Peppersass wrote:You're right. I hadn't noticed the correlation. The profile slope flattens slightly starting at 7:29. I checked my last nine roasts, and that's where RoR starts declining in almost all of them. I have a couple of roasts of Yemen Al Haymap Natural where RoR started a distinct decline well before 7:29 -- as early at 7:00. Not sure what happened there. But in general, yes, the decline starts where the exhaust profile declines. Duh.

I noticed with a lot of their profiles that they match a slope change with a phase. The cool part about the Ikawa is how darn repeatable the profiles are even with completely different coffees.

The problem I was trying to solve was roasts going a little darker than I wanted when I marked 1C when I heard two or more pops, or even at the very first pop. If I pulled back on DT to lighten the roast, I'd get extremely short times, like :15 after 1C, or really low DTR, like under 10%. I had better luck with marking it at roughly 7:29, where the profile declines. Sort of a self-fulfilling prophesy.

At the same time, though, I started looking at temperature rise after 1C and drop temperature. What I've found is that it's really the drop temperature that determines roast color. I can simply roast a given bean to, say, 417F and get the color to within +/- one point on my RoastVision color meter every time. I don't have to mark 1C or look at time after that or DTR, etc.

An interesting set of profiles that may help you further determine your final color variables are the Morten Munchow sample profiles that achieve the same final roast color but use two different methods. One is a long development but lower temp and the other is a faster profile but a higher final temp. I tried them out on some sample coffees that I had and it gave me some good insights.

One issue I seem to have run into is that perhaps some of the older profiles (maybe the home profiles?) had very high temperatures. So high that I'd think it would scorch the beans. Perhaps these are inlet air temps? I believe the Home uses those instead of exhaust temps? Anyway, I was trying to find a good full city plus profile to mess with for a family member and stumbled upon a post showing their profiles. I never tried it because I thought I'd start a fire...

But how to determine the drop temperature for a new bean that hits the color and development I want? The way I do it is to go for at least 10F temperature rise after I mark 1C, as some have recommended (probably for drum roasts.) This tends to get me into the right ballpark. If the roast is too light, I can bump the drop temperature. I have yet to encounter a situation where after waiting for a 10F rise the roast is too dark. I suspect it works for me because I'm marking 1C early, at the start of the profile decline, not the first pop or pops. If I did that, I wouldn't get 10F rise before the roast got too dark.

Bottom line, if I mark at the first pop or pops, presumably the true 1C, then the development time, DTR and temperature rise after 1C will be wildly shorter/smaller than the conventional wisdom. Or, if I use the conventional metrics, the roast will be way too dark. This makes me think that either fluid bed roaster timing after 1C is drastically different from drum roaster timing, as it is for dry time, or it's a peculiarity of the Ikawa roaster configuration and very small batch size (the Pro 100 could behave differently!) Or both. I guess it boils down to "know thy roaster" and "roast for taste, not curves".

Your last sentence is very true. I struggled with the Ikawa for the first 20 roasts or so trying to replicate drum profiles before finally letting the beans talk. I roasted 10 different profiles with the same beans and did a cupping with friends and family. My wife that hates black coffee said she would drink and enjoy the roast that came from the Ikawa Sample Profile 1 black! I about fell out of my seat. After some tweaking such as the fan needed to be stepped up a bit at the end for some beans to stay fluid and some beans needed more development time, it has been a great profile to expand from.

I am eager to try the espresso profile again and try to dial it in. I'd like to have 3-4 different go-to profiles for different coffees and roast levels as I sample many types.

That seems almost AI-ish to me, and I'm not sure it would work with the hardware they have. I wonder if simply adding more control points would help to smooth the curve and require less drastic temperature changes.

FWIW, I don't see overshoot in my roast logs, except a few seconds after TP when it's trying to get the exhaust temperature to settle into the profile. After that, I can see some tiny ripples in exhaust temperature, but no space between them and the curve. The ripples are magnified in the inlet temperature curve, though, so I can see that the PID is oscillating to keep the exhaust profile on track. Whether those frequent temperature changes are problematic is hard to know. It's possible that you see more overshoot because your batch size is larger, requiring more drastic action on inlet temperature to maintain the profile.

What I've been thinking is if the software could apply interpolation between the data points using basic numerical methods. The amount of interpolation could perhaps be adjusted via a slider. This, in practice, would automatically add more control points using the basic curve set by the user without the user having to manually add more points. Another way to say it is there are fitting functions that take data points and fit a curve to them. Then use that curve to determine more points for the Ikawa to use. It isn't really AI, just somewhat advanced mathematics. I'm not sure what type of computing hardware the Ikawa has on-board, but it could certainly be programmed into the App as a pre-roast feature. The only limitation I could think of would be how many total points the Ikawa hardware can load onto its memory. I think that would go a long way to curing my only complaint so far with the Pro100.

