Relative humidity measurement. Best practices.

[miro123]

Hello,
I have questions - How can I accurately measure RH? Short list of requirements are listed below.
   - accuracy +-2% -> 10years
   - working temperature range - T=5...50^oC
   - storage temperature  T=0...40^oC
   - normal indoor condition - no harsh environment and exposure to elements excluded in sensor datasheets.
   - re-calibration period. TBD - How often and how?
   - initial calibration - Do I need it and what technique/ e.g wet bulb, NaCl_75% or reference instrument? test condition https://www.astm.org/e0104-20a.html

Why I asking this questions?
 - I'm busy with building of component models. Voltage and resistors references,ADC and AFE boards. Models relay on accurate data.
 - The data is gathered in real time from local sensors and internet.
  - I'm struggling to get consistent RH% data. I have tried many consumer and DIY sensors/BME280/SHT 1x/2x/3x. Just added HDC1080 and considering experimenting with SHT45 EVM
  - the trends data is ok, but absolute value has StdDev= 7..10%

Best regards,
Miro

[mendip_discovery]

I can't think of the documents at the moment but from what I have read RH is a bit of a vague thing. It is not an exact science.

[Kleinstein]

The best accuracy is likely from a cooled missor type instrument, measuring the dew point and difference to the room temperature.
The absorbing sensors like BME280, SHT..  have the problem that they can also be effected from other gases in the air. This can be temporary, but also causing drift over time - here even relatively low levels could be an issue. These sensors would need a more frequent re-calibration - especially initially to get an idea of the drift and scattering.

For the calibration a defined salt solution (not sure of NaCl is the best choice, one may want more test points at least for the initial calibration) should be good enough. The cooled mirror type may get away without from getting a zero power temperature reading to get the RT to dew point zero.

Accurate RH only makes sense if the temperature is also uniform and the sensor also gets the right temperature. 2% or the RH should correspond to some 0.3 K in the temperature. Central Europe usually has relatively high RH - no need to worry about single digits.
With a person in the room the RH can actually scatter quite a bit and 10% variations may be real. The problem is with the temperature and humidity source, especially when relative dry.

[dietert1]

miro123, would you show some data?
What is your humidity standard? Using extreme points of 0% RH and 100 %RH at different temperatures?
Or are you logging various sensors against each other?
I think one meaningful test would be logging humidity and temperature in a hermetically closed box. At intermediate values of RH there should be a pretty obvious correlation. I have been using SHT35 and SHT45 sensors and they measure both temperature and humidity to calculate relative humidity.
I have seen 10 or 12 % deviations from another sensor at the "dry" extreme. While the other sensor would yield about 10 % an SHT35 stayed at exactly zero with fresh desiccant for several weeks, which isn't probably true either.

Regards, Dieter

[bck]

There are some publications on that topic:

NPL 1996, A Guide to the Measurement of Humidity
https://www.npl.co.uk/special-pages/guides/a-guide-to-the-measure-of-humidity-gpg103

If you want to use a wetbulb-instrument as reference:
ASCE 2005, THE ASCE STANDARDIZED REFERENCE EVAPOTRANSPIRATION EQUATION
https://www.mesonet.org/images/site/ASCE_Evapotranspiration_Formula.pdf

German NMI guide to calibrate Hygrometers
PTB 2019, Kalibrierung von Hygrometern zur direkten Erfassung der relativen Feuchte
https://www.ptb.de/cms/fileadmin/internet/dienstleistungen/dkd/archiv/Publikationen/Richtlinien/DKD-R_5-8_2019.pdf

If you really need the 2% accuracy, i really think you need recalibration cycles of 6mo or less if you intend to use regular humidity sensors. (capacitive hygrometers)
If you use an dewpointmirror as reference instrument with these uncertainties: Tdew: 0.15K Tair:0.08K, the resulting RH uncertainty is at 10%RH ~0.15%RH, at 50%RH ~0,55%RH and at 90%RH ~0.95%RH (assuming Tair = 23°C)

The drift of cheap sensors can be >4%RH in a year at low RH, or they just go down to 20%RH and then stop measuring further down (keep indicating 20%RH, or loose the sensitivity)

maybe look for humidity sensors with higher surface area, they are more stable, but generally, all capacitive humidity sensors need initial calibration.

