Ikawa Home thermal performance - Page 8
i hate to be a pain but i see you run all your batches with 60g. could you do me a massive favor and run the same profile (pick any bean you want ) but use 100g and then post the temp curve on the same graph as the 60g doses. i love your stepped profile and want to see how much different the readings are for the 100g size as this is what i use.
-
- Supporter ❤
And I probably said this before, but with this much data so generously provided, we can SO derive the ODE satisfied by the bean temperature given the intake curve. I'll do it now and get back...
-
- Team HB
I believe this work started with my posting of the profile I use on my Pro and I only roast 100gm baches, so it should be capable, though maybe with increased airflow to increase the heat available for the extra beans.
In this particular case, I only had 60g left of one of the greens...sambuist wrote:i hate to be a pain but i see you run all your batches with 60g. could you do me a massive favor and run the same profile (pick any bean you want ) but use 100g and then post the temp curve on the same graph as the 60g doses. i love your stepped profile and want to see how much different the readings are for the 100g size as this is what i use.
Typically, I add beans in the funnel until they reach the level of the glass top - that's usually around 65g or so. The IR camera tests in earlier posts 64/65 made me a fanboy of the spinning/corkscrewing donut I get when underdosing.
FWIW, 100g batches have essentially the same temperature readings and so should be essentially the same curvature-wise. Yet for some reason 100g comes out a bit lighter - could be an airflow thing, or a total heat delivered thing, I dunno. I'd just bump the end temperature up a few degrees and recalculate the curve.
This set of curves might be relevant, showing the temperature drops from setpoints (inlet) and in-chamber readings:mgrayson wrote:And I probably said this before, but with this much data so generously provided, we can SO derive the ODE satisfied by the bean temperature given the intake curve. I'll do it now and get back...
I might be reading too deep into the tea leaves, but note there's a subtle dip in the (Setpoint - Bean temperature) about 30 seconds before I heard audible cracks...
-
- Supporter ❤
Makes sense, as audible FC is preceded by the exothermic reaction (or so I'm told - never looked up just what reaction is producing the heat).
I can get good matches to the above data and to the earlier stair-step set point curve. The equations have the same form, but the coefficients are different. bean'(t) = a*setpoint(t) + b*setpoint'(t) + c*bean(t). It turns out a constant term isn't needed. That coefficients a and c are not of equal magnitude creates a level-dependent offset. What puzzles me is that the coefficient b doesn't seem stable - it is constant throughout a roast, but changes from one roast to another.
If I can get a stable model, then one can really back out the set-points for a desired bean temp with a single numerical integral. Yeah, if.
I can get good matches to the above data and to the earlier stair-step set point curve. The equations have the same form, but the coefficients are different. bean'(t) = a*setpoint(t) + b*setpoint'(t) + c*bean(t). It turns out a constant term isn't needed. That coefficients a and c are not of equal magnitude creates a level-dependent offset. What puzzles me is that the coefficient b doesn't seem stable - it is constant throughout a roast, but changes from one roast to another.
If I can get a stable model, then one can really back out the set-points for a desired bean temp with a single numerical integral. Yeah, if.
TLDR: update on the thermal behavior of the IKAWA home, reduced to an equation. This is a couple steps removed from any actual roasting data, so as such may set a new record for fastest exit off first page of topics...
This is a first step towards the ultimate goal of simulating roasts. I'm currently using a prototype setpoint-to-BT simulator, and it's already been crazy-useful. For example: I had a roasting session where first crack was later than I wanted, so I used the simulator to calculate new setpoints while keeping the same start/stop time and temperatures. The first-crack timing simulation estimate was 30 seconds earlier, and the actual roast result was 25 seconds earlier. Nice. I'm looking forward to the prospect of no longer having to gather and pour over post-roast log data.
I ran several (actually, a whole lot of) profiles through my IKAWA Home and measured the empty-chamber temperature with a fast-response thermocouple located right where the beans would normally be spinning around (post #1). The profile was a simple step function: ramp to a set temperature in 30 seconds and hold that same temperature out to 10 minutes. Goal was to measure the thermal 'lag' between the set temperature and the actual temperature in the chamber.
The empty chamber response is a significant component of the overall response between the setpoints and the actual bean temperature (BT). The thermal lag of BT will be pretty significant for the first 2-3 minutes of the roast, before ultimately reaching the empty-chamber temperature by the end of the roast.
This chart shows the data (green lines) overlaid on the calculation (black lines) fitted to the data:
The thermal response follows a simple exponential curve, the coefficients of which are function of the initial temperature step:
Ts = Initial step temperature, Tr = roast time
Chamber temperature = Ts - ((0.15 x Ts - 5) + (0.1 x Ts - 3.5) x exp(-0.01 x Tr))
Seems like a pretty decent approximation, even after I simplified the coefficients to look more aesthetic.
