Hi everyone, I have done over 100 roasts on the Bullet R2 V1 since I purchased it in June of this year. Coffee has tasted fine for the most part but I’m trying to analyse the data from these roasts for optimisation. The roast times, pre-heat and other settings have varied quite a bit but the IBTS ROR seems to be the same regardless. I’ve downloaded the roast data from RoastWorld/Time but I haven’t learned much from it. So far, I did a slope analysis, standard deviation , etc on all the roasts to get some indicator of a consistent roast versus non-consistent … but nothing has come out of it. Is there some metric in all this data that would indicate a good or bad roast. Now, I do realise when the taste is good, I should accept that but given the era we are in, I really don’t see any problem with trying to get the data to assist or tell some kind of story. Has anyone had any success with this approach?
If you attribute a sensory rating to each roast and then see if you can find common themes for the best sensory performing roasts in the data.
Some starting data points:
Colour: This is a huge one if you’ve not yet established a consistent roast degree for each recipe, i.e. you’re not prioritising colour as to when you drop your beans, then look at your 100 roasts and see which roasts achieve the highest sensory value and see if they share a common colour. It’s likely you’ll favour a particular roast level.
Development Time %: A lot of roasters will rely on hitting a particular window for development time. For example you might find that your best roasts land between 18-23% development time.
I love the fact you’re doing this analysis, I’m sure there will be other useful info from your data set. Perhaps you will discover an optimal charge weight and temperature for your top performing sensory roasts. I think I might have to follow your lead and get a spreadsheet going.
Having said all this, each coffee is unique, and behaves very differently according to varietal, process, origin, micro climate, drying method, however I think it’s a fun little project with some value, despite all the variables.
Thanks for your comments Ben. Yes, it’s the sensory ratings may be the trick. I don’t have a colouring meter though so it’s be difficult to accurately measure. Plus, yellowing and even first crack are not exact numbers either but focusing on known quantities such as pre-heat and charge weights may be the best avenue to start. We’ll see
I bought some bead boxes off Amazon which from which you can create your own roast level chart, and use this as a reference to ensure you match your desired colour accurately every roast. Do a roast and use the trier to pull off beans every 5/10 seconds until you have a range of colours that are PRE and POST your target colour. The bead boxes have several small compartments that you fill. I also a couple of boxes that I use to collect and catalogue green coffee defects and one for coffee roast defects. It’s a great teaching aid.
Air temperature is a good thing to measure too. I noticed an improvement on consistency when measuring it.
You can see how big of a difference airflow can behave between roasts in overlay mode, even when the IBTS and Bean Probe appear to be very similar to the previous roast curve, the air temperature can be very different, especially in the development stage.
Overlay, a great tool when repeating a roast and applying a parameter change to see the consequence in real time, or checking consistency with back to back identical roasts.
Exactly. As you can see the IBTS and Bean Probe are on near identical curve paths to the previous roast, but the air temperature departs dramatically from the previous roast, and that was caused by using different applications of heat to prove that IBTS and BP are not telling the whole story.
Sorry, but how do I get that Air Temperature curve. I don’t see the option on Roastime?
I’m not sure how useful it will be for you, as it is pretty nerdy control freak sort of stuff.
If you are serious about it, I can send you info about how to get it working.
IBTS is good at marking two things we can do with our eyes and ears: mark yellowing phase and first crack. So it is good as an A.I. tool or predicting those markers when using automation across batch sizes and bean types. It is pretty amazing and useful.
Airflow is better for predicting how well the heat or energy is maintained and penetrating the beans.
Airflow is roasting.
People may try to discuss how airflow is not important in roasting on the Bullet, however, if you are not measuring the air in the roaster, you are not really recording the most important factor in roasting evenly.
Hi, Jimmy,
Do you have a way to pull all the raw data from the system, including the notes?
Thanks
@jimmybulletroaster
Did you figure out a way to get an air temp trace directly into the Roast Time profile?
@garyliu09.H1k8 and @pmahon1998.nTac
on a mac I load individual JSON files from from roasTime via the following location:
/Users/USERNAME/Library/Application Support/roast-time/roasts
if you are handy with Python you can probably make use of my code here:
I’ve been on and off working with it for quite a while and close to digging into an analysis of phase times and average RORs in phases. But I’ve been a bit to busy to get it done. I don’t know if my last work was pushed to that repo but we should talk more if you are trying to do more analysis in python.
You can see below that I have several derived or engineered features beyond what is recorded in RoasTime.
When combined with origin, density, and processing type, it expect to be able to gleam a little more insight to guide charge temps and aiming points.
Yes.
It is part of the aillio design, but the feature was never fully implemented and phased out.
This is super helpful, So, the json files are manually downloaded from RT? I did something similar with Python, albeit less sophisticated, but I’m thinking that unless I get the taste data it’s going to be difficult to predict /fine tune a profile. Someone suggested the bean colouring in lieu of the cupping which might work if I can get a process to accurately determine the exact bean colour.
Hi, I’ve been digging into my roast to see how I could improve them and I would be very interesting on how to get that air temps curve in RT