Load Cell/ Sensor Recommendations for Weight Shift Detection

[am1]

Hi all,

I have a project for a medical device that is to detect when weight shift occurs from right to left foot and from heel to toe of each foot. I only need relative load (e.g. right foot bearing 70% load, left foot 30%) so the high precision of a load cell or strain gauge is not needed, which is why I was leaning towards FSR sensors. Off the shelf have very poor repeatability from sensor to sensor and require calibration. The most important aspect for the selected sensor to have is that is is ready to integrate to our PLC system without needing individual in-house calibration and signal conditioning.

I have gotten quotes for some custom FSRs with PCB boards for signal condition included and ready for integration with my +/-10V analog input module for my PLC, however these seem pretty pricey. Does anyone know of alternatives or where to find low-cost load cells? The initial research I have done seems pricey also and I don't need the precision or complexity of these sensors, but I'm curious if this will be cheaper. The issue I see with any sensor is that they will require calibration for repeatability as I will be using 4 per machine, unless these load cells have relatively high repeatability from one sensor to another right out of the box (FSRs can have up to 40% variation from one bath to another which is why I would need signal conditioning and calibration). Any advice would be appreciated.

[moffy]

I would advise against using the FSR sensor because of their lack of stability and matching. What you save in sensor cost you will more than spend in chasing stability and accuracy which you might never achieve. Go with something like, just an example: https://www.sparkfun.com/products/14282
a  200kg load cell and just accept the sensor cost, far more stable, repeatable and comparable, you will end up with a better solution. I don't know if that sensor is suitable but something of the sort. As they say, we don't need to reinvent the wheel.

[matb]

Totally agree with the previous reply.
To detect heel to toe shift you will need quite a good signal which FSR are not going to give you. Too much drift.

You could go for an analog sensor then use PLC weighing modules or directly buy a 10V sensor.

where to find low-cost load cells?

Sellers : Scaime, Dini Argeo, HBM, Flintec, Arpege Master K, Mettler Toledo, Precia Molen

These are industry oriented, maybe they have cheap sensors that would meet your requirements.


Just be aware that is you go for 4 sensors the total weight of the person is divided by 4 on each loadcell. So you would not need a 200kg sensor x4, maybe a 100kg would be enought, just try to estimate the structural weight.
You will also need some filters as depending on the physical condition of the person they might not be able to stay stable enought.

[Doctorandus_P]

Industrial type load cells will be relatively expensive, but the type used on person weighing scales are very cheap and easily available. They cost less then EUR1 and usually just have a half bride configuration.

[Kleinstein]

Even if the load is ideally shared between the 4 sensors, the worst case load can still be mostly on one sensor. In addition to the weight there can be some dynamic load, accelerating the body when taking a step off the sensors.  So a 200 kg range can well make sense.

[Randy222]

Hi all,

I have a project for a medical device that is to detect when weight shift occurs from right to left foot and from heel to toe of each foot. I only need relative load (e.g. right foot bearing 70% load, left foot 30%) so the high precision of a load cell or strain gauge is not needed, which is why I was leaning towards FSR sensors. Off the shelf have very poor repeatability from sensor to sensor and require calibration. The most important aspect for the selected sensor to have is that is is ready to integrate to our PLC system without needing individual in-house calibration and signal conditioning.

You stated the requirement twice, but contradictory.

1) weight shift from heel to toe of each foot
2) I only need relative load (e.g. right foot bearing 70% load, left foot 30%)

#1 and #2 needs perhaps 4 sensors, one at each corner of a square (square easier for math).
To get the results of #1 and #2 you apply some math to the sensor data in realtime.

There have decent specs, easy to implement, and low cost. Largest one is 200lb, but they have other models too.
https://www.digikey.com/en/products/detail/te-connectivity-measurement-specialties/FX292X-100A-0010-L/11205656

[am1]

Hi all,

Thank you all for your response. I initially eliminated load cells as I will only need relative load and not a high accuracy lb/force readout and since the FSRs will be simple to integrate mechanically.
To clarify my system requirements, I need to show % load read out on each foot, and have a simple indicator for whether the load is in the heels or toes. The device has elliptical-like footplates, so the sensors will be going on a flat form surface, so I plan to put 2 sensors on each footplate (one near the back of each footplate for heel recognition and one towards the front for toe recognition) with 4 total sensors per machine.

I have made a rudimentary "mock system" using off the shelf Flexiforce A502 sensors. It works relatively well with detecting heel/toe loads, and a relatively reasonable % load readout for each foot. These sensors are potentially produced from the same batch, which is a contributing factor with repeatability, but it makes me consider if a reliable system can be made with these sensors.