[RoGeorge]

Many electronic sensors have some gel-like cover that absorbs the water vapors and then a circuit measures the resistivity or capacitance.  The problem (in my experience) is that the gel-like material tend to dry/crack/change its properties in a few years.

The chilled-mirror method (based on the dew-point https://en.wikipedia.org/wiki/Dew_point) seems better for long term reliability and precise humidity measurements.
https://www.yesinc.com/resource/products/hygrometrytemperature/humidityds.pdf

[Andreas]

Hello,

rH is rather difficult to measure.
In some ranges a temperature difference of 1 deg C equals to a 6 % change in rH.
So self heating of the sensors is very critical.

with best regards

Andreas

[mzzj]

Your requirements are actually rather strict and not trivial to archieve.

Vaisala products are pretty well proven and one of the best options before going total nuts with cooled dew point mirrors.
I'd go for Vaisala HMP7 probe, (with Indigo 200/500 display if needed) and Vaisala HMK15 saturated salt calibration kit with 11% and 75% salt solution.
https://www.vaisala.com/en/products/instruments-sensors-and-other-measurement-devices/instruments-industrial-measurements/hmp7
https://store.vaisala.com/en/products/HMP7
https://store.vaisala.com/en/products/HMK15#Configuration

6 month calibration/verification cycle, full calibration at manufacturer every 5 years or whenever there is sudden large error on 6mo. saturated salt calibration.
And fresh salt solution every 3 to 5 years.

HMP7 is specified accuracy is ±0.8 %RH (0 … 90 %RH) and ±1.5 %RH 90 … 100 %RH)

[Overspeed]

Hello

You can perhaps recover humidity sensor from Vaisala weather probe fit on weather balloon , there are sometime available on Ebay

https://bardagjy.com/?p=1370

Vaisala RS80 Radiosonde

Regards
OS

[dietert1]

SHT35 sensor is specified with 2 %RH typical accuracy (except the extremes) and 0.25 % RH typical drift per year.

Here a little log from an SHT35 and a BMP280 that i use for ambient monitoring:
SHT=24.70°C 72.3%
BMP=24.69°C 927.1hpa
SHT=24.70°C 72.2%
BMP=24.69°C 927.1hpa
SHT=24.68°C 72.3%
BMP=24.69°C 927.2hpa
SHT=24.71°C 72.3%
BMP=24.70°C 927.1hpa
SHT=24.68°C 72.3%
BMP=24.69°C 927.1hpa
SHT=24.68°C 72.3%
BMP=24.70°C 927.1hpa
SHT=24.71°C 72.3%
BMP=24.68°C 927.1hpa
SHT=24.70°C 72.3%
BMP=24.70°C 927.1hpa
SHT=24.68°C 72.3%
BMP=24.69°C 927.1hpa
SHT=24.68°C 72.3%
BMP=24.69°C 927.1hpa
SHT=24.71°C 72.3%
BMP=24.68°C 927.1hpa

This is with one second sampling. These integrated sensors with digital calibration represent a rather advanced technology level. I think they are good for scientific work, maybe better than some used sensor from ebay, where you don't know the history and calibration may be lost due to age.

Regards, Dieter

[Conrad Hoffman]

2% is tough! That said, it isn't very hard to build a chilled mirror hygrometer if you're handy. The peltier and control loop are easy; use an LED and photodiode to reflect off the mirror. The biggest problem I had was the mirror itself. It needs to be very conductive, so something like polished copper, but it needs to resist oxidation. I don't know what's done commercially, but gold plating sounds like the answer. Bury a temperature IC in the copper and use another one for the ambient. A small muffin fan can do dual duty for cooling the back side of the peltier and maintaining some airflow over the mirror. The beauty of a chilled mirror design is, done properly, it's a primary instrument.

[dietert1]

Yesterday i found an improved version of SHT45 that is specified with +/-1 % RH tolerance, SEK-SHT45-AD1B, at rs-components. It is announced for February 9th.
https://de.rs-online.com/web/p/entwicklungstools-sensorik/2436411

I am intentionally linking the test kit, as it is a convenient method to avoid errors while soldering the sensor. One can extend the flexible board with some flatcable or stereo audio cable, in order to have the two I2C lines shielded. One shield is Gnd, one is supply. Depends a bit on cable length.

