Rob Hoos: Thermocouple lesson

Discuss roast levels and profiles for espresso, equipment for roasting coffee.
User avatar
Team HB
Posts: 10527
Joined: 13 years ago

#1: Post by TomC »

Rob wrote a great article for Roast Magazine about thermocouples and how we process the information they provide. It's a solid reference that deals with understanding their limitations. Worth a read, and several clear graphs make his points. ... eyhole.pdf
Join us and support Artisan Roasting Software=

User avatar
Posts: 844
Joined: 16 years ago

#2: Post by MaKoMo »

Thanks Tom for linking to this nice summary of some core issues in comparing roast profiles. Love his suggestion to Treat coffee like jazz music as I love Jazz.

Regarding the take away to Consider moving to a thinner thermocouple I want to stress that thinner probes have the disadvantage of delivering a noisier signal. As correctly noted by Hoss, thicker thermocouples tend to naturally filter the data, or smooth the line, because they are not able to react as quickly. This is mainly due to the larger thermal mass of their shielding. This analog filtering is more effective and produces a lower time-lag than every filtering added later in the chain by either the thermometer electronics or a roast logging software working with a relatively slow sampled digital variant of the original signal. Those thinner probes have also the disadvantage that the material within the probes shield, electrically isolating the actual sensor from the metal shield more easily breaks by the bean movements and turns an ungrounded probe into a grounded one that may produces all type of ground loop issues and thus a horrible noisy signal. I am quite happy with my 3mm k-type probes, which seem to be still ungrounded, and offer a good balance between speed vs noise.

One thing that is obvious in all graphs depicted in that article is that the shape of the RoR signal as given does not correspond to the real RoR. As Rob writes correctly, Where the thermocouple reaches a zero-degree-per- minute RoR, we have a theoretical thermal equilibrium between the bean mass and the thermocouple (modulo the error caused by also measuring the air-flow and air-temperature). However, exactly at this turnaround point (TP), the bean temperature (or the values reported by the probe) start to rise again and thus the RoR should start to rise again. However, independent of the probe, the RoR rises above 0 almost a minute later than the TP as reported by the probe (cf. Fig. 4). So here the ROR curve is shifted to the right on the timeline or putting it the other way around, at an point in time you see the RoR as it was a minute before.

Rob did not document how those graphs where produced, but users of roasting software should be aware that the RoR is a computed value taking the current and a previous readings into account. Thus the resulting RoR value corresponds to the moment between those, the two readings were taken. Smoothing applied by the software to either the temperature signal or the RoR signal or both adds further delays. During recording there is in principle no better way to come up with the RoR, essentially as the future is usually unknown. Only in the offline case, there are algorithms that can avoid the shifting of the RoR computations completely by taking at any moment also readings "of the future" into account. Artisan implements such an algorithm in the offline mode, but its RoR also suffers from a time lag in the online case. Even worse, the current version of Artisan applies the offline mode to compute the RoR of a background profile, which then is shifted w.r.t. the RoR of the current roast what makes it hard to duplicate a roast by following the previous RoR. The next version of Artisan will improve on this and others...

User avatar
Posts: 1022
Joined: 7 years ago

#3: Post by EddyQ »

The article is very good. I recently figured this all out when trying to compare my Quest BT profiles to other profiles posted with the same coffee, but different roasters. The turn around or turning point of mine was typically shorter time and lower temps. My conclusion was my probe diameter was likely smaller.

The article shows many different things going on that are a little confusing. The turning point differences is solely due to different response times of thin vs thick probes. However during development, when RoR is very low, I would expect the thick probe temps to approach match with the thin probe. But the profile curves show differently. So something else other than probe response time is going on. Rob nicely explains this type of thing being caused by slight placement differences and other things.

I haven't roasted for very long. But have seen a lot of thermal things going on in my Quest that are rather difficult to explain. And this is before I dump in the green coffee. So, I totally get Robs comments on how precise one should expect from probes.
LMWDP #671

User avatar
Posts: 3612
Joined: 10 years ago

#4: Post by Almico »

Great article. I picked up on this:

Whether at any given point the beans are absorbing heat more quickly or slowly than the thermocouple is up for debate and needs to be seriously investigated in future scientific studies; however, we should not assume they are the same, merely that the process thermocouple's RoR is a proxy for that of the actual bean mass.

