mortenfyhn / coffee-scales Goto Github PK

For adafruit feather with battery

The weight reading is very unstable.

Stupid and pointless but fun: Estimate the bias based on something like "if rate of change is very low assume it should be zero"

Adafruit feathers seem nice because they have the charging circuit integrated:

• Feather 328P: same mcu as the old arduino
• Feather M0: more power and a bit more proto space

• Decide which to order --> M0 lol
• Order comp
• Order battery

I'll just use a photoresistor. Keep it simple. With the random ldr I have, I measure approx

• 1 ohm (raw 633) outside on a bright but overcast day (should give 100 % brightness)
• 2k ohm (raw 540) in a bright room (should give 100 % brightness)
• 10k (raw 350) ohm in a medium bright room (should probably give less than 100 % brightness)
• above 2M ohm (raw 36) in a really dark room with no lights on (should give minimum brightness)

So it seems that with a circuit like [3v3]---[ldr]---[Vmeas]---[10K]---[gnd] and the ldr I have, a good starting point is: max brightness above Vmeas = 2.44 V (500 raw analog with Vref = 5v) and minimum brightness below Vmeas = 0.73 V (150 raw analog).

• Get suitable photoresistor.
• Figure out suitable fixed series resistor. --> 10k
• Test separately (write a small program that outputs 0-100 depending on ambient light).
• Integrate with everything else.

Using https://platformio.org/lib/show/6582/SevenSegmentTM1637 now, but I think https://platformio.org/lib/show/258/TM1637 is better. It's certainly smaller.

• Make it run again
• Make it pass again
• Update badge

Seems to happen after it's been used with a particularly heavy item. The weight starts to alternate between 1156 and 1565 at something like 3 Hz. The timer still runs fine.

The old display modules are too big. Gonna develop new ones that

• have a PCB that is the same height as the 7-segment itself,
• are the same size (0.56") or smaller, and
• probably use the HT16K33 instead of TM1637 (HT seems to require fewer external components).

• Make schematic.
• Make layout.
• Order from osh park or something.
• Order 7-segments.
• Order HT16K33s (can get them here)
• Assemble and verify.

Maybe it's better if I always set the hysteresis limits to the threshold between two values (say if value is around 0.1, then thresholds are 0.05 to 0.15, and so on).

Should be possible to make it "prepare for tare" so that when hitting the tare button the reaction is immediate. Consider ideas from #15, too.

It feels better somehow. Removes the need for non-const globals.

Maybe a macro to toggle it. Should output valid csv to serial.

Still going for sandwich-based design, probably with shifted hexagons. Need to figure out dimensions that work with two 7-segments and the Adafruit Feather.

• 4-digit 7-segment module: 19 * 50 mm.
• Adafruit Feather: 51 * 23 mm.

When I do a read call, does it block?

Displays 0.0 instead of 1.0 grams.

Clang tidy probably.

• Run and fix current clang tidy errors.
• Integrate with CI.

😢

Stopped working after March 17.

Here's the last good build: https://github.com/mortenfyhn/coffee-scales/actions/runs/659453634
And the first bad build: https://github.com/mortenfyhn/coffee-scales/actions/runs/662921725

Which seems to indicate that v5.1.1 broke my build: https://github.com/platformio/platformio-core/releases/tag/v5.1.1

Update: Yep, v5.1.0 works fine.

I'm guessing it's caused by platformio/platformio-core@7c271c8

• Program it with platformio with some dummy program
• Program it with the current program (patched if necessary)

The current taring just reads a raw value n times and sets the offset equal to the average of the readings. What if I instead take a single reading and set the offset repeatedly until for instance the change in offset is very small. Then taring a) might go faster and b) won't fail if it happens while the applied weight changes.

Seems that the HX711 outputs slightly wrong data first. Maybe we can adjust the tare function to wait for it to settle.

Add a physical tare button, and implement it in software. Should probably just be a simple momentary button. Even I can understand the circuit:

• Choose a button that fits and feels nice.
• Integrate it.

Gotta decide what to include and what to skip.

• Need to get a switch (a nice toggle)
• Figure out layout
• Need larger board?
• Solder it up!

Probably the easiest issue yet

Should run at midnight

The measurement sometimes jitters between adjacent tenth of gram values. Need better measurement filtering to avoid this.

A good opportunity to try out doctest

• Get any test example running
• Share dependency specification

Often ends up at something like -0.2 grams after taring. Should fix it so it hits 0.0.

Use the logged data I have.

Spec / goal:

• No jitter during steady state
• Very quick response when adding weight
• No overshoot

Maybe bias estimation? In the log data, after finishing the pour, the data falls from 200.45 to 198.98 over approx 90 seconds. But maybe that's just evaporation!?

• Take another log, maybe 10 minutes, with nothing on the scales.
• Re-implement the current filtering and run on log data to assess
• Implement in such a way that I can use the same code for filtering on the arduino as for assessing?

The scales cannot currently display negative numbers on the 7-segment display. Fix it.