The 1st joint industrial laboratory in artificial intelligence


In early 2020, EDF, Thales and TotalEnergies have opened SINCLAIR (Saclay INdustrial Collaborative Laboratory for Artificial Intelligence Research) on the EDF Lab Saclay site.

Its research program is dedicated to the development of artificial intelligence methods and tools (M&O) that meet the shared needs of these three companies. The pooling of R&D personnel from the three companies aims to increase the research effort needed to develop mature artificial intelligence (AI) capable of improving the design and management of complex industrial systems. The overall objective of such AIs is to use modern learning tools to accelerate and improve engineering tasks that are currently costly and repetitive, or even unattainable by conventional means, and to allow specialists to concentrate on higher value-added actions.




By working on explainability, industrialists want an automatic system to be able to give an account of the reasoning that led it to the proposed answer. This aspect plays an essential role in the trust that users can place in AI, especially in critical systems.

Reinforcement learning consists in teaching the system a behavior by rewarding the AI, positively or negatively, as it learns.

Finally, by putting the AI in front of the simulation, the experts generate interactions between the AI and a physical model, in order to optimize the behavior of a system in front of a set of parameters. AI can thus considerably improve the tuning of complex physical systems, a step that would take much longer - or even be impossible - to achieve with a classical optimization method.

Each of these three axes should allow to enrich the results provided to engineers and thus reinforce the decision they will take. AI is thus placed at the service of humans to increase the value of their decisions.

> 23
Permanent researchers
> 5
PhD students
> 2
Post-doctoral researchers

The challenges of AI

Control of production units

Industrial production facilities maintenance and design and production forecasting could greatly benefit from automated assistance in performing certain costly and error-prone verification tasks, such as predicting clogging or cracking in industrial components during the design and usage phases or analyzing manufacturing lines.

Alarm systems

The refinement of control, monitoring and alarm systems, made more refined by the ability of AI to generate new situations and anticipate the complexity of risky situations, can be based on the integration of different types of data (sensors, videos, analysis reports, etc.). Such systems would allow site engineers to focus on delicate situations and exercise their expertise in the best possible way.

Digital twins

The development of multi-objective digital twins, simulating increasingly complex phenomena. They are intended to be used for advanced engineering tasks (e.g. design, anticipation), typically by coupling with optimization tools, and for training. By reproducing real behaviors as closely as possible from a learning process that hybridizes physical and logical knowledge and data observation, they make it possible to optimize the use of simulations in the search for operational gains.

Scientific & technical ecosystem

SINCLAIR welcomes interns, doctoral students and post-doctoral students in collaboration with renowned laboratories (from Université Paris Saclay, Ecole Polytechnique, INRIA, Institut de Mathématique de Toulouse, etc.) and participates actively in the scientific life of the Saclay plateau, within a competitive ecosystem.



Yann Semet
Nicolas Bousquet
Sébastien Gourvenec