Skip to main content

About

Scientific Objectives


The ASTEC Software Package

The ASTEC (Accident Source Term Evaluation Code) software system makes it possible to simulate all phenomena that take place during a water-cooled reactor meltdown accident, from the initiating event to the discharge of radioactive materials (called the “source term”) out from the containment

The main scientific objectives of the project are the following:

Developing a “basic-principles” severe accident simulator for a generic western-type PWR

  • Interfacing ASTEC with Tecnatom’s simulation environment
  • Achieving real-time execution for training purposes
  • Developing a proof-of-concept to prepare more complex simulators in the future

Improving severe accident codes to meet the requirements of a simulator

  • Improving ASTEC performance through the optimization of its algorithmics and implementation
  • Optimising ASTEC reactor models nodalization
  • Developing surrogate models based on Artificial Intelligence for ASTEC and MELCOR

Supporting the adoption of machine-learning approaches in nuclear science

  • The severe accident sequence database used to train AI models will be in open access
  • Methodologies can be applied to other multi-physics codes

Many challenges are associated with the above-mentioned objectives. 

Severe accident modelling is based on strongly non-linear models, that can exhibit chaotic behaviours in some conditions. Specific machine-learning algorithms will be necessary to develop fast-running surrogate models that capture those specificities.

ASTEC, like other nuclear simulation software, is a legacy code. It results from the coupling of several modules that have been developed for more than 30 years. The evolution of the hardware architecture, relying more and more on multiprocessor CPUs (central processing units) and GPUs (graphics processing units), calls for adaptations of its algorithmics.

ASTEC is still being actively developed by IRSN with the support of KIT and CS Group, which gives the consortium a unique opportunity to explore hybrid machine-learning approaches. Many similar projects are limited to so-called “black box” approaches, because they cannot have access to the source code of the software that must be emulated.

Finally, ASSAS will bring together two different research communities: severe accident specialists and machine-learning experts. This bi-disciplinarity will undoubtedly result into fruitful collaborations.