How coffee beans move within a revolving drum. (Also: Why you should be anticipating a revolution in the coffee industry.)

An interview with coffee researcher Mark Al-Shemmeri

15 min readJul 10, 2023

SPOILER ALERT: This is a long, geeky article! And although it specifically addresses research regarding coffee bean motion within a roasting drum, it’s about much more than that.

Don’t expect any new “rules” for roasting better coffee. But grant us fifteen minutes of your time and you just may view coffee roasting in a different light, and return to your workspace armed with some new ideas to try out.

A Bullet R1 in the lab.

Across the world of specialty coffee, from farm to cup and back again, we are all asking ourselves what works best and why? And yet our collective tried-and-true best practices are, more often than not, merely best guesses — methods and principles that are accepted as dogma only because they seem to be effective.

Us coffee people are all-too-eager teachers, and so it’s quite easy to find someone to show us how to do things the right way. For example, the “correct” grind size for cold brews, the “perfect” water temperature and coffee ratio for a proper pour-over, or how to “control” your roast and limit off-flavors with an ever-declining Rate of Rise (ROR).

And yet, many (most?) of us have experienced the unique pleasure of enjoying an incredible cup of coffee that broke some rule we were explicitly taught never to break. Why?

Why, indeed! The problem is that without peer-reviewed studies, we lack the theoretical underpinnings to explain why one approach should be preferred over another. Instead, we are left to speculate, which can at best provide us with an incomplete understanding, and at worst mislead us entirely.

These gaps in our knowledge are missed opportunities — and potential tinder for a coffee revolution. Speaking in these terms, Mark Al-Shemmeri is a bit of specialty coffee arsonist, gleefully setting our assumptions on fire.

We (virtually) sat down with him to pick his brain about his latest research into the motion of beans within a drum roaster, the implications of his results, and his motivations for undertaking the herculean task of challenging popular assumptions within the world of specialty coffee.

But before we get into the interview, a brief explanation of what he’s been up to in the lab and a summary of his initial findings are necessary to set the stage.

Who the hell is Mark Al-Shemmeri and why should we care about his research into coffee bean motion?

Good question! For starters, like you, he’s into adult beverages, having worked part-time as a barista while pursuing his chemical engineering undergrad, and then following it up with three years in craft beer production after his graduation.

He’s now returned to academic research — this time doctoral, and focused coffee. While working on his Engineering Doctorate thesis titled “A Data Driven Approach to Process and Product Simulations of Coffee Roasting,” he published a paper in Food Research International in which he utilized a technique called Positron Emission Particle Tracking (PEPT) to study how coffee beans move within a drum roaster — specifically, the Aillio Bullet R1 V2.

Why study coffee bean motion? Because while those little nuggets of flavor are getting tossed around in the hot drum, they will heat up “differently” depending on their location within it.

For instance, if they are pinned against the wall by centripetal force, they will be exposed to more rapid, localized conductive heating.

More time hanging in the air? That means relatively more convective heat.

Therefore, if one can identify variables affecting bean motion in predictable ways, one can also predictably alter the amount of time the beans spend pressed hard against the steel, flying solo in the air, or snuggling up to their little friends in the ‘bean bed’, where they can more equitably share each other’s warmth.

In short, physics matter. And although we don’t yet possess a good understanding of how these differences in heat application ultimately affect flavor, Mark’s work shows that by manipulating three key variables — batch size, roast degree, and drum speed — we can gain more control over roasts, and therefore experiment with flavor in ways that are deliberate, reproducible, and grounded in science.

It’s the kind of research that points the way forward without pretending to have all the answers. And that’s why we like Mark.

PEPT for Non-Physicists

So, we realize that most of you reading this aren’t scientists, and “Positron Emission Particle Tracking” probably doesn’t mean anything to you.

Fortunately, you don’t need to be an expert in physics to understand PEPT’s utility in coffee bean research. PEPT involves labeling a particle (in this case, a coffee bean) with a radioactive isotope emitting a unique signal that can be detected using technology similar to what is used for PET scans in medical diagnoses.

But unlike PET scans, which capture a snapshot image of the body to diagnose illness, PEPT tracks the movement of a single particle over time as it travels within a particular space, such as a drum roaster. Using this technique, Mark was able to render maps of bean trajectories — how quickly they move, where they tend to cluster, and how much total time they spend in each region of the drum — with great accuracy, allowing us to “see” inside a coffee drum in a way never before possible.

