IKAWA Home - profiles - Page 7

#61: Post by **mathof** » January 17th, 2022, 4:27 pm

mgrayson wrote:Great to see these experiments!

Pardon my unfamiliarity with these plots, but where do the numbers (I assume they are cutoffs to the AUC) 167.4 and 160.3 come from? The relevant numbers look like 150 and 148.

Actually, that puzzles me too. The numbers represent the moment I registered the end of the drying phase. They always appear after the roast has ended lower on the graph than they seemingly should. I assume it is something wrong with my settings but it has never bothered me enough to investigate.

#62: Post by **Iowa_Boy** » January 17th, 2022, 5:12 pm

mathof wrote:I have taken up GDM528's invitation to use his stepped and RoR profiles. I have the previous 50g Home roaster, tricked out with a thermocouple placed in much the same position as his and run through an Amprobe TMD-56 to a MacBook Pro running Artisan. Here are the results of today's roasting (I used Guatemala Poaquil beans):

image

image

As you can see, the AUC of the two roasts differs markedly, with the stepped profile measuring 240C*min and the RoR profile 313*C min. This is reflected in the colours of the roasted beans. The former measures 78 on my Tonino (interpreted by the Tonino as City) and the RoR a stage darker at 70 (interpreted as Dark). With these measured differences, they will certainly taste different.

This is so interesting - I am really impressed with GDM528 and your data!
Are you using a bare wire thermocouple as well or something different? If you are willing, could you show a picture of your placement as well?
I am getting a lot of noise on my Artisan readings, so could definitely do a better job with the thermocouple and/or placement.
Also, just curious what you using for your curve smoothing settings?

#63: Post by **GDM528** » January 17th, 2022, 6:08 pm

mathof wrote:I have taken up GDM528's invitation to use his stepped and RoR profiles. I have the previous 50g Home roaster, tricked out with a thermocouple placed in much the same position as his and run through an Amprobe TMD-56 to a MacBook Pro running Artisan. Here are the results of today's roasting (I used Guatemala Poaquil beans):

As you can see, the AUC of the two roasts differs markedly, with the stepped profile measuring 240C*min and the RoR profile 313*C min. This is reflected in the colours of the roasted beans. The former measures 78 on my Tonino (interpreted by the Tonino as City) and the RoR a stage darker at 70 (interpreted as Dark). With these measured differences, they will certainly taste different.

Firstly, thank you! So much to unpack here:

Despite two different beans in two different models of the Ikawa Home, we hit first crack within 10 seconds of each other. Whoa...

Speaking of different models of the Ikawa Home, is the chamber smaller in the previous version? The latest Home chamber is about 90mm / 3.5in diameter. Differences in chamber volume and nominal fan speed could significantly change the heat density the beans will see.

I matched area under the curve for an empty chamber. Loaded with beans, the Stepped profile was about 4% higher AUC above 160C - still pretty close. Your temp curves are much 'softer' than mine, and your RoR of the stepped profile never levels off like mine does - that will significantly affect the AUC calculation. Is this a bean thing, or a software thing?

The Artisan curves are really(really) smooth - I'm jealous, but also a little skeptical. I've never used Artisan (yeah, maybe I should), but I got the impression they may do some curve fitting to the data to improve smoothing for the RoR calculations. If they are curve fitting, they're making assumptions about the general shape of the data that may not comprehend the multiple inflections in the stepped profile - that may explain your sawtooth-wave RoR curve.

Despite my concerns over software differences, you did see a measurable difference in roast level, so it actually is a bean thing... or a machine thing. I'm betting Ikawa recalibrated the control loop for the newer Home. Perhaps the newer version is more aggressive when following rapid changes in temperature. You could check that by running without any beans in the chamber and comparing to my results.

I think I've been too gonzo in my data collection: 7 samples per second of raw readings - that's over 4,000 data points that I copy into a spreadsheet for post-processing and charting the data. Despite my high-performance PC, the lag time is driving me nuts, as there's typically over 20,000 active calculations for smoothing and calculating RoR. I've taken some inspiration from the Artisan curves to reduce the amount of data I collect. What's typical for Artisan, 1 reading per second?

Again, thank you for running the test, and I hope both batches are 'not-bad'.

