Roast Vision - a new and affordable roast level sensor for $299 - Page 8

Read about current and future offerings from the site sponsors.

#71: Post by beallpatrick »

My wife (a QC chemist) is curious as to why the device never needs calibration since that's sort of blasphemy in her lab.

Is there something particular to the tech/method you're using to measure roast level that ensures it isn't going to drift over time, and we'd get the same readings 4 years from now that we would today?

Nice work, btw! Appreciate the transparency in components you're using as well for those inclined to DIY.

espressovision (original poster)

#72: Post by espressovision (original poster) »

Hi Patrick,

Thanks! And great question! I previously answered a similar question over on a Reddit post here, but I'll explain it here a bit more thoroughly.

The main factor for drift in devices like this would be electronic wear due to aging over time, which is typically in proportion to the intensity and duration of the power being applied to those circuits over time. For the Roast Vision specifically, the LED is set to a very low pulse amplitude (regulated on the breakout board), and the LED and sensor are only powered for a similarly VERY short period of time for each measurement cycle.

In fact, the animation between ----oooo----oooo---- only demonstrates that the sensor has detected coffee is present (to inform the user), and at the end of that animation, it takes the measurement (multiple for averaging) that is used for the output. The Roast Vision only takes measurements at the end of that ----oooo----oooo---- animation and before it prints out the instructions, as that is when it looks for the presence of coffee, and triggers the "reading" animation.

Another important factor is the resolution of the output scale in comparison with the resolution of the sensor itself. The Roast Vision output ranges from 0 to 35, whereas the sensor raw values typically range from ~70,000 to ~230,000. While we chose the 0 to 35 scale for specific reasons (symmetric, intuitive, not overwhelming), this also contributes to the long term stability, as even if there is a very small drift in the raw sensor values over a long period of time, this wouldn't be reflected in the 0 to 35 scaled output for an even longer time.

The photodetector could lose sensitivity due to external light exposure over time, which is why we recommend to avoid direct sunlight on the device. However, acrylic sheets (not directly tested here, so this is based on general theory) block UV-C light, which is often the cause of most sensor damage (cameras for example), as UV light has a much higher photon energy than visible or near infrared light.

For a future iteration of the Roast Vision, we're considering adding in a calibration mode just for peace of mind for the users that want to be able to confirm at any time that the unit is in proper calibration. While not necessary, this could be added, but adds some complexity regarding sourcing the calibration standard (often a matte grey disk for NIR sensors) and ensuring a long term supply that wouldn't change from the producer.

Sorry for the delayed reply as well!