PID Roasters Article

Discuss roast levels and profiles for espresso, equipment for roasting coffee.
User avatar
SAS
Supporter ❤
Posts: 322
Joined: 14 years ago

#1: Post by SAS »

Roasting

Stronghold, a South Korean startup has been manufacturing small sub-kilo electric roasters for a few years. Their software is fairly advanced, but not yet as good as an experienced human. The Alilio Bullet R1 is a similar story. As is the Iwaka sample roaster. These roasters are great, but like all that have come before, the software lets them down.

In recent news, neural networks and deep-learning software have become extremely good at recognizing images and even beating a world champion at the game Go. This kind of software requires immense amounts of data to "teach" the computer; orders of magnitude more than a human requires. Here are two excellent summaries by the Economist and the AlphaGo team team at Google.

PID's and simple algorithms are not going to solve the roasting problem. Instead, I'm hoping someone will use machine learning. Think of how many roast curves a company like Cropster have on file. All of them with gas changes and multiple temperature probes and sometimes even sensory data. Just imagine if they applied some machine learning to it; a roasting AI that has learnt from millions of batches on thousands of machines


Taken from Barista Hustle/Matt Perger Article via email; April 4, 2016.

Subscribe here http://www.baristahustle.com/
LMWDP #280
Running on fumes.

User avatar
FotonDrv
Supporter ♡
Posts: 3748
Joined: 11 years ago

#2: Post by FotonDrv »

I think the PID home roasting is in its infancy and such a database would be neat if it could be applied uniformly.

Sight and texture of the bean surface could be done but how do you get smell into the sensory data? Smoke, sure, but smell could prove very elusive.
That Light at the End of the Tunnel is actually a train