with one instrumentation amplifier. I would then do away with the multiplexer. Would you have used one stage with such a high gain, or split up the amplifier into multiple low-gain stages? The datasheet specifies a maximum gain of 1000 V/V, but is it wise to use the IA at so near to its maximum value (assuming sufficient gain/phase margins)?

I also notice the addition of 4k7 resistors in series with the load cell output. What is the purpose of these resistors?

Kind regards,
John (Hartley)

 

hi John,
On some installations air bourne electrical noise can be a problem, so some form of filtering is required at the INA inputs , also the 4k7 resistor protect the INA inputs from any over voltage.

I usually connect all 4 LC's in parallel, add a summing circuit if required as shown in the PDF.
This allows you to adjust out any variations in the individual LC's

I expect you are using a micro to do the amplified signal ADC conversion?

As you know the ADC's are ratio metric, so stable tracking reference voltages are important, in supplying the LC's and the micro reference.

I use the USB 5V to power the micro/ circuitry and LC's, that should ensure good ADC tracking.

Can you post a sketch showing the micro wiring?

E

Update: Which version of the STM32 are you using???

 

I am not familiar with operating an instrument amplifier with a single supply. I am not sure that is a problem, but it certainly seems like it could contribute to a problem with a wandering zero offset. Next, how well shielded are the cables from the load cells to the IA?
Aside from that, are the load cells each full bridge load cells, or do they use external bridge completion resistors??
I see in the simulation circuit that a simulation signal is applied across the IA input. I would not ever consider doing that because of the probability of introducing an unknown unbalance condition

 

Can it be common mode coupling mixed with aliasing that is poorly rejected by the ADC. Some ADC will have digital filter to reject the 50/60Hertz from the power systems.

 

Over the years of my career the strain-gauge bridge problems with both load cells and pressure transducers have been from inadequate isolation of the connections between the bridge and the instrument amplifier. That has included non-connected shield sections, non-shielded sections, shields tied to noisy grounds at the sensor, and bridge excitation circuits connected to other circuits.
I would suspect noisy power from the USB connection in this case. USB power often is tied with the system +5 volts power, and while 50 millivolts of noise will seldom upset digital systems, it plays havoc with low level analog signals.

 

Hi,
The TS is using Load Cells, not just strain gauges. Read post #6

Load cells are supplied with attached screened cables of a fixed length.

The TS is currently not using the USB 5V power.

Most ADC's are ratio metric, which means if the TS uses the same 5V for the ADC and Load cells, the two devices will track correctly.

As the TS is using 3.3V powered STM32 micro, it will mean that the Load cells will have to be powered from 5V or 10V and the STM32 using its voltage internal reference.

The noise he initially complained of was most likely due to using 4, INA's driving into a switched analogue selector, then into the single ADC input.

E

 

From Post #10 :" I use the USB 5V to power the micro/ circuitry and LC's, that should ensure good ADC tracking. " I am guessing that "LC's" is the load cells. And most bridge type load cells are arrangements of strain gages.

My experience has been to power the (full bridge) load cell from either the load cell power conections on the IA package, if it has them, or the same bipolar supply used to feed the IA. I also use the much less efficient linear regulated DC supply to power both the load cell bridge, and the IA. That avoids two sources of problems. A big advantage of using a higher voltage to excite the load cell bridge is a greater output voltage as the load is applied, and that benefits the signal to noise ratio quite a bit, and allows using less gain.

 

Hi everyone, please forgive my very delayed reply. I have had other very pressing deadlines at work which have kept me quite occupied.

I greatly appreciate the input given here and have learned a great deal from it. It appears that my poor performance was due to a number of factors. I now have the opportunity to give time to the revision of this design.

I was passing the output of the IAs through the multiplexer directly to the onboard ADC of the STM32F469NI. The voltage supply of the LCs and IAs were also being shared with other circuitry and was only regulated with a 7805 Vreg and filtering/decoupling capacitors. The STM32's ADC was indeed being supplied from a separate Vreg. The 7805 was also located on the indicator/controller and its output was passed to the junction box (with the IAs and multiplexor) via a shielded cable.

As per the advice given here regarding ratiometric measurement, I am considering using an integrated solution such as the TI ADS1234. I would then place the ADC and voltage regulation for the ADC/amplifier and LCs on the same junction box PCB, using the parallel configuration of the four load cells with signal trimming as described in the AN that Eric Gibbs shared. I would forego the use of a multiplexer.

Thank you once again for your input Eric, Bill and Denis. A have learned a great deal from your input as well as from stumping my head with this first iteration of the ADC frontend design. I will give updates on the new design as I progress.

Kind regards