The impact of filtering on refractometer TDS and extraction yield measurements - Page 3

Beginner and pro baristas share tips and tricks.
User avatar
NoStream

#21: Post by NoStream » Feb 14, 2019, 9:51 pm

For what it's worth, I collected some data and noticed some difference. Not a scientific study but enough to convince me to filter. I also didn't find the effect from filtering to be consistent. So you can't just subtract or multiply to correct. I use generic filters, so my cost is not very high.

https://docs.google.com/spreadsheets/d/ ... p=drivesdk

nuketopia

#22: Post by nuketopia » Feb 14, 2019, 10:17 pm

Yeah, for me, the point of filtering is to achieve consistency in my measuring process. This allows consistency of measurement over time.

I don't do refractometer measurements all the time, usually when I have a question or trying something new. So it isn't a major cost and a bag of VST filters doesn't represent a lot of cost.

I'm not worried about different arguments of TDS or how it is measured so much as I'm interested in maintaining consistent results in my process. It is important to note though, when discussing results of the refractometer with others, that filtering is or isn't used so we're all comparing apples and apples.

CwD

#23: Post by CwD » Feb 14, 2019, 10:21 pm

NoStream wrote: I also didn't find the effect from filtering to be consistent. So you can't just subtract or multiply to correct.
That's definitely one of the biggest things. With my data, the MAD for centrifuging or filters in my testing was .01, but nine times less precise at .9 for unfiltered. And a .11 spread in how far apart from VST filters the unfiltered was, so trying to subtract the systematic variation of .38 could still easily put you way too high or way too low.
AssafL wrote:I think filtration to achieve precision is very clever. Too bad it is at a cost of accuracy.
No, unfiltered is is no way shape or form more accurate. A refractometer isn't measuring the amount of stuff in the liquid, it's only capable of measuring the dissolved stuff. The dust isn't changing the refractive index to get an appropriately higher reading. It's just blocking the light and causing the reading to spit out total nonsense that's not reflective of what's in the sample in any way.

Filters, centrifuging, or unfiltered sampling all give up some accuracy versus using a full on lab grade drying oven every time. But filtering or centrifuging give up no accuracy versus unfiltered.

User avatar
AssafL

#24: Post by AssafL » Feb 15, 2019, 3:33 am

CwD wrote:No, unfiltered is is no way shape or form more accurate. A refractometer isn't measuring the amount of stuff in the liquid, it's only capable of measuring the dissolved stuff. The dust isn't changing the refractive index to get an appropriately higher reading. It's just blocking the light and causing the reading to spit out total nonsense that's not reflective of what's in the sample in any way.

Filters, centrifuging, or unfiltered sampling all give up some accuracy versus using a full on lab grade drying oven every time. But filtering or centrifuging give up no accuracy versus unfiltered.
It depends what is meant by EY. If EY is the stuff dragged from the puck - it isn't very accurate if you filter some of it out. Hence the idea of a homogenizer which converts the suspension/emulsion of variable sized particle into a homogenous/uniform one. To achieve accuracy.

Perhaps it shouldn't be called EY. Perhaps it should be EY@<25uM. So that it is EY adjusted for particle say over 25uM. Or whatever the filtration size is.

Yet another assumption made is at what size particles do flavonoids stop being flavorful. Even after the filter there is particulate. I assume it has little flavor yet Contributes to Brewster's angle.

One can RO the coffee or distill it, or use a ceramic membrane filter to get rid of all particulate albeit I am not sure what would be left. I definitely sense (with no data points) that measuring the EY of Perfectly filtered or distilled coffee that (1) It will have the highest precision of measured RI and (2) Totally wrong EY.

Hence the homogenizers (looked them up on eBay and wow those are 1000$ and up).

BTW. Don't know where the argument that solids in a suspension don't contribute flavor. I looked it up in my books but couldn't find the reference. It may very well be that a 24u particle that passes the filter is chock full of flavor yet a 26uM one is a tasteless brick (or whatever the paper filters pore size is) - but I find that hard to believe without some hard evidence.

I make ice creams as a hobby (I use a pacojet) and solids contribute much of the flavor. It may not be applicable but ice cream is an emulsion suspension foam. Espresso is also an emulsion suspension but only the crema is also a foam. Is that enough to make the solids tasteless? Dunno.

In any case, we are measuring an ill defined "EY", using some "filtering" that is an uncontrolled variable, and doing so with a particularly accurate device for measuring Brewster's angle.