Milligan wrote: I'm not sure what type of computing hardware the Ikawa has on-board, but it could certainly be programmed into the App as a pre-roast feature. The only limitation I could think of would be how many total points the Ikawa hardware can load onto its memory. I think that would go a long way to curing my only complaint so far with the Pro100.

Per a teardown thread on this forum, the Ikawa Home is controlled with an Atmel ATMega644P - a pretty decent microcontroller but with only 64K of flash memory. Should be plenty for managing the wireless interface, thermal sensors, motor, heater (PWM), safety, etc. I bet this same board could run the Pro models as well. Chip cost around $5 in volume - if you can find them...

No sign of a discrete PID controller chip, so likely handled by the microcontroller, which seems reasonable given the Ikawa's thermal model and continuously-changing thermal targets. Many (most?) PID controllers are based on a static goal and try to get there quickly yet calmly - but the Ikawa UI is based on temperature ramps. Some of the thermal response testing I've run on my inlet-controlled machine starts to significantly overshoot setpoints above 2C/sec.

I recall seeing a spreadsheet data dump from a Pro model, and it had columns for what looked like PID parameters - does that mean Pro owners can tweak those parameters? That would be interesting and a perhaps a bit like literally playing with fire

luca wrote:I scratched my head a bit at what there is to misunderstand, but I guess you are saying you don't understand whether I'm saying you should compare the Ikawa probe readings or the TC that you have retrofitted into it. I didn't direct that suggestion to you personally; I was trying to articulate a general principle: if you compare two temp reading graphs, of whatever temp reading, and the resultant roasts are of the same beans and they taste different, then that temp reading probably isn't useful for delivering a predictable taste outcome and controlling the roast. Is this a controversial statement?

Thank you for your response. I may have misunderstood your point, but phrased as you do here, I have nothing to disagree with.

I've been messing a bit trying to get the RoR more stable. I had been using 100g charges on my Pro100 and decided to drop it down to 80g as a comparison. I used the same profile for 100g and 80g. Quite a difference in stability.

100g roast:


80g roast:


I dropped the 80g roast a little earlier because the 100g developed a bit more than I wanted, but the graphs speak for themselves. Something else interesting is that the 80g dried faster (seems obvious) but went into 1C later. Perhaps a more stable RoR helps elongate the Maillard phase? I'm not sure what to make of that just yet. It is a pretty substantial difference in 1C, not just a few seconds. 1C marked not at the first pop but the first cluster of pops.

I'm going to also roast something close to pepperass's profile and cup everything tomorrow. This is a washed Colombian Supremo I'm working with here.

Milligan wrote:I've been messing a bit trying to get the RoR more stable. I had been using 100g charges on my Pro100 and decided to drop it down to 80g as a comparison. I used the same profile for 100g and 80g. Quite a difference in stability.

Wow, and, yikes. The change in inlet temperature is remarkable. Is there some way to superimpose setpoints on the curves? It looks like the inlet temperature for the 100g batch approached 550F - were any of the bean scorched?

In this thread Ikawa Home thermal performance I saw little change in the temperature curves versus dosage for the inlet-controlled Home version. However, that meant the same amount of heat energy was distributed across a different quantity of beans leading to a difference in final roast level. Does the Pro app adjust the profile if you tell it the dosage?

Over the course of many roasts I've developed a sense for how navigate the control loop dynamics of the Home model. I presume the same holds for the ET controlled Pro models. I also wonder if there's a way to adjust the PID parameters with the Pro app.

I took IR videos of the beans during a roast (posts #64 and #65 in aforementioned thread), which convinced me to stick with 50g batches.

I had a chance to taste them. I did the roast with pepperass's profile as well using 80g and 100g.

100g:


80g:


The above are profiles similar to pepperass's which are based on the Ikawa Espresso profile. I cut it back like he suggested and I let the final temp be the guide. Interestingly there is 12s difference in when the Ikawa reached 423F. In the 100g roast, the Ikawa overshot
the temp profile target by quite a bit and I'm not sure why. It overshoots well after 1C, so that small perturbation shouldn't have caused that. The RoR isn't out of whack too much but you can see the intake temp fall quite a bit as the Ikawa tries to correct the high temp. Perhaps it has something to do with the fan speed increasing pushing more residual heat through the beans? The 80g sample shows that small bump in temp above the target path at the very end but not nearly as much. RoR is, again, flatter with the 80g profile.

I didn't do a full on cupping protocol. I brewed each in my Moccamaster and took notes as I tasted. 30g, medium coarse grind, 500ml of water. Coffee is a problematic washed Colombian from the Huila region. I haven't gotten a great cup out of this bean yet, so I thought it would be a good one to try since I know its poor qualities quite well.... I have a different washed Colombian that is very good, but didn't want to waste it on this experiment.

