Temp is not temp, so how do we use it?

[Steve]

Another alternative is for the heat to be locked or held to prevent it changing for a portion of the roast. This would prevent the control system from dropping the heat if the roast went above the line, and from increasing the heat if the roast went below the line. I'm a bit doubtful that this would be advantageous, but it can be easily experimented with. In a zone (using engineering settings) make the zone multipliers both zero. This effectively turns off PID control and locks the power at the value it had at the start of the zone. Once the end of the zone is reached it will recommence adjusting power to turn towards and get back on to the line, provided that it has not got too far away from the line during the zone.

Thanks for this suggestions Chris, it's actually very useful in controlling the a bit twitchy / trigger happy control system.

The following is not trying to make definitive claims, just where my current experimenting has landed me.
Which has also involved finding the minimum fan profile required to keep proper bean agitation with a flatline coming into and through first crack to provide more useful ROR data. Comparison to Default Classic below.

CompareDefaultFan.JPG (72.8 KiB) Viewed 5391 times

Currently my thinking is leading me to believe that the black power line is quite important to not introducing baked notes, in that a steadily INCREASING line is required throughout the entire roast to avoid baking the coffee.

The two attached screen shots are a good example of using Chris's new suggestion for the zone effectively stopping the PID control from overreacting and ruining the roast.

Observations for K1
4 blue lines = one for each point below to illustrate the cascade of events.

Roast is going well.

Blue line 1: At 6:25 - temperature profile starts to drift above line very slightly. Control reacts by starting to flatline power briefly.

Blue Line 2: At 6:45 - 6:50 - Because of previous PID overreaction the temp profile dips below the line briefly and the control ramps the power quickly / too high, which causes a temp profile spike above the line coming into FCS. This forces a slightly premature first crack.

Blue Line 3: At 6:55 - 7:15 - Start of first crack, we do not want any abrupt changes to power here! which we can assume negatively affects the Environment temp. Instead, the control drops power abruptly trying to bring the temp back on profile from previous overreactions. Additionally this previous power increase then drop in power has caused the ROR to crash quite badly going into negative ROR.

Blue Line 4: At 7:18 - 7:30 - another sharp increase in power, trying to restore the crash. The power eventually settles back into a steadily increasing rise but by this time its is far too late and the roast has been stalled and baked.

K2: has a zone with both multiplier set to zero spanning 6:36 to 7:11, keeping the power steady as Chris suggested it would.

Cupping these blind with another coffee for contrast, the differences are not small.
K1, lacks aroma, lacks sweetness, thin mouthfeel, acidity tastes coarse/ less developed, persistent bitter aftertaste.
K2, pleasant aroma, good upfront sweetness, rounded purple acidity and body, long aftertaste with minimal bitterness.

K2 is by no means perfect and I would like to work on reducing my enforced power flatline and try and get the MAI to 2:30 but in comparison to K1 it is like VHS vs FHD3D.

K1.JPG (104.38 KiB) Viewed 5391 times

K2.JPG (103.72 KiB) Viewed 5391 times

[kaffelogic]

I'm a bit doubtful that this would be advantageous,

Well, I am delighted to be proved wrong - very impressive results. The power line goes flat and I can see potential advantages in being able to specify a power line gradient during the zone?

[Damian]

The first graph is with PID control, the second with PID off from 6:45

Unfortunately turning off PID and holding power constant isn't going to work. But I would still like to explore profiling heater power.
Basically, the goal we are chasing, is to follow the ROR curve as accurately as possible, maybe we can use the profile ROR curve as the target for PID to control power?

5.jpeg (145.4 KiB) Viewed 5384 times

4.jpeg (133.39 KiB) Viewed 5384 times

[Steve]

I'm a bit doubtful that this would be advantageous,

Well, I am delighted to be proved wrong - very impressive results. The power line goes flat and I can see potential advantages in being able to specify a power line gradient during the zone?

Oooh now you are getting very tricky! Yes i think that would be great, although how best to implement that is getting way above my pay grade.

[Luca]

If the absolute temperature reading doesn't really mean the same thing at different fan speeds, then what is the relevance of the ROR if the fan speed is changing? Isn't the one thing that we can basically conclude that, so long as the fan speed is changing, a straight line linearly declining ROR is the one thing that cannot produce the results that Rao writes about?

