Prediction of fermentation activity

The challenge

One of the main challenges in agro-food production is to manage the variability of raw materials to deliver a final product that meets quality specifications. Champagne production is a typical example: the quality of the main substrate, grape juice, depends on the annual weather, wine-growing practices, and the cultivation area. This means that fermentations will also behave differently depending on the composition of the juice (amounts of sugar and nitrogen, turbidity, among others).

Alcoholic fermentation involves the sugars in the juice being converted by the yeast into ethanol and CO2, and the flow of CO2 generated has been observed to be a reliable representation of fermentative activity. This is the main monitoring measure.

The challenge for the great champagne houses is to ensure that all fermentations take place correctly (golden batch!). The harvest takes place in less than a month and therefore hundreds of fermentation vats must be monitored. The variability of the components of the grape implies that each fermentation is different and that many aspect that will influence it. The volume and urgency of production do not allow oenologists to effectively control each tank during this period and some fermentations may be outside of the specifications.

The solution

A “live” and automated management of these fermentations is therefore of interest, so that the fermentation behaviour is anticipated and corrected if necessary. This will ensure the quality of the wine and may also have a positive impact on the production flow.

In the first phase, we only received 6 fermentation curves to develop a model. Differential equations have been developed to represent them. Physical modelling makes it possible to develop a representation of the process with some data, which is an advantage over statistical approaches. For the “live” management of the measurements, a decision tree has been designed to identify the context and guide the model to be used for the curve prediction.

The second phase, fed by more data (fermentations with and without corrections), made it possible to refine the modelling and integrate the impact of the corrections on the fermentation behaviour.

The next step is integration into the supervision of the process.

The potential gain

The operational gain can be easily translated into financial gain because it leads to better management of fermentation and less losses.