MN - Modèles Numériques

Bayesian Methods for the Diagnosis and the Probability of detection supported by Simulation – ByPASS

Submission summary

The context of ByPASS is the numerical simulation for Non-Destructive Testing (NDT). More specifically, the project aims at developing new computational methodologies and software tools dedicated to two crucial industrial uses of the simulation. The first concerns the performance evaluation of inspection methods through the calculation of Probability of Detection (POD). The second concerns data inversion for the characterization of defects which are at the origin of the detected signals. Automated diagnosis is the perspective in this second case. Both of these applications have in common that they are strongly influenced by the treatment of uncertainty and by the "confidence" that can give the simulated results. This notion of confidence based firstly on the validation of the models involved, and secondly on the ability of simulation tools to provide quantitative information expressible in terms of confidence (in the statistical sense ), integrating fluctuations and uncertainties affecting factors of influence. To address this common issue, it is proposed in this project to implement Bayesian approaches in order to provide a level of confidence associated with information from simulated data.
For POD, the introduction of Bayesian methods will fulfill a crucial requirement that is to have a methodology combining simulated and experimental data in order to provide values of "POD supported by simulation" robust because consolidated by experience, while maintaining a reduced number of experimental data, and therefore reduced costs.
For the inversion of data in perspective to automatic diagnosis, these probabilistic methods are proposed for producing results with confidence regions which will help the engineer to take the appropriate decision.
The implementation of such approaches, in addition to methodological aspects, requires the production of a large number of simulated data in a sufficiently short time to make them usable in practice. This requires calculation methods adapted to this context such as the factorization of recurring calculations (search for invariants in the field of exploration, factorization) and the implementation of massively parallel computing strategies. These developments should allow to produce enough data to build surrogate models ultimately giving access to simulated data extremely rapidly across the domain of exploration.
To meet these objectives ByPASS is a project of industrial research over a period of 36 months and involving a consortium composed of:
• Two industrial end users CND: EADS for the aerospace industry and EDF for the energy.
• The CEA LIST, research laboratory that develops platform CIVA simulation
• A SME company Phimeca, specialist treatment of uncertainties in the physical models.
• An academic laboratory, the Laboratory of Signals and Systems (L2S), specialist techniques and Bayesian inversions.
• An Italian academic laboratory, research center ELEDIA (Information Engineering and Computer Science Department) of the University of Trento, specialist of inversion and optimization, an associate partner in the project.
The expected results of the project are a set of methods and tools that will be implemented in the software tools developed by the partners. These tools will allow the end user to rely on simulation to construct POD curves, or to go forwards automated diagnosis.

Project coordinator

Monsieur Pierre Calmon (Commissariat à l'Energie Nucléaire et aux Energies Alternatives - Institut LIST)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.


CEA LIST Commissariat à l'Energie Nucléaire et aux Energies Alternatives - Institut LIST
EDF R&D Electricité de France R&D
Phimeca Phimeca
ELEDIA ELEDIA, University of Trento

Help of the ANR 873,056 euros
Beginning and duration of the scientific project: October 2013 - 42 Months

Useful links

Explorez notre base de projets financés



ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter