Artisan Version 7.0 - Page 4

Discuss roast levels and profiles for espresso, equipment for roasting coffee.
User avatar

#31: Post by MaKoMo »

The two above are listed as related products on looking at the phidget 1048. Would you recommend using one of these or should I be getting a different one all together.
I do not feel I have enough expertise on probes to answer this and leave it therefore to others to guide you.



#32: Post by boublanc »

Have a look at this probe I think it would be a better choice. ... 3a886999fd
The 3109 would be too flimsy. The 3108 is grounded it will be susceptible to spike on Artisan...


#33: Post by Barrie »

Version 7 is now available on the Artisan website, and there is a guide entitiled "Controlling the Hottop Roaster with Artisan: The Basics" available for download in Word format on the HRO website. ... load_id=24

Version 2.0 of the guide was posted 2013-12-20 ... load_id=25

2013-12-22. v. 3 of the guide (pdf file) is now on the Artisan website along with Jack Hannon's Arduino firmware installation instructions, and version 7.2 of Artisan

The guide was written for Windows platforms but I am told it also applies to Mac and Linux. Just haven't tested those. :D
If anyone has trouble with Word format, send me a PM with your email address and I will forward a pdf version - not quite as good resolution but serviceable.

User avatar

#34: Post by MaKoMo »

Zaneus wrote:Is there an option to smooth the deltaBT graph while its graphing, but not have the smoothing be applied again once the roast is finished? If i leave the smoothing option on low it comes out good once i save the profile but during the roast it shoots all over the place in an up and down zig-zag. If i apply enough smoothing to get an accurate straight line while im in the middle of the roast then it over-smooths it at the end of the roast and i lose an accurate graph of that roast.
First of all, sorry for answering just now. Your post escaped my eye before.

I understand your dilemma and am unsatisfied with the situation myself. But let me try to explain below, why I think there is not much room for improvement here.
Zaneus wrote: Increasing the sampling interval helps but i like having a more reactive line to follow during first crack.
Yep, see below.
Zaneus wrote: Alternatively, would anything else be causing the erratic delta readings? The actual BT curve is accurate and straight, its only the deltaBT that jumps around.
The BT curve is actually not that smooth as you think (indicated by the DeltaBT;) and I do not think that the delta readings are erratic.
Zaneus wrote: Regardless, ive been absolutely loving artisan over the last few weeks. Big improvement over the Probat program they send out with the big shop roasters!
Thanks for that positive feedback!

Let me try to give some explanation on the RoR and the related smoothing issue.

The roast master wants to know at any time during a roast the exact RoR of the BT curve to make decisions on how to adjust the air flow and especially the heater settings. In some cases he might have loaded a previous profile in the background in order to repeat that roast by duplicating its BT RoR curve.

The BT RoR (or DeltaBT) curve represents the first derivation of the underlying BT curve. As the BT curve is only given as discrete values on has to approximate the DeltaBT curve by calculating for each measurement a corresponding RoR value. That is done by taking the time points and values of two measurements and calculate the corresponding raise in C or F per minutes. This can done easily for each measurement and one ends up with an (approximation of) the BT RoR.

Live Smoothing
As you observed the RoR curve might get spiky to a point were it turns useless. One way out of this is to smooth the RoR curve. Smoothing the underlying BT curve helps too. Smoothing can be done by simply considering not only the last value, but a number of previous values, let's say n, and computing a (potentially weighted, preferring later values) average over them. However, the larger n get's the later the smoothed RoR reacts on RoR changes in the underlying BT data (best seen at the point the RoR becomes again positive after the TP in a drum roaster). Note that here the number n of measurements count and not the time. Therefore, the smoothing of the RoR is sensitive to the sampling rate too.

Static Smoothing
A second method to smooth the RoR is based on convolution of a scaled window involving also reflected copies of the signal. A parameter m (should be uneven to avoid shifts) defines the size of the window and with that the amount of smoothing applied. This method has the advantage to avoid any temporal shift, by taking the "future" (w.r.t. every measurement) into account. As consequence this method can only be applied in retrospect once the future is know, but not during the recording.

Smoothing in Artisan
Artisan applies the first method during roasting and the second one (since v0.6) on loading a already recorded profile (holds also for the background RoR curves).

Side Note: there is a natural upper limit to smoothing as this process is not only potentially responsible for a shift in time, but also for changes of the values. In overdoing it, one might end up with a flat line, but smooth line. Not very productive.

