Refractometer FG Readings

Homebrew Talk - Beer, Wine, Mead, & Cider Brewing Discussion Forum

Help Support Homebrew Talk - Beer, Wine, Mead, & Cider Brewing Discussion Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

a10t2

Well-Known Member
Joined
May 24, 2010
Messages
557
Reaction score
14
Location
Leadville, CO
I've mentioned it in a few threads before, but since it affects so many home brewers I thought it was worth a new discussion. For the past year or so I've been tracking my FGs using both a hydrometer and a refractometer, and found that the correlation used in JavaScript calculators, included in software, etc. is wildly inaccurate (at least for me). I've come up with a much better correlation, and I'm curious to hear what other brewers think. In particular, if anyone else has both SG and °Bx data for multiple batches, I'd love to see how well your findings correlate.

Blog post with more info: Toward a Better Refractometer Correlation
 
How do you launch the open office version?
I don't see a document in the zip, just supporting files.

Here are some readings from recent batches.
OG OG brix FG FG brix Beersmith FG
1.052 13.0 1.010 6.4 1.009
1.045 11.2 1.009 5.0 1.005
1.045 11.0 1.010 5.2 1.006
1.049 12.0 1.016 6.6 1.012
1.046 11.5 1.008 5.6 1.008
 
Hmm, if it's downloading as a zip your PC may be doing some kind of processing. It's just a .ods file. (direct link) Maybe disable anti-virus software for a second? Or send me a PM with your email address and I'll email it to you.

Anyway, I plugged your data in and while it's an improvement, it isn't as good as I saw. The mean discrepancy for the old correlation was -2.9 points, and with the new one it's 0.7 points. The standard deviation is 3.8 points though, so it isn't a great fit.

One thing you might want to look into is that based on those five OGs, your "wort correction factor" looks to be 1.00, and you may have it set to the default 1.04. I don't know if you can change that in Beersmith.
 
I turned off Norton Internet Security and I still can't see a .ODS file.

I don't have a lot of confidence in my readings.
They just don't make any sense sometimes.
 
refractometer's are not very accurate for measuring finished
beer IMHO, even with the correction formula's.Great for measuring
wort and O.G during the brew. Just my .02$. Cheers!!!
 
You based your fit on 12 datapoints. Without wanting to be discouraging (from my experience, this is definitely something that needs looking into), your fitting equation has nine terms. I don't think a close fit with the data actually indicates much when you have only slightly fewer terms in your equation than datapoints to fit. ie, you could likely fit completely random data fairly well with that many terms. I don't have sufficient background in statistics to calculate the statistical significance of a fit like this, but I suspect there isn't much of one. With that many terms the closeness of the fit becomes pretty meaningless.

The equation will accurately predict the datapoints you have, but I don't think it's statistically clear that it will predict future datapoints with much accuracy.
 
Yup, overfitting is the main concern at this point. That's why I'm looking for more data. The only reason I'm sticking with the full cubic set so far is that it makes it easy to drop terms and play around with fits later on, rather than do a new regression every time. That the linear equation remains a pretty good fit is encouraging.

Also, with these ranges (1.036-1.106 OG, 1.007-1.022 FG) hopefully there won't be too much need for extrapolation.
 
Here are some additional FG data. The first FG goes with the first nD etc. All nDs are corrected to 20 C (WITHOUT the use of the refractomer's ATC - don't use it for beer that has started to ferment!)

These (or rather the "points": 1000*(FG-1)) are fit quite nicely (r= 0.9499 - note that's r, not r-squared) by a linear function with an rms residual of 1.1 and a peak residual of -1.7 points which isn't half bad but 7 data points isn't a lot to draw global conclusions from. The fit is fg_points = -2046.4 + 1531.6*nD The beers include stout, ESB, bock,Vienna, weizen, Pils and alt.

FG's
1.009
1.009
1.02
1.011
1.013
1.01249
1.01525

nD's (Relative to Air)
1.34153
1.34237
1.34877
1.34429
1.34575
1.34363
1.34534

OG's
10.52
12.82
18.57
13.43
15.90
13.86
14.74

ABV's
4.20
5.45
7.44
6.20
6.68
5.63
5.68

[Edit] Added ABV and original gravity data. Note that OG's are calculated from ABV and true extract.