If I were to look at VST vs the Atago approaches (their response), the VST one is about precision at all costs (Both EY and TDS are wrong but consistent) whereas with Atago the answer is "TDS in what's in the cup and your mouth" and your process (filtering, centrifugal acceleration, homogenization) will affect the results. And it is up to you to decide how many EY points are you willing to sacrifice to make the process easily repeatable.

A more honest approach would be to avoid TDS and EY altogether (which are well defined terms and could therefore be measured pretty accurately by more patient hobbyists) and call them X and Y. At that point one would compare VST 12,9 and we'd all know what that meant. Just like Agtron and Tonino have their own scales and don't try to abuse well defined color charts. Is expecting honesty just too much?
Scraping away (slowly) at the tyranny of biases and dogma.

jpender

#25: Post by jpender » Feb 15, 2019, 4:26 am

AssafL wrote:Don't know where the argument that solids in a suspension don't contribute flavor.
That's not the argument. The whole point isn't to determine an "honest" measure of what has been extracted from the coffee beans, it's to find some measure that correlates to taste. Whether or not undissolved solids affect the taste, mouthfeel, or both is beside the point.

It's possible that using total solids as a measure would work just as well and that the reason we use TDS is because that's what the most convenient tool we have measures. If you were able to demonstrate that total solids provides a significantly better correlation to taste than TDS then you would be onto something. If you couldn't then there's no value in it.

CwD

#26: Post by CwD » Feb 15, 2019, 4:39 am

AssafL wrote:It depends what is meant by EY. If EY is the stuff dragged from the puck - it isn't very accurate if you filter some of it out. Hence the idea of a homogenizer which converts the suspension/emulsion of variable sized particle into a homogenous/uniform one. To achieve accuracy.

...

Yet another assumption made is at what size particles do flavonoids stop being flavorful. Even after the filter there is particulate. I assume it has little flavor yet Contributes to Brewster's angle.

...

If I were to look at VST vs the Atago approaches (their response), the VST one is about precision at all costs (Both EY and TDS are wrong but consistent) whereas with Atago the answer is "TDS in what's in the cup and your mouth" and your process (filtering, centrifugal acceleration, homogenization) will affect the results. And it is up to you to decide how many EY points are you willing to sacrifice to make the process easily repeatable.
You're missing the point. The point is that it's fundamentally impossible to measure suspended non dissolved solids with a refractometer. Failing to filter them out doesn't give you a measurement that's more accurate to true ey as found by drying. It gives you an entirely useless number by jamming an instrument with something it's fundamental incapable of measuring. The higher tds given from an unfiltered sample is in no way whatsoever from the device measuring and including the suspended solids. It is entirely a measurement error caused by something physically in the way, and has no relationship with what amount of the drink is undissolved solids. It needs to be filtered out because it gives nonsense with no relationship to the sample. It gives you something much higher than the real tds+tuds because it's just blocking the lens and screwing up the path of the light. It is not a tradeoff. It's just not based on any real part of the drink. The instrument cannot check undissolved solids, not VSTs, not Atagos, nothing based on any loosely related principle ever fundamentally could. You're setting a weight on top of a ruler because you want to know how heavy it is too in addition to length, but rulers can't measure weight so you just end up not knowing the weight and messing up your length measurement by bending the ruler.

Filtering isn't perfect since we'd like to know how many undissolved solids are there too. But not using a filter does not give you a reading including undissolved solids. The only way we can find that number with existing technology is a full lab grade drying oven. Which would be ideal to do for each shot, but is just too far out of reach for almost all users. Without access to that, filtered refractometry is the closest we can get.

User avatar
AssafL

#27: Post by AssafL » Feb 15, 2019, 7:34 am

jpender wrote:...it's to find some measure that correlates to taste. Whether or not undissolved solids affect the taste, mouthfeel, or both is beside the point.

It's possible that using total solids as a measure would work just as well and that the reason we use TDS is because that's what the most convenient tool we have measures. If you were able to demonstrate that total solids provides a significantly better correlation to taste than TDS then you would be onto something. If you couldn't then there's no value in it.
Yes! So why market the accuracy of the refractometer? It is like marketing a 1 ppm voltage reference in a 4 digit AC voltmeter. Useful voltmeter, stupid waster of a reference.

So there are two "sets". There is Total Solids (say TS) which (ignoring water chemistry) will be EY x Dose.

And then there is TDS which is measurable with a refractometer. It would work the same way as EY but for accessible solubles only.

And then the conversion from the TS EY set to the TDS set is where I am not clear on the theoretical model on. Now it may be empirical, but in that case one must derive the numbers from a whole host of tests.

