Bel Builds a Digital Twin of La Vache qui rit, Still Experimental

In Lons-le-Saunier, in the Jura region, La Vache qui rit’s historic plant is celebrating its 100th anniversary. Opened in 1926 to make processed cheese portions, it hosted the announcement by the Bel Group on Tuesday, October 6, of a research project for a digital twin of a cheese, which the company presents as the first of its kind. The principle can be summed up in one sentence: test a recipe on a computer before making it for real.

In practice, the twin is a mathematical model that links two ends of the chain: the ingredients on one side, the manufacturing process on the other. Bel’s research, innovation and development teams can use it to simulate a large number of combinations before keeping only the leads that merit physical trials. The group expects shorter development cycles, more precise formulation work and continuous improvement of the nutritional profile.

The project is being carried out with Dassault Systèmes, a partner since an agreement signed in the fourth quarter of 2023 and announced in July 2024. The agreement provides for the deployment of the “Perfect Production” solution, built on the 3DEXPERIENCE platform, in 11 plants across the group. The amount invested has not been disclosed, with Bel describing it only as “significant.” Group chairman Antoine Fiévet defends a technology that serves human expertise. Bel says it has “proven the model” on sensory predictions and textures, while acknowledging that “work remains to be done.”

The limits are clear. The consumer model is in the validation phase, while the one concerning processability, meaning a recipe’s ability to be manufactured on production lines, still needs about a year of work. The digitalization of the plant, with its sensors and analyzers, remains at the experimental stage, and the group’s returns on its data investments “remain to be demonstrated.” Yet a quote reported by L’Usine Nouvelle claims: “We can divide product development times by three.” The available sources give no measurement supporting this at this stage.

For Bel’s teams, the expected gain is to weed out recipes that will not hold up in production earlier, even before physical trials. The deadline is 2028: industrial-scale rollout is targeted for that year, and the group hopes to launch its first recipes developed with the tool by then. The scope is wide, with nearly 200 recipes sold in more than 120 countries. The site, which exports two-thirds of its output to 60 countries, also benefits from a €10 million modernization plan spread over three years.