Filter Syringes for TDS Coffee Measurements seem to be Unnecessary | by Robert McKeon Aloe | Towards Data Science

Robert McKeon Aloe

Follow

Towards Data Science

--

1

Listen

Share

As I poked around the World Wide Web, I came across a small study looking at filtering an espresso sample before taking a TDS (Total Dissolved Solids) measurement. TDS is widely used to measure coffee extraction as a more objective measurement of the quality of a coffee. This small study examined using a centrifuge for sample filtration as compared to a syringe filter. They compared centrifuge, syringe filter, and unfiltered samples across 20 shots split between two coffee beans over the span of a few weeks. They were meticulous about making sure the samples were good.

However, I disagree with their findings, and I think they missed an opportunity because of the way they plotted the data. On the bright side, they seemed to have uncovered something interesting: there is a direct correlation and transformation between filtered samples and unfiltered samples. This means that even when measuring TDS without a filter, you can estimate the filtered TDS with good accuracy. In essence, using a costly centrifuge or costly syringe filters is unnecessary for measuring TDS at home or in cafe settings; the unfiltered measurement is reliable.

Previous data from Socratic Coffee suggests filtering results only improves precision. I took previously found filtering not to be useful enough to justify the cost. Ultimately, a centrifuge or syringe filters are quite expensive for a hobby that’s already very expensive.

This study took 20 samples across multiple days, multiple shots, split between two coffees. They then computed probability distributions to show the unfiltered samples were problematic. They showed the data in a table, but they didn’t plot the data as a scatter plot.

I was curious about the data, so I pulled it to take a look. If their data proved filtration was necessary, I would be more inclined to filter all of my results. However, there is a major cost hurdle to filtering samples which is syringe filters are expensive, and even if a centrifuge wasn’t expensive, it takes more time.

They made the raw data available, and I started with some scatter plots.

A simple best fit linear line reveals an R^2 value of 0.68, which is quite high. So let’s cut the data again to the two separate roasts to see how that changes these trends.

Now the linear fits for each coffee is much stronger, which indicates a transformation will get you from filtered samples to unfiltered samples. Let’s line up samples and sort them but split by coffee:

Again, they trend with each other, so we should calculate correlation. Correlation is a metric to help understand how closely two metrics trend with each other:

For the Sump Ethiopia Roba, the correlation coefficient is too high to ignore. Even the Sump Alma Negra correlation coefficient in the high 80’s is telling that this link between unfiltered and filtered is just an transformation.

Applying the transformation to each data group, we can compute the percent error (|x — x_GroundTruth|/x_GroundTruth) for the regular measurement and then corrected based on the best fit line transformation.

Assuming the centrifuged samples are ground truth, the actual percent error after correction isn’t much to be concerned about.

While the author of the study did very well controlling variables and collecting the data, their analysis did not reveal the real findings which is the evidence that filtration of coffee samples for TDS measurement might be unnecessary. The uncertainty is the possibility that every bean has a different correction curve, and if you were quite concerned, a larger study across multiple beans would help confirm this theory.

If you like, follow me on Twitter and YouTube where I post videos of espresso shots on different machines and espresso related stuff. You can also find me on LinkedIn.

Coffee Bean Degassing

Deconstructed Coffee: Split Roasting, Grinding, and Layering for Better Espresso

Pre-infusion for Espresso: Visual Cues for Better Espresso

The Shape of Coffee

To Stir or To Swirl: Better Espresso Experience

Spicy Espresso: Grind Hot, Tamp Cold for Better Coffee

Staccato Espresso: Leveling Up Espresso

Improving Espresso with Paper Filters

Coffee Solubility in Espresso: An Initial Study

Staccato Tamping: Improving Espresso without a Sifter

Espresso Simulation: First Steps in Computer Models

Pressure Pulsing for Better Espresso

Coffee Data Sheet

The Tale of the Stolen Espresso Machine

Espresso filter analysis

Portable Espresso: A Guide