Maillard Profiles Phase 3

[aaronbeach]

At this point calling it "Maillard Profiles" is a bit of a misnomer as I'm testing many variables in tandem. I did this test using Columbia Los Naranjos Heirloom Caturra

Summary:
In phase 1 I convinced myself I could test multiple variables systematically (maillard% and Roast Level). In phase 2 I generalized that to include total roast time. In phase 3 I generalized it further to the absolute times for each phase and the final drop temperature. While in phase 2 I identified the % of the roast in Maillard as the most important variable for predicting my enjoyment of the coffee (more than Dev time or total roast time), this experiment revealed what it was about the maillard % that was so important, and in particular how important "sweetness" is to my overall enjoyment.

Variables:
v1: Dry Time (time spent 20C - 170C) from 90-270 seconds
v2: Maillard Time (time spent 170C - 207C) from 65-240 seconds
v3: Dev Time (time spent > 207C) from 70-180 seconds
v4: Final Temp (Drop temperature) from 216.5-225.5 C

7 Tests:
To keep variables uncorrelated and evenly distributed I used a Halton Sequence to generate the values for the 7 tests.

Below you can see the correlations between the variables:

Screen Shot 2023-06-20 at 9.47.19 AM.png (208.17 KiB) Viewed 5381 times

I had to adjust the values slightly to meet certain limits or requirements. In particular, I required that RoR not increase, this meant no phase could proceed at a faster rate than its preceding phase. These are the tests I settled on and have attached them, some are clearly non-standard profiles:

Screen Shot 2023-06-20 at 9.51.17 AM.png (54.83 KiB) Viewed 5381 times

You can see these curves are all over the place:

Screen Shot 2023-06-20 at 9.52.55 AM.png (1.49 MiB) Viewed 5381 times

Here are the 7 tests:

Test.Profiles.zip

(7.25 KiB) Downloaded 306 times

Results:
The most interesting result was that in making the timing of the Dry phase and the Maillard phase independent of one another it revealed that the main driving factor in my preference for certain Maillard % was actually the Dry time, not the Maillard time itself (within a fixed roast time the Maillard % was controlling the Dry time). While the Maillard time does contribute to my perception of chocolate and sweetness, the strongest relationship of all was that too much dry time or too little killed the sweetness, below you can see my rating of "sweetness" and dry time in seconds:

Screen Shot 2023-06-20 at 9.57.05 AM.png (22.47 KiB) Viewed 5381 times

Interestingly, while total dev time and drop temp clearly impacted the darkness and flavor of the coffee (this Columbian was best around 2 min dev time finishing at 221C) I still enjoyed coffees across the range of drop temps. However if the dry time was too short (<100 sec) or too long (>220 sec), the coffee was unpalatable.

A 2nd order polynomial regression identified the following as optimal. I tested it this morning and both I and my wife agreed it was the best we'd had:

Opt.LosNaran3.6.kpro

(1.79 KiB) Downloaded 303 times

Screen Shot 2023-06-20 at 10.02.55 AM.png (253.07 KiB) Viewed 5381 times

Up Next:
I suspect my low fan profile may be a limiting factor as I've looked at other profiles (e.g., firestarter) that push the fan to the limit of the heater wattage at certain points. I am testing this profile with higher/lower fan and will report back. It seems the minimum is just keeping the beans moving and the maximum is when the heating power can't keep up. I will also be narrowing the search space and using a normal rather than uniform distribution across that range. This next phase may be the last of these tests. Once settled I will share:
1. a final set of test roasts
2. A google doc you can copy to calculate the "optimal" given 7 scores
3. Start a thread to track "optimal" roasts from tests

[aaronbeach]

The optimal profile was indeed better at a higher fan speed, more developed, darker, sweeter and more body.

Hi.LosNaran3.6.kpro

(1.77 KiB) Downloaded 339 times

On the next test I will have a fan speed variable, potentially a start and finish speed.

I noted that by the end of the roast an 8% increase in fan speed required 7% more power.
However, up to 110C the high fan required equal or less power and really wasn't a big difference until > 150C

[aaronbeach]

Phase 4 will explore fan profiles in relation to dev time and dev % staying within the sweet spot for Dry/Mai ratio from the last tests. This is what the test profiles look like (temp and fan profiles):

Screen Shot 2023-06-21 at 10.57.14 AM.png (1.42 MiB) Viewed 5332 times

[fnq]

Hi aaronbeach,

I might ask you revisit your assertions around drying time ( best window of time).
BUT
I would need to see your log files.