Regards, Dieter

[mendip_discovery]

How about the SHT85

https://sensirion.com/products/catalog/SHT85/

[Conrad Hoffman]

That SHT85 looks like the way to go; nice part and they do a good job of specifying the accuracy in the datasheet. Now, for the less sensible, here's the chilled mirror device I built. In the middle is an aluminum block with a circle of holes drilled to match the fan blades. On top is a peltier with a copper top hat mirror. A little plastic U-shaped piece presses the mirror against the peltier. You can see an angled LED on top that reflects off the mirror onto a photodiode. Out of sight on the back of the aluminum block is the TO-220 power device that controls the peltier. An LT1013 takes care of the control loop and signals. The connector on the right carries the ambient temperature and mirror temperature to a DVM or ADC. The ambient sensor is floating in air in front, with a small IC heatsink attached. The wires are short lengths of constantan for thermal isolation. I need to find the schematic for the thing, which has been lost for a while. We used this at work to confirm readings inside an environmental chamber and it worked well, but it's way overdue to have the mirror polished and maybe plated.

[Anders Petersson]

Get a reference instrument. For example Rotronic HC2A-S 0.8 %RH accuracy, or extra-accuracy 0.5 %RH:
https://www.processsensing.com/en-us/products/humidity-probes-hc2a-s-sh-hh.htm

Send it for calibration regularly. Yearly is most common but you may do it more often in the beginning to get a calibration history that gives confidence. For me the calibration cost was around $100 (but they only calibrated at 33% and 75%, to about 1 %RH accuracy).

Build a few boxes with saturated salt/water solution, or silica gel for 5-10 %RH. I built a perforated floor to put sensors on. The floor is attached to the lid so all sensors follow the lid when I move the lid to measure in another box. Attached under the floor, a PC fan just above the water mixes the air. See attachments.

I find the humidity has mostly recovered a few minutes after inserting the sensors to the box. But the RH level is not so dependable that a reference instrument can be omitted. I'm writing an article to explain this method to do RH calibration.

I've tested with about eight different humidity levels. Sensors can definitely have errors, sometimes 10-15 %RH. So far all sensors I tested have been fairly linear (about 1-3 %RH nonlinearity), so ordinarily I expect testing at two points to suffice. I attach one such measurement series. The curves have some common features that I suspect are artifacts from the reference.

For that reason I've also bought several chilled mirror hygrometers to upgrade my test setup. It's definitely extra work (pumping the air through tubing etc) so I suggest to start with a capacitive reference. I plan to sell some of my chilled mirrors once I tested them thoroughly against each other.

[miro123]

Thanks for helping me understand the problems associated with humidity measurements.
In addition to above mentioned method there is also  wet bulb method. It was used in the past before - petier/dew point meters becomes popular. There ware some tables back in those days, without computers and limited functions calculators.
https://us.flukecal.com/blog/how-accurately-calibrate-hygrometer

@Anders Petersson - Where can I buy MgCl₂? Is it the food grade or industrial?

[mendip_discovery]

Extech do some 33%RH and 75%RH humidity standards for notalot.

Extech RH300-CAL iir
https://www.test-meter.co.uk/extech-rh300-cal-humidity-solutions-calibration-kit

Prices vary, I got ours for 30ish.

[mzzj]

@Anders Petersson - Where can I buy MgCl₂? Is it the food grade or industrial?

I have used MgCl from Vaisala (Link I gave earlier has MgCl as option) and from laboratory chemical distributor.

Both worked fine for my purposes but Vaisala salt kit comes with calibration certificate that sort of gives you traceability. Without one it is bit of quesswork what the generated humidity is.
Even if you have purity information available like >99% it probably makes big difference what compound the last 1% is. 

[Anders Petersson]

@Anders Petersson - Where can I buy MgCl₂? Is it the food grade or industrial?

It's available on Ebay or various sources. I bought from an aquarium shop (I guess they use it to regulate the chemistry in salt water aquariums). That particular store seems offline now. The price was EUR 11 for 1 kg. I don't expect it has any particular purity.
But when you use salts for high-accuracy calibration purposes, would you dare to assume the nominal humidity even with very pure salts? With some salts, the nominal humidity varies with temperature. And in a relatively large box like these, the high-humidity salts (>90 %RH) take hours to recover the last few % of the equilibrium humidity so doing an early reading saves time. MgCl2 isn't that bad, but it's easiest to always use a reference sensor. And using lower-grade salts saves hundreds of dollars. Heck, I even have a completely empty box to get a free comparison at the ambient humidity level of the day.

