Joris, I am only writing this because I am a geek and like to play with reading accuracy. The 18B20s are not precision temperature sensors and should not be relied on to given anything better than .5-1C accuracy and .25-.5C repeatability. There is more than just a single calibration run that goes into making a system capable of being trusted at precision levels.
I have tried any number of things to do this. I finally ended up going with the traceable standard that is the basis of all metrology. The idea is that there is some standard, and everything is traced to that standard. There are reference units that are very precise, which then are used to measure the difference for the next level of reference units and so on. Precision thermistor systems can get you to about .2C accuracy and class A platinum probe/meter can get you .3C. Either will run you a few hundred dollars. How this plays out in temperature measurement is that there are various types of sensors and different accuracies within those types. Ideally, the probes would all be placed directly in a nonconductive liquid bath. Since I don't have access to a vat of clean fire suppressant and don't want to deal with cleaning up oil, I put the probes in latex gloves and push that into water. The water pushes the air out of the glove and you have fairly good conductivity to dissipate the heat built up during measurement. I also read less frequently than you did and turn the rtd meter on and off to prevent heat buildup. Sensor heating should not be ignored. jerry On 04/06/2014 03:05 PM, Joris Meijerink wrote: > Hi, > > Anybody here tried to calibrate the 1-wire temperature sensors > additional to the factory calibration? Only reference I've found sofar > is this one: > http://yosemitefoothills.com/Thermometers/ThermometerCalibrations.html > > At the moment I've got 30 DS18B20 measuring inside a glass jar in a > little box of polystyrene foam with water and ice. Connected to a > Raspberry Pi which takes a measurement of all the sensors every 2 > minutes. This takes about 30 seconds. After 8 hours the water is still > pretty cold in a room which is approximately 20°C. Measurements: > http://nl.tinypic.com/r/2mz9wed/8 > > The results are a little spiky. That's probably the resolution of the > sensors? Can this be changed? Or do I have to take an average over 5 > measurements the smooth things a bit? > > When using the average of the 30 sensors as being the "true" > temperature, I can use regression analysis to compensate each sensor. > Sofar i've found that linear regression (-0,04 / +0.06) gives me a > better fit then quadratic regression (-0,14 / +0,25). Probably because > of the small range of the temperature at the moment. > > Linear: http://nl.tinypic.com/r/2u8ur1w/8 > Quad: http://nl.tinypic.com/r/23wpeme/8 > > How can one identify the sensors? Is the id code written on the DS18B20 > itself? Or do I have to plug them in one at the time to ID them? > > Kind regards, > Joris Meijerink > > > ------------------------------------------------------------------------------ > _______________________________________________ > Owfs-developers mailing list > Owfs-developers@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/owfs-developers ------------------------------------------------------------------------------ Put Bad Developers to Shame Dominate Development with Jenkins Continuous Integration Continuously Automate Build, Test & Deployment Start a new project now. Try Jenkins in the cloud. http://p.sf.net/sfu/13600_Cloudbees _______________________________________________ Owfs-developers mailing list Owfs-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/owfs-developers
