The Great Optris Debate

  • It is as accurate as the generic advertised calorie count reported for a McDonalds chicken burger, compared to actually measuring the calories in a randomly selected McDonalds chicken burger.


    @PGM : I completely agree: Optris is no replacement for exact calorimetry. Only if you have two identical objects, with identical heat generation, then you may compare the "calories".

    Or simply said: With Optris there are to many free parameters, that you must keep an eye on.


  • MFMP do indeed say that in this paper, along with their recalculation of Lugano performance which is very similar to mine based on their thermal modelling.


    I asked Bob Higgins for clarification, because what they say here is contrary to what the Optris camera software actually does - as you can easily validate for yourself. Wyttenbach - you seem most interested in this - I strongly suggest you get your hands dirty and have a look at the Optris software as I suggest above (with a link to the relevant downloads).


    Bob replied to my e-mail as below (his remarks in italic). You can see that MFMP were making assumptions about the exponent of 3 for all temperatures, since they did not test this. They do not get a matching COP between the two higher power runs, as would be expected, and as TC got. So their results do not have the internal self-validation that TC's do: probably this is a tell-tale sign that more accuracy with the exponent in the higher temperature run was needed.


    Of interest in the MFMP paper is the fact that the Lugano thermal modelling results were way off (by a factor of more than 100%) for the dummy reactor. Specifically the temperature claimed by them for this would be reached at a much lower input power. this result is independent of the possibly flawed exponent, since it is matching claimed Lugano input power against actual measured temperature.


    This is a mystery because it appears to indicate:

    (1) The Lugano testers had some experimental issue that meant they were undermeasuring input power to the dummy run by a large amount. (There was that mystery change in electrical setup, you will remember. It should not have altered anything).

    (2) The Lugano tester theoretical calculations for the dummy run, which match the input to within 5%, were very wrong.


    if you look at their assumptions you will see that to get even this 5% match they had to make some assumptions. It is a sign that here something was indeed wrong. Such wrong results can be generated when people try (in all honesty) to retrofit theory to results instead of predicting results from theory. We do not know what is their error, though there is a good candidate in the convection calculations.


    TC did not attempt to reanalyse the dummy results. I agree with this (still) because the convection data is so full of unsafe assumptions it is very difficult to tell whether correct or no, and for the dummy this constitutes most of the power budget. The convection theoretical analysis could be wrong by 100% or more making the dummy runs unsafe. For the active runs convection is a small part of the total power budget so even if this is very wrong the results (corrected) still have some merit.


    So my best resolution of this is:

    (1) Some unknown error resulting in wrong input power measurement for the dummy

    (2) A retrofitted convection calculation which is much too high that matches the (wrong) input data.


    Further interesting (to obsessives like me) checking would be to test this hypothesis. Suppose the convection powers are all too high. Match them with the MFMP thermal data (thereby making them much lower). Check to see what effect that has on the active power budget (it will bring COP down slightly) and on the corrected COP for the two active tests. this should remain pretty well identical. I'd expect this because although we get some difference in convection budget between the two cases it is mostly linear - and only the non-linear component of this will produce a difference in COP. But it needs doing properly - I've found that expectations here do not always work.


    if the power budget matches less well after this convection corrections we have a continued mystery. Otherwise this new information strengthens TC's (and MFMP's) conclusions. Note that what we know for sure is that the MFMP 1.3 reanalysis of the higher temperature test while broadly correct will be a bit high due to the assumption they made that exponent stayed at the same value for the two tests.


    I'm almost motivated to write this up properly...


    My e-m to Bob Higgins, with Bob's reply in italics

    -------------------------------------------------------------------------------------------------------------------------

    So that I can understand this.
    I am using the PI Connect software you have posted, and a .ravi file of
    raw camera readings together with a layout that marks areas and assigns
    emissivity to each one. I'd expect this is what they would have done,
    since they use the marked areas in their photos?
    So that changing the emissivity can be done at any time, for any one of these areas, and the calculated temperature changes.
    I have posted the raw data for this on LF and you can see that this
    software does exactly what theory says - an exponent varying from approx
    9 to 2 over the range 60C to 1200C (remembering that this is T in K
    that the exponent applies to.
    When MFMP took the data, it was adjusted live while the Optris was imaging
    the dummy reactor at high temperature (~900C as I recall). It was on
    this basis that the 3.0 exponent was selected, not upon an evaluation
    over large temperature range. Though as you point out, it was likely an
    error to use it over such a large range in the analysis. However, it
    is approximately correct in the range where it was used.