I found MgCl2 to clump badly even before dissolving in water and it could dissolve much less water than expected, so I guess it already absorbed moisture from the air. Not a problem -- just add a little water at a time until you have a partially liquid solution.

[dietert1]

Apparently some humidity sensors include a heater to handle high humidity or condensation. So let's say relative humidity is above 90%. One would turn on the heater to dry the sensor and then make another measurement. Using another temperature sensor not affected by the heater one could then calculate (the higher) relative humidity for (the lower) ambient temperatue. Any experience with this?
Also i am wondering about how relative humidity depends on barometric pressure. As far as i understand those integrated sensors won't measure barometric pressure but assume a default value.

Regards, Dieter

[miro123]

Apparently some humidity sensors include a heater to handle high humidity or condensation. So let's say relative humidity is above 90%. O

I've used this feature on my HDC1080 sensors. I found it only useful,  when I measure outdoor conditions. My experience is that  gel based humidity sensors has the highest and the longest humidity hysteresis. BME/BMP280 and SHT2x/3x are returning faster to normal operation once the outdoor condition get beter. Mostly humidity jamming occurred at night, once the sun shines they are returning to normal operation
I don't seen any use case for this features for any indoor conditions.

Also i am wondering about how relative humidity depends on barometric pressure. As far as i understand those integrated sensors won't measure barometric pressure but assume a default value.

No idea how big is the influence of the air pressure. Even more - i have oposite experiance -pressureless SHT35 gives more consistent results that pressure equipped  BME280 .
I saw that air quality have huge impact - e.g indoor painting , or applying silicone sealant  drives the  humidity sensors crazy. It is expected behavior - datasheet describe it. That is the reason why i avoid buying of such sensors via e-bay.

[mendip_discovery]

Dust etc is a big issue for humidity sensors I have found. If I have ever had to fail a unit its becuase its drifted big time and it's just not worth repairing and just taking a look at the sensor there is quite often a significant amount of dust and dirt on and around the sensor.

[dietert1]

As an example we are in Sao Paulo at a height of about 800 m above see level. The pressure measurement gives 924 hPa, while the local meteorology reports 1015 mb (value normalized to see level).
In this case the pressure effect is about 10 % and i'd guess the density of air and water contained therein may vary similar. I'd guess condensation (100 % RH) will occur at a lower water density if the air density is less. Then the question arises whether a sensor really measures relative humidity or water per air volume. Also concerning metrology i don't know whether a humidity effect on plastic parts or resistor bobbins will really be proportional to relative humidity or water density.

Regards, Dieter

PS: I started to compare SHT35 measurements with those of the local weather report and found an average difference of 1.2 % RH and a standard deviation of 2.5 % RH. Data in a range of 60 to 80 %RH. This includes a calculation of water density and redetermination of %RH according to the measured lab temperature.

[miro123]

As an example we are in Sao Paulo at a height of about 800 m above see level.

Interesting question. I have never has such problem. I have the luyxury that I leve in The Netherlands. My house fundament is 12m above see level.
Country is highly populated - I have at least 3 profesional/goverment meteological station in radius of 12km . One semi professional station is just 1km from me.
I  have 3 BME280 - the absoulte atmosferic pressiure seems to be prety close - best time to do this by steady state wether - the most meteo stations give the same air air presure - normalized to altitude /Happily the altidude diference here is just 10..15meters./.
It was interesting that the absolute air pressure value is prety close to the government stations data.
------
I cannot say the same for humidity - forest,my small vilige, agriculture area, lakes and rivers has different RH - I experiance it with naked eye in foggy weather
I cannot say the same for temperature - I cannot achieve equal temperature even at 10mm distance

-------------------
I'm not expert in this areay but fast google search shows that relative humidity calculation from dew point is independent from air pressure
  - the only requirement is that dew point chamber mus have the same . 
  - mathematical explanation  - The two part of equation neutralize the air pressure
  - engineer/my/ explanation
1. RH is function of air pressure. Ideal air P*V=const     
2. Dew point occurs when liquid pressure >= air pressure
1 & 2 neutralize each other
https://lambdageeks.com/dew-point-and-pressure/
https://www.orionkikai.co.jp/english/technology/kuatsu/dewpoint-temp/
https://en.wikipedia.org/wiki/Dew_point

[Kleinstein]

The RH is equal to the water partial pressure relative to the vapor pressure over clean water.
There is very little interaction between the gases. So the pressure does not have an effect.

