we always had planned to ramp up such a setup you apparently did right now. Very cool, thanks! Actually, we talked with @tiz (btw. he is also a Martin) about the very same thing you are showing us right now, a DIY climate chamber – would it actually be the correct phrase for it?
Good luck to you and all others who might join here to draw further conclusions from that and to improve our shared knowledge to improve the respective implementations. Thanks again for going up there and keep up the spirit!
Unfortunately, I’m not that much of a hardware guy to be able to give reasonable recommendations about that topic. Saying that, I’d like to encourage all others who feel ready to contribute valuable things here.
P.S.: Feel free to start an appropriate thread named “DIY climate chamber” or similar, where specific details about this endeavor could be discussed related to the temperate compensation thing we are coming from. [edit: @clemens did already, thanks!]
Very cool, ahm, hot! :-) There are a lot of assumptions regarding load cells and temperature. So it would be really nice to have some facts. What is the influence of
the load cell
the electronic part
For testing it would be also nice to have the cable outside the climate chamber or have load cell and electronics at a constant temperature but heat up the cable e.g. with a bit hot air from one side only to simulate sun radiation.
Would it be possible to justify your scale first! A difference from 25 000 to 50 000 can mean 50 g so neglectable or 5 000 g so a significant change! Without this calibration the raw values are a bit useless or at least less meaningfull. :-)
thanks for the cheering words, it is nice to see that my work is also interesting to others and may help you to build better scales. I will open an thread named “DIY climate chamber”, the setup is very simple.
First of all, i will also do all measurements with a ADS1232. to see the difference between the HX711 and the ADS1232. But it will take a while, i just ordered the parts, and delivery will take 6-8 weeks :)
I will do the test with a calibrated weight of 20kg. Then we will see how much the delta in raw-values means delta in weight. With both datasets, we can justify the scale offline.
I keep you informed.
I have the first data with loaded cell – the heating phase is still in progress, I have data up to +15°C
A hysteresis-effect is clearly visible (upper right plot): The cell itself or more likely the cooling/heating gradient is too big. Temperature-Sensor, load-cell and electronics doesn’t have the same temperature. Less heating power will slow the heating gradient, but I can’t change the cooling part. But since the correlation is very linear, probably one steady state point at -20°C will be enough. After the heating phase is done (40°C is reached) it turn off the heating and let the system cool down by its own. This is a very slow gradient, let’s see what’s happening.
I played with the numbers, here my thoughts,results.
3 possible ways to get from the raw-data to the actual weight in grams.
(1) Classic way data is transformed with one scalar offset and one ratio value.
(2) like (1), but with a correction-term
(3) both, the offset and the ratio is a linear function of the temperatur
On the left side the cell is unloaded, right side loaded with 20000g.
first line -> raw value vs. temperatur
second line -> weight calculated with formula (1)
third line -> weight calculated with formula (2)
fourth line -> weight calculated with formula (3)
Comparing (2) and (3), shows, formula (3) is the better way.
I also tried a polynom third-order, but this doesn’t get better results. So keep it as simple as possible, a linear function is doing the job quite good.
regarding the question : Which part is causing the temperature dependencies?
I wasn’t expecting that, bit it is clearly the load cell! A bit disapointing, since i thought the BOSCHE Cells are better. cell is unloaded, i have 3 more sets of scales one by one is going in the chamber the next days.
In the previous diagram the ordinate shows the raw value of the HX711.
Here in grams :
I just have measurements with the cell unloaded, which is not optimal (since there should always be a minimum load on the cell), but still, you clearly see :
The temperature-dependency is not coming from the HX711!
I am disappointed, I thought Bosche cells are better.
I probably will do the measurments also with loaded cell, but i don’t know when, since i don’t have a lot of time right now.
I asked Bosche, but i just got the test specs and the norm (Metrological regulation for load cells) and i don’t understand the reading :) r060-e00(1).pdf (1006,3 KB)
No, look at Bosche’s H30A datasheet, they clearly state, that they don’t require a minimum load (Mindestvorlast (E min) = 0 ), so for this load cell it is a legitimate load case. Second, this is how you can measure the temperature drift of zero (aka offset) - also needed as Bosche unfortunately does not state these drifts but instead only refers to the OIML… 8(
Better than what?
Your results are good: over a temperature range of about 58 K (-20…+38°C) the scale shows an abs. offset change of about 520 grams (could be shown as a differential error of ±260g). This is 9g / K error (for comparision: @thias ’ H30A shows about 8,7g / K). These results were achieved while stressing the cell with an ever wider temperature range than specified as compensated range (to -20 rather than to -10°C) - and no additional non-linearity has been observed. This also remains a good result.
What is the Emax for this measured H30A cell, 150 or 200kg ?
It is a 100kg Cell, so the verification interval is 100kg/3000=33grams.
I am not sure if interpret that right, but in the linked PDF
22.214.171.124 Temperature limits
Excluding temperature effects on minimum dead load
output, the load cell shall perform within the limits of
error in 5.1.1 over the temperature range of – 10 °C to
40 °C, unless otherwise specified as in 126.96.36.199 below.
The error is defined in 5.1.1, which is about one verification interval (Table 5 Maximum permissible errors (mpe) on pattern evaluation), so max 33grams over the temperature rang of -10 to +40°C.
This is my interpretation, i am not sure if i understand that correct.
Now, all 4 scales are calibrated, the difference between the setups is not negligible.
So without individual temperature compensation - at least for me - the accuracy with the H40A is not sufficient!
first plot : without temperature compensation offset and ratio calculated @ 15°C
second plot : with temp. comp. offset(T) and ratio (T)
third plot : same as second, but y-axes is zoomed to see the actual error
Sorry to bother you again, @zmaier. ;-) You have in the first graph 4 lines, purple, yellow, orange and blue, this are the same setups? All with a HX711 and a Bosche H40?? So why has the purple one nearly no temperature effect and the yellow one such a big? I fear I do not understand the difference of the 4 “settings”. May you clarify this, thanks!
Is the only difference that you are using different load cells (but all H40 from Bosche with the same max. weight?).
Can you imagine what was special / other on the purple setup? Other power source, flux cleaned vs. not cleaned? screw terminals vs. cable soldered? Cut / shorted cable? Really no difference??! The purple setup is great. So in case we can find “the error” – and it is not inside the Bosche H40 this would be great! ;-) Many thanks again for the effort and time you spent!
I have absolutely no idea.
I started the thread below, because i have seen those huge differences between the setups.
To figure out why, “built” the test-chamber.
All setups have the same build-up.
Flux is cleaned
HX 711 and Load Cell is soldered with the original cable (not shortend)
Power-Supply is the same
I have no idea why?
It seems the Load Cells are really that “weak”, see my posting where I compared the temperature behavior just of the cell and just the HX711. It clearly comes from the cell itself.
What are your experiences? Should i contact Bosche with this issue?