Announcement

Collapse
No announcement yet.

Roast evaluation model: help wanted

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Roast evaluation model: help wanted

    As some of you may know, I have a strong interest in the thermodynamics of coffee roasting and have been playing around with an algorithm that models the roasting process. If it works as well as I think it does, it distills the roast profile into three numbers which will allow a simple comparison between different profiles and hopefully a clear path to manipulating profiles to achieve a desired result.

    The key part there is "if it works as well as I think it does" and that's where I need help. I need more data, based on different roasts.

    Ideally, I want for each roast: the bean source, the roastmonitor CSV file (or at least the "time" and "Value1" columns from the CSV file) or equivalent, your assessment of flavour profile and some indication of final colour: Agtron values would be perfect but failing that any other moderately objective colour indication.

    PM me if you are interested in helping out.

  • #2
    PM sent! Sounds like some fun.

    Comment


    • Lyrebird
      Lyrebird commented
      Editing a comment
      Wrong place, see below.

  • #3
    OK here's a basic outline of what I'm doing.

    Some of you may be familiar with the concept of the Pasteurisation Unit (PU), a method of calculating the amount of heat that has been applied to something that is being heat treated for microbial stability.

    One PU is the equivalent heat dose to that received when the process is held at 60 oC for one minute. Total heat dose in PU = 1.39(T-60) x time (with T in Celsius and time in minutes).

    Very commonly used in process industries, works a treat. As temperature in the process usually changes with time, you can either derive the integral or take a piecewise linear approach at finite time slices, the latter is simpler and accurate enough for practical purposes.

    It occurred to me that we might be able to quantify a coffee roast profile using a similar approach, so I invented the Caramelisation Unit (CU), one CU being the equivalent caramelisation to that occuring when coffee is held at 200 oC for one minute.

    There are two major caramelisation processes going on in coffee, the first being the familiar Maillard reactions, the second being simple sugar caramelisation which becomes predominant later in the roast. This presents a complication but also an advantage as the change in the CU value through the roast tells us something about what reactions are occurring.

    There is enough literature extant on the thermodynamics of coffee roasting to be able to derive appropriate activation energies to build a model using Arrhenius kinetics, so I came up with the project title Coffee Roasting Evaluation Method by Arrhenius. Gotta love a backronym.

    Arrhenius kinetics are a bit cumbersome so I then simplify into a pair of equations similar in form to the PU equation above and use these to derive three numbers that define the roast profile.

    As an example for a recent roast of mine the numbers are 4.9, 9.9, 14.8. I think I know what these numbers mean and thus how to manipulate the roast profile if I want to change it.

    As an example the third number is the total CU through the roast and tells me that the roast is on the dark side of medium. Obviously increasing the number increases the degree of roast but the advantage is that I can predict this before dropping the roast by tracking the CU number and also (hopefully) reproduce it with some precision

    I need more data to test out this theory. That's where you come in.
    Last edited by Lyrebird; 15 March 2021, 10:50 PM.

    Comment


    • ethanio12345
      ethanio12345 commented
      Editing a comment
      Sure, I was thinking more about interplay between bean temperature and air temperature (and how heat is transferred between air-beans and drum-beans) may result in different caramelisations for a given temperature. So, 1 minute at 200C for pure direct irradiation will not result in the same colour change or CU vs 1 minute only drum roasting vs 1 minute more air roasting. It was more I'm not sure what your benchmark is because caramelisation is a very complex mechanism (more so that PU where the measure is cell death).

      Point above being, you may have temp vs time plots for the entire roast, but that may not tell enough of the story (or it may, hence the question about how well your model fits the experimental data).
      Last edited by ethanio12345; 16 March 2021, 03:35 PM. Reason: Words bad. Need more coffee.

    • Lyrebird
      Lyrebird commented
      Editing a comment
      That would be the case if different heating regimes lead to different intrabean heat fluxes.
      If you read Basile's excellent paper on modelling heat transfer in different roasting regimes,
      https://www.semanticscholar.org/pape...1e466a21ae74f6

      He comes to the conclusion that in all cases the Biot number is sufficiently low that the adherent air film is the major factor that influences heat flux into the bean. In turn this should mean that any differences in caramelisation rates will be small.

