Need help with counting scale project

[DaveEvans]

One way would be to continuously calculate a moving average and say that a stable value is achieved when the average does not change more than X in time period Y. You choose: the moving average method, X, and Y.

[DaveEvans]

@maillemaker

From what I am reading, this (connecting E- to ground) is not necessarily required.

Read the datasheet? Its block diagram and reference design schematic show E- connected to analog ground.

[Wawa]

@mancera1979

I can't see why any soldering is required...
No continuity between E- and ground is expected.

There is a batch of HX711 boards out there with a manufacturing defect.
Single-sided boards have been used, so several ground connections are missing.
That includes E- and the resistors for the 4.3volt load cell supply.
The faulty boards are transparent (with strong light).
Boards with bottom ground plane are fine.
Leo..

[mancera1979]

@Wawa

The faulty boards are transparent (with strong light).

I was not aware of that. All my boards are like this
No connection between GND and AGND

Thanks for the heads up

[maillemaker]

I don't know how to compute a moving average; I'll have to research that.

I tried putting in a 10 second delay after detecting weight, and then taking 10 readings and taking the average. You can see from the results that I'm still catching the upswing on the first set of readings.

But even when it "stabilizes" the averages still deviate by .34 grains.

This is a 1kg load cell (the lightest weight I easily found on Amazon). I wonder if these values are just too light for the cell?

I have ordered a higher-quality HX711 with filters and ground plane and also a 100 g load cell.

[DaveX]

How fast do you need to do the measurements -- that output looks like you are doing an average in 10s or 35ms, depending on the loading.

An Exponentially Weighted Moving Average is a good choice:

@maillemaker

const float alpha = 0.01; // EWMA smoothing constant per https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average
float ewma_weight_grains = 0;

weight_grains = CheckScale();
ewma_weight_grains +=  alpha * (weight_grains - ewma_weightGrains) 

Where you'd choose alpha between 0 and 1, with smaller numbers being smoother, and large numbers being more responsive. It is good to know enough about your measurements to know how fast things settle, how autocorrelated the samples are to choose the sampling period and the smoothing parameter. If you are sampling every 35ms, and the noise between samples is correlated, you should sample slower.

Since EWMA smoothers are cheap, you can use a couple at different smoothing parameters and determine "stable" by whether they are close or not.