Sense Hat Rubbish temperature accuracy

[Roaders]

I've just got my Sense HAT after waiting about 2 months for it to be delivered from china and am VERY disappointing with the temperature accuracy. I think that it is probably reading temperature at around 10 degrees too high.

What a waste of time and money...

[MarkHaysHarris777]

I've just got my Sense HAT after waiting about 2 months for it to be delivered from china and am VERY disappointing with the temperature accuracy. I think that it is probably reading temperature at around 10 degrees too high.

What a waste of time and money...

Any temperature measuring system needs to be calibrated (and because you control the software, that's really not bad).

You don't expect a PI HAT to be calibrated for you, do you? I mean, its going to be in the (as we say on this side of the ocean) in the ball park, more or less, but you'll have to make sure that its temperature readings are mapped to linear calibrated units that are useful for science; and that's not hard to do, for a scientist.

( which we all are, right ? )

... and while you're working at it, make sure its calibrated in both the C and F scales...

[bensimmo]

I've just got my Sense HAT after waiting about 2 months for it to be delivered from china and am VERY disappointing with the temperature accuracy. I think that it is probably reading temperature at around 10 degrees too high.

What a waste of time and money...

Any temperature measuring system needs to be calibrated (and because you control the software, that's really not bad).

You don't expect a PI HAT to be calibrated for you, do you? I mean, its going to be in the (as we say on this side of the ocean) in the ball park, more or less, but you'll have to make sure that its temperature readings are mapped to linear calibrated units that are useful for science; and that's not hard to do, for a scientist.

( which we all are, right ? )

... and while you're working at it, make sure its calibrated in both the C and F scales...

ah but that is not the problem with the SenseHAT.
It's problem is there is too much heat around the two sensors, they are actually pretty accurate and both collaborate with each other. Unfortunatly everything around them is warming the ambient temperature it sees up.


You need to use some cheap DS18B20 1-Wire to get a good result away from the Pi and HAT (and can be used at the same time and are easy to implement).

There are some er bodges (sorry I mean scientific calculated mappings) using internal CPU temperature, but hit the SenseHAT forum for that, which is still unfortunatly called the AstroPI forum viewforum.php?f=104

[DougieLawson]

viewtopic.php?f=104&t=148103

[alphanumeric]

The heat transfer from the PI CPU will artificially raise it above ambient. I added a tall stacking header, https://www.adafruit.com/products/1979 soldered to a proto board https://www.adafruit.com/products/2310 in between the two. Problem solved.

[icedcougar]

I've just got my Sense HAT after waiting about 2 months for it to be delivered from china and am VERY disappointing with the temperature accuracy. I think that it is probably reading temperature at around 10 degrees too high.

What a waste of time and money...

That's basically what im doing with mine at the moment is trying to correct the data (just doing a huge ton of collecting at the moment).
Needing to buy a real themometre and possibly run it alongside a UPS one as well to see if i get same results after error correcting.

As some have said, it is unfortunate that the HAT is only 1-2cm off the main board... unfortunately the CPU is basically right under the pressure/humidity chips which do the temperature.

Some ways of dropping temp is having the pi on it's side to allow some air flow.

At the moment im playing around with something like:

###CPU Heat

Code: Select all

output = subprocess.check_output("cat /sys/class/thermal/thermal_zone0/temp", shell=True)
cpu = int(output)/1000

###Avg temp from humidity & pressure

Code: Select all

temperature = (sense.get_temperature_from_pressure() + sense.get_temperature_from_humidity()) / 2

^The above is very close to a 10 degree difference in general. You'll get roughly 40C cpu and an average of around 30C from sense HAT.

doing something like:
temperature = avg temp - (CPU - avg temp) #Basically minus the difference from the sense hat.

Theres another one of:
avg temp - ((CPU - avg temp)/5.466)-6
I dont understand exactly what the /5.466 is meant to be doing; the -6 brings it closer to realistic, this ends up usually 2degrees higher than the above formula.

But it does provide some 'fun' in trying to work our and error correct temperatures. The alternative is getting a female-male 40 pin GPIO so you can extend and move the sense hat away from the pi... but that kind of defeats the purpose of having a tiny pi (given you could of just added that to your normal PC).

I wont lie... it is a very odd set up, and i dont see how the one they sent to the international space station could of been of any valuable use, esp with the case they had which would of restricted airflow almost entirely.

[Davespice]

Yeah we've had some teams base their Astro Pi experiment on calibrating our the error here, some of them have gone to huge lengths to work out how much the CPU load increases it's temperature and all sorts.

Because the Sense HATs that went to the Space Station had the fault we decided that everyone has to have it too, but if we make a rev 2 Sense HAT this will certainly be one thing we will look into fixing.

[richrarobi]

This has been rumbling on ever since the sense hat first appeared.

I have tried all sorts. including a "black-hat hacker".
Currently, I have the hat mounted by use of a few bits of "fake" meccano on some stand offs.
The hat is NEXT to the pi(2), and at a slight sloping angle for visibility, and ROTATED 180 degrees compared to the gpio..
With the pi LAN connections to the left, the sensehat is now facing you and the right way up....
The hat connector is close to the pi gpio, and connection is by use of 13 wire jumpers (m to F) that connect the relevant pins (see raspi pinout), leaving other pins available..

This appears to provide an accurate temperature reading (!) - i.e. absolutely no heat from the pi.