Given this assertion that our readings are based on the TC rate of heat absorption and not that of the beans, there should be an adjustment made for bean density, since higher denity beans absorb heat more quickly than lower density. It would be nice to be able to calibrate the TC to the bean based on g/1000cc.

Posts: 514
Joined: 9 years ago

#5: Post by dale_cooper »

This is an amazing article; something I wish I could have read 4 years ago when I started home roasting. The rabbit holes you can go down trying to "match" profiles on here, and takeaway lessons, are endless. I do believe there are still some great nuggets of information you can pull out by looking at profiles with regard to "energy".

Interestingly, I believe my quest utilizes a BT probe similar in size to the 3.2mm hoos has shown, and my "data points" are very similar to table 1 in the article - very interesting. Additionally, my MET probe is smaller and I should take that into account with regard to the responsiveness (or lag) of MET vs BT.

Sidenote - figure 1-3 at the beginning of the article are somewhat ridiculous; the Y axis scaling for BT ROR is completely different among the figures.

I wonder if the trend of "third wave OJ" could be due to roasters having thicker probes and trying to mimic a certain profile but the reality is they're really not using sufficient energy and thus are baking their roasts.

User avatar
Posts: 253
Joined: 9 years ago

#6: Post by millcityroasters »

I'll go out on a limb and disagree with my good friend Marko.

Although our roasters currently ship with stock 5mm BT probe, we have ungrounded 2mm probes specially made that exhibit a minimum of noise. Aside from that, I've recently taken the advice of Scott Rao and eliminated smoothing entirely. He's convinced me that the granularity of the raw (unmanipulated and unsmoothed) data allows the operator more reaction time and better control.

Personally, I've tried it both ways. When I was learning to roast and learning to control the machinery, smoothing provided greater benefit as I wasn't constantly reacting to readings I didn't understand. Later as I grew more accustomed to the response of the bean mass to changes in gas and airflow, smoothing became much less of an issue. Now, I've discovered a marginally better level of control by almost maximizing noise by going to 1 second sample rates and setting smoothing to 0 in Artisan.

As crazy as it may seem I'm getting marginally more successful roasts and marginally better coffee, but it's remotely possible that I've been doing it wrong all along. :)

Posts: 36
Joined: 9 years ago

#7: Post by wrakocy »

millcityroasters wrote:Although our roasters currently ship with stock 5mm BT probe, we have ungrounded 2mm probes specially made that exhibit a minimum of noise.
Been meaning to ask you guys about those 2mm probes for ages Steve.

I see you've got three different lengths (100mm, 75mm and 50mm) for sale on the site. Which would you recommend for BT and ET placements on a TJ-067?

User avatar
Posts: 844
Joined: 16 years ago

#8: Post by MaKoMo »

millcityroasters wrote:I'll go out on a limb and disagree with my good friend Marko.
On what does my good friend Steve disagrees exactly? Optimal bean probe diameter? Is 3mm too thin or too thick from your perspective?

User avatar
Posts: 207
Joined: 10 years ago

#9: Post by AZRich »

Not mentioned was bare wire TC's. I built my current (10-12oz usual charge size) solid drum roaster 3+ years ago and used very inexpensive bare wire K TC's for all 3 locations BT,ET,MET - initially because they were so cheap (~$1 ea). Using a phidgets + Artisan for that time and maybe 200 roasts, I have had absolutely no problems with noise and have been very happy with them! In line with the article table showing faster turns as they get smaller, I always get very fast turns in about 30+ sec's, and first crack is usually around 400F. Surprisingly, the same 3 TC's have held up perfectly well for me - even the BT probe in the beans. I do get much higher ROR numbers just after the turn as BT climbs very quickly. Small air or heat adjustments can have a very fast chart effect. But overall, the Artisan charts do not look all that different from others here.

User avatar
Posts: 253
Joined: 9 years ago

#10: Post by millcityroasters »

MaKoMo wrote:On what does my good friend Steve disagrees exactly? Optimal bean probe diameter? Is 3mm too thin or too thick from your perspective?
If anything, only just having recently done away with all software based smoothing.

This might be a case of premature interjection. I didn't notice you've been usinga 3mm probe until after I posted.