The Start of Something…
Our New Infrared Bean Temperature Sensor (IBTS) is poised to forever change the way you roast.
Aillio’s Taipei office is near the main train station, a part of town rife with vendors selling electronic components. Capacitors, diodes, resistors, transistors — all can be bought at any number of stores just across the street. Needless to say, as our CEO and resident Geek-In-Charge, Jonas Lillie is a frequent visitor to these little shops.
And so when, on a rather humdrum weekday winter afternoon, Jonas quietly slipped out of the office to purchase a tiny 5v fan no bigger than a dime, I didn’t think much of it. The incessant whirring of the 3d printer behind my head also failed to set off any alarms — its robotic whistling had been a constant presence for weeks. For all I knew he could be printing out a purple stand for his computer monitor, or (another) model of Yoda’s head.
But when, at the end of the day, I found him unmounting the control panel on our Bullet R1 and gluing an odd looking contraption — some sort of custom housing containing an infrared (IR) sensor and the aforementioned fan — directly onto the faceplate of the roaster, I took notice.
I asked him what he was doing. He told me it was a new design for the IR sensor in the Bullet, one that would both eliminate the need for a protective germanium glass, and also allow us to use infrared to measure bean temperature during the roast.
In response, I probably said something along the lines of “Woah.”
When the glue dries, he explained, we’ll know if it works.
We have a big bag of old specialty beans from Honduras that we only use for testing. We measured out 500 grams of past-its-prime goodness, dropped the batch into our friendly office Frankenstein Bullet, and together watched a *real* bean temperature curve being plotted for the first time.
Not so humdrum after all.
Death to the Old Ways?
Roast profiling, or the plotting of bean temperature data over time, has revolutionized the way people think about coffee roasting. For many, rightly or wrongly, shapely profile curves have become synonymous with a good roast. The data used to generate those curves is produced by temperature probes placed within the coffee roaster. These probes come in a few flavors, including thermocouples, Ntc (negative temperature coefficient) thermistors, and RTDs (Resistive Temperature Device.)
But the readings from all of these probes are, ultimately, only a proxy for the temperature of the beans themselves. In other words, the curves on your roast profile graph are a plot of the bean probe’s temperature, not the actual bean temperature. This creates a number of problems for people hoping to rely on profile data.
For instance, because a bean probe is primarily heated by conductive heat from the beans, and because a smaller batch of beans may do this less efficiently, it often appears that smaller batches reach first crack at lower temperatures, even though they don’t. The data is, in fact, objectively false. This makes it impossible to meaningfully compare bean probe data across different batch sizes.
Another important variable is the placement of the temperature probe within the roaster, as where the probe sits within the churning bean mass can also affect how quickly the beans heat it up.
The way the roaster is designed, too, may affect the readings. For instance, in some roasters, thermal energy may migrate from the front plate to the probe — or vice-versa — which will also affect the readings.
And it’s important to note that different probes will react to temperature changes at different speeds, as well. If the casing around a probe is thicker, the metals inside the probe will be more shielded from the heat, and thus will heat up more slowly than probes with a thinner casing. In other words, two probes of different thickness placed in the same place in the same roaster will generate wildly different looking curves. And even among a batch of probes from the same supplier, you may discover that there is a great deal of variation. (See above photo for an example.)
Regardless of how thick or thin the casing around the probe is, there is always a lag between the temperature of the probe and the actual temperature of the bean itself. Even in the best scenarios, this lag is not inconsequential, and it often exceeds a full minute. “Real-time” profiling, then, is not very real at all. You are instead peering into the bean’s recent past, and can only guess at what changes are happening in the moment, while the roast is still “live.”
So while there is no doubt that traditional bean probes can be a very helpful reference, they also come with many limitations and caveats. This is particularly true when it comes to sharing data among roasters— with so much bad data, and so many variables at play, armed only with a profile, how can one expect to duplicate another’s roast?
Infrared sensors work differently. They measure the amount of light of a specific wavelength (known as infrared) emitted by an object. All infrared probes must first be adjusted to the ‘emissivity’ of the material they are measuring, as different materials can be more or less effective at emitting thermal radiation as they heat up. But once the emissivity is set to match that of roasting coffee, the temperature readings will be very accurate, and also very, very fast. And, perhaps most exciting of all — they will not be influenced by batch size. No matter what your charge weight, you’re going to be watching your beans crack at the same temperature. That makes it much simpler to anticipate changes before they happen during the roast, and also to compare data from different roasts afterwards.
You may wonder, then, why the rest of the world is not using infrared sensors in their roasters. That’s a good question, and the answer is simple: if an infrared sensor is left exposed to the coffee roasting environment in the drum, oil and dust particles will quickly coat it, at which point its readings will become useless.