    That is an easily verifiable (and verified by several people) fact, so to square this with your observations:
    (1)
    There must be some different and incompatible way to enter area
    emissivities to the camera other than PI Connect with a ravi file from
    the camera. There might be a direct online connection, but it is highly
    surprising that it would be so obviously incompatible?
    or



    (2) Optris must have radically changed this software between the version you used and the version currently posted 2.9.2147.0
    or
    (3)
    I was testing with a .ravi derived from 60C e=0.95 measurement. It
    could be that for some reason at larger radiances corresponding to higher
    temperatures they change (largely) the algorithm.
    or
    (4) Something else.


    Do you have an idea which of these might be true? The difference here
    is not small. For example, at 60C (a common temperature) a 10% change in
    emissivity makes a 1% difference in temperature, rather than a 3%
    difference as would be true for exponent 3. The result would be
    radically different temperatures for an emissivity of 0.5.


    I suspect that the calibration is stored in the Optris camera itself and
    read out by the PI Connect software. I also bet the calibration is
    contained in the .ravi file so that the correct calibration gets used
    for the camera that produced the data. Thus, I suspect that the
    operation of the PI Connect software on re-loaded .ravi data is correct
    and behaves the same as it would with the live data with regards to
    behavior over temperature and E.



    TC's calculations based
    on the correct correction (and this is the same as what the PI Connect
    software does on the MFMP posted sample file) neatly give identical COP
    for the two high temperatures to within 0.5%. this result is robust
    against any of the parameters being changed within reasonable bounds
    and so represents significant internal consistency checking. Your
    recalculation does not do this, so I've no a priori reason to prefer it.


    The first data comparison in the JCMNS-21 paper, in matching the lower
    temperature Lugano thermal state to that of the dummy reactor, was very
    close to where the exponent of 3 was extracted. Because of the failure
    of the Optris during the dummy measurement, the exponent was not
    measured at the higher temperature, and the exponent of 3 was used in
    the extrapolation to somewhat higher temperature thermal state. The
    first, lower temperature thermal state match, is likely very close to
    correct. The second, higher temperature extrapolated thermal state may
    be in error due to the presumed exponent of 3. It would be worth
    recalculating.


    Perhaps
    we can find out how the calibration coefficients are stored in the
    .ravi file and how they are applied. The calibration may include the
    exponent as a function of temperature for the particular camera.

  • Search around for the Padua Reheat Google folder (there should be a link on the MFMP site). I found it without too much difficulty.

    In there are .ravi files for a significant range of temperatures (some are rather large files).

    Here are the ones I downloaded a couple of months ago (image)


    So: I can google this and get an image. The MFMP experiment list does not seem to index older experiments. As always I find their site impenetrable when it comes to researching and checking specific older results. the format is optimised for live comment, and does not well suit reflective research. I'm sure it is all there, just no easy way to navigate or search - but maybe I've just not worked it out.


    Can anyone help me over this? I'd like to check these higher temp ravi files.


    Regards, THH

  • MFMP do indeed say that in this paper, along with their recalculation of Lugano performance which is very similar to mine based on their thermal modelling


    THHuxleynew : Thanks for digging in!


    For me the mfp calibration power measurements show, that it is not easy to correctly model an unknown reactor. Nevertheless, the Optris errors could be reproduced and a simple procedure for COP recalculations/adjustments could be given.

    If an experiment shows a huge COP > 4, then it might be OK to use thermal calorimetry just for verification purposes. But there are many caveats: Single band measurements are sometimes inaccurate especially, if a vendor does recommend to use an other frequency... Convection is difficult to calculate in, if the surface is irregular. Convection, in some bands, leads to mirror-effects and random fluctuation of the signal.

    Conclusion: Any serious measurement must relay on two independent systems. I personally would never accept a final result, that is based on a single type of T measurement. This includes at least 3 TC's that must be used!