Busy, busy, busy...

Matt-the-Code-Warrior has been keeping me busy with changes to version 3 of the RoasTime software for our Aillio Bullet coffee roaster. This most recent release was a major step in promised features that have been years coming.

The model V2 roaster which I have contains infrared temperature measurement of the drum (during the pre-heating period before beans are added) and of the beans during the roast as they fly past the view port of the IR sensor. The roaster has an infrared sensor mounted in the front panel which is aimed at the beans as they tumble in the drum. Since the beans are being heated by the drum they absorb that heat which in turn causes them to radiate in the (not visible to the human eye) infrared spectrum. Anyone who has had their temperature taken as part of the current COVID pandemic has most likely had that measurement done using an infrared thermometer (preferred measurement because there's no contact).

The device has a space age ring to it but the reality was that the raw data was plagues with anomalies related to the sampling rate of the internal measuring device vs. the rotation speed of the drum. The signal appeared to be modulated with a low frequency sine wave (in the cycles per minute range) on one roast and then the next it was some sort of repeating bump every few seconds. The excursions weren't so large that the data was unusable, but trying to use the data for projecting forward temperatures at the current power settings required an advanced degree in mysticism or possibly alchemy.

So that job got put in the too-hard-to-do pile for a couple years. Matt managed to make sense of the projected temp data using "splines" (a software algorithm specifically for that purpose) which predict the temperature looking forward 4 minutes.

After that bit of hocus pocus he managed to calculate (well... approximate) the 1st derivative of the raw infrared data- it is used to estimate future temperature in F° or C° per minute. That sort of information is the basis of trying to manage power, drum and fan settings and is provided through a smoothing algorithm which is an adaptation of an exponential moving average often used in technical analysis of stock data. The displayed graph of that derivative, called Rate of Rise, is similar to the same measurement provided by the bean temperature probe. That measurement uses a conventional thermocouple but isn't encumbered by similar wild gyrations since it has metallic mass which has to be heated/cooled as the temperature of the beans changes the reading and the mass dampens the data swings.

So now after 2 years of operating the roaster I have software that displays everything available during a roast. The only question remaining is... how the hell do I use all this info! (Where's Sister Stanislaus when I need her?!)

So roasting-life going forward will be interesting as I (try to) develop an understanding of how this info interrelates in order to create a consistent roast.