A brief overview of the study.

Take a look for yourself. Below are graphical representations (beautiful eye-candy!) of how your beans will be moving within the drum of a Bullet R1 at the end of a 900g roast, after first crack, on D1, it’s slowest drum speed setting. Figure A represents individual particle trajectories, Figure B represents the total time spent in regions of the drum (occupancy or residency time) and Figure C represents the velocity of the roasted beans in specific regions of the drum.

Visualization of PEPT data: (a) Cartesian coordinates, overlayed with individual Lagrangian particle trajectories, (b) Eulerian occupancy profile and © Eulerian velocity profile — data correspond to 30 mins of data for 900 g of roasted coffee with a drum rotation speed of 42 rpm (clockwise).

It’s okay if you can’t make much sense of these renders at first glance. Those who want to get super geeky should head over to the study itself, or read Mark’s breakdown of his results on his blog.

For our purposes, however, it’s enough to understand the summary of his results below, looking at the three main variables he investigated.

Batch Size

Roast Degree

Drum Speed

Seems simple enough. But it’s probably got you wondering how you can use this information to roast better coffee, right?

As you will soon hear from Mark himself, the answer is that it is all up to you.

In this study, you looked at three major variables — batch size, drum speed, and roast degree. Can you explain how these parameters affect the coffee’s behavior in the drum?

“The research data show us that each of these variables can be used to inform roasting best practices. The study firstly aimed to understand the effect of batch size and drum speed within realistic operating conditions. But because coffee’s mass, volume and density change during roasting, the study also aimed to show that these changes may have a more profound effect on particle motion than the latest anecdotal evidence suggests.

One of the key learnings here is that fill volume (how much space the beans occupy in the roasting chamber) has a greater impact than bean mass on particle motion, and that there are two factors influencing fill volume — initial batch size (more mass, more beans) and roast degree (greater bean volume from green to part-roasted to roasted). Knowing how these two factors interact with drum speed to affect particle motion offers an additional level of control over coffee’s roasting development.”

It’s clear from the study that the fill volume of the drum has a dramatic impact on the way the beans move within it. When you first introduced your research to me, you mentioned that you felt as though we may be missing the point by measuring batch size in terms of mass — in grams — rather than as volume. Can you talk a bit about that?

“Coffee is an agricultural product, whose properties depend on growing conditions and post-harvest technologies, so coffees often vary in mass, volume and therefore bulk and packing density. A coffee with a greater packing density (i.e., higher number of beans per kg) will have a lower fill volume for a given batch mass but will have a greater total surface area (available for heat transfer). If for a second we consider roasting conditions with a specified thermal load (i.e., burner setting), the batch with greater surface area will develop at a slower rate.”

Why might a batch with greater surface area develop at a slower rate?

“If we consider convective heat transfer rates, there are two major considerations: (i) the heat transfer coefficient and (ii) the heat transfer area. The heat transfer area can be approximated using the number of beans and the individual bean surface area, but there will also be an effectiveness factor that accounts for how much (or how little) of the bean’s surface is heated at a given time. If we then think about a coffee that is smaller in size, it will have a lower bean volume and so, a given batch mass (let’s say 1kg) will occupy slightly less volume, be more densely packed in the roaster and will have a greater surface area.

Think of the bean bed as having a permeability like in espresso percolation. It’s easy for water to flow through a puck of coarse grinds. For finer grinds, even though they are easier to extract, excessive channeling might lower your overall extraction yield because the water didn’t flow over it. It’s similar here in roasting.

With there being a larger total surface area, but also a reduced availability of the beans, greater heat transfer coefficients (i.e., inlet air temperature or burner setting) are needed to maintain equivalent heat transfer rates.

As the roasting system scales, the difference in size of a coffee to some reference (or within a blend) and the difference in available heat transfer area becomes more significant. Of course, it is difficult to untangle the effects of packing density and bean size whilst there are so many other variables involved, such as intrinsic (cellular) density and chemical composition, although I am working on a study of size-sorted beans from the same lot that could help answer this…”

Where does drum speed fit in?