There is a relationship between fan speed and temperature probe offset from true bean surface temperature. The faster the fan, the bigger the offset. The existence of this offset should not be a cause for concern - after all there is also an offset between bean surface temperature and bean core temperature. The probe offset is not just influenced by air speed, it is also changing as the relationship between air inlet temperature and bean surface temperature changes. We have to manage with the offsets, and without certain knowledge of exactly how big they are at any point in time. Rao writes about steadily declining ROR. If the temperature profile curve exhibits steadily declining ROR, then a changing air speed will alter that ROR, but it will retain its steadily declining nature provided the air speed is also steadily declining. A steadily declining fan speed will almost always be compatible with a steadily declining ROR. On the other hand, a steadily increasing fan speed could potentially shift the offset in such a way as to cancel a steadily declining temperature curve causing it to go flat. For this reason, the rule of thumb is to use a steadily declining fan speed.

Chris, the bold part makes no sense to me. Let me try and explain the issue again, because maybe I'm not being clear.

I think what you're saying is that the machine will adjust the heater temp with whatever airflow is present such that the bead probe reads the target temperature. If the air speed is higher, then it might need a higher temperature. This means that the air temperature at the inlet to the roast chamber will be higher. Since the bead probe is on the exit side of the roast chamber, in this circumstance the bulk of the green coffee will have been exposed to hotter air and will thus reach first and second cracks at lower temperature readings. (Or maybe I got the relationship between airflow and temperature reading the wrong way around and it's the opposite, but that's still fine for the purpose of this discussion - the point is that the beans in the roast chamber are actually experiencing different temperatures at different fan speeds at the same temperature reading.)

The ROR reading is based on the temperature reading. But if the fan speed is different, then the temperature reading actually means different things. So if I'm correct that there's a lower inlet temp with a lower fan speed, and you lower the fan speed after first crack and end up with a perfectly linear ROR decline, in fact the ending temperature would have been even lower, had the fan speed stayed the same. Hence, the straight line ROR is in fact dipping if you adjust for fan speed. So isn't the position actually something like if you are aiming for a straight line decreasing ROR as per Rao's book, then the kaffelogic graph only in fact gives you this over the portion over which the fan speed is constant?

The extent to which the straight line fan speed adjusted ROR dips, I don't know, since this is information that neither the machine, nor the software, provides to the user. Note that Ikawa ended up providing both an inlet and an exit temp on their pro machine. Maybe you don't want to do that to keep costs down - and having a machine that can deliver similar (or perhaps better) results to the Ikawa Pro at 1/4 of the price is admirable, so don't let me sound too critical of this! Since the machine seems to have a lot of power and a lot of flexibility, it seems to me like there's a very good chance that we should be able to get excellent results out of it, once we figure out how to use it. But is there some way to work backwards to work out what the air temp is when going into the inlets in the roast chamber? On most drum roasters, you get some reading of an environment temperature that's hotter than the bean temp, often the maximum environment temp. The BT inexorably asymptotes towards it. Not having this information feels really disorienting ... though I guess that not being able to make meaningful changes on the fly, nor see the roasting beans is also disorientating.

There's a further issue, which is that the fan speed on a regular drum roaster is calculated from BT information from a 3mm-6mm probe, which will have its own buffering, lag and smoothing by virtue of the physical properties of the sheath. So I guess what I should be asking is how the temp readings displayed by the software compare with what one would get from such a probe (and, again, with how the derivative is calculated in artisan, cropster, etc). Again, all of this is just maths, so presumably there's some way for the data to be transformed to be similar, if that's necessary or desirable. At the moment, I can't make head nor tail as to how similar it is or isnt. For example, do the numbers on a drum roaster lag more? Would one need to shift the kaffelogic ROR graph 15s to the right to make it more equivalent to a drum roaster?

Now, I do have to add to this that, as Damian has observed, it may be that the fan speed, damper settings and convective heat transfer are changing in drum roasts, but if they are, there are a number of reasons why they are likely to be less dramatic; there tend to be only a few changes, there's a mix of conductive and convective heat transfer, and the maximum amount of airflow is probably waaaaay lower than a fluid bed roaster to start off with.

Finally, I guess I'd like to know, are we making this all too difficult for ourselves? Do we actually need a fan curve at all? Why can't we start high enough to agitate the green and just leave it there? What do we lose by having it that high? I guess you might say that the inlet temp would be too high, and might cause charring. Or you might point out that the beans might be hurled around ridiculously towards the end. However, I'd observe that we have neither inlet temp/MET data, nor any sensible way to observe the agitation of the hot beans during the roast, so I think these are pretty legit questions for users to ask. We are literally roasting blind on these issues. I understand you're working on a transparent roasting chamber replacement, and that will hopefully help with at least one of these issues.