I observed the thermal measurements were abruptly more erratic when the step temperature went above 200C. Weird. Might be related to IKAWA's predilection for spiking the temperature early in their curated roast profiles. Perhaps there's some sort of shift in their control algorithm to trigger a different response to the spike. FWIW I'm not planning to start any of my roasts that hot.
Next up is to account for the setpoint temperatures actually ramping over the course of the roast, rather than holding steady. Then, finally account for the thermal lag between chamber temperature and BT, which I expect will follow a similar exponential trend.
Shout-out to the things that have enabled me to get this far:
The IKAWA's physical structure is simple, and therefore so is its thermal response.
The inlet-control loop is closed at the air temperature just before entering the chamber - fast, simple, precise, repeatable.
Froze some key variables: initial ramp from start always 30 seconds, fan speed always 80%.
IKAWA may have taken away the in-chamber thermal probe, but science and spreadsheets can make that irrelevant.
This is a first step towards the ultimate goal of simulating roasts. I'm currently using a prototype setpoint-to-BT simulator, and it's already been crazy-useful. For example: I had a roasting session where first crack was later than I wanted, so I used the simulator to calculate new setpoints while keeping the same start/stop time and temperatures. The first-crack timing simulation estimate was 30 seconds earlier, and the actual roast result was 25 seconds earlier. Nice. I'm looking forward to the prospect of no longer having to gather and pour over post-roast log data.
I ran several (actually, a whole lot of) profiles through my IKAWA Home and measured the empty-chamber temperature with a fast-response thermocouple located right where the beans would normally be spinning around (post #1). The profile was a simple step function: ramp to a set temperature in 30 seconds and hold that same temperature out to 10 minutes. Goal was to measure the thermal 'lag' between the set temperature and the actual temperature in the chamber.
The empty chamber response is a significant component of the overall response between the setpoints and the actual bean temperature (BT). The thermal lag of BT will be pretty significant for the first 2-3 minutes of the roast, before ultimately reaching the empty-chamber temperature by the end of the roast.
This chart shows the data (green lines) overlaid on the calculation (black lines) fitted to the data:
The thermal response follows a simple exponential curve, the coefficients of which are function of the initial temperature step:
Ts = Initial step temperature, Tr = roast time
Chamber temperature = Ts - ((0.15 x Ts - 5) + (0.1 x Ts - 3.5) x exp(-0.01 x Tr))
Seems like a pretty decent approximation, even after I simplified the coefficients to look more aesthetic.
I observed the thermal measurements were abruptly more erratic when the step temperature went above 200C. Weird. Might be related to IKAWA's predilection for spiking the temperature early in their curated roast profiles. Perhaps there's some sort of shift in their control algorithm to trigger a different response to the spike. FWIW I'm not planning to start any of my roasts that hot.
Next up is to account for the setpoint temperatures actually ramping over the course of the roast, rather than holding steady. Then, finally account for the thermal lag between chamber temperature and BT, which I expect will follow a similar exponential trend.
Shout-out to the things that have enabled me to get this far:
The IKAWA's physical structure is simple, and therefore so is its thermal response.
The inlet-control loop is closed at the air temperature just before entering the chamber - fast, simple, precise, repeatable.
Froze some key variables: initial ramp from start always 30 seconds, fan speed always 80%.
IKAWA may have taken away the in-chamber thermal probe, but science and spreadsheets can make that irrelevant.
Late to the game without anything of substance to add, but I'd like to mention after sticking a bunch of probes in different positions I observed the temperature readings were nearly identical 45 seconds to 1 min 15 seconds into the roast after preheat with or without beans. The probe in the bean mass and the one fixed to as close as I can get it similar to the Ikawa pro were just about the same at every stage +/- 2C, whereas the probe in the little trench was +/- 3-6C. Correction from earlier. After testing from a cold start the plastic vent tube area the temperature difference observed is around 50-60C lower and after each roast it hovers around 35-40C lower than the chamber area.
I'd bet your picture looks close to what I imagine might have been in IKAWA's test lab when they were figuring out where to locate the ET sensor - minus the plywood work surface however 
Since I was hand-building the thermocouple, I had the freedom to easily place it anywhere. Putting it directly into the spinning beans minimized the mixing with air that skirts around the greens and/or is cooled by the chamber walls. And there was an added benefit of additional mixing of the beans by the probe tip, kinda like a second 'ramp'.
I never tried other locations, so it's useful to see your survey results, thanks!

Since I was hand-building the thermocouple, I had the freedom to easily place it anywhere. Putting it directly into the spinning beans minimized the mixing with air that skirts around the greens and/or is cooled by the chamber walls. And there was an added benefit of additional mixing of the beans by the probe tip, kinda like a second 'ramp'.
I never tried other locations, so it's useful to see your survey results, thanks!