Automated first crack detection and analysis - Page 4

Discuss roast levels and profiles for espresso, equipment for roasting coffee.
User avatar
drgary
Team HB
Posts: 14347
Joined: 14 years ago

#31: Post by drgary »

Following up for independent audio detection I looked at what's available for my iPhone in the App Store and found a free audio DB and spectrum app that can supplement my hearing. It's VisualAudio. Screens include some you see here, and it records. I would expect the display to jump in a certain frequency range when first crack is happening.



Gary
LMWDP#308

What I WOULD do for a good cup of coffee!

btreichel
Posts: 141
Joined: 8 years ago

#32: Post by btreichel »

10 frequency bands over that range is possibly too course.

User avatar
drgary
Team HB
Posts: 14347
Joined: 14 years ago

#33: Post by drgary »

The top screen is the equivalent, with finer delineation. I'm less skeptical. I think this will work. There will be an overall increase in decibels and it will show up more in one of the bands. I'll post more when I test it.
Gary
LMWDP#308

What I WOULD do for a good cup of coffee!

Kaldi23
Posts: 32
Joined: 6 years ago

#34: Post by Kaldi23 »

A word of caution that automatic detection of 1C using frequency analysis may not be as easy as we'd like it to be!

I'm a frequency geek (background in acoustics and vibration). Frequency analysis is very well-suited to tonal analysis b/c we'll see distinct peaks in the frequency plots that are easy to auto-detect and report, hence the proliferation of cheap auto-tuners for musical instruments. Broadband sounds and vibration can be more difficult to identify. Broadband sounds include both white noise and sharp impacts. I'd classify 1C as sharp impact-like. So its signature in the frequency domain will be spread out over a range of frequencies. And may be better suited to a frequency plot with large bands, like the second one shared above. OR it may be easier to detect in the time domain by applying a low pass filter, and looking for the spikes associated with the pops of 1C.

Not an easy problem, but probably not insurmountable either, as long as background noise is kept in check.

User avatar
drgary
Team HB
Posts: 14347
Joined: 14 years ago

#35: Post by drgary »

^ ^ ^ Good info.

It would be great to have a refined device that would plug into Artisan and have good discrimination in detecting first crack.

As a practical matter a roaster equipped with a bean temperature probe lets one know by its readings alone that we're in first crack territory. With my reduced hearing I can still usually hear first crack. Observing my phone display when first crack is starting and gets rolling may show me a changes in the visual display I'm able to correlate with temperature and sound. Beans that crack more quietly may show a similar signature on the phone and correlate with temperature.

It's another thing to include first crack detection in roaster automation, and I'm not that ambitious.
Gary
LMWDP#308

What I WOULD do for a good cup of coffee!

User avatar
Balthazar_B
Posts: 1724
Joined: 18 years ago

#36: Post by Balthazar_B »

drgary wrote: It's another thing to include first crack detection in roaster automation, and I'm not that ambitious.
No doubt if an average human (with decent hearing) can detect FC, it's practically certain that machine learning can enable automating its detection. The main questions would be:
  • Assuming this would currently require cloud processing, can it be quick/responsive enough?
  • Can it and Artisan (or other roasting solution) sufficiently communicate so as to drive automation on the roaster?
Just find the right coffee geek at Google and it could be yours sooner than you can imagine. :mrgreen:
- John

LMWDP # 577

Post Reply