So we have to deal with the trade-off smooth and late vs spiky and accurate timing. The matter is further complicated as smoothing is done not only on the BT and RoR values by Artisan, but also by the firmware within each meter (btw. this is configurable in the case of the TC4). Some meters apply smoothing only internally to achieve more stable display values and still deliver rather raw values via their serial connection (e.g. the Amprobe). Others return smoothed values via the serial link. Finally, smoothing is also done by the probe itself as effect of the physical mass of the shield material. Again different from setup to setup. Note, that the smoothing done by the probe and the meter are of this first type of, inevitably producing a shift in time (and they differ from setup to setup).

Hope it became now clear why the RoR curve during roasting can differ from the one displayed after roasting especially in terms of smoothness, but also w.r.t. time shifting.

I received one suggestion on the issue that the RoR of the profile loaded in the background is accurate in time, but does not fit the timing of the live RoR during roasting due to that time shift. Why not just shift the background curve such that it fits the time lag of the live RoR? Now if you consider the assumption stated above it becomes quickly clear that this is counter productive. The control (even if manual as in the case of non-PID controlled roast) of a mechanism (in our case the application of a certain amount of heat) based on a signal with a significant time delay can easily produce a swinging system. This is usually not what the roast operator wants, right?

So in summary, I still have to be convinced that the RoR curve is useful for the case of coffee roasting (as this information comes either late or in an unstable form). Further, I do not see much hope to be able to improve the RoR implementation in Artisan.

Please prove me wrong,


#35: Post by rgrosz »

Marko - that is a very clear explanation. Thank You!!
LMWDP #556
Life is too short to drink bad wine - or bad coffee

User avatar

#36: Post by MaKoMo »

One think I want to add. From my perspective, smoothing of the temperature curve (which helps to get a smoothed RoR curve) should be done as "early" as possible. So best by the probe itself. So looking for a probe that reacts as fast as possible does not make sense to me. Next the meter should do average smoothing based on some form of "oversampling" to minimize the time lag and maximize the effect. By oversampling I mean sampling faster then Artisan does. Most meters (also the TC4) basically continuously sample their inputs, while Artisan requests samples only at an interval of minimally 1sec. I am aware that some meters are way better in doing this than others. My 4 meters all behave differently, and the oldest, slowest and most expensive one (an Omega) returns the smoothest curves on the same probes, while others like the CENTER/Voltcraft ones are quite spiky.

So smoothing in Artisan should only be the very last step and best be turned off.

Still, I plan to experimenting with oversampling (as time allows) within Artisan by sending several requests to instruments per sampling interval and applying averaging on the answers. However, this might only be an option for fast responding meters like the Phidgets. Anyhow, I will report on the outcome.



#37: Post by 10ett »

Thanks for this. It is greatly appreciated.


boublanc wrote:Have a look at this probe I think it would be a better choice. ... 3a886999fd
The 3109 would be too flimsy. The 3108 is grounded it will be susceptible to spike on Artisan...


#38: Post by rgrosz »

I've experienced issues where my logging with artisan randomly seems to stop working. My first roast today was fine - but not the second one. The LCDs on the screen show the ET / BT temperature values, but artisan does not draw the ET / BT curves on the screen. Has anyone else experienced this problem?

I am running Windows 7 Pro 64 bit. Based on past experience, this problem usually occurs after "patch Tuesday", when the Windows Critical updates are released. I do backups every week, and after restoring the prior week's backup image, then everything works properly.

Never mind ... user error. I had the same problem after I restored from backup. While trying to capture a screen shot, I figured out the cause of the problem. I had loaded a prior profile, and the horizontal axis was shifted. The lines WERE being drawn - they weren't visible on the screen for the first 3 minutes :oops:
LMWDP #556
Life is too short to drink bad wine - or bad coffee


#39: Post by rgrosz »

When I roast, I enter a few events by hand - when I adjust the fan speed and the power prior to 1st crack. I have not figured out how to get these to appear on the graphs reliably.

Something seems incorrect in the latest version of artisan. I entered two events - one for the fan, then a later one for the power (which is shown at the bottom of the screen). If you look carefully at the legend at the upper left, it has two entries - one for Fan, and one for None - but I expected Fan and Power instead.

Rick G
LMWDP #556
Life is too short to drink bad wine - or bad coffee

User avatar

#40: Post by Italyhound »

Using a phidget, would there be any potential Artisan issue using it with the wifi adapter?
@evshack on Twitter