My biggest concern is that I am part of a small operation, so it will be difficult for me to do in-house calibration for each sensor for every production device. So I want to avoid individual circuitry and/or calibration factors for each sensor in software for version control. The more repeatable and ready-to-integrate the sensors are more my PLC, the better. I spoke with some custom FSR companies that include PCBs with signal conditioning so it is ready to integrate with my +/- 10V analog input module, however these were quite pricey. Another company I spoke with said it could meet all our specifications and will have high sensor-to-sensor output repeatability upon us receiving it, without us being additional circuitry/signal conditioning, which would be great, but slightly suspicious, haha. But I wanted to get some samples from them for testing. 

These low-cost 200 kg load cells would be the preferred option for us, but it brings me back to my concern of being unsure what the off-the-shelf repeatability/variation will be from one sensor to another, because avoiding additional circuitry from sensor to PLC controller is important with our limited resources and space available. I am also concerned about exceeding rated load (within a reasonable range, say ~100-200lbs), with overloading load cells damage it? I'm sure there is some safety factor built in, but that is another thing for me to consider.

Does anyone have thoughts on this? Do these custom FSRs seem like an option worth considering? And if not, how can I go about off-the-shelf load cells to ensure repeatable outputs? I know custom and pre-calibrated load cells can get pricey... but I will start looking into some of the options you all recommended, Thanks!

[Randy222]

So maybe some flat piezo sensors since they are flat and you can wedge them between two plates?

As for matching sensors, I think best practices would be to have a calibration routine.

[Kleinstein]

A test calibration for a relative change would be relatively simple with weights. For the heel / toe part there is anyway a limited accuracy as feets have different size and the position of the foot would matter.

The flexi force sensors are rather low accuracy amd high drift, so they may need a relatively frequent recalibration and check of the temperature. It may still need extra effort to couple the force to the sensor, e.g. spread it out over the area and avoid damage from local peaks. So the sensor look like ready made, but it still needs a machincal interface. This applies to essentially all sensors to some degree: one can hardy directly step on the sensor. An uneven ground could be an issue too.

The shelf strain gages should be more accurate even without an individual calibration other than maybe the initial zero. The thickness of the sensor can be an issue - many of the strain gages would add quite some height.

Piezo sensors may work, but they are less standard parts and only for the dynamic part. Over time the reading drifts and one needs a periodic zero. The advantage is very good dynamics (low noise with high maximum load) over short time.

If the mechanics has to be done anyway, one could also consider capacitive distance sensing and using the deflection of the mechanical part. This would however be more like a DIY solution - not many ready made sensor. There are capacitive measurement chips (separare "capacitance sensing ADCs" or inside a PIC µC).

[am1]

Thanks for the replies. The Flexiforce ones are just to mock up a theoretical FSR system, due to the poor part-to-part repeatability and drift I will avoid those. The custom FSR vendors I have spoken, however,  do not seem to have concern about their sensor's repeatability, drift, etc.

My concern/issue with in-house calibration is storing the calibration factors/ corrected value. I can do a simple 2 point calibration, but storing it in a microprocessor or saving all 4 values for each sensor for each production device in the software is my challenge. For easy version control purposes, we try to deploy the same software to every machine, so that is why I prefer to use a pre-calibrated sensor and hit a road-block when trying to consider other non-custom, calibrated sensors. The sensors our device currently uses do not require calibration out of the box, so this is new for me, so if anyone has advice on this I would greatly appreciate it.

Ill look into button load cells and piezo sensors options some more as well. This device will be used on patients with various neurological conditions so I cannot guarantee the loading would be equal, so considering the worst case scenario of how the sensors are loaded, I would want each to be rated for the 330 lbs, so they get a bit pricey at that range.

[Randy222]

A test calibration for a relative change would be relatively simple with weights. For the heel / toe part there is anyway a limited accuracy as feets have different size and the position of the foot would matter.

I am not so sure it would be trivial. More than one sensor under the pad? One sensor only, then what if the weight is not dead center? Calibration routine yes, but it can get tricky with certain setups if the goal is to cal each sensor in the system.

I thougt I read somewhere long ago perhaps, some force pads that are like a sheet of sensors that form a matrix, from there software reads all the sensory to determine total weight and force point data. Like what they use in the shoe places where you step on and the display shows your foot in color gradient of force, mapping your pressure points. The thing being stepped on is some sort of mat that has sensory in it.

