Modeling of alcoholic fermentation in champagne-production context

Share this post:

By Technord, IBM partner “This is champagne”

One of the main challenges in the agro-food production is to handle the variability of the raw matters to deliver a final product meeting the quality specifications. The champagne production is a typical example: the quality of the main substrate, the grapes juice, depends on the yearly weather, the vineyard practices, the culture soil. It means that the fermentation will also behave differently following the juice composition (sugar and nitrogen amounts, turbidity, among others).

It is therefore of interest to monitor the fermentation and anticipate their behaviour to ensure the workflow in production. The dynamical modeling and representation of the main measurement, the CO2 flux, is the first step towards an automated ‘live’ management of these fermentations.


Figure 1

Hexose sugars are transformed by the yeast into ethanol and CO2 and it has been observed that the generated CO2 flux is a reliable representation of the fermentative activity. CO2 flux increases with the exponential growth of biomass reach a peak of activity and then decreases as the ethanol produced by the yeast will paradoxically inhibit the latter.

The fermentation ends when sugar is depleted and there is consequently no CO2 flux anymore. An illustrative example can be found in Figure 1.

Moreover, corrective actions can be taken to avoid sluggish fermentation and preserve future wine quality. Micro-oxygenation has a positive impact and is studied to better understand the metabolic mechanisms that are involved.

Automated monitoring of the fermentation

The challenge for big champagne houses is to ensure that all the fermentations proceed properly. The grapes harvest is done in less than 1 month and therefore hundreds of fermentation tanks have to be monitored. The variability in the grape components implies that every fermentation is different and a lot of parameters will influence this latter. The volume and the urge of production do not allow oenologists to check efficiently every tank during this period and some fermentations can be failed or mediocre.

Automated monitoring of the fermentation associated with predictions is, therefore, a valuable tool for the producers. The operational gain can be easily translated into a financial gain.

The IBM partner Technord has developed a mathematical simulator that can provide predictions of CO2 flux in ‘live’ conditions:

  1. The first upcoming measurements are analysed to determine when the fermentation starts.
  2. When this start is detected, the following measurements are used to follow a decision tree that will indicate which set of parameters to use for the model equations;
  3. The model provides predictions of the CO2 flux for the whole fermentation and allows to anticipate if this fermentation will present issues or not, based on the predicted peak of activity. The end of fermentation is also predicted to prepare the next steps of the workflow.

    TechNord figure 2

    Figure 2


IBM Watson IoT allowed to calculate fast enough the predictions and provide a ‘live’ web dashboarding to monitor and predict the behaviour of the fermentation for the end-user, see Figure 2.



The smart predictive sensor that Watson provides is an example of innovative AI into the agro-food field!


Head of Data Science/Analytics/Modeling, Technord

More Industries stories

Rethinking the future with Victoria Bunyard, CTO IBM Benelux

In July 2020, IBM Benelux welcomed Victoria Bunyard as their new CTO. British by birth, Ms Bunyard moved to the Netherlands in 2004 and has found her second home close to the beach in Noordwijk. Her biggest passions? Rowing on the Amstel, skiing, and, of course, driving innovation.   When you ask Victoria Bunyard about […]

Continue reading

Unlocked: how the best businesses will emerge from the pandemic smarter and stronger

In this blog series, we’ve explored the challenges organisations have navigated as the COVID-19 pandemic and lockdown have unfolded. The future will look wildly different depending on your industry perspective. But many people are now beginning to make sense of the changes and rebuild their businesses to accommodate the different normal. As we move into […]

Continue reading

What do you really need to do to get your AI/ML models into production?

  Companies and institutions that take the step towards applying artificial intelligence (AI) and machine learning (ML) often overlook the technical integration and internal alignment that the use of these capabilities require, which means that activities around AI/ML are not taken into production. With the right insights and techniques, this promising technology can be put […]

Continue reading