top of page

ENGINEERING

Our goal is to speed up design processes and to reduce time to market:

  • NeurEco creates Reduced Order Models (ROMs) from a small amount of simulation data

  • It goes beyond ROMs by creating a real-time copy of a complex high-fidelity simulation tool, from a few simulation results

  • Beyond learning the data, NeurEco learns the underlying rules; It happens that the NeurEco model is more in line with the laws of physics than the data itself

  • This is the basis of a robust long-term dynamic prediction

shutterstock_125887775_edited.jpg

OUR SUCCESS STORIES

Engineering_edited.png
Model reduction
techniques
Nuclear core cooling system simulation
Neutronic fluxes prediction
Optimal parameters of an antenna

Model reduction techniques : Nuclear Energy

Model reduction techniques allows companies to reduce the time needed to undertake complex computations as these computations are replaced with real time automatic AI models.
 

Model reduction techniques.png

Context

  • We are proposing a unique dynamic modeling tool.

  • The goal is to go further and to take into account the contact and friction in the reactor.

  • Examples: Pellet clad interaction in a nuclear core, seismic motion.

Challenge

  • The time needed to model build the model is reduced as the data needed for parsimonious NN models is significantly reduced without compromising the accuracy.

  • Nonlinear dynamic modeling is challenging because it is not addressable with traditional methods.

  • Dynamic modeling is more challenging due to this being an almost chaotic scenario.

Results

This model reduction for nuclear power production was thought to be an impassable limit, however, NeurEco easily overcame this limit and provided Framatome with an accurate, reliable and efficient solution.

Ancre 3

Nuclear core cooling system

Context

Model reduction of complex dynamic simulation of a nuclear core cooling system.

Input

Boundary conditions, system parameters

Output

The dynamic flow

Challenge

  • This use case is not suitable for a convolutional approach.

  • The dynamic solution is living in a large dimensional space.

  • Classical compression methods (SVD) reduces the dimension space to 100 within a 1% error.

  • NeurEco reduces these 100 compression coefficients to 3 nonliear coefficients.

Results

  • Adagos’ dynamic simulation is unique on the market.

  • Real time simulation when the traditional simulation takes weeks.

  • This allows decision support in a critical context.

  • NeurEco is intensively used by Framatome.

Ancre 1

Optimal parameters of an antenna

Optimal antenna.png

Context

Produce a reduced order model of an antenna that will help to optimize antenna’s performance.

Challenge

  • Classical model reduction methods are failing because of the presence of resonance.

  • Parsimony is the key to solve this model reduction.

Benefits

  • Speed up the design process - real time data flow available.

  • Reduce the time to market.

  • Dramatically increase the performance of their antennas proving a better experience for their clients.

Ancre 2

Neutronic fluxes prediction

Neutronic fluxes prediction.png

Context

Predictive risk management and uncertainty quantification.

Input

The enrichment distribution uncertainty in the nuclear core

Output

The uncertainty on neutronic fluxes

Challenge

Strong non linearities: prediction in a hardly predictable context

Methodology

We are using a parsimonious convolutional approach.

Benefit

NeurEco produced a predictive model which was reliable, robust and accurate. The predictive nature provided the client with a better risk management.

Ancre 4
bottom of page