Personally I would expect VST and Atago to provide the testing they did for the conversion between the sets for peer review. Do dark roast light roasts robusta arabica humidity affect outcome?

For example, can two results be compared if the drop temp in 2 degrees off? Or does "set conversion" artifacts completely mask any real EY differences? Can we even say precision (let alone accuracy) be maintained between two different coffees? Or two different grinders?

Filtering helps make two successive measurements precise and repeatable. Does it make measuring two coffees comparable?
Scraping away (slowly) at the tyranny of biases and dogma.

Mrboots2u

#28: Post by Mrboots2u » Feb 15, 2019, 8:19 am

NoStream wrote:For what it's worth, I collected some data and noticed some difference. Not a scientific study but enough to convince me to filter. I also didn't find the effect from filtering to be consistent. So you can't just subtract or multiply to correct. I use generic filters, so my cost is not very high.

https://docs.google.com/spreadsheets/d/ ... p=drivesdk
I don't have the data I collected from 3/4 years ago ish. My experience was similar to yours, that there was a noticeable difference that I could not correlate accurately. this was done over 25 ish shots filtered to non filtered with a VST refractometer. Also i would get outliers that simply bore no retaliation to the filtered sample.
Again if you wish to use non filtered samples to chart where you yourself are impacting on extraction then fine, but to estimate of have a formula that arrives at a " filtered sample reading " and posting as such doesn't work for me.
I have not tasted a 25-27% shot, , I think a 18 % tds shot would blow my had off. The closest I got was 24% ish from memory, some were good, some were not, the limiting factor was the roast of the coffee itself.

CwD

#29: Post by CwD » Feb 15, 2019, 8:38 am

AssafL wrote:Yes! So why market the accuracy of the refractometer? It is like marketing a 1 ppm voltage reference in a 4 digit AC voltmeter. Useful voltmeter, stupid waster of a reference.
Because it's accurate at measuring the tds of the coffee. It tells you how much of a sample that has been properly prepared to contained no undissolved solids is not water. It isn't accurate or inaccurate at measuring the total contents of the liquid or the total material removed from the puck since it's just not a tool that measures that. A ruler isn't an inaccurate scale, it's just not a scale. Speaking of scale, you might as well complain that it doesn't tell you the total shot weight too. Just like with weight and tds you're going to need different tools to measure dissolved and undissolved solids.

Then we can extrapolate from that how much of the dissolvable material in the puck made it to the coffee and how much was probably left behind. We don't know how much non dissolvable material made it, so we compare the coffees based on what we do know. I'll argue here that "extraction yield" is actually extremely appropriate here, because we are measuring what's extracted from the coffee. The sediment does contribute to experience, but it's not really extracted as much as just moved. It's not what we're measuring or talking about, nor do we have a way to measure it on the timescales needed for dialing in shots.

It would be ideal to also know total solids and record it separately from total dissolved solids even so that we could know more about the composition of the drink, but we still get valuable information about the coffee that's comparable between all coffees and grinders from properly performed refractometry. Not comparable in a "one coffee extracted to 1.42% tds and 24.5% so I should always do that with no regard to how this flavor profile does at 1.42% tds or if this coffee brings out any unpleasant aspects at high extraction", but comparable in that you can know which is more soluble. It's not complete information as it's only a measure of one thing (and neither would having total solids too, or mass spectroscopy, or anything short of the damn quantum wave function of every quark in the drink) but it's valuable information.

namelessone

#30: Post by namelessone » Feb 15, 2019, 8:50 am

CwD wrote: Not really seeing that. The precision and accuracy wasn't any worse than VST filters, and I don't see any reason to think we'll be getting the same actual "numbers on the display" systematic difference rather than similar margins as a percent of the actual reading, which would put it below the resolution of the .01% screen. I'll probably do a whole experiment for it eventually, but I'm not planning it very soon since it's going to take like $80 worth of VST filters and I don't really like any immersion methods in the first place.

Maybe in the next few days I'll do a quick one with just like 3 trials instead of 20 just to see if it holds up to basic scrutiny, and can set up a big test later.
What I meant was, from your data (kudos for the experiment by the way - seems much better performed than Socratic's data) the *absolute* difference between centrifuged and VST filtered TDS would be significant at cupping ranges (i.e. difference of 0.05% to 0.1%). This could make QC difficult if you're comparing filtered results to centrifuged results for example.

I'm not saying that centrifuge wasn't consistent with other centrifuged results in itself.