AI Sommelier: Russian Scientists Teach AI to Identify Wine’s Chemical Makeup
Scientists at Moscow State University have developed an artificial intelligence system that can analyze wine’s chemical composition in less than a minute. The algorithm measures five of the wine’s main components. The technology could help lower the cost of software for rapid chemical analysis and may also be used to test other food products.
Researchers from the Department of Physics at Lomonosov Moscow State University (MSU) and the D. V. Skobeltsyn Institute of Nuclear Physics developed a machine-learning system to analyze wine’s chemical composition. It measures the concentrations of ethanol, sugars, acids, glycerol, and sulfur dioxide using a sample’s infrared spectrum. The findings were published in the scientific journal Scientific Reports.
The technology addresses a practical challenge for the wine industry: getting information about a wine’s composition faster and potentially paving the way for more affordable quality-control instruments. The researchers also avoided having to develop a separate algorithm for each substance. Instead, the system was trained to recognize the spectral signatures of several components at once.
Every Detail in a Drop
The method relies on infrared spectroscopy. When infrared radiation interacts with a substance, its molecules absorb energy at specific frequencies. This produces a spectrum – a distinctive pattern of signals linked to the sample’s chemical composition.
But wine contains many substances at once, and their signals overlap. That makes it difficult to determine the concentration of an individual component from the spectrum alone. This is where machine learning comes in. The algorithms identify patterns in complex data and estimate the concentrations of the substances researchers want to measure.
The researchers trained the system to recognize mid-infrared spectra. They tested several approaches, including neural-network models. These models process the spectral data and calculate the concentrations of components that quality-control specialists need to track.
How the Analysis Works
The analysis requires less than 50 microliters of wine – literally, a single drop. Measuring the spectrum takes less than a minute, and if a prepared reference database is available, the result can be produced almost immediately.
The researchers tested the algorithms on 1,734 samples. The dataset included model solutions designed to mimic the composition of white and red wines, as well as actual wines. This approach allowed the team to assess how well the system handled both carefully prepared mixtures and real beverages containing many different compounds at once.
According to the researchers, the reported accuracy for determining ethanol content was 0.36%, 8 g/L for sugars, 2 g/L for acids, and 3.4 g/L for glycerol. For sulfur dioxide, measurement accuracy reached the maximum permissible concentration level.
Quality Control
These results matter beyond the research lab. Winemakers need to monitor a product’s composition at different stages of production, and rapid measurements can give them the information they need to make process decisions sooner. For example, sugar and acid measurements can help assess a wine’s characteristics, while data on ethanol and sulfur dioxide can be used to check whether the product meets specified parameters.
One of the development’s main practical benefits is the potential to reduce software costs for instruments that use infrared spectroscopy to analyze wine quickly. Such equipment is useful to producers and laboratories that routinely test large numbers of samples. Faster measurements mean specialists can get information about a product sooner. And when spectral analysis software costs less, more businesses may be able to use these tools as part of their everyday operations.
From Wine to Other Products
The researchers say their method is as accurate as foreign commercial counterparts. That means the Russian technology could replace software currently used with spectroscopic instruments. The spectral-analysis software can be adapted to different equipment and production settings. The researchers propose extending the same approach to juices and carbonated beverages in the future.
