(10/28) Update: We've added some experiment results to our writeup. Since we're effectively playing with 9 response variables (data points), this is very inconsequential to what statements we can make about airflow and moisture loss. The initial results do beg two larger questions though: 1) are we measuring properly? I am curious about batch size as a result of this work along with what other measurements might ensure accuracy (batch moisture content for example). 2) does pressure buildup in the barrel have a significant influence on weight loss? This comes from an observation of two "outliers" -- not statistically proven but visually observed -- that had weight loss results different to our expectation. We'll share sensory feedback in the next few weeks along with some loose plans for future airflow work to add to these results.
Intro
Airflow has continued to be a foundational but mysterious component in roasting to me. At its extremes, I think we have a great deal of understanding of what fully releasing the hot moisture filled air out of our drum roaster does to the taste and weight loss in a batch of roasted coffee. There are also pretty obvious moments in our roasting careers – usually sample roasts that go too fast or have too much airflow – when we unfortunately cook a batch so uneven that we’ve replaced the character of our coffee with “green coffee” character. I don’t know where the information comes from, but I’ve been told up to 70% of my heat application in a roast comes strictly from conductive heat – hot air.
However, where I feel inference takes over is when you begin to ask what sensorial qualities are associated with those extremes, and at what modulation stages of our roast. I’ve seen and been told, “Open your airflow towards the end of your roast, or else introduce smokey qualities into your roast”. This might be true, but in practice, folks launching their air valves open to 100% near first crack does not correlate with the rest of how we practice roasting. Surely, if we care about preserving chemical reactions in incremental steps in heat application and reduction, the same rules would also apply for airflow. My thoughts apply to the fixed airflow – “set it and forget it” practice too. Won’t this have adverse effects at different stages of our roast curve. I also began to ponder, “was the rate of weight loss as a result of airflow even the same across our roast?” Surely, it will follow the logarithmic decline we know exists in total moisture loss of a roast.
Nonetheless, even if someone has done all of this, tested it, and been sharing those takeaways, its been too closeted for me not to want to explore this myself. I needed to confirm it. And, I was still curious what that all might taste like. And so, this study became some of those questions.
Experiment Design
In order to test this, I wanted to really keep some parameters of our roast consistent. If we remove hot air application, it needs to be replaced with convective heat. We still need to be cooking the coffee “the same” – whatever that means – across the total roast time so I wanted to follow the same rate of rise (RoR) across the roast. Bean end temperature needs to be consistent so we’re tasting the coffee that should be (externally) the same level of cooked. Total roast time, modulated roast times, and specifically development time I wanted to keep as fixed as physically possible too. What would change would be our gas adjustments to compensate for our variable airflow settings.
Below I share the baselines and profiles tested in this study. They will be called by their codes for the remainder of this post.
In table 2 below, variance is shown in relation to our near 0% airflow profile as it had the most potential for heat application throughout the entirety of the roast. FC refers to First Crack and TP refers to turning point. TBDs will be replaced with values once calculated.
Finally, here are the curves and modulation stage times for each profile (minus P6 which will be added in the final blog).
Here is Profile 6 (pink) included.
Experiment Results
To analyze weight loss as a function of airflow, I’m creating a single value to represent the degree (0-100%) and time (0 seconds to about 7 minutes) that airflow could let moisture escape. In one case, I’m using a simple linear or uniform distribution to represent moisture loss. This implies that throughout the roast, we’re assuming that moisture escapes the coffee seed evenly throughout the roast length. Then, we’re looking at moisture loss from a coffee seed as a beta distribution and creating our average based on the cumulative density function of a Beta distribution with the alpha of 2 and beta of 4 (see chart below).
Note that this is an arbitrary approximation of what we might expect moisture release from the coffee seed to look like through the entirety of a roast. There are a few limitations of this airflow score we’re creating. One of the largest being that it does not include how the build up of moisture in the barrel plays a role, how moisture can find other ways to escape, how pressure plays a role, or how flow rate of air through the valve can affect loss.
Outliers in Scatterplot 1: Baseline 3, Profile 3, Profile 6. Outliers in Scatterplot 2: Profile 1 and Profile 3.
Long story short, we have very inconsequential results at this point and a few concerning measurements of weight loss in a few batches. The unexpected weight loss scores of profiles B3, P1, P3, and P6 end up being outliers in fitting a basic linear trend line. We’d expect a linear line to fit our second scatter better since we normalize the airflow score to a beta distribution, which should remove the effects of differing moisture loss throughout a roast.
There is not much to draw from this amount of data and we are inferring far too much from too little. Beta distribution does seem to model airflow valve scoring to a better degree than a linear uniform distribution can. This implies moisture loss might be better modeled by a Beta distribution than linearly, which we were expecting to be the case. It’s disappointing that our weight loss was less than our baseline 1 for profile 3. This is where I get skeptical of experiment design, weight loss as a measurement at this batch size and the between batch moisture content to begin with.
Sensory Results
Coming soon!
Limitations and Future Work
I'd be lying if I said there were not many. To name a few:
-
9 data points is far too little to make any major claims about weight loss in relation to airflow.
-
We’re testing one coffee: varietal, region, country, altitude, moisture content and density could all be a unique factor to these results.
-
Batch size might be too small for accurate readings of weight loss (our between batch moisture may be significant at this size or our bean loss at drop or variable chaff loss… etc. could all be influencing our weight loss % a bit too much. The difference in weight loss for these batches ranges from 5 to 10 grams.
-
Too much variation between roasts (roast profiles not similar enough)
-
Beta distribution is not a good approximation of moisture loss in a roast (or airflow score is not a good approximation of the release of moisture’s effect on weight loss)
Future work should include agtron scoring as a response, moisture content readings before and post roast, as well as increased quantity of batches. It would also be helpful to categorize by coffee traits – density, elevation, and varietal to name a few.
Another factor to consider is that Profile 3 is actually an accurate measure of weight loss as a function of airflow and that the underlying unmeasured variable could be pressure. Since we maintain drum pressure in the roast longer in our baseline, perhaps that causes convective heat to secrete more moisture from the coffee seed than releasing that pressure early and attempting to rebuild presume for the remainder of the roast when there is far less moisture to play with. In the scatters, Profile 3 and Profile 6 appear to be outliers using our method of calculating airflow score.
In general, it was expected that these should have had a greater weight loss. It makes me wonder about the effect of pressure build up in the initial phase of the roast (when the most moisture is available) on the ending weight loss.