In the framework of a collaboration with DIAM Bouchage, the Carnot institute CEA List has developed a technology to detect defective corks. This solution uses artificial intelligence to perform an X-ray analysis of the cork.

An automatic sorting solution for defective corks

As wine connoisseurs know, sometimes a wine is corked or oxidized when a bottle is opened, even though a cork is considered the ideal choice for preserving it. For a wine to be preserved, the cork must be able to let in just the right amount of oxygen to allow it to mature and guarantee its taste quality over time. If this is not the case, it loses its flavour.

However, it is impossible to tell with the naked eye if a cork is properly sealed or if it is allowing too much oxygen into the bottle. It is in this context that the partnership between the Carnot institute CEA List and the company Diam Bouchage intervenes. The latter has asked the researchers to develop an automated and reliable method for sorting corks made from cork oak bark.

Artificial intelligence to automate the tool

X-ray tomography is used to produce images of the cork to be analysed, so that the cork can be viewed in its entirety. To carry out this analysis, the researchers developed a machine learning algorithm adapted to cork classification.

The solution compares the characteristics obtained by imaging, such as growth lines, cork density or the number and distribution of lenticels, with data on the rate of oxygen transfer carried out over the long term. Thus, the model automatically assesses the tightness of the cork and classifies it as a quality cork or not.

Initial tests of the platform show that 75% of the corks are correctly classified in a matter of seconds, as the tool only analyses two images per cork. The researchers are continuing their research in order to improve their model with a goal of reaching 100% correct classification, a mandatory criterion for their algorithm to be used to cork great wines.

Translated from Le CEA List élabore un modèle de machine learning pour classer les bouchons de liège