What may effect the dew point measurement are water soluble gases, as they lower the vapor pressure of the condensed water. CO2 would be such an example.  Still the effect should be small.
One has to really look at the details one what is meant with RH to tell if it is relevant.

[dietert1]

Today i supplemented my ambient monitoring kit with a TI HDC1080 sensor. When it arrived its results were about 8 % high in comparison to the SHT35, e.g. 81 %RH instead of 73 %RH. Datasheet specs are +/- 2 % typical for the SHT35 and +/- 3 % typical for the HDC1080 in this humidity range. In the HDC1080 manual they write that a systematic shift is expected after an extended period of high humidity (above 70 %RH).
Right now i am using the built-in heater with about 4.5 °C temperature increase. As a result the HDC1080 reads less than 70 %RH and is slowly recovering. The first 2 % of the initial 8 % shift is already gone after about one hour.
Appears like in a high humidity ambient the heater should be normally on and turned off only for taking a valid measurement. For the time being this would be my conclusion.

Regards, Dieter

[Anders Petersson]

Some relevant Sensirion documents.

Recommended RH testing conditions (stringent!):
https://sensirion.com/media/documents/17D5076D/6350F402/Sensirion_Humidity_Sensors_Testing_at_Ambient_Conditions.pdf

Using the built-in heater of SHT4x for decontamination:
https://sensirion.com/media/documents/FEE9F039/62459F54/Application_Note_Heater_Decontamination_SHT4xX.pdf

[dietert1]

Thanks. As they propose about 1 hour at 110 °C as temperature for reconditioning, i will try with an external heater for the HDC1080 to accelerate the process a bit, using soldering iron. The risc should be small, as the sensor is monitoring its temperature. Currently it's running at about 70 °C, measuring 20 %RH.

Regards, Dieter

[dietert1]

I think i want to put a stronger heater onto the HDC1080 and operate it at a higher temperature adjusted such that it outputs a relative humidity of 50 %RH. The 50 %RH temperature will be somewhat similar to the dew point except not at 100 %RH. Then a little calculation should give relative humidity at ambient temperature. Maybe that's a good method to use those sensors, at least in a high humidity environment like here.

Regards, Dieter

[CorporateReference]

Hello, this may help with your quest. We calibrate around 60 Fluke 2626 sensors a year (Used with the 1620A).  The 2626-S and -H sensor are exactly the same, they just use tighter calibration points. They both use the Honeywell HIH-4000 Sensor. Fluke does not pre-age these, however we have found after a year they are more stable (The Fluke manual references the initial calibration should be 2x in the first year to support this).  When our contract started, we had some sensors that had not been calibrated for 10 years and they were only out of spec by <5%.  From our experience after soldering, the Sensirion SHT45 and SHT85 require a bit more processing for stability (drying out and re-humidifying) vs the HIH-4000.

For calibration in field, we use a check/reference probe (HC2A-SH calibrated w/ Thunder Scientific PP Chamber) together with saturated salt solutions.  We use Lithium Chloride, Potassium Acetate, Potassium Carbonate, and Sodium Chloride - All high purity ACS/Reagent grade mixed with DI water. Make sure to have more solute in the solution to maintain the equilibrium in case of excess moisture.  As long as you are able to keep them at a fixed temp (We keep and use these in a temperature chamber or dry well calibrator) and allow 30min-1hr time to settle after inserting the probe this method has proven very stable and repeatable within <0.5%RH.

Refer to the NIST article https://nvlpubs.nist.gov/nistpubs/jres/81A/jresv81An1p89_A1b.pdf for your solution of choice.

We use Pyrex 1399 Bottles to keep the salt solutions. During calibration, we use the 195-45HTSC caps with a silicone septa that has had a hole punched through to seal the sensor. For odd sensors, we also have this silicone seal (DWK Kimble Silicone Lid) that can be put over - see photo.