“Data in the study show that for a specified mass of beans, drum speed can be used to vary the motion of coffee beans in the roaster and therefore could be varied during roasting to optimize their motion, as bean density and volume shift, and modulate heat transfer phenomena.

It’s something to consider when roasting different batch sizes in a roaster — it’s obvious to most roasters that the thermal load should be adjusted to maintain a reasonable time-temperature profile, although heat transfer depends on particle motion, which in turn depends on batch size, so drum speed could be varied to realign particle motion and support batch homogeneity.

Increasing drum speed for a specified mass of beans will propel more beans into the region of the drum where convective heat transfer dominates, but also increases the amount of time at the wall, near the main heat source for drum roasters, where conductive heat transfer dominates. So, in reality, drum speed is a lever that can help bias convective or conductive heat transfer rates!”

Experimental PEPT data detailing (a)-(b) Cartesian data overlayed with individual particle trajectories, ©-(d) occupancy profiles and (e)-(f) velocity profiles for 900 g of part-roasted coffee operated at (a),©,(e) 42 rpm and (b),(d),(f) 78 rpm, depicting the impact of drum rotation speed.

That’s an interesting finding. Many of us Bullet users (myself included) barely consider drum speed if we consider it at all. We pick one speed and stick with it. Are you suggesting we should consider changing it up?

“Go and explore as much as you can. It’s easy to get drawn into a single way of working. Risk a few batches to roast with a single power setting throughout, starting at a drum speed setting of 5 up to 200°C (392°F) and then change the drum speed to 1, and repeat this setting for 5 and 9 (be careful with the rate of development though).

You can also try roasting two distinct coffees with different bulk densities using the same roasting profile and matching total batch surface area (rather than matching batch mass). See how this approach compares to matching batch mass in grams. See how well they match as compared to matching mass with grams.

I don’t intend to prescribe new ways of working, but to implore people to explore, compare and taste every batch to guide their exploration.”

You mentioned that some effects may become more or less prominent as the roasting system scales. Do the concepts in this research scale up to drums with bigger diameters? Would individual beans in a much bigger roaster behave so differently that we wouldn’t be able to extrapolate and model the behavior of the beans?

“As many scientific studies are specific to certain experimental conditions, as was this one, it is not recommended to chase a specific number or quantity. But with that said, the study can be directly used to inform roasting strategies.

There is nothing extraordinary about the physics of the Bullet’s drum; it’s a rotating drum with fixed vanes like so many others. It just so happens that it is easy to use and fits inside of a scanner, which made it ideal for this research.

So the conclusions of the study are relevant to most rotating drum roasters, the main difference to other roasting systems and batch sizes will be in the relative magnitude of the parameter’s effect on particle motion and therefore heat transfer rates.

For those who want to explore, load an ambient temperature roaster on P0 or with the burner off, and with different batch sizes of green and roasted coffee, vary the drum speed and take a look into the sight glass to see the effects on particle motion.

On the other hand, with regards to modeling, we’ll need to have more data about such things as friction coefficients and restitution coefficients from this system, but we should be able to scale it up for modeling different drums. And to be honest there is a lot of data out there already about the physics of rotating drums.

That’s a lot of what my research is about, trying to find out how these roasters actually work.”

Besides avoiding roast defects, you haven’t discussed how bean motion may affect flavor in the cup. What have your own personal roasting and cupping experiences told you about varying drum speed on the Bullet or other drum roasters?

“This is something that I didn’t have time left to fully explore, so I don’t have robust enough data to confidently share. However, from the few roasting trials I did, and what is implied from the concepts we’ve spoken about earlier on and in the articles, is that if you are operating in a flow regime that causes more conductive heat transfer to the bean, more localized, surface heating will occur. Similarly, if your convective heat transfer rates are very high, then you’ll also see rapid surface heating. In both of these instances, the difference in temperature will result in a difference in moisture, porosity, color and therefore flavor and aroma. You’ll end up with a batch that’s not particularly uniform. That’s not necessarily a bad thing, but it can bring both harsh acidity and excessive “roasty” attributes that clash, making for a bad cup. But if done in the right way, it can provide balance. This is the reason blends often work so well!”

What first got you interested in this kind of research?

