Robots to decide on beer quality
Researchers look to AI to cut brewery product development time and costs. Jeff Glorfeld reports.
The international beer market was valued at more than $US590 million in 2017 and is projected to top $US685 million by 2025, according to business analysts Allied Market Research.
With so much at stake, producers are looking closely at any development that might help them gain an edge in this highly competitive industry.
Rather than rely on the vagaries of ordinary consumers to determine a popular beer, a team of researchers from the University of Melbourne, in Victoria, Australia, has called in a mechanical expert to help them identify the attributes that are most pleasing to consumers.
Writing in the journal Beverages, Claudia Gonzalez Viejo and colleagues describe a robotic pourer, dubbed RoboBEER, built in part with components sourced from Lego toys. The machine is programmed with a series of computer vision algorithms, which allow it to scan and assess beers for 15 colour and foam-related parameters.
The researchers suggest that the main benefit of such a system is a reduction in time and cost for brewers when developing new products. Current practice involves conducting large sensory tests using human volunteers.
These exercises require time for preparation, data gathering and analysis, as well as financial resources to cover recruitment and stock.
“This will offer to the beer industry a completely automated process to predict liking and acceptability of different beers by consumers,” the authors write.
Viejo and colleagues constructed several training algorithms to develop artificial intelligence networks capable of predicting preferences based on four sensory attributes: carbonation mouthfeel, bitter taste, flavour and overall liking.
To obtain the initial data, they conducted sessions with 30 consumers, who rated 17 types of beer according to the categories.
The field covered brews produced using top, bottom and spontaneous fermentation, and included a porter from Poland, a sparkling ale from Australia, a Mexican lager, and a kriek lambic from Belgium.
After comparing the models developed using the training algorithms, the best model was selected.
When it came time to analyse the beers for colour and foam-related parameters, RoboBEER was called in to ensure uniform pouring.
The machine works with two Lego servo motors and has three sensors to measure temperature, alcohol and carbon dioxide gas release. It also incorporates a smartphone, used to record five-minute videos of the pouring, in order to judge foam behaviour.
The researchers say their model allows them to accurately predict consumers’ liking of specific beers.
They add that consumers are able to judge beer quality and acceptability from visual first impressions based on foam and colour characteristics.
Bitterness, derived from hops, also contributes to the amount and stability of foam. Hops also influence the development of aromas and flavour, which are released as the foam bubbles burst.
The advent of RoboBEER and its algorithmic outcomes in one sense represents a triumph of economics over romance.
For some people, the idea of being recruited to sample beers – and, heavens, even being paid to do so – is a delightful prospect.
From the point of view of the world’s big brewing conglomerates, however, having actual people involved in the beer research business is a messy and lengthy matter.
AI-trained robot drinkers, on the other hand, Viejo and colleagues conclude, will “aid in the optimisation of costs and time for breweries to assess beer acceptability without the need of recruiting consumers and running sensory sessions, being able to get the results within minutes”.