#64: Post by **GDM528** » January 17th, 2022, 6:24 pm

Iowa_Boy wrote:This is so interesting - I am really impressed with GDM528 and your data!
Are you using a bare wire thermocouple as well or something different? If you are willing, could you show a picture of your placement as well?
I am getting a lot of noise on my Artisan readings, so could definitely do a better job with the thermocouple and/or placement.
Also, just curious what you using for your curve smoothing settings?

My setup is described here: Ikawa Home thermal performance

All smoothing was done in an Excel spreadsheet using a 10-sample moving average nested in 2% exponential moving average (EMA) for the temp data, followed by a 5% EMA for the RoR calculation. I'm really regretting the amount of data I'm crunching however, so I'll be working on methods to reduce the volume of data without distorting the overall system response. My goal is for the thermocouple readings being at least 10x faster than the ramp rate of the Ikawa - and it can ramp really fast

January 17th, 2022, 7:28 pm

You could prune the data by keeping only the points where the value has changed by some small amount. The smoothing algorithms get a bit messier with uneven spacing, but it keeps the total computation livable.

#66: Post by **mathof** » January 18th, 2022, 1:20 pm

Iowa_Boy wrote:This is so interesting - I am really impressed with GDM528 and your data!
Are you using a bare wire thermocouple as well or something different? If you are willing, could you show a picture of your placement as well?
I am getting a lot of noise on my Artisan readings, so could definitely do a better job with the thermocouple and/or placement.
Also, just curious what you using for your curve smoothing settings?

Bare wire thermocouple. I'm afraid my tc mounting is not very neat and is covered with roasting dust. You can't really see, but it is secured by a layer of kapton tape overlaid with aluminium tape.

My curve smoothing settings on the images I sent yesterday were 40 and the smooth spike box was ticked.

#67: Post by **mathof** » January 18th, 2022, 1:57 pm

GDM528 wrote:
Speaking of different models of the Ikawa Home, is the chamber smaller in the previous version? The latest Home chamber is about 90mm / 3.5in diameter. Differences in chamber volume and nominal fan speed could significantly change the heat density the beans will see.

I matched area under the curve for an empty chamber. Loaded with beans, the Stepped profile was about 4% higher AUC above 160C - still pretty close. Your temp curves are much 'softer' than mine, and your RoR of the stepped profile never levels off like mine does - that will significantly affect the AUC calculation. Is this a bean thing, or a software thing?

The Artisan curves are really(really) smooth - I'm jealous, but also a little skeptical. I've never used Artisan (yeah, maybe I should), but I got the impression they may do some curve fitting to the data to improve smoothing for the RoR calculations. If they are curve fitting, they're making assumptions about the general shape of the data that may not comprehend the multiple inflections in the stepped profile - that may explain your sawtooth-wave RoR curve.

I make the diameter of my roast chamber also to be around 90mm. The depth is about 110mm. Comparing the outer measurements of the whole machine to the published figures for the new machine, width (130mm) and depth ((240mm) seem to be the same, but the new one may be slightly taller. Mine is 305mm from table top to top of glass plate, and 325mm to top of cork edged loading device. The published figure for the new machine is 350mm, which I suppose is to the top of the cork handle. (Perhaps you could measure the height of your machine.)

Here is the RoR curve I sent before, but with no smoothing:

And here is the empty chamber results on my machine (no smoothing):

I hope these images mean more to you than they do to me. Although, I keep Artisan records of my roasts, I don't find them of much help when I try to modify roast profiles: which remains for me mainly a matter of hit and miss.

#68: Post by **GDM528** » January 18th, 2022, 3:29 pm

mathof wrote:I make the diameter of my roast chamber also to be around 90mm. The depth is about 110mm. Comparing the outer measurements of the whole machine to the published figures for the new machine, width (130mm) and depth ((240mm) seem to be the same, but the new one may be slightly taller. Mine is 305mm from table top to top of glass plate, and 325mm to top of cork edged loading device. The published figure for the new machine is 350mm, which I suppose is to the top of the cork handle. (Perhaps you could measure the height of your machine.)

For my newer (100g) Home:
Chamber depth (outer edge of bottom to glass cover) = 110mm
Tabletop to top of glass = 316mm
Tabletop to top of loading port = 349mm

The rubber feet on the bottom contribute 6mm to the overall height, so the metal shell is 305mm.