Quick profile:
80g: Bright acidity, slightly sour, slight bitter finish. Medium body. Papery, lemon aroma. Slight vegetal, bland taste, lemon finish. Does have decent sweetness. Negatives mute as it cools into a bland coffee.
100g: Tart, sour, more bitter. Medium body. Papery aroma. Bland taste, bitter sweet after taste with the lemon. Astringency builds with sips. Bad.

Comparison: 80g wins hands down. Although underdeveloped and not something I'd ever want to drink again, the 80g was noticeably better. Interestingly, the 100g had 20s more total time yet the 80g still had smoother acidity. So one could begin to see that RoR does matter at this range.

Pepperass's:
80g: Low, citrus acidity. Medium body. Much more balanced. Slight chocolate aroma. Bland flavor.
100g: Low acidity, slightly more than 80g. Medium body. Balanced. Bland flavor.

Comparison: This one was more interesting to me than the last. I think this particular Colombian needs the additional development to overcome the intense acidity. Still doesn't have a good flavor. I just need to reconcile that it isn't a good bean. That's fine, because my other Colombian I have is 10x better. Anyway, the 80g did have brighter, more "flavorful" acidity. While not much flavor, it did make me want to go back for another sip compared to the 100g. The difference wasn't as staggering as the last round, but it was noticeable.

Overall, for this bean the Pepperass profile was much better. Take away is that the Ikawa seems to be very sensitive to dosage and the Pro100 does seem to do better at lower doses. I'll likely experiment with 90 to see how that react. As of right now, I think I will go to 80g for my normal sampling due to how much of an easier time the Ikawa has at managing its temperature at that dose. I think it is too early to say, but I'm not so sure I'd spring for the Pro100 if it needs to be dialed back so much to match the consistency of the Pro50.

GDM528 wrote:Wow, and, yikes. The change in inlet temperature is remarkable. Is there some way to superimpose setpoints on the curves? It looks like the inlet temperature for the 100g batch approached 550F - were any of the bean scorched?

In this thread Ikawa Home thermal performance I saw little change in the temperature curves versus dosage for the inlet-controlled Home version. However, that meant the same amount of heat energy was distributed across a different quantity of beans leading to a difference in final roast level. Does the Pro app adjust the profile if you tell it the dosage?

Over the course of many roasts I've developed a sense for how navigate the control loop dynamics of the Home model. I presume the same holds for the ET controlled Pro models. I also wonder if there's a way to adjust the PID parameters with the Pro app.

I took IR videos of the beans during a roast (posts #64 and #65 in aforementioned thread), which convinced me to stick with 50g batches.

I checked today in the sunlight and didn't notice any scorching. I would say the beans are slightly more mottled on the 100g samples, so perhaps they dried more unevenly. I'm kind of tempted to send an email to the rep at Ikawa I had when purchasing and ask if this behavior is normal. As far as I can see, there isn't a way to enter the dosage for a profile until after it is roasted just for record keeping purposes. If someone knows how to change more settings then I'm all ears.

So after someone mentioned the PID's slow response and the oscillation at the start of the roast, I decided to play. These are Pro graphs, but since you can see the inlet temperature, it might give you a hint of what's necessary to tame the oscillation. The first one is peppersass's modified Espresso 1 profile, the next 2 are my first 2 guesses I made to try and tame the beginning of the roast. No comment on roast quality. These are 100gm roasts of the Ethopian Boui Bora that came with my Home 100. Preheat to 375F and then down to 271F at 28 seconds. I added the one step at 28 seconds and the rest of the profile is untouched.

ira wrote:These are Pro graphs, but since you can see the inlet temperature, it might give you a hint of what's necessary to tame the oscillation. The first one is peppersass's modified Espresso 1 profile, the next 2 are my first 2 guesses I made to try and tame the beginning of the roast.

Interesting - consider adjusting the timing between the setpoints and addition to the temperature.

In post #23 of this thread: Ikawa Home thermal performance I found that the clever peeps at Ikawa had figured out how to adjust the timing between radically different setpoint temperatures to almost completely suppress oscillations. The loop delay for the Ikawa Home is much shorter than the Pro, so the 40ish second spacing with the Home might have to be closer to 90ish seconds for the Pro, and would be strongly dependent on batch size.

Does the Pro allow running empty? That would allow experimentation without the greens sacrifice, and give a baseline response to compare with 50g and 100g batches.

So further playing with that profile got me here, almost completely smooth. Here is the roast and the profile settings I used. For 100 gram roasts on the Pro 100.

ira wrote:So further playing with that profile got me here, almost completely smooth. Here is the roast and the profile settings I used. For 100 gram roasts on the Pro 100.

So it looks like you increased the T0 setpoint and lowered the T1 - correct? The peak inlet temperature is really getting up there, looks about 525F - how does that compare to the preheat temperature?