jalpert wrote:For simplification purposes, why not break your roast into phases and define ideal curves for each phase, which you can then vary based on what you're roasting?

Thanks. That is a good proposition specially towards the finish phase, which I think is most critical because of the temperature window.

jalpert wrote: As with the other posters, I would be uncomfortable trying to define an ideal curve without a suitable model of the underlying chemistry. I feel like the tail is wagging the dog here a bit.

For the finish phase, there could be a model based on the ROR, bean properties (inner bean temperature gradient), and temperature window. As far as I know the ROR should be decreasing on the finish phase based on empirical results. The bean properties would establish the length of the finish phase for inner bean development. And the temperature window would model the reaction window (~400F+). There are more variables but these three would make things easier to work with. It would also be roaster dependent as each roaster is different. If you know what are the three most important variables let me know.

Cheers

I think trying to derive a model profile from underlying chemistry will be a fruitless task; there are just too many variables at play, so that the curve you generate will probably be far more constrained by your modelling assumptions than by your data.

The better approach, I think, is to base your model off of past roasts involving similar beans, and for which you compared a reasonable number of possible profiles before choosing a favorite based on taste. That is what I have been advocating for here, as well as Marshall and Patrick (I think). Since "similar bean" is such an important caveat, there will be no single "best model profile" for all roasts, but rather, samples that work well for different types of beans or preparations.

For the record, I think Patrick's approach is the best that has been posted to the thread so far. He properly incorporates the assumption that ROR falls faster in the beginning than at the end, which means that he uses a cubic function rather than my quadratic one. It might be simpler to just plug temperature data into excel and model that rather than modelling ROR data and then integrating it, but either way should get you to a similar answer in the end, I think.

Thanks. I think it would help to understand the underlying chemical process so as to adjust the heat to get the best possible results. This does not deny or contradict the experience gained from previous roasts (not exclusive). Roasting by imitation (background profile) does not have to conflict with roasting by hypothetical optimization (ROR, T. Window, Beans). The background profiles could also have subtle differences that cannot be seen with the naked eye.

Cheers

Arpi,

In my (not so humble) opinion, the natural log approach is not the "natural" way to simulate a roast profile. To obey the natural logarithm function, you need to assume 1. The temperature difference gradually decline and 2. The heat capacity of the bean throughout the roast is a constant. And we all know, practically the two assumptions are false.

Let me try an analogy.
If we mount a thermometer on the surface of a hot stove (maintained at low heat), and move a kettle of water (at room temperature) onto the stove. If the heat input is very, very low, the water would never boil. It may heat up to said, 200F, and the effect of the heat escaping into the surrounding environment would play a dominant role and eventually equilibrate at 200F. What do we expect at the thermometer readouts? A deep, a turnaround, and yes, then natural log increase of temperature.

However, we don't roast this way.

We control the heat input during the roast. It's not a constant inflow of heat. (The assumption 1 is false.)

The heat capacity of coffee bean is not a constant. What if we put a kettle of ice cubes and boil it? It needs heat to melt the ice, and it needs heat to evaporate water into steam. While heated, coffee beans may undergo various phase transition steps. There is a term "glass transition temperature", and cellulose wall would physically rearrange within this temperature range. This is phase transition, and I suspect the typical "flick" prior to the first crack and the second crack may have something to do with the phase transition. Your natural logarithm approach couldn't account for it.

For the record, declining RoR is not the only way to roast. A good model should be able to simulate various profiles.

Hi.

These are some points that drove me in favour in the ln() curve model:

1 The ln() curve has an always positive slope (never sinking) that decreases steadily (ROR incrementally goes down with time). This models the current trend in roasting in terms of ROR and best results.

2 The concentration of chemical products in a first order reaction follows a logarithmic curve (ln). Most reactions speed up as temperature increases and the concentration of reactants decreases. There could be some parallels between ROR and speed of the chemical reaction.

http://www.csus.edu/indiv/m/mackj/chem142/kinetics.pdf


As far as your arguments, they are welcome but they show that they are also based on assumptions. That is, your assumptions about my assumptions are victims of also being assumptions. Guess what, nobody could do anything without first assumptions. Roasting is complex and there is not enough info out there as it encompasses many disciplines.

Cheers

Arpi wrote: http://www.csus.edu/indiv/m/mackj/chem142/kinetics.pdf

On the discussions on assumptions, dont these slides assume constant temperature in each reaction? Similar to half-life decomposition, it's an exponential decay of a compound turning into its chemical ingredients. How is this related to RoR in roasting, when our goal is <insert yours here; mine has to do with maximizing sugars>?

Hi.

There are many variables that affect the whole process. In the slides, there is a discussion of how temperatures affect the reactions at the end. It says that temperature affects the rate of the reaction exponentially for a first order reaction. In the whole process, one has to consider the inner temperature gradient (bean shape, size, and density), and the dominating heat transfer method. The reactants are also dynamic, limited, and activated within a window of temperature, pressure, and time across an uneven 3D bean. These obstacles, and other, make roasting difficult to turn into an exact science and formulas and profiles will not work equally for everybody. However, they could be used as a starting point and adapted with experience.

Cheers

Based on Rafaels idea I added some support for Natural Roasts to Artisan.



The approach and new feature is described in a short post on the Artisan blog

http://artisan-roasterscope.blogspot.de ... oasts.html

A prerelease version of Artisan v0.9.8 to play with this is available on the projects GoogleDrive at

https://drive.google.com/drive/folders/ ... FNJYjRUTlE

Marko

Thanks Marko,
This made me take my first look at using the designer..
I really appreciate Artisan and think roasting would be much harder for me without it..