I don't want to sound overly critical. I do think that this roaster has tremendous potential, but I also find it difficult to use. Some of these difficulties are just the inherent difficulty of roasting coffee.

[kaffelogic]

Thanks for all the thought going into this topic. To do justice to the deep questions raised will take me a while. Expect a post later in the week.

[Luca]

Hi Chris, no worries, and thanks for taking the time. And sorry for being a pain with my stupid questions.

Do you have some like nasa-level frankenrig for gathering temp information? What I'm thinking would be really cool to gather data would be a kaffelogic with 10 probes; bead probes and 3mm sheath probes at 4 locations, being pretty close to the element, right at the inlet to the roasting chamber, halfway between the existing probe and the bottom of the chamber, right at the existing probe location and then maybe another one just a little bit higher up, just above where the beans ever jump to. It would be really interesting to see (a) if they track each other with an offset, or if there is a different relationship and (b) if any of the data from that rig looked any closer to what one expects from a drum roaster BT or ET probe.

[kaffelogic]

I'll try to break the fan speed issue down and tackle it one point at a time.

First, anything to do with aerodynamics can become quite complex, as can anything to do with fluidised particles, and here we have both interacting.

Taking the fan on its own, one nice simple principle is that air flow (cu m/sec) is usually more or less proportional to RPM. However, this nice relationship only holds if the pressure is constant. In the Nano the pressure is not constant. The air flowing out of the fan is pressurised by being forced to pass through a bed of coffee beans. We load 120g of green beans into the roaster, at the nominal speed of 16,000 RPM the fan generates an airspeed of 2.45 m/sec through the roast chamber (0.00943 cu m/sec). During the roast cycle the weight of those same beans drops to 100g, dropping the air pressure the fan has to overcome and the same 16,000 RPM generates an airspeed of 2.63 m/sec (0.01012 cu m/sec).

So to "Temp is not temp" we can add "RPM is not air speed".

On the face of it we can expect a fan profile that goes from 14,700 RPM at the start to 13,694 RPM at second crack to be doing no more than compensating for moisture loss during the roast and delivering the same air speed throughout.

However, observing the bean mass it is obvious that there are other factors at play than air speed. The beans have expanded to almost double their size and a correspondingly much lower density by the end of the roast. At the start of the roast they barely circulate, with small perimeter spouts and slow central turnover. At the end of the roast the whole bean mass bounces with vigorous central turnover. We are looking at a completely different fluidisation state. At the end of the roast there is more bean surface in contact with air (more fluidisation) and the air is in contact with the bean mass for longer (greater load volume). This means potentially more heat is transferred to the beans bringing the bean surface temperature closer to the bean probe temperature - although the internal bean thermal conductivity is lower, so this effect might even be cancelled out or reversed.

So when we are talking about the relationship between fan speed and the difference between probe and bean surface temperatures, we are forced to accept that it is not a simple linear relationship. Keeping the fan speed constant won't keep the relationship constant.

So the short answer to the question "temp is not temp, so how do we use it?" could well be: pretend temp is temp and get on with cupping the coffee!

As physical systems go the situation is not too complex that it is not amenable to mathematical modelling, however there is sufficient complexity in that modelling to warrant a decent PhD thesis. For now the best I can offer are generalisations like "you will generally be safe with a steadily falling fan speed." I suggest you use fan speed when you want more or less circulation, evenness, cleanness, etc. and bear in mind that sudden or significant increases in fan speed could jeopardise your goal of a steadily falling temperature rate of rise (unless compensated with other simultaneous changes).

All other things being equal, faster fan speed will result in the bean surface temperature being further behind the probe temperature. This is useful to know when comparing two roasts at more or less the same stage of the roast, e.g. at first crack. It is not so useful to know when thinking about different stages of the same roast when all other things will manifestly not be equal.

On the subject of Scott Rao's principles, he talks about steadily declining, not linearly declining. The only way to get a pure linear decline in ROR is to start with a parabolic roast curve (calculus and all that). Basically, if you can see a steadily declining (albeit curved) ROR of the probe temperature curve you can assume there is a steadily declining (albeit curved a bit differently) ROR of the bean surface temperature. Besides, all of Rao's data was probe temperature curves anyway. So back to my point: go with the probe and keep the fan speed changes steady to prevent air speed from messing with it.

