We have been somewhat spoiled by the detailed charts afforded by laser particle analysis, but there is still useful information to be gleaned from sieve analysis, for those on a budget. Sieve analysis for coffee grinds has been carried out for at least the last 70 years.
The standard method is to shake 100g (or maybe 10g for a small sieve, or any weight that translates easily to a percentage) for five minutes, rotating and tapping the sieve. This can be done by hand, or a mechanised sieve shaker used. There are several manufacturers of sieve shakers but W.S. Tyler’s “Ro-Tap” has become synonymous with sieve analysis, much like Hoover has to vacuum cleaners. Such machines are cheaper than laser diffraction, but still expensive for a home user.
The Single Sieve Test.
We can use a single sieve, of a known mesh, to help us establish our median particle size. E.g. if 45-55% of our ground weight passes through a 707 micron sieve, then we know that 707 microns is around our median particle size (we still don’t know our ‘mode’, which is the most commonly occurring particle size, but that is beyond the remit of this particular test).
A single sieve can also be useful if certain assumptions are made about the anticipated grind distribution. For example, the SCAA’s cupping grind is described as “25-30% passing through a #20 (840 microns) sieve”. Assuming the grinder’s output produces a ‘normal distribution’ or bell curve, this percentage implies a median particle size of around 700 microns.
Even if the sieve mesh is not known, a kitchen sieve can still be used to assess relative fineness & coarseness, for a given grinder setting & coffee (some coffees grind up tangibly finer/coarser at the same grinder setting). Or, a sieve may be useful for removing an unwanted portion of a grind – say you are getting lots of silt in a metal filtered/unfiltered brew, but grinding coarser leads to under-extraction? Grind coarser to mitigate fines, then sieve out the largest boulders to reduce the median grind size & give extraction a lift.
The Two Sieve Test.
Ideally used for analysing grind when the median is known at, or close to, the mid point between two selected sieve meshes. This is the methodology described in the SCAA home brewer certification http://scaa.org/?page=cert2 (scroll down to the section on “Grind”).
The idea being, to catch around two thirds of the ground weight between two sieves, one twice the mesh size of the other. If these two sieves capture 68% of the ground weight it may be assumed that the median is x1.414 the finer sieve & /1.414 the coarser sieve. For example, if we used a 500 micron sieve and a 1000 micron sieve, capturing 68% of the ground weight between these sieves, the median may be close to 707 microns. This can be confirmed with, yes…you guessed it…
The Three Sieve Test.
Using a 500 micron, 707 micron and a 1000 micron sieve, knowing our median grind size was close to 707 microns, would indicate a ‘normal distribution’, to one standard deviation of the mean, if we catch 34% above & below the 707 micron sieve. Perhaps we would like to have some idea about how we stood with respect to two ‘standard deviations’? Drum roll please for the…
Four Sieve Test.
Probably the most common & standardised test. In this test we would add another sieve at 1.414 times the 1000 micron sieve (in this example, or twice the size of the median sieve as 1.414 is the square root of two). We would then hope to see 2.5% or less of our ground weight being caught in the 1414 micron sieve. This would equate to 95% of our grind falling with two standard deviations of the mean. Of course, if the grind is coarse enough, a fifth sieve (354 microns in this example) can be added above the pan to confirm the two standard deviation percentage at the finer end.
For more detail, or to test completely unknown grinds, a full twelve sieves can be used, each being x1.189 (the square root of 1.414) smaller/larger than the next.
In the example above sieves were selected at 298 microns, 421 microns, 595 microns & 842 microns. 31% landed in the pan, so was smaller than 298 microns. 38% was captured on the 298 micron & 28% on the 421 micron sieve, with 3% sitting on the 595 micron sieve, 0% on the 842 micron sieve. This is absolutely known, no assumptions (see the red lines on the graph).
This might not seem to tell us much to start with, but let’s look closer. 38% was captured between the 298 & 421 micron sieves (69%-31%). In a normal distribution (bell curve, blue dashed line on the graph) we would expect to see 19% either side of the median at half a standard deviation. Hey presto! That is what we have as x1.189 = half a standard deviation and 298 x 1.189 = 354 microns, also 421/1.189 = 354 microns.
Our median is therefore ~354 microns. We know we have 3% of the grind caught at 595 microns (one and a half standard deviations from the mean), so it’s pretty much a certainty that we have 2.5%, or less, of the ground weight at 707 microns, or at two standard deviations from the mean. A high likelihood, therefore, that we have 95% of the ground weight falling between two standard deviations of the mean.
Problems arise with sieving very fine grinds (less than 250 microns), so this remains a strong assumption at the minus two standard deviations mark.
Problems also arise from there not being a standardised convention in plotting distribution graphs derived from sieve analysis. The graph above could just as easily be presented as a curve like this…
…or, even like this:
In other words, you need to be aware of the convention being presented and whatever normalisation has taken place in order to make comparisons between graph data from different sources. This data should be explicitly stated, or data points shown.
There can also be glitches where the intervals between sieves are not consistent, or where lots of sieves are used to try and conjure the effect of greater detail…remember, sieve data is useful, but is somewhat of a blunt instrument compared to laser diffraction.
In the graph below, the thick block line is sieve data, taken from seven sieves, the first six are factors of 1.189, the seventh jumps from 1000 microns to 1415 microns (x1.414). It looks great & indeed, when compared to other data from the exact same method/coffee/grinder, it is informative. However, it doesn’t make the basic information easy to read. If just three sieves of had been used (595, 842 & 1189 microns), we would much easier see the relationship between the revised sieve results (thick dashed line, peaking at 32% & 842 microns) and and a normal distribution (thin dashed line, peaking at 50% & 842 microns.
So, this grind does not seem to conform to a normal distribution to within one standard deviation of 1.414, there is a higher proportion of small particles relative to the median (or, a high incidence of large particles is dragging up the median, depending on which way you look at it) therefore, the standard deviation is larger than 1.414. But don’t panic, it doesn’t mean that anything is wrong, or the grinder is useless! Far from it, the grinder might still produce delicious brews (this one does), we have just identified the grind profile characteristics…which was precisely the point of the exercise.