CEMEF internship 2019 : Data-driven multiscale strategy for the prediction of void closure in material forming processes

Subject: Data-driven multiscale strategy for the prediction of void closure in material forming processes

Context & Objectives of the present work:

After casting, metal ingots may contain voids of different shapes and sizes, which need to be eliminated in order to deliver a sound material. Hot metal forming processes are regularly used in the industry to reach this goal, but the calibration of these processes to get a complete closure of internal voids is still an issue.
Recently, a scalar void closure model was developed based on a multiscale strategy [1, 2, 3, 4]. This model accounts for equivalent strain, stress state (stress triaxiality ration and Lode angle) as well as shape and orientation of the initial ellipsoidal void. However such metal forming processes are usually based on multiple stages with rotation of billets between each stage, which conducts in modification of the shape and orientation of voids with respect to the compressive loading direction.

Knowing the void’s geometry evolution would allow to handle multi-stages forming operations where a rotation is applied to the ingot between two forming stages. Numerous parameters are involved in the void shape evolution and volume decrease, namely: void geometrical parameters, process parameters and associated thermo-mechanical conditions and material parameters.

Due to this huge number of input influencing parameters, it is particularly difficult to define the mathematical form of the void closure model as well as its dependency on each parameter. Previous studies have been conducted on a limited number of parameters with pretty good prediction for a scalar void closure model for restricted number of configurations [1-3]. An attempt to extrapolate this approach to a tensor prediction of void shape, orientation and volume has shown some limitations due to the complexity of parameters evolution for such high number of input influencing parameters [4].

The objective of this work is to build a data-driven multi-scale strategy to define an enriched tensor void closure model that would be able to predict void shape evolution and void volume decrease with respect to geometrical, thermomechanical, material and process input parameters.
Image illustrative stage Data-driven multiscale strategy

Program work:

  • Bibliographical study on deep-learning techniques and applications to data-driven computational mechanics and materials science.
  • Definition of all input parameters and identification of their variation range for metal forming processes.
  • Definition of an extensive campaign of full-field simulations with variation of geometrical, thermomechanical, material and process input parameters. These simulations will be carried out on representative volume elements (RVE) of appropriate size and will define the database that will be used to train machine-learning algorithms. For each RVE computations, the database will contain the list of input parameters defined previously as well as important output parameters (void volume, inertia terms).
  • Choice of appropriate machine-learning algorithms with respect to the present problem, training of these algorithms based on the RVE database results and analysis of machine-learning parameters on accuracy and efficiency of void closure predictions.
  • Validation of the void closure model on industrial processes based on the comparison between full-field simulations (with voids meshed explicitly) and the mean-field model defined during the project.

Internship framework:

This 6 months internship is funded by a consortium of industrial partners including Timet, ArcelorMittal, Industeel, Ascometal, Aubet&Duval and Framatome. Due to computational resources, the internship will be based at the Center for Material Forming Processes (Sophia-Antipolis). Regular meetings will be scheduled with the industrial partners and a close collaboration with Transvalor is also expected for future integration in the Forge Finite Element Software.
logos of industrial partners

In this internship you will learn: multiscale finite element modeling, machine learning strategies for materials science, Influence of stress states on voids closure to improve materials behavior, working on advanced scientific methodologies with clear direct industrial applications.

Références :
[1] M. Saby, P.-O. Bouchard and M. Bernacki, Void closure criteria for hot metal forming: a review, Journal of Manufacturing Processes, 19:239-250, 2015.
[2] M. Saby, P.-O. Bouchard and M. Bernacki, A geometry-dependent model for void closure in hot metal forming processes, Finite Elements in Analysis and Design, 105:63-78, 2015.
[3] A. Chbihi, P.-O. Bouchard, M. Bernacki, and D. Pino Munoz. Influence of lode angle on modelling of void closure in hot metal forming processes. Finite Elements in Analysis and Design, 126:13-25, 2017 (DOI information: 10.1016/j.finel.2016.11.008).
[4] A. Chbihi, Modélisation de la refermeture de porosités pour des procédés multi-passes, Thèse Mines ParisTech, 2018.
Data science, Void closure, tensor model, Finite Element
6 months, starting in March
Data Sciences, Finite Element modelling, Solid Mechanics
Ability to work in group, motivation and sense of initiative and capability to report regularly on his/her work.
MINES ParisTech - CEMEF, Sophia Antipolis (06), France
Industrial consortium including Timet, ArcelorMittal, Industeel, Ascometal, Aubet&Duval and Framatome
The position is for an undergraduate student, at the level of Master or French Engineering School. Please send your CV, your marks from the last two years and a recommendation letter or email of a reference to:
Daniel Pino Munoz:
Pierre-Olivier Bouchard:
Marc Bernacki:
B.P. 207 - 06904 Sophia Antipolis Cedex
logo S&CC
logo Transvalor