[kaffelogic]

Hi Chris, no worries, and thanks for taking the time. And sorry for being a pain with my stupid questions.

Do you have some like nasa-level frankenrig for gathering temp information? What I'm thinking would be really cool to gather data would be a kaffelogic with 10 probes; bead probes and 3mm sheath probes at 4 locations, being pretty close to the element, right at the inlet to the roasting chamber, halfway between the existing probe and the bottom of the chamber, right at the existing probe location and then maybe another one just a little bit higher up, just above where the beans ever jump to. It would be really interesting to see (a) if they track each other with an offset, or if there is a different relationship and (b) if any of the data from that rig looked any closer to what one expects from a drum roaster BT or ET probe.

Not at the moment, but this is a worthwhile research approach that we will work towards.

[kaffelogic]

So isn't the position actually something like if you are aiming for a straight line decreasing ROR as per Rao's book, then the kaffelogic graph only in fact gives you this over the portion over which the fan speed is constant?

See my post https://kaffelogic.com/community/viewto ... rt=10#p858

But is there some way to work backwards to work out what the air temp is when going into the inlets in the roast chamber?

There is, but at present I don't have the mathematical modelling or the experimental data to make this exact. The controller works off the basis that in ambient temperatures of 20 deg C, with 1050 Watts, 14700 RPM, and 120g green beans, the inlet temperature will be 240 deg C. This is a nominal figure of 220K above ambient. You could adjust this proportionately as power changes to get a nominal figure at other power levels.

On most drum roasters, you get some reading of an environment temperature that's hotter than the bean temp, often the maximum environment temp. The BT inexorably asymptotes towards it. Not having this information feels really disorienting ... though I guess that not being able to make meaningful changes on the fly, nor see the roasting beans is also disorientating.

The Kaffelogic Nano has been designed on the idea of designing a profile, running a roast, studying the log, making changes to the profile, and re-running the roast. This is contrary to traditional practice which involves monitoring and modifying settings during a roast. So it is understandably disorientating. As we develop accessories like the glass chimney, additional temperature sensors, and Bluetooth connectivity the monitor and modify approach will become more practical.

There's a further issue, which is that the fan speed on a regular drum roaster is calculated from BT information from a 3mm-6mm probe, which will have its own buffering, lag and smoothing by virtue of the physical properties of the sheath. So I guess what I should be asking is how the temp readings displayed by the software compare with what one would get from such a probe (and, again, with how the derivative is calculated in artisan, cropster, etc). Again, all of this is just maths, so presumably there's some way for the data to be transformed to be similar, if that's necessary or desirable. At the moment, I can't make head nor tail as to how similar it is or isnt. For example, do the numbers on a drum roaster lag more? Would one need to shift the kaffelogic ROR graph 15s to the right to make it more equivalent to a drum roaster?

The Kaffelogic control system follows the profile 8.5 secs behind. The mean temperature is further behind at 12 secs, and the ROR a full 19.5 secs behind the profile. These delays are created by the smoothing and intrinsic lag. I can't tell you how these compare with a typical drum roaster, except that every drum roaster is bound to be different. If you construct your profile by recording an exemplar roast in Artisan, then the delay is actually built into the profile curve already, however there will be a further delay introduced when you use that profile curve to drive a roast replication. I believe that general practice is to pretty much ignore the delays on the basis that the shape is the same whether or not it is delayed.

With regard to attaining Scott Rao's objective of steadily falling ROR, remember his data was bean probe data from predominantly drum roasters - not actual bean surface temperate data. His objective is steadily falling probe ROR.

Finally, I guess I'd like to know, are we making this all too difficult for ourselves? Do we actually need a fan curve at all? Why can't we start high enough to agitate the green and just leave it there? What do we lose by having it that high?

The air speed affects power required. Any departure from the default fan curve, including a straight horizontal line, is a potential option provided there is sufficient power available to hit the temperatures. At this stage experimentation is the key as I do not have enough of the required data to provide better guidance.

However, I'd observe that we have neither inlet temp/MET data, nor any sensible way to observe the agitation of the hot beans during the roast, so I think these are pretty legit questions for users to ask. We are literally roasting blind on these issues. I understand you're working on a transparent roasting chamber replacement, and that will hopefully help with at least one of these issues.

For careful observation of the bean circulation I study the beans in manual mode with fan only. That means I look at green beans, and then roasted beans, with just the fan running. During roasting I lift the lid briefly and shine a torch into the roast chamber. All a bit hacky I know, and the glass chimney will make this whole process a lot easier.