I decided to investigate.
Part of this investigation was the result of buying a really old espresso machine, an Enrico of Italy.
I bought this on a whim because I originally was going to buy a commercial Gaggia on the cheap, but it was most likely stolen.
I found this Enrico machine for pretty cheap, and I set out to South San Francisco to buy it from an old man on an old, dirty boat.
Luckily, I didn’t get murdered and eaten.
The Enrico didn’t come with a filter basket, but I figured I would find one easy.
I was wrong; this was a common problem for the Enrico.
I did find a solution in another basket, and a friend ground down the lip so it would fit.
I also had to add 3D printed piece to keep it set in the portafilter.
In the mist of this, I decided if I keep up my bad habit of buying coffee machines, I should at least know if they are worth a crap.
A lever espresso machine is made up of three main components: a water boiler, the lever, and the filter/portafilter.
In the case of the Enrico, the water boiler was in great shape, so I was less concerned.
Espresso needs a good lever puller, but a good artisan can pull good shots from a crap machine.
At my old office, there was a Mr.
Coffee, and I pulled good shots from that piece of junk.
The filter can usually be replaced, but some times in old machines, the filters are weird sizes, so it is tricky.
OverviewFor data, I had a few machines or access to some, and I thought of a good way to measure attributes of the filters was through image processing.
My aim was to measure the following:Hole sizeVariance of hole sizeDistribution of holesVariance of the distribution of holesDistribution of distance to closest holeTotal open area of the filterI back lit each filter using an iPad on full brightness, and I then annotated each filter with a blue circle to more easily find the filter.
ProcessingI first took an image with multiple filters, but later, I found an individual image with the filter centered in the image and on the iPad was better for the accuracy and comparison.
However, this first image serves as a great platform to show the image processing done to be able to measure what I so desired.
The distance between pixels was estimated using the length of the iPad.
Processing the ImageZoom and AnnotateFilter Outlines ExtractedFilter Outlines Filled and LabeledFilters OnlyFilter Holes ExtractedSingle Filter Magnified:There is an odd pattern almost like the filter was made on a roll and flattened out.
That pattern only shows up after keeping pixels above a certain threshold.
Data Analysis1.
Hole DiameterFirst, we will look at hole size per filter.
Originally, hole size was calculated by determining the area of pixels above threshold per each hole and determining the diameter.
However, the results did not show the detail I was looking for because it was based on whole pixels.
This is a legend for those unfamiliar with box plots:Below is a box plot of hole diameter determined only by pixels above a threshold (th).
As you can see, some filters don’t show any variance in size simply because the method is ineffective in terms of resolution.
Instead, I determined an area weighted by the intensity of the light coming through the hole.
If a pixel was completely white, it was weight as 1.
This allowed for a sub-pixel estimation of hole diameter.
The colored graph is the same data as the box plot, where the scheme is a jet color scheme, so red is the maximum count in the distribution and dark blue is near zero.
The final list extended to all of my machines and a few extras.
There is a wide variance, and my standard is the VST filters.
They claim +/- 0.
02mm difference in hole size across the filter.
They certainly achieve this according to these measurements.
Let’s adjust the graph to look specifically at the distribution and not worry about specific hole diameter.
This chart shows why my Kim Express filters were terrible compared to the VST ones.
However, the Odea Giro is quite impressive.
I didn’t not expect as much.
Kompresso and Flair also performed very well, which is an ode to quality machines.
2.
Hole DistributionFor hole distribution, I computed a Delaunay triangulation between all the centroids of all the holes, and then looked at the distance for all the edges connecting filter holes.
One could assume similar distances would allow the coffee to flow evenly across the filter.
Below on the left is a VST filter’s triangulation from each hole.
The only issue is that there are two different edge lengths: square and diagonal.
The right plot is the distribution of distances between points, and this bimodal distribution doesn’t quite help present a single metric to use.
I would be more interested in the standard deviation of each modal.
Let’s see if we could use some color distributions to understand a comparison and multiple distributions on the same graph.
The box plot, on the left, is deceiving as many filters are clearly bimodal.
On the right, we can see the differences in distributions clearly.
The square grid filters are all bimodal, but Flair and the stove top ones (Moka and Keleidos).
3.
Distance to Closest HoleThe idea behind this metric is that espresso is all about fluid flux.
Water comes in, espresso comes out; every path of espresso exiting is one of the holes.
So if water is passing through any part of the puck and makes it to the bottom, one would want the distance to the nearest hole to be even across the filter.
Let’s visualize this with a single filter: VST Double1.
On the LEFT, is an image of all the holes on the filter colored by relative hole size in the jet color scheme where red is larger, blue is smaller.
2.
The MIDDLE is the same image on the left but all the holes are dilated for easier viewing.
3.
On the RIGHT is an image of the distance to the closest hole in the jet color scheme where blue is closer, red is further away.
This metric should provide a distribution easier to understand at face value as show below with a boxplot and a colored distribution.
What’s interesting here is that Flair has such a low distance to the nearest, and the Kim Single/Double have far more variance, which could potentially explain the best espresso from those filters is worse than the VST filters.
When looking at a higher level view, we can see some differences pop out.
First, there is a wide variance in what makes a good espresso filter in the difference metrics selected.
Second, the distance to the nearest hole provides particular insight for good and bad machines as does hole diameter.
Finally, the objective quality of an espresso shot is hard to determine across these machines in a fair way, but this is a good start in my opinion.
Overall, there are many more differences to espresso machines than meets the eye.
The filter has been one of the keys for me to make consistent espresso.
Stay tuned: I plan to publish all the filters in the color representation as above.
If you like, follow me on Twitter and YouTube where I post videos espresso shots on different machines and espresso related stuff.
You can also find me on LinkedIn.
.. More details