[PCB.Wiz]

My concern/issue with in-house calibration is storing the calibration factors/ corrected value. I can do a simple 2 point calibration, but storing it in a microprocessor or saving all 4 values for each sensor for each production device in the software is my challenge. For easy version control purposes, we try to deploy the same software to every machine, so that is why I prefer to use a pre-calibrated sensor and hit a road-block when trying to consider other non-custom, calibrated sensors. The sensors our device currently uses do not require calibration out of the box, so this is new for me, so if anyone has advice on this I would greatly appreciate it.

If you cannot find a good enough 'better sensor',  an admin alternative would be to include an EEPROM, or even simpler to manage, a small MCU with each sensor.
Parts like EFM8BB52 have 12b ADC, 10b DAC in a 3x3mm  package, and enough flash to be able to self calibrate in a bench fixture.

[PCB.Wiz]

... The device has elliptical-like footplates, so the sensors will be going on a flat form surface, so I plan to put 2 sensors on each footplate (one near the back of each footplate for heel recognition and one towards the front for toe recognition) with 4 total sensors per machine.

I have made a rudimentary "mock system" using off the shelf Flexiforce A502 sensors. It works relatively well with detecting heel/toe loads, and a relatively reasonable % load readout for each foot. These sensors are potentially produced from the same batch, which is a contributing factor with repeatability, but it makes me consider if a reliable system can be made with these sensors.

Is this static, or dynamic use ?
You could do a rudimentary auto-calibrate knowing the total weight, but you would hope the sensors drift the same over time/temperature, and do not diverge randomly.

addit: I presume you have already seen this ?

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9331244/

gives

Linearity             ±3% of full scale            Line drawn from 0 to 50% load
Hysteresis           <4.5% of the full scale   Conditioned sensor, 80% of full force applied
Repeatability   <±2.5%                           Conditioned sensor, 80% of full force applied
Drift               <5% per logarithmic time scale   Constant load of 111 N (25 lb)

[matb]

I am also concerned about exceeding rated load (within a reasonable range, say ~100-200lbs), with overloading load cells damage it? I'm sure there is some safety factor built in, but that is another thing for me to consider.

That question should be in the datasheet of the sensor. From the ones I use (which are industrial ones) it's typically 150% to 200% of the full scale. You could ask the sellers for that information.

[am1]

I am not so sure it would be trivial. More than one sensor under the pad? One sensor only, then what if the weight is not dead center? Calibration routine yes, but it can get tricky with certain setups if the goal is to cal each sensor in the system.

I thougt I read somewhere long ago perhaps, some force pads that are like a sheet of sensors that form a matrix, from there software reads all the sensory to determine total weight and force point data. Like what they use in the show places where you step on and the display shows your foot in color gradient of force, mapping your pressure points. The thing being stepped on is some sort of mat that has sensory in it.

Yes, I have been looking into matrix FSRs that are commonly used in those types of applications. I don't necessarily need the pressure mapping for our application as a simpler UI would be preferred, but the custom sensors would be a matrix sensor.

Is this static, or dynamic use ?
You could do a rudimentary auto-calibrate knowing the total weight, but you would hope the sensors drift the same over time/temperature, and do not diverge randomly.

addit: I presume you have already seen this ?

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9331244/

gives

Linearity             ±3% of full scale            Line drawn from 0 to 50% load
Hysteresis           <4.5% of the full scale   Conditioned sensor, 80% of full force applied
Repeatability   <±2.5%                           Conditioned sensor, 80% of full force applied
Drift               <5% per logarithmic time scale   Constant load of 111 N (25 lb)

It would be a dynamic application. The patients would have their feet strapped down to footplates, which are on a free hinge to allow ankle movement as the machine carries them through a gait-like movement pattern (similar to elliptical).

One of the manufacturers I am speaking to for custom FSRs said they can meet the specs you listen above and said the following regarding drift and life cycle: 

" Drift is affected by the following factors: (1) the amount of pressure; (2) the duration of continuous pressure; (3) changes in temperature environment
This project is to compare the difference in force between two sensors, so even if there is drift, the two sensors change simultaneously, so the function will not be affected too much.
The service life is 1-3 years. "

If you cannot find a good enough 'better sensor',  an admin alternative would be to include an EEPROM, or even simpler to manage, a small MCU with each sensor.
Parts like EFM8BB52 have 12b ADC, 10b DAC in a 3x3mm  package, and enough flash to be able to self calibrate in a bench fixture.

I personally have not worked on any embedded systems using MCU and PLCs, so planning the circuitry while avoiding custom or in-house PCBs is what I have been trying to consider when thinking about load cells/strain gauge application. Thank you for these recommendations, I will look into them some more.

Any other thoughts or advice on this would be greatly appreciated!