Friday, October 1, 2021

Probe Arrays

There are a number of reasons why using an array of probes to measure temperatures inside a roasting machine, instead of using single probes, seems to be a good idea. The intrinsic redundancy of probe arrays allows a machine to handle probe failures silently and to report them early on to the operator, avoiding down time. Another advantage is using averages of all readings per sampling time. This increases the stability compared to single probes. Averaging of readings gathered in parallel avoids also the main disadvantage of common software based smoothing, the introduction of delays caused by averaging readings gathered sequentially. Are probe arrays worth it?


To estimate the potential noise reduction of probe arrays we measured and compared the Idle Noise of two setups with up to 4 probes. One setup is based on the best performing measuring system from our Idle Noise study, the Phidget TMP1200 module using RTDs probes. The other one is using Thermocouples (TCs) probes and is based on Phidget TMP1101 modules. Artisan can connect directly to both probe arrays and take advantage of the noise reduction using the corresponding configuration.

To make the results comparable we used the same Artisan settings from the On Idle Noise study: 0.5sec sampling interval with software smoothing turned off. The Phidget data was harvested in synchronous mode. As we did in On Idle Noise, we took the Standard Deviation (StdDev) of the raw rate-of-rise (RoR) signal over a 2 minutes period. 

We kept the standard Phidget module setups of Artisan which instructs to harvest the probes in synchronous mode (menu Config » Device).

Next we added extra devices to harvest the individual temperature channels from the connected modules (here the case of an array with three RTDs).

Finally, we defined symbolic formulas to map the average over all temperatures to ET and the temperature of the first channel to BT.


Below are the resulting BT RoR signals for arrays with 2, 3 or 4 elements for the RTD as well as the TC setups. The single probe case (RoR data taken from probe 1) is rendered in blue and the RoR from the average over all probes in red. To both RoR curves we applied a constant offset to avoid overlaps. Note that the RoR scale on the right side of the graph is tighter for the RTD setups (-2!C/min to 2C°/min) then for the TC setups (-15°C/min to 15°C/min, equal to the one used in On Idle Noise). Note further that the measurements reported here were taken with probes connected. This is in contrast to the original work in On Idle Noise where the idle noise was measured without the probes connected. However, we learned that connecting probes does not add relevant extra noise.

Note that we are only comparing the Idle Noise behavior of those different setups. The advantage of using a probe array to take bean temperature readings while roasting is expected to be much higher than the results reported here, as probe arrays reduce also the additional fluctuations produced by the beans bumping against the probes and increase the stability of the readings better reflecting the average bean surface temperature in the drum. 

A proper installation of a probe array within a roasting machine would be done such that all probes of the array are well covered inside the bean mass at any time, but slightly in distance of each other to max out the effect of the averaging. Still, the usual installation rules should be followed like keeping the probes in greatest possible distance from the drum wall. Still we expect, but did not test, positive effects from replacing a single probe by a dual probe to form a simple 2-element probe array. Even inside a dual probe the sensor elements are usually in some distance which should already allow for some noise reduction in addition to the reduction in idle noise by using two separate meters to read in their signal, as analysed here.

Idle Noise – RTD Arrays

For the 4x RTD array we repeated the experiment replacing the one HUB0000 module by two, each connected to two probes, to see if that makes any difference. We got the identical results from this setup.

Idle Noise – TC Arrays

All of the above data was gathered using individual temperature meters, one per probe, TMP1101 in the TC case. The TMP1101 features 4 inputs and allows also to connect all probes to just one meter. Below is data recorded with 4 probes connected to a single TMP1101.

As in the previous report, we computed the StdDev over this data to ease the interpretation of those results. Higher values correlated with more noise. The results for the one probe setups are similar to those we found before for using those modules in synchronous mode.


Should coffee roasters have probe arrays?

Absolutely, as all this comes with zero signal delay and the expected noise reduction on roasting should be considerably higher than just the reduction on idle noise reported here. The idea to take advantage of probe arrays was noted in the original 2019 post on Idle Noise. Now we are looking forward to roasting machines equipped with probe arrays from factory.