We did discover a workaround for this. After all, one of the main selling points of the original Bullet R1 was the ability to set a preheat temperature with an infrared sensor. We accomplished this by shielding the sensor with a germanium glass. Why germanium? Because if we used normal glass, we’d be measuring the temperature of the glass itself, rather than the drum behind it. Germanium, however, is transparent to the infrared wavelengths, allowing the light to pass through to the sensor, and therefore enabling us to “see” how hot the drum is.
Of course we wanted to use the infrared sensor as a bean temperature sensor too, but we soon learned that the germanium glass becomes coated with moisture during the roasting process, making it impossible to give accurate readings. We tried to make it work — we really did — but in the end, it became apparent that we could only use the IR sensor for drum temperature readings.
Nonetheless, the ability to set a preheat temperature using the IR sensor made the Bullet R1 unique among its peers. It allowed for greater roast-to-roast consistency than would have otherwise been possible. There was just one minor drawback. The germanium glass used to shield the sensor will eventually become coated with all the materials it protects the sensor from. Thus, cleaning the IR glass became an integral part of routine maintenance for version 1.0 of the Bullet R1. The process itself doesn’t take more than 10 or 15 minutes and only needs to be done occasionally, but we (and our users) always wished it could be a little less troublesome.
We knew that if we could just figure out a way to roast without the glass, then we could have our cake and eat it, too: a sensor that performs double duty as a drum AND bean temperature sensor; one that rarely (if ever) needs to be cleaned.
About that cake…
After our first roast with Jonas’s prototype, we immediately checked the sensor for any signs of dirt or grime. It was spotless, and that gave us the green light to put the new infrared system through its paces. We roasted different batch sizes (from 150g to 1.2kg) and different bean types (from chaff-happy fruity Ethiopians to chocolatey Papua New Guinea peaberries). We roasted them Cinnamon-light and Charbucks-dark and everything in between. We cycled fan settings, drum speeds, and roast times. We threw everything at it that we could, everything we imagined our users might do.
What we discovered was truly exciting — revolutionary. All of the beans were reaching first crack at comparable temperatures, and different batch sizes of the same bean were cracking at almost exactly the same temperature. The key to it all — the new low-flow, high-pressure air barrier that Jonas 3D-printed — did its job. The sensor never got dirty. In essence, we had found the holy grail of roasting: instantaneous bean temperature readings that were unaffected by batch size.
Algorithmic Gymnastics or “How to stick the landing.”
Of course, the work didn’t end there. We began making adjustments to the emissivity to more accurately reflect real first crack bean temperatures. That, in turn, skewed our drum temperature readings. We had to rewrite the firmware to allow us to increase our preheat temperature setting to account for the change.
The very first profiles with the new IBTS were made with the original digital smoothing algorithm that we had been using for drum preheat temperatures, which we knew was less than ideal. As you can see below, when we reverted back to unfiltered raw data, we discovered that the sensor was also picking up the temperature of the drum vanes, resulting in some wild temperature readings, particularly early in the roast.
It fell upon Jonas’s brother Jacob, who was already building a new induction board from scratch, to develop a smarter filtering algorithm to reduce the noise. This algorithm is still a work in progress, but we’re already very happy with the results we have now.
The 300g roast profile below employs the new filtering algorithm, and is a good example of an infrared bean temperature sensor curve. The IBTS curve is plotted alongside a traditional probe reading from the same roast. (Click to enlarge.)
The first thing experienced roasters will notice is that the “turning point” (the point when the traditional bean probe temperature begins to rise following the charge) has essentially vanished. The next thing you’ll notice is that the IR reading is much higher than the traditional bean probe reading, with first crack in the above roast occurring at 204.3°C on the IBTS and 165.9°C on the traditional probe. Both of these differences are due to the lack of thermometric lag in the IBTS, whose digital filtering delay is a mere ten seconds. No thermometric lag means no data artifacts. In other words, you are seeing a real bean temperature profile curve for the first time.
To understand the greater implications of this, take a look at the 1000g roast profile below.
Both these profiles are using different charge weights of the same coffee. Between them, the first crack readings vary by almost 30° on the traditional probes, making it very difficult to meaningfully compare them to each other. But on the IR they are essentially the same, varying by only a few degrees. This is because bean mass does not influence the IR temperature reading in the same way it does traditional probes.
First crack temperatures are consistent, data artifacts such as the “turning point” are no longer present in roast profiles, and for the first time an accurate reading of the bean temperature is readily available for roasters. No more guessing.
Infrared is the future of roasting.
Toward the future.
The new IBTS is a technological step-forward that we believe will open new windows into the coffee roasting process. We’re so happy to share this with Bullet R1 owners — both present and future. Honestly, we feel downright giddy just thinking about what you all will do with it.
As for our part, we will continue improving the filtering algorithm, and we will also be building new IBTS-based features into the RoasTime 2.0 software and Roast.World website, including IBTS RoR values and access to a treasure trove of really, really good bean data that the whole world can learn from.
We hope you’re looking forward to it as much as we are.