Engineers Gone Wild

One of the things that has been troubling me for a while is better understanding the differences between roast profiles. A change in how a coffee has been roasted, even when the final temperature is the same, can be dramatic. However, looking at the temperature data doesn't always tell the story. Here is a good example:


This is two Kenya beans that were roasted differently. The differences in the curves are quite subtle even though what I was changing in terms of temperatures, times, and fan speeds were very different. Even roasts that were absolutely terrible, like the experiments I did on high drop temperatures, show barely a difference in the roast curves. Some of this is due to the fact that the scales are large since you cover a sizable temperature range in the roast, but some of it is just subtlety in the temperature profile producing large effects in the taste of the coffee.

I was playing around in Excel the other day trying to determine if I was just fooling myself that I have any control over the roasting process. I decided that I needed to look at the data another way, so I decided to take the derivative of the temperature data and see if anything illuminating happened.

Without getting into the mathematical details, the 1st derivative is essentially the rate of change of temperature as a function of time. Taking the same two coffees, plus a third example, you can see the data this way:


The x-axis is still time, but now the y-axis is the change in temperature per unit of time. You can see when the coffee is dropped in at time=zero, there is a huge temperature drop (this is why it is a negative number), and eventually it recovers to something between 10 and 50 degrees of temperature increase per minute until the end of the roast.

One interesting fact immediately popped out after looking at a few charts like this. On the experiments I did with a higher drop temperature, the rate of temperature decrease when the beans are added is substantially higher than when the coffee is dropped at a lower temperature. I am not sure why this is (or if it matters), but I could imagine that that heating element is still on and going strong for the lower drop temperature, and in the other case the environmental temperature and thermal mass have stabilized such that the heater is not on as much. Anyhoo....

The more interesting data is when you zoom in around the region from first crack to the end of the roast:


Don't worry too much about the noisiness in the data; that is typical of this sort of data analysis and the sample size I have used. A couple of things are interesting here. First, I strive to keep a roughly 10 degree per minute temperature rise from first crack to end of roast. You can see that these roasts are doing pretty well. Looking at the earlier times, you can see my attempts to ramp down the heat as first crack approaches in order to have a nice slow, steady progression through that phase of the roast.

The dots on the graph indicate where first crack began on each roast. You can see that the roast of the Kenya Tambaya had a slower ramp rate at first crack than the other two. This is exactly what was intended. I am trying to develop a roast profile that very gently enters first crack. By looking at these curves, it is evident that it is working. The other two Kenya roasts were roasted with the same profile, hotter going into first crack, and the data shows this clearly.

I'll be looking at these derivative curves more as time goes on. It really hones in on important parts of the roast. Amusingly, I came across some other posts on the web where people are actually building micro-controllers for their roasters that track this same 1st derivative data as a hands-on tool for roaster control. Mediocre minds think alike and all that.

Mathematics aside, you are probably wondering how the coffee tastes with the slow entry into first crack. The initial data is extremely promising - very sweet coffees with well integrated acidity and no grassiness. Stay tuned for more coffee reviews.