ASSAS project

The ASSAS (Artificial intelligence for Simulation of Severe AccidentS) project, funded by the Horizon Europe programme, brings together 14 partners from the European Union, Switzerland and Ukraine, under the coordination of IRSN.

Characteristics

  • Completion dates: 2023-2027
  • Budget: approximately €4 million, of which €3 million is funded by the European Commission
  • Project coordinated by: IRSN and ENEA Consulting
  • Partners: Belgium - France - Germany - Italy - Netherlands - Spain - Sweden - Switzerland - Slovakia - Slovenia - Ukraine

Context and objectives

ASSAS (Artificial intelligence for Simulation of Severe AccidentS) is a Horizon Europe funded project targeting the development of nuclear severe accident simulators. It gathers 14 partners from the European Union, Switzerland and Ukraine, under the coordination of IRSN.

The main output of the project is to develop a prototype of a severe accident simulator based on IRSN’s calculation code ASTEC (Accident Source Term Evaluation Code) and Tecnatom’s simulation suite TEAM_SUITE®. It will consist in a proof-of-concept of realistic industrial simulators targeting not only phenomenological training but also potentially accident management training, management guidelines assessment and emergency preparedness and response. This proof-of-concept will be evaluated by the project partners as well as end-users.

One of the challenges of the project is to run the simulator in real time, or even faster, without significant impact on modelling accuracy. To achieve this goal, different solutions will be explored that include both the improvement of the numerical solvers of the current severe accident modelling and the integration of surrogate models developed using machine-learning techniques.

This last option, which is the most innovative and promising, will be the second pillar of the project, bringing together a large community of severe accident researchers and data scientists. Various data-centred approaches will be tested with different levels of hybridization between AI (Artificial Intelligence) models and the original code. Different generic reactor designs will be considered. The developed methodology might be adapted in the future to other applications in severe accident research and beyond. 

An additional outcome will be the database of severe accident scenarios generated during the project to train machine-learning models that might also serve other applications in the future.

Progra​m overview and areas of research

​The project has started on the 1st of November 2023 for a duration of 4 years. The budget is approximately 4 million euros, including 3 million euros funding from the European Commission. Apart from the management of the project led by IRSN and the dissemination led by the ENEA, the workpackages (WP) of the project and the institutions leading them are described below. 

  • The first task of this WP is to define the strategy to accelerate calculations in ASTEC more precisely. Since ASTEC has a modular structure, the strategies mentioned above (improved numerics and machine-learning) will be applied to different parts of the code, depending on the expected outcome.

    The second objective is to provide a methodological support for the development of AI-based surrogate models in WP 4.

    Finally, this WP will define validation criteria for the improvements / developments of ASTEC made during the project.

  • Within this workpackage, a large database of severe accident scenarios will be generated, which will be used to train machine-learning models. Criteria will be defined to create a homogeneous, unbiased, and representative database. An infrastructure will be designed to host the large amount of generated data. Partners will run calculations and check their validity before sharing the results through the centralised database. 

  • In this WP, partners will develop machine-learning models using the training databases developed in the WP 3. These models are meant to replace some parts of ASTEC or the whole severe accident code (such as MELCOR), to achieve higher computational efficiency with a controlled accuracy. 

  • In this WP, the performance of ASTEC will be improved thanks to the development of more efficient numerical schemes and solving approaches including parallelisation. This WP also includes the integration of machine-learning models into ASTEC, to form a hybrid tool using simultaneously classical and AI models.

  • This WP is devoted to the actual design of the prototype simulator (human-machine interface) and its connection with ASTEC. It will start with a more complete definition of functional specifications.

​Pa​rtners

IRSN (France), Tecnatom (Spain), CIEMAT (Spain), JSI (Slovenia), KIT (Germany), KTH (Sweden), ENEA (Italy), TU Delft (Netherlands), Phi-Meca Engineering (France), IVSTT (Slovakia), Energorisk LLC (Ukraine), Bel V (Belgium), CS GroupGROUP-France (France) and PSI (Switzerland).

Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or European Commission-Euratom. Neither the European Union nor the granting authority can be held responsible for them.