MaKoMo wrote:Therefore the relatively low sampling rate in Artisan (which can be maximally 2 samples per seconds with a 1sec rate + oversampling) should not play a role here, right?

Marko,

Took me a while to think about this more. I haven't dealt with sampling converters for a while.

Below I generated some data close to where a digital thermometers might produce if no analog filtering occurred ahead of its internal analog to digital converter (ADC). The first plot, I added 1 standard deviation of noise (zero mean) to an ideal increasing temperature. Then I sampled it at 2 second rate. The resulting samples are also 1 standard deviation (zero mean) which likely could be smoothed nicely with Artisan.


Next, I replaced the added noise with a 60Hz sinewave of similar amplitude. The 60Hz is a signal common to our roasters due to high power electric heating and/or line voltages. As seen, the resulting samples also have a sinewave shape of a much lower frequency. These samples would not likely be something Artisan could filter. If aliasing like this occurs, Artisan cannot correct it with smoothing or oversampling.


So, I feel if folks are having issues that do not look correct, it is best to try Artisan with no smoothing and no oversampling to see the magnitude and shape of the "noise" being sampled. Does the shape of samples look periodic or random? Then, do things that reduce amplitude and any periodicity and then apply smoothing and oversampling.

If noise is low as you can get, seems random, try slightly different sample rates. 2 seconds to 2.5 seconds may make the sampled sinewave look a lot like noise where Artisan would nicely smooth out.

In the case above, the 60Hz noise likely could never be related to the probe diameter, but rather coupled with probe grounding. But it is possible (not sure likely) there could be much slower periodic temp variations due to a spinning drum or air turbulence that a thin probe may sense and a thick probe will not. Hence the thicker probe will have somewhat less error. I think this case may not be something happening to most of us, but it does show a vulnerability with digital sampling systems. There is not much you can do about it with software.

All good and thoughtful discussion.

As a side note, I think this is one of the big reasons automated roast quality is almost invariably poorer than a competent operator relying on sensory cues.

I'm no Luddite, but in the battle of technology vs craft: craft usually wins.

millcityroasters wrote:As a side note, I think this is one of the big reasons automated roast quality is almost invariably poorer than a competent operator relying on sensory cues.

I'm no Luddite, but in the battle of technology vs craft: craft usually wins.

I don't feel I have the experience to comment on the quality with and without technology. But when my beans are crackling away and my temp probe says they shouldn't be, the craft wins hands down.

I am a beginner roaster and would be rather lost without the thermocouples and live plotting vs time. I'm not saying I don't notice all the sensory cues and make corrections based on them. But I am saying my sensory skills are not precise enough to pinpoint certain aspects of roasting (such as declining RoR, momentum into FC). But as I learn my roaster, I am finding with specific heat/air adjustments vs time I can sit back and coast (not even look at plotting) with confidence that my temps are in fact declining nicely without risks of stalling or worse. At this time, I am focusing mainly on what the beans are doing.

EddyQ wrote:Marko,

Took me a while to think about this more. I haven't dealt with sampling converters for a while.

[cut]

So, I feel if folks are having issues that do not look correct, it is best to try Artisan with no smoothing and no oversampling to see the magnitude and shape of the "noise" being sampled. Does the shape of samples look periodic or random? Then, do things that reduce amplitude and any periodicity and then apply smoothing and oversampling.

If noise is low as you can get, seems random, try slightly different sample rates. 2 seconds to 2.5 seconds may make the sampled sinewave look a lot like noise where Artisan would nicely smooth out.

In the case above, the 60Hz noise likely could never be related to the probe diameter, but rather coupled with probe grounding. But it is possible (not sure likely) there could be much slower periodic temp variations due to a spinning drum or air turbulence that a thin probe may sense and a thick probe will not. Hence the thicker probe will have somewhat less error. I think this case may not be something happening to most of us, but it does show a vulnerability with digital sampling systems. There is not much you can do about it with software.

Dear Eddy,

thanks for those detailed explanation and suggestions on how to work-around those periodic swings. I am anyhow in the same boat with you on choosing not too thin thermocouples, especially for the bean probe. Problems should be resolved as early in the chain as possible...

Marko

I've done a few more roasts with both the 6.4 and 3mm probes and am finding a little difference in response time between them. In these graphs the blue is BT and measured by the 6.4mm located inside and the roast chamber. The 3mm (BT) is mounted outside and poking through:



I wrapped the 3mm probe in paper to try and isolate it from the roast chamber a bit.

Here are some roasts using this configuration:



I bumped the heat twice early on for this roast:


I made a small adjustment around 3:00. The 6.4mm probe reacted, not so much the 3mm:




For this last one I turned off all smoothing for the deltas:



So it seems the black curve (6.4mm) is more responsive than the 3mm, at least placed in these locations. I'll have to switch them around and see what happens.

When I switched them around, putting the 3mm inside and the 6.4mm poking through the bottom I got this:



The hump at around 9:00 is from reducing airflow to prevent the beans from flying out.

It seems the 6.4mm is more stable in the bottom location that the 3mm, but surprisingly, the 3mm is more stable inside the roaster where the 6.4 was jumping around, although it does register over 20* hotter by roast end.

I really don't want a jumpy probe where I'm tempted to tweak heat all through the roast. Remember, what we're seeing in Artisan is the probe temperature, not what is actually happening to the beans. They are far less responsive than both probes!

But this is a good exercise for me in finding the best BT configuration for probe type and location.

And it appears we're still far away from being able to share detailed roast data between roasters. I'm getting a wide range even on the same roaster.

Alan, I don't see anything in the last plot (smoothing off) that would indicate high frequency interference type noise. As you said, it looks a lot more like fluctuations due to probe placement.

According to the data provided by Omega. If your 3mm probe is internally grounded and in air, it will respond in roughly 11 seconds. The 6.4mm probe would be 45seconds. If these probes are not internally grounded, then expect the times to be 1.5X longer.
So, if your 3mm probe is not internally grounded and the 6mm is internally grounded, the 3mm should still be faster.
However, your probes are not in air and beans are hitting them, so they both likely respond to changes in bean temps faster than Omega's tests in air. http://www.omega.com/techref/Thermocoup ... eTime.html

All info seems to indicate probe placement is your path to more or less response from either probe. I personally don't like the fluctuations in your 2nd to last black probe (unless you were making adjustments to cause the steps making them real).

Both of my probes are ungrounded. I just did a continuity test and neither lead is connected to the probe shell.

I'll play more with tis weeks roasts.