Finite Element Simulation and Data-Driven Statistical Approaches for Predicting the Variability[...]

PhD Position - Digital Twin for Predicting Fracture Variability in Mesosegregated Forged Steels

Charpy impact tests are widely used in the nuclear industry to certify forged components. However, for some very large components produced by Framatome, these tests exhibit significant variability in impact toughness. Previous studies have shown that the forged steels involved are highly heterogeneous at the mesoscopic scale, due to the segregation of alloying elements during the sequence of thermomechanical operations involved in forging. These heterogeneities are a major source of the observed scatter in fracture properties.

The objective of this PhD project is to develop an efficient numerical simulation framework to better understand and predict the variability of fracture properties, in particular impact toughness and fracture toughness, in mesosegregated forged steels. The project will combine finite element simulation, probabilistic modelling and artificial intelligence, with the long-term goal of building a digital twin methodology for forged materials.

Statistical representation of microstructures using generative AI

Experimental segregation fields are generally obtained from two-dimensional images produced by Nital chemical etching. The goal will be to generate representative three-dimensional microstructures from these 2D observations. Generative approaches such as SliceGAN, diffusion models and Gaussian random fields will be investigated in order to produce realistic 3D mesosegregation fields while improving model stability and reducing training cost.

Fracture modelling using the finite element method

Mechanical simulations of Charpy specimens will be performed in finite strain plasticity using the Zset finite element code. These simulations will then be post-processed using statistical fracture models of the Weibull-Beremin type. The project aims to extend existing methodologies to brittle fracture, accounting for the competition between cleavage and intergranular fracture, as well as for the spatial dependence of fracture parameters on local alloying-element concentrations.

Accelerating numerical simulations with AI

To enable large-scale statistical studies, structural zooming strategies will be developed. In parallel, convolutional neural networks will be investigated to correct simplified mechanical models, such as two-dimensional approximations, and provide fast and accurate three-dimensional predictions.

The resulting tools will enable statistical studies aimed at understanding the origin of low-toughness outliers. In particular, the project will investigate the respective roles of heterogeneous mechanical behaviour and heterogeneous fracture resistance in the tails of the impact toughness distribution.

Required skills

• Strong background in computational mechanics and nonlinear finite elements.
• Proficiency in Python and/or C++.
• Knowledge of and interest in statistics, geostatistics and artificial intelligence.
• Interest in industrial applications and ability to develop technological solutions.

Application

Applicants should send their application to , including a detailed CV, a copy of an identity card or passport, a motivation letter, academic transcripts for L3/M1/M2 or equivalent, two recommendation letters, the names and contact details of at least two referees, and proof of English proficiency.

The official application page should be viewed here