[miro123]

I would like to everyone for valuable feedback.
I already start improving my system. Results are promising. Lessons learned.
 - temperature and humidity probing are the main source of errors - Applying knowledge from my project "sub mK TC oven" was helpful.
 - consumer grade IC are not robust enough e.g humidity creeping and many other side effects
https://sensirion.com/media/documents/A88858C9/629626D4/Application_Note_Creep_Mitigation_SHT4x.pdf
 - trick learned from local meteo administrator - use forced air to measure Temperature and humidity.  It achieve at least two goals 1.dry the humidity sensor polymer. 2. Equalize the ambient air.
 - industrial grade capacite sensors are step up from consumer garde BMExxx/SHTxx, HTH seems to be somewhere in the middle.
 - Ulitimate performes is aschived with dew point/chilled mirror/ hygrometers - proposed from Conrad Hoffman. COTS Michell S4000
 - the heater on consumer sensors solves one problem but introduce another - small heater works only in thermal insulated configuration. Result is wrong temperature measurements and we are back to initial problem again
  - last but not least - frequent calibration and calibration history is a must.
-------------
Some hyperlinks
https://sensirion.com/media/documents/A88858C9/629626D4/Application_Note_Creep_Mitigation_SHT4x.pdf
https://www.knmi.nl/kennis-en-datacentrum/publicatie/knmi-humidity-sensor-test
https://cdn.knmi.nl/system/data_center_publications/files/000/068/576/original/p325_bloeminkms_netherlands.pdf?1495621276
https://www.vaisala.com/en/search?k=&content_group=products%20and%20services&product_type%5B0%5D=2171&measurement%5B0%5D=776
https://www.rotronic.com/en/humidity-measurement-feuchtemessung-temperaturmessungs/humidity-measurement-feuchte-messung/probes-filters
http://www.michell.com/us/products/s4000_rs_and_trs.htm

Best Regards,
Miro

[dietert1]

While waiting for the real thing, i made two test bottles with desiccant bags. In one bottle the desiccant generates about 20 %RH (after drying bags in oven), in the other one there is about 55 %RH. These bottles maintain sensor measurements constant within +/- 0.5 % over several days.
Today ambient has 46 %RH, everything at temperatures of 19°C. The diagram shows how fast the sensors react after going from one bottle to the other and how slow nominal humidity in the second bottle recovers after mixing with ambient air.
Hope it will be faster with liquid chemistry.

Regards, Dieter

[Anders Petersson]

I guess the last seven data points are the steady-state after a long delay.

The sensor response time and the chamber RH recovery time are somewhat sped up by adding a fan. See the fan in my earlier post https://www.eevblog.com/forum/metrology/relative-humidity-measurement-best-practices/msg4670443/#msg4670443
But the recovery time will still take minutes or hours also with chemicals. I don't bother waiting until perfect equilibrium occurs.

I don't think the sensor response time is a property that most people care about. I do however... attachment shows my module with a fast-response RH sensor. The large sensor area gives a 0.3 second response time. By the way, PTB has built a test rig for RH response time.

[dietert1]

I guess the last seven data points are the steady-state after a long delay.
..

Yes, after about 2 hours.
I posted the diagram to show how the steady state of the SHT35 agrees to the 1 minute response of the HDC1080 (samples 32..34). The observed differences may result from the HDC1080 calibration procedure.

Regards, Dieter

[dietert1]

Last week i made a small setup with some I2C sensors currently not in use. Also i got five new SHT45 sensors (the ones i mentioned above). They agree very well to each other. The SHT35 (about 2 years old ) is about 5 % off and matches the HDC1080 results better.
The table is from 6 hours of data logging with the setup inside a box. 1 sample per second, no apparent self heating.
Of course this can't replace a calibration, but it's a preparation. Chemistry will arrive soon.

Regards, Dieter

[CorporateReference]

Airflow/small movements can affect most of these sensors in the fractional percent range, as well as sources of bright light. For more stable readings you can try placing a small amount of foam over the sensor (as Fluke does) or some type of plastic shield/tubing around the sensor such as in Ander's photo. If it will be in a sealed/protected area then it doesn't matter.