Coffee Data Science
Coffee Density and Weight Loss in Roasting
Casual Data Exploration
I have been collecting roast statistics for a few years, but I haven’t used them for anything. I didn’t feel like I had enough data until now to plot it. So I have initial data on density and weight loss from roasting green coffee. All of this data should be considered exploratory data, and I hope they inspire you and me to better collect data on roasting.
Most of my roasts were blends of two bean types, so it isn’t worthwhile at the moment to split the data by roast.
All roasting was done on a Hottop using the same profile. The only thing I changed was the end time. I typically kept roasts between 1 minute and 2 minutes past the first crack.
Measurement
I collected a few measurements on time and ambient temperature, but density and weight loss seem to be more important than other data points.
For density, I used a simple graduated cylinder of 80ml. I filled it, shook up, and filled it more if needed. Then I weighted it (density = mass / volume). This method could have been done more accurately, but I accepted the error as long as I used the same technique for all my roasts.
For weight loss, I measured the roast before and after using a scale accurate to 0.1 grams, and the roasts ranged from 200 to 350g each.
Data
I plotted this data for regular roasts, yeast, and robusta. Regular means roasts that are arabica. Then I have anything that used yeast to process the beans and robusta.
Yeast seems to be the wild card, and robusta doesn’t have enough data. So let’s focus just on regular arabica beans:
There’s a pretty decent linear fit, and there could certainly be some error in measurement.
I also looked back on data collected using the metal cup, and the technique seems accurate in fit to itself, but it had an error relative to the yellow tube.
Looking back at the metal cup for measurement, I had some roasts that were around 9% robusta. So I plotted these mixed roasts, and they sit pretty well with the trend, probably because they were mostly arabica.
Accuracy
To check on the accuracy of the measurement, I took a single roast and took ten individual measurements on density. There is a larger than desired standard deviation (STD) on the weight which is around 4 to 5 beans but that doesn’t have a big impact on density because the test tube is large enough.
There are other methods out there to provide a better measurement. For my purposes, as long as I keep the same technique across roasts, it is useful enough from a data standpoint.
This long-term experiment has been really interesting to me. I hope to continue, and I hope for the day when this information is more useful to me in the cup.
Future experiments should include:
- Green Bean Density
- Green and Roasted Hardness
- Moisture Meter Readings
- Calibrated Color Readings
I’m also interested in any data anyone has on their own roasts especially if they have a high volume of data from roasting coffee daily.
If you like, follow me on Twitter, YouTube, and Instagram where I post videos of espresso shots on different machines and espresso related stuff. You can also find me on LinkedIn. You can also follow me on Medium and Subscribe.
