Three-wire load cell and HX711 weight compensation

Question

In my project I use a 3-wire load cell (like this (3-wire 50kg), but mine were purchased from eBay) and an HX711 for weighing.

My wiring is like this (but I use an Arduino Mini Pro 3.3 V):

Apparently it should be a full-bridge circuit where the temperature is fully compensated and temperature changes are minimal. However, my last-day charts are like this (loaded statically with 18 kg):

The graphs show that the weight varies with temperature. Therefore, I put all temperatures and changes from the calibration value into a graph, where the X-axis is the temperature and the Y-axis is the change from the known value (18 kg).

This chart showed me that the weight is probably not ONLY dependent on temperature as the first charts might seem to indicate.

I'm really frustrated about it, because I've tried everything, but the result is always the same, even though this wiring should compensate for itself.

Has anyone encountered anything like this?

Edit 1:

Answer 1

Why so many fans of HX711 module, here? You have also the AD7190 24Bit ADC module on Ali. If the weigh scale has three wires, than it can't be full bridge, aka Wheatstone, these have four or six wires. Also, you can't do a Wheatstone temperature compensated bridge from two pairs of these sensors, since one diagonal of the bridge is orientated to measure elongation due to applied force, the other diagonal pair gauges are orientated 90 degrees off, so they measure the elongation due to temperature change. The compensation is achieved because the material elongates isotropic in all directions VS. temperature, so all four gauges sense this, while only two of them sense also the weight.

EDIT:

You chart gives the visual proof of relationship between weight and temperature, however the analytical proof would be a cross correlation between those two values. The other discrepancy could be noise, voltage fluctuation ...etc. Note that cross-correlation is a frequency analysis, no DC drift will be included.

EDIT 2:

IO, you also did connect wrong your cells. If your schematics represent forur cells connected in the bridge, then it's wrong.

^{simulate this circuit – Schematic created using CircuitLab}

To connect four cells, then connect the other two each parallel to these two, with same wires.

^{simulate this circuit}

Answer 2

Replace the load cell with a pair of resistors and see if you still see the drift. Then you'll know if it has anything to do with the load cell. My hunch is that the measurement circuit is temperature-sensitive.

It'd also help to get a 6.5 digit voltmeter and measure the cell supply voltage over a day, and then the differential input voltage on another day.

I have no experience with HX711 and it wouldn't be my first choice for this application. Any decent 24+-bit ADC with differential input would work, as long as it can do ratiometric measurements, i.e. has an external reference input or is referenced to AVDD.

I haven't worked with 1kΩ strain gages, only with lower resistance types, so I'm not sure how the one you got will behave vs. temperature.