Case study: fast coffee bean profiling in partnership with Lavazza
The goal of Volatile AI
Ingredient chemical profiling for quality control is time consuming and highly complex requiring professional chemists. At Volatile AI our goal is to reduce the analysis time from 30 min+ to 3 min making it straightforward to run for anyone.
Coffee quality control today
For coffee monitoring today, one would engage a fully-featured lab running a gas chromatograph – mass spectrometer (GC-MS) or a professional tasting panel. A lab analysis using GC-MS takes around 30-60 min of running plus the warm-up time. When accounting for the manual interpretation of the acquired peaks, the whole lab process can run into several hours to generate an insight for the business. Such complexity results in fewer screening points and centralised quality control.
Lavazza’s approach to quality
Coffee roasters survive and thrive on the flavour and quality of their blends. To ensure a consistent and top quality coffee blend, our partner Lavazza spend considerable effort in analysing and monitoring the incoming coffee beans. While the ultimate quality control usually rests with the tasting panel, lab testing is done to determine the processing method and to check for blend ratio irregularities. Knowing these properties guarantees full control of supplied materials and provides uniformity in the company's blends.
Volatile AI developed electronic nose “Scout2”
Coffee testing results: chemical composition screener vs. electronic nose
Of the two instruments, the chemical composition screener (ion mobility spectrometer) has shown better detection ability and can also generate faster results when profiling ground coffee beans. 80% was achieved in determining if roasted ground coffee beans were processed using a washed or a natural method. For the determination of a blend ratio of arabica vs. robusta, the instrument managed to reach 66% prediction accuracy and had a margin of error of 30%.
Electronic nose struggled significantly more as it is highly affected by the overall abundance of VOCs from the sample. For coffee, the general level of aroma changes very quickly once exposed to oxygen leading to a consistently decreasing signal. Nevertheless, the electronic nose has managed to reach 60% in prediction accuracy in distinguishing washed vs. natural coffee.
All results were validated using a method called “by day cross validation” ensuring that results correspond as closely as possible to real on-site conditions.
How the achieved result can benefit the partnership
The achieved 80% accuracy in differentiating washed coffee beans from naturally processed opens up an immediate opportunity for fast on-site coffee screening, which could also be performed remotely. It also gives a chance to continue building up the knowledge and database for more detailed coffee bean insights, such as its geographic properties.
Coffee testing process with Volatile AI solution
During the test ground coffee beans were received weekly throughout a period of several months for Volatile AI to analyse directly from a major roaster.
Two different instruments: an in-house built electronic nose Scout2 and a chemical composition screener (an adapted ion mobility spectrometer) were used. Both instruments do not require a trained chemist, have a straightforward plug-and-play sampling routine, and can be used with a few clicks of a button, producing a result as quickly as 30 seconds.
2g of ground coffee beans were placed into a sample cup and left for the sensors to analyse them for several minutes each time. Once the sampling routine was complete, the heavy lifting of result interpretation was carried out by an AI algorithm trained specifically for coffee analysis.
Volatile AI adapted chemical composition screener screening coffee within seconds
