Odd Si7021 humidity readings over 6 nodes in same jar

[CallOne_Phil]

As the title says I am seeing odd reading from the Weather Shield(Si7021 sensor) running on 6 Moteinos inside a large glass jar.  Everything is running with the same weathermote example code on a mix of R3 and R4 Motenio boards powered by AAA batteries.  I first suspected something was wrong when my dehumidifier was reporting about 20% less humidity than one of my nodes, but I assumed it was the sensor on the dehumidifier.  However, after collecting every node deployed on that floor of my house and putting them into a large jar I was very surprised at the the results. 

My test includes two calibrated LCD screen meters which both read 63% at the time the attached screen grab was taken. I have read over the datasheet for the Si7021 about how it has a film that can become dried or saturated and throw off the readings.  That seems to be what is happening here, but strangely none of these sensors were exposed to extreme humidity.  I do not have the equipment to perform the bake process from the datasheet to dry them out, so I will be leaving them in this jar hoping they correct themselves over time.

My questions at this point are:

1.  I own 11 Weather Shields with the Si7021 sensor.  Has anyone that deployed around that number or more experienced this issue?

2.  Do the new Weather Shields show any early signs of this humidity drifting with the other sensor?

3.  Has anyone found a simpler way to "bake" the sensors and dry out the film?

[joelucid]

Funny you post this today - I don't have weather shields, but a couple of motes built around the same si7021 chip. I, too, was getting some pretty unlikely results so I did a test yesterday with 4 si7021 based nodes, 2 dht22 and an honest to god human hair analog hygrometer.

Results were the si7021 showed about 10% more relative humidity than the hair hygrometer and the dht22 about 10% less. All si7021 were in a +-3% range of each other.

I have a feeling the si7021 are off. Maybe the si7021 is not as good a sensor as I once thought.

Joe

[WhiteHare]

IIRC, the datasheet says that it's sensitive to how it's reflowed. 

[Edit: also, there's required a rehydration step after reflow, and I wouldn't be surprised to learn that many forget to do it or simply omit it.  From the datasheet:
"The measured humidity value will generally shift slightly after solder reflow. A portion of this shift is permanent and
is accounted for in the accuracy specifications in Table 4. After soldering, an Si7021 should be allowed to
equilibrate under controlled RH conditions (room temperature, 45–55%RH) for at least 48 hours to eliminate the
remainder of the shift and return the device to its specified accuracy performance."]

There are certain salt solutions that can produce an accurate humidity level inside a sealed jar.  I don't know whether any of the common ones produce a 45-55%RH.  If so, that would probably be the easiest way.  Regardless, they would offer an accurate standard for  testing the accuracy of your humidity readings once your sensor goes operational.

The datasheet also has some vague warnings about the sensor loosing accuracy, some of it permanent, if it's kept at too high a humidity for too long.

[joelucid]

Here's an article showing the weaknesses of the si7021. The author also finds the chip systematically measuring too high a relative humidity.

http://www.kandrsmith.org/rjs/misc/hygrometers/calib_many.html

[Felix]

@joelucid,
I've seen that article in the past, it's a really great analysis.
The BME280 is declared the winner:

Within the data obtained here, the BME280 is the undisputed winner. It is the most accurate at the specified 25°C and it shows the least dependence on temperature.

[WhiteHare]

Here's an article showing the weaknesses of the si7021. The author also finds the chip systematically measuring too high a relative humidity.

http://www.kandrsmith.org/rjs/misc/hygrometers/calib_many.html

At least for me that's a useful paper because I have some si7021 breakout boards that look just like the one that he tested.  On the other hand, they're cheap boards, and I have doubts that the manufacturer did the rehydration cycle after making them.  Who knows whether the breakout board manufacturer stayed within the bounds of the sensor's reflow profile.  In general, though, the impression I get is that by default most of the ultra-cheap generic board assemblers know nothing at all about the parts they are putting together aside from "part A" goes into "position B" with a certain orientation.

The only (meager) datapoint I have to offer is that the humidity reported by a  si7021 breakout board was in large disagreement with my Lennox thermostat, which does humidity readings so that I can set it to a "feels like" temperature rather than an absolute temperature. 

[Felix]

I would expect the sensor to give readings within the accuracy tolerance of each other if the nodes are identically built and tested. Not sure if how you place them in the jar matters, or if it can be assumed the humidity is spatially uniform in that jar? Just a thought...

If I recall correctly, for humidity measurements there's a heating element on the die to heat up, maybe that's something to look into to stabilize the readings.
Anyhow the new WeatherShields have the BME280 sensor now.

[TomWS]

Odd.  All the motes I've had/have that use the Si7021 (approximately 10-12) have had plausible humidity readings.  They might be off by as much as 5%, but nothing along the lines I've seen cited here and typically within reasonable range of other sensors in the similar vicinity (including different components, ie HIH-6120). 

I wonder if it has to do with possible contamination of the sensor.  I've been very careful to keep all liquids, including 'noclean' solder flux away from the sensor port, even when cleaning the board with IPA, I've stayed away from the sensor.  Given that 'noclean' flux is hydroscopic, ISTM that this wouldn't be a good contaminant. 