      I am hoping that they are small enough to ignore but as always I am prepared to be wrong.

    • ethanio12345
      ethanio12345 commented
      Editing a comment
      If that's the case, wonderful, and sounds fair! I wouldn't have thought they'd be low enough with the temperature as conventional and fan forced baking result...still, I'll trust the science more than my intuition, thanks for the reference. If the values are low enough to not create a statistically significant result, then I'd agree bean temp should be a dependent measurement. I would have thought that having environment temp would have been useful for a) what you'd expect the beans to asymptotically approach and b) help match out rates of change.

      That point r.e. if it is actually small enough to ignore should show up if/when you plot your model based on the calculated values. I.e. if the turning points (thinking more time than temperature) and gradients match up (to within a given uncertainty) at certain relevant points. But again, if the model matches the data well enough, may not even be necessary. ?

  • #4
    Most of that is Greek to me but i love the concept.
    Are Kaffelogic Nano roast logs any good to you?

    Comment


    • #5
      Lyrebird,, fascinating stuff and well above my pay grade, but good luck testing your modelling. I was reading an article the other day that portrayed the gentleness of heat application at around FC for Ethiopian beans as more of an Art than a Science,(my inference of the presentation) - probably more a testimony to the skills of the roaster than negating the science. Again best of luck.

      Comment


      • Lyrebird
        Lyrebird commented
        Editing a comment
        Heat application at around the inflection point between the two caramelisation regimes would appear to be very important.

        The second number in my triplet above is an indication of where this inflection is occuring. The differences between the numbers then indicate how much heat is being applied either side.

        As I see it, first crack ismuch less important than is commonly supposed. I think it occurs as a byproduct of other processes that aremuch more important: CO2 evolution and the structural polysaccharides reaching their glass rubber transition temperature.

        This last is a guess as there is precisely no published research on the polysaccharide transitions in coffee roasting even though these processes possibly dominate the late phase of roasting.

    • #6
      Originally posted by amberale View Post
      Most of that is Greek to me but i love the concept.
      Are Kaffelogic Nano roast logs any good to you?
      As long as it has bean temperature vs time in a format I can read, I can use it.

      Comment


      • Barry O'Speedwagon
        Barry O'Speedwagon commented
        Editing a comment
        It seems that you can export a Kaffelogic log as a csv file (Export to Artisan, select csv, seems to be tab delimited so 'text to columns' it if needed). This file just has a couple of time variables and 'BT' (bean temp...via a thermocouple in the middle of the bean mass). You could theoretically match it the roast profile used if desired I guess.

      • Lyrebird
        Lyrebird commented
        Editing a comment
        Very helpful, Thanks

      • ethanio12345
        ethanio12345 commented
        Editing a comment
        Just thinking as well, you may need different calibrations for different diameter thermocouples. Thinking more along early in the roast (when beans and thermocouple temp equilibrate), your variables in the model may not represent the different heatflow and temperature as well.

        I know I've suggested a few things, but it's all to be taken with many grains of salt as I haven't seen the code. ?

    • #7
      One advantage of 100g batches is the high number of repeats( for your purposes).

      Comment


      • #8
        Would you have a Google drive folder/Dropbox people can add the roast files too? When I get around to properly writing down observations I'll send them through .

        I.e. could drop any relevant files in (i.e. test photo/roast logs/observation files - thinking if you could use the CS cards photo or something so everyone is using a similar measure... wouldn't account for printer variation, but allows for link between camera, eyes and objective colour)

        Comment


        • Lyrebird
          Lyrebird commented
          Editing a comment
          My Google account (Lyrebird Cycles) will set something up.

        • Lyrebird
          Lyrebird commented
          Editing a comment
          I meant I will set something up using my business Google account. Google accounts aren't sentient...yet.
      Working...
      X