Also noteworthy is the height to the exit chute - for the new Home machine it's 70mm, and the new bean deflector they added is exactly at the halfway point in the chamber/exit port span. That bean ramp may be why they boosted the claimed capacity - it ameliorates the bean-mixing issues with the larger loads. It looks like you left the insulation jacket on the thermocouple probe you mounted in the chamber, which I imagine is acting a bit like a 'mini-ramp' that may improve roast uniformity.

#69: Post by **GDM528** » January 18th, 2022, 3:36 pm

mathof wrote:Here is the RoR curve I sent before, but with no smoothing:

And here is the empty chamber results on my machine (no smoothing):

I hope these images mean more to you than they do to me. Although, I keep Artisan records of my roasts, I don't find them of much help when I try to modify roast profiles: which remains for me mainly a matter of hit and miss.

Thank you! The clouds have parted, and I see what happened:
For the stepped profile I programmed the temperature steps to complete in 30 seconds - but in your profile they take 60 seconds. That will significantly reduce the AUC, especially in the final development step, which looks nearly a minute too short, hence lighter roast result.

Any idea how the ramp rate changed - was the profile edited, or did the firmware in the machine limit the slew rate?

#70: Post by **GDM528** » January 22nd, 2022, 7:10 pm

Follow-up on posting #58 RoR versus Stepped profiles:

I have patiently waited 6 days before tasting the results of the RoR vs Stepped experiment. Around the third day both roasts started showing a smattering of light oil spots on roughly 10-20% of the beans:

Perhaps about a half-step in roast level advancement after resting for a few days. The sounds of first-crack were subtle for both roasts and tailed off very shortly before the drop. Never a hint of anything like second crack at the end of the roast - despite briefly hitting 230C on my thermocouple reading. Also worth noting that my thermocouple readings could be offset (probably too high) by a few degrees, despite my efforts to carefully calibrate. I'm learning to not be so judgey immediately post-roast. This is yet another reason for me to keep batch sizes small, to limit the damage from any rash decisions.

Another axiom is starting to sink into my head:
"Trust the Temp" If the beans hit first/second-crack temperatures, they reached first/second-crack - even if you didn't see or hear anything.

"Temp" will be different for each thermocouple calibration and placement for any given machine, so everyone's mileage will vary - but not by too much IMHO, and once you learn your particular setup you should be good to go. The Ikawa is so freaking precise that it bodes well for recipe-sharing. Any given genus of bean, its processing, and age may also affect the specific temperatures at which cracking sounds may occur, but the thermodynamics of the chemicals common to all bean types should be close enough to get a good roast result with just a couple iterations.

As for how they taste:
I pulled consecutive shots with each roast as same-y as I could: ground 18g single-dose with RDT and WDT. I use precision-cut filter paper disks at both top and bottom of the puck and tamped the H outta that bad boy into a VST basket in a naked portafilter. Pulled at 95C/6.5bar to 2:1 in about 25 seconds after a 15 second-ish low-pressure pre-infusion. No signs of channeling.

My tasting skillz are largely non-existent. I use language like "Homer-Simpson-eating-a-donut", "dog-jumping-into-a-pile-of-Autumn-leaves", or "burnt-dirty-tractor-tire". Fortunately this is all about differences. I normally don't like drinking straight espresso, but both were pleasantly drinkable for me - and I could not detect any difference between them. So my tasting notes are: "Same Same".

I consider this good news. I think it implies that I don't have to emulate drum roasting profiles. I can 'deconstruct' the roast profile into distinct, individually-optimizable steps. Similar to sous vide cooking, I can focus on finding the perfect time and temperature for each step - that's harder to accomplish with the steadily-declining RoR technique, as the different phases blend into and interact with each other.

NOTE: mathof in post #59 tried running the stepped-profile recipe, but for some as-yet-to-be-determined reason couldn't match the 30-second temperature ramps with a prior-generation 50g Home machine. I have a theory that other Ikawa models might also have a slew-rate limit, which significantly affects the roast results with the stepped profile. Dunno if I should be considering slowing down the slew rates for the stepped profiles...