I don't know and I haven't tested hundreds, but, as I say, I haven't had an issue with this sensor.

Tom

[WhiteHare]

In fact, there may be nothing wrong with the sensor per se.  Silicon Labs is a reputable company, and I have no reason to suspect that the sensor falls short of what's promised by the datasheet.  Rather, I was just pointing out that the paper joelucid cited seems to confuse the sensor itself with an inexpensive breakout board (BoB) that appears to have the sensor on it.  The difference is that the BoB manufacturer may not have  properly adhered to the precautionary directions contained in the datasheet during the course of making the BoB. 

[joelucid]

Yeah the issues I'm having with the sensor obviously could have to do with my homegrown brachial assembly methods. Still if the bme280 is more robust against this type of abuse it might be a good one to try out. I think I'm going to order a couple.

What definitely doesn't work is the dht22. I don't have one that has survived a year with still valid readings.

Joe

[WhiteHare]

Yeah the issues I'm having with the sensor obviously could have to do with my homegrown brachial assembly methods. Still if the bme280 is more robust against this type of abuse it might be a good one to try out. I think I'm going to order a couple.

Joe

I've never heard the term "brachial assembly method" before,  but I'm guessing you are referring to manual assembly?

@Felix
When you previously did your weather shield node using the si7021, did you do the rehydration cycle, and if so, how were you controlling your humidity during the rehydration phase?

BTW, I'm reasonably sure that humidity within a jar is pretty much uniform, due to Brownian motion.  I say that because I've moved some pre-made humidity sensors around my house to a variety of different locations, and they seem to register pretty much the same dewpoint regardless of where they're positioned.  RH might be different, though, because that is influenced by temperature, which may in fact vary from one location to the next inside the house, often influenced by the location and  amount of heat load that the sun is impinging on that area of the house.

[Felix]

@Felix
When you previously did your weather shield node using the si7021, did you do the rehydration cycle, and if so, how were you controlling your humidity during the rehydration phase?

No, I did not do anything special afterwards. My lab is a pretty constant 50-60% humidity though and that would be plenty of opportunity to stabilize while they wait for depanelizing and testing.

[WhiteHare]

@Felix
When you previously did your weather shield node using the si7021, did you do the rehydration cycle, and if so, how were you controlling your humidity during the rehydration phase?

No, I did not do anything special afterwards. My lab is a pretty constant 50-60% humidity though and that would be plenty of opportunity to stabilize while they wait for depanelizing and testing.

Have you tested to see if your old si7021's are in good agreement with your new BME280?  If so, maybe your lab humidity--which sounds favorable for rehydration--would explain why. 

@JoeLucid
Do you happen to know what the humidity level at your lab was in the 48 hours after your si7021 fabrication?

[Felix]

Have you tested to see if your old si7021's are in good agreement with your new BME280?  If so, maybe your lab humidity--which sounds favorable for rehydration--would explain why. 

No, I only have a handful of si7021 based nodes and they are deployed throughout the property and I would not consider them test worthy.

FWIW now with this si7021 post-reflow requirement discovery (I was not aware myself, and I have not seen anything like this post-reflow special rehydration in my otherwise limited experience manufacturing electronics) I am quite glad to depart the 2 sensor design anyway.

[CallOne_Phil]

It would seem I kicked off a fun little chat with this post

I've been a bit slammed with work, but wanted to attach a newer shot of all my humidity readings.  I am very confident that the air is/was uniform in the large glass cookie jar I am using.  To verify this I briefly added a mini fan inside the jar and while the temperature did slightly rise from the running fan it seemed to confirm my thinking and has since been removed.  Four of the nodes under this test have an identically build as Felix was wondering.  I have also checked for any splashing or damage and found no detectable signs with my simple USB microscope on the sensors in question.

After reading the link posted by Joe and the input by WhiteHare I believe my issue is the "wetting" of the polymer inside the sensor.  Looking at my newest readings you can see my sensors are lowering their reading to get more inline with the true 63% reading.  The main observation from this is that the sensors reading the highest humidity and the slowest to recover all came from higher humidity sections of my home.  I had the the 2 worst sensors in the basement near my fish tanks which can drive the humidity up to 70ish% on bad summer days.

TomWS, what is the average humidity where you have your sensors deployed? I have a sinking feeling that these sensors work perfectly if the humidity stays in the 45-60% range and anything above that will cause them to creep up.  These sensors were used to help avoid mold from extra moister by my fish tanks, but how fast are they creeping up?  I will continue looking into this and try to keep anyone else interest updated on my findings about this humidity reading creep I seem to have.

As for the last chapter of this way too long post... I have just placed an order for 3 of the new weather shields to test along with some of the trace antenna moteinos(been waiting to try them out) hopefully this will give anyone in my situation more helpful info. I would love to keep using the old si7021 boards, but until I can get a handle on how fast the higher humidity effects the overall readings they will have to be benched for unreliability.

Thanks for all the input everyone,
-Phil