“There are many people in the industry asking fundamental questions about coffee roasting. These questions are quite often the right ones to be asking and are necessary to drive the industry forward, yet without the appropriate tools or methods, even though these intentions are necessary to drive the industry forward, anecdotes and misinformation might lead the community into rabbit holes and inhibit progress. It’s tough to know what to focus on without more empirical research to point the way.”

Can you give any examples of such rabbit holes?

Defining the ‘phases’ of a roast is a great example. How many phases do you consider? What events are happening over discrete time intervals? What events are happening simultaneously? What is even happening? I’m going to ask some more questions…

The drying phase…it starts when the beans reach a high enough temperature for diffusion of liquid and then the phase change of water from liquid to vapor but, when does drying finish? For most people’s definition, probably at around 170C (338F), when the beans actually still have 3–5% moisture (down from 8–12%). If you look at how the moisture of coffee changes during roasting, you’ll see that it decreases for the entirety of a roast down to 0.5–1.0% depending on how dark you go. So, is the drying phase the only phase? Well, kind of, but no.

Yellowing… anyone got a color meter on their roaster yet? Maybe there’s an app for this out there somewhere? What if we use the pantone color sample cards? Sunflower yellow, or sunshine yellow?

The Maillard phase…Maillard reactions need the right environment to be created before getting started — at what temperature, moisture and time do these reactions start? When do they end? Are we focusing too much on the macro-scale and forgetting what’s really happening at the micro-scale?

First crack is a great process indicator because we can hear it. But what else is happening during first crack? Data suggests that we’re over-celebrating first crack…which leads me nicely into the final phase that we’ve recently started to romanticize.

Development time… what does the word develop mean? When does the bean start developing? Many people define development time as the time after first crack. So, does that mean that the bean doesn’t develop before first crack? Does time to first crack even matter?

Hmm…I’m going to need a bigger spade…”

You use the word “explore” frequently, and when we talked before, you were adamant about not simply “chasing a number.” Many people believe “data driven” means exactly that, chasing a number…

“Coffee has been researched for more than 50 years and I know there’s going to be more than 50 years of coffee research to come. That’s the beauty and the curse of coffee: there’s always a new variety, an exciting post-harvest process, a new roasting technology, a new flavor space we didn’t know people would enjoy (I’m looking at those peaberries and super-naturals that were considered defects until recently). We could spend years researching a specific coffee and a specific roaster (which I did for four years) and still not have all of the answers on why that coffee behaves the way it does in that roaster (trust me, I don’t).

It’s really important to be okay with the fact that we might never know if we got the best out of the coffee or not. The main thing is that we tried, right? It’s difficult to explore extreme roasting conditions when you’re in a production setting but that shouldn’t stop you from continuously improving or developing your approach in small steps.

The truth is that roasting is more forgiving than many of us would like to admit. If you have a 4 minute temperature profile and your roaster can perfectly replicate it, then chances are that the difference between 3:45, 4:00 and 4:15 along that profile is not going to be noticeable for a consumer once you’ve dialed that coffee in for an espresso destined for an oat milk flat white.

Hang on, how many of you read 4 minutes and got confused? Honestly, I’ve had really tasty coffee roasted in 4 minutes. It’s all about context. I roasted 50g of an anoxic washed pink bourbon (from El Carmen (Colombia) via Raw Material) in a sample roaster and it was an absolute juice bomb. Stop getting hung up on the number!

When writing scientific articles, the methodology needs to be described in enough detail so that someone could replicate the study and get the same results. For coffee, that’s a little difficult. These studies are meant to be informative not instructive, which is why validation of roasting studies is really important. For some things, we will have to trust others and their diligence, especially if they used really expensive (time and/or money) methods to test their hypotheses. But, more often than not, seeing/tasting is believing, so try it for yourself…”

Are you working on any new projects? Where do you see your research taking you next?

“I’ve now finished my studies for the EngD and I’m still finalizing my thesis, which will document the major outcomes of the last four years of roasting research and will eventually be open access. I’m also writing several articles that will help prevent people from falling down the rabbit holes that frequent the specialty coffee community.

Aside from academia, I have recently moved to the Netherlands, where I’m working as an R&D Technology Specialist at JDE Peet’s, working on the implementation of new coffee technologies with a huge platform to invoke change.”

*** For a deep dive into Mark‘s research, take a look at his blog on Medium as well as his ResearchGate profile, where all of his works to date are open access.