For me I would almost discard your definition of drying time ( 20 degrees to 170 degrees) and replace it with:

For your consideration only ......../ please feel free to ignore

In my experience many good roasts on the Kaffelogic have profile curves and roast logs that converge at around 150degrees in a time period of between 90 seconds and no later than 2 minutes. ( so it can actually be a moving point in differing profiles but all can still get a tick of approval)


I would add qualifications that it is not just how synchronised the roast log and the profile lines are above this starting point ( basically preheat and putting heat into the bean mass) BUT crucially the way in which the log parameters are kept in step with the profile. Too much preheat can lead to hot and fast roasts, which are not inherently good or bad BUT need to be wary of heavy handed PID action or power throttling to tame the rate of rise ( across FC and into development)
Not enough energy ( i'll call it momentum) and the fear is the roast crucially stalls.

So for me , when i look at logs, most are just a cursory check to see rough alignment is okay UNLESS the convergence isnt there near that time AND then i look for power flatlining or throttling and or make some assumptions about PID heavy handedness trying to keep the log file on the rails and not overshoot.


I tip my hat to people like Benjamin ( Raost files) and Steve ( MD and other profiles) whose cleverness I basically steal and then just add tiny simple tweaks to adapt to my setting. If you go through some of the Steve threads , he actually produces 2 near identical Roast charts but explains the world of difference in final results.

Here is a log file where convergence is near 150 degrees and 90second and the tracking is well underway at about 180 degrees at 2 minutes.
I could probably slightly weaken the preheat parameters in my adaption of this profile, but i am too lazy to. Note that the heavy handed adjustments ( by the machine ) are all before the 2 minute mark and then much finer adjustments are required for the rest of the profile ,,,, including how the power ( in my case 230volts) copes without flatlining and indeed feeds in nicely to help keep that momentum going

Anyway good luck on your continued quest

[aaronbeach]

That sounds like what you've observed agrees with what I'm seeing. You said good roasts reach 150 between 90-120 seconds. It takes about 20-40 more seconds to get to 170 - so the time I would see at 170 would be 110 - 160 seconds. While I did see good coffees a bit slower than that, most of the range identified above aligns with that (110-220).

Couple of questions related:

1. Is 150C a better benchmark than 170, if so why?

2. Why are you trying to track by 180? (I try to adjust my preheat to track as early as possible)

I attached the related test roast logs:

log0044.klog

(51.54 KiB) Downloaded 304 times

log0045.klog

(80.93 KiB) Downloaded 303 times

log0046.klog

(61.3 KiB) Downloaded 300 times

log0047.klog

(55.16 KiB) Downloaded 320 times

log0048.klog

(90.36 KiB) Downloaded 283 times

log0049.klog

(74.47 KiB) Downloaded 282 times

log0050.klog

(62.32 KiB) Downloaded 287 times

Opt:

log0051.klog

(81.57 KiB) Downloaded 290 times

[fnq]

Hi Again

In reponse to questions. I quickly gauge if the roast log is quickly synching to the profile using the approximation of 150 degrees between 90 seconds and 2 minutes. 150 is no better than any other,,figure ,, but in my mind what happens below the 2 minute mark is almost irrelevant ( similar to a preheat in a drum roaster..... so i was questioning whether using 20 degees ( clearly in that warm up/ preheat) was a good figure to use.
In my analogy there is no set point so long as both the parameters are met. (temp and time) ( So I am happy to look at that metric on a single curve/ log BUT i don't want to necessarily compare absolute figures with other curves.

My tracking by 180 degree comment was a bit of a misrepresentation , sorry . I was trying to track at closer to 150 but this log came in a bit heavy and had i been worried about the outcome, or had I been less lazy, i would have tweaked the profile a little bit just to better mimic my rhetoric. I use this profile on a number of beans and these settings seem to work well in the cup for the differing beans so i kept it.

On you logs , I am nicely surprised at how well your log tracks your profile - Particularly on your steeper /hotter start offs). If they drink well, then great but for example if log 49 drank quite muted - i would look at HOW your PID and power application is reigning in the roast and slowing the development after FC.

Two things that I might suggest>
Try and log FC on important roasts, that way you may also find trends around your roast flavours and development time ( and the machine keeps calculations for you)

AND

use the Kaffelogic Studio TOOLs - Compare files functions with some roasts that you like the taste of or are known good producers.

So for example for the Ethiopian beans you used in Phase 2
maybe compare to Core 1500-2000 REST file and a Raost v 5 ? to see the similarities and differences.

With so much knowledge and experience out there ( roasting in general and roasting using the K)L - I think it wise to use that body of knowledge to inform your curves.
EXample - looking at the profiles for low grown less dense beans and seeing the differences for high grown SHB gives some good background underpinning knowledge.

Anyway , so long as you are having fun and drinking tasty cups , it doesn't matter.