[Edit] If you want to convert the nD's to Brix you can use

Brix = ((4997.6025*nD - 21822.2749)*nD + 32237.6805)*nD - 16304.345

(Doubt all those decimal places are needed but thought I'd leave them in)
 
Thanks for sharing, AJ. Based on that data set, it seems like a correlation could be worked out based only on the final refractive index. That doesn't intuitively make sense to me - in a three-phase solution, I would think that at least two combinations of sugar and alcohol levels could yield the same nD. Do you have the OG data for those beers?

I definitely can't get as good a fit to my data without the second independent variable. It isn't terrible though (r^2 = 0.78).
 
I edited the previous post putting OG and ABV data in there so it would all be in the same place.

As to the goodness of the fit: I haven't got that much data here and, while it is true that they span a fair range of OG, I don't really put too much faith in a model based on so little data. My impressions of refractometry at this point is that it works great except when it doesn't i.e. it's generally pretty good but occasionally will have you off by more than 1°P.
 
Thanks again for the update. It does seem like your data don't correlate very well with mine. I'm back to being frustrated now. It may just be that refractometers aren't all that accurate when it comes to FG. Then again, since your OG data are based on the ABV readings, maybe the innacuracies come at at that point. Like I said, there should be a simple solution based on two IVs. If nothing else, it seems like I have a good correlation for my own beers. The last three have all been <1 point off the curve.

BTW, would you mind me mentioning your results "publicly"? James Spencer has asked me to discuss refractometers on his podcast.
 
If you delve into the subject a bit you'll find that refractometers have their uses in the brewery but they are pretty restricted. Refractometry is not, in general, considered applicable to binary mixtures except in those cases where a measurement of some other type is available. For example, in the sugar industry, they sometimes use polarization measurements. Here you are trying to use the OG and that works up to a point. More Beer has a spreadsheet that calculates AG from OG and RI. When I try that against my 7 beer data set I get within a point or 2 for most of the beers but am off by over 4 points for one of them. I think this is pretty good as considering the potential differences in sugar spectrum for a given OG and variability in yeast attenuation it doesn't seem that OG is going to be that great an indicator. Not to mention that there are lots more than 2 optically active substances in beer.

On OG: Is what you measure with the hydrometer or especially a refractometer really the OG? If you measure before pitching the starter did you adjust for the starter (including the fact that it is fermented)? If after pitching how do you treat the alcohol in the starter? How much water wast lost to evaporation during fermentation? Should OG be adjusted for that? Was the wort absolutely uniformly mixed? This can be a factor if you sparge the hops in the kettle (don't want to leave any extract on those hops) and/or push the last of the wort out of the chiller into the fermenter. I do both so the hydrometer reading I take in the fermenter are at best approximate. That's why I back calculate as do many commercial brewers. As an example, the last beer in the list of data was the first Alt out of a new brewpub around here. The brewer said the OG was 16°P but it back calculated to 14.75. That's a pretty typical spread. Note that this doesn't mean that 14.75 is spot on. It's based on Balling's formula. If more or less than 0.11 grams of yeast are produced per gram of alcohol the OG calculation will be off.

The place where a refractometer can be used in a brewery is in the determination of alcohol but to use it effectively you need to calibrate against the beers you are brewing and that's where my interest lies. If I can demonstrate a tight cal on that Alt, for example, the brewer should be able to estimate his ABV to within a tenth of a percent or so and thus back calculate his OG. There would be a separate curve for the Kölsch and another for the APA etc. The catch is, of course, that one needs to get the calibration (the ASBC method uses ABW and AG) data and that's a lot of work (unless you have an Alcolizer setup into which you put the beer and out of which all these numbers pop).

Yes, you are certainly welcome to use the data any way you like. I just wish there were more of it.
 
Great. It would be nice to have someone check the math. Note that I have edited the post with the data because the conversion to Brix polynomial was in error (I fat fingered in 30°C instead of 20°C and, as it that weren't enough, 0 for vacuum when I should have put in 1 for air). Or, if you use your own formula be aware that the nD's are relative to air.

Over the 7 beers I found the spreadsheet gave average error of 2.75 points too big. RMS error was 3.24 points and the max error was 6 points (1.5°P).

The refractometer data were taken with a Reichardt hand held digital refractometer, corrected manually to 20 °C (the ATC tables in refractometers are not suitable for beer). The FG measurements were taken with an Anton Paar DMA 5000 M. So the data should be good but, as you see, I can fat finger most anything when it comes to the keyboard!
 

Latest posts

Back
Top