Digital Materials - An industrial Reality In Motion – RealIMotion

Ambitious groundbreaking objectives are aimed in the RealIMotion Chair and will be reached through the following steps:
.To develop new very promising front tracking approaches allowing to improve by a factor 100 to 1000 the computational efficiency of the better existing 2D and 3D full-field formulations in mesoscopic simulations of hot metal forming processes.
.To enrich the number of physical mechanisms compared to what is taken into account in existing numerical strategy and develop them in the new front tracking approach (ToRealMotion algorithms). This objective goes hand in hand with the enrichment of the materials and process conditions or processes under consideration, guided by the industrial needs and priorities.
.To speed up representative volume element (RVE) computations with a large number of grains allowing the use of RVE simulations in a large number of integration points of macroscale simulations by keeping reasonable computation time to ensure the compatibility with R&D computational facilities.
.By using our new RVE capabilities, a large numerical database concerning materials/mechanisms/ process conditions will be built. Supervised deep neural network (DNN) strategies and deep re-inforcement learning strategies will be trained on this database. Thanks to this, the acceleration of R&D calculations by coupling RVE computations with automatically proposed RVE results coming from the trained DNN will be investigated. Moreover, automatic interpretation of some microstructural singularities (e.g. abnormal grain-growth) or mean-field model calibrations will also be tested.
.All along the project, the developments will be validated thanks to optimized targeted experimental data collection and optimized model parameter identification procedures.
.The relevant numerical algorithms will be integrated, throughout the project, to the industrial DIGIMU software in order to be effective in the R&D departments of the industrial partners and validated in industrial configurations.

The project officially started on 01/10/22. Three PhD theses, one master's thesis, and one postdoc are in progress, with two more PhD theses starting next October. Two PhD theses focus on developing machine learning methods to speed up mesoscale simulation calculations or replace high-fidelity calculations for some metallurgical mechanisms, with promising early results. The postdoc supports these efforts with experimental data. The third PhD thesis aims to adapt existing mesoscale models for aluminum alloys, while the master's thesis tests slice-GAN methods for reconstructing 3D microstructures from 2D data.

The numerical formalism related to the chair has been enhanced to model solid-state phase changes for simple compositions, with plans to extend its scope in 2025/2026. I also contribute to the development of the ToRealMotion full-field methods to increase their versatility and populate the necessary databases for machine learning methods. A new medium-field model formalism applicable to high deformation rates has been proposed and tested on austenitic steel, with plans to extend this work to a nickel-based superalloy in 2025/2026.

Various industrial partners have worked on data acquisition to support and feed the developments for different materials. All planned tasks and deliverables have been completed as detailed in part B. The progress of the work is on schedule, and the dissemination/communication of the initial results is important. Key achievements in the past 18 months include:

.Initial tests of LSTM methods for predicting grain growth mechanisms based on the full-field model developed in the chair.
.Delivery of the new ToRealMotion numerical formalism for integration into DIGIMU software.
.An article on the numerical methods developed within the RealIMotion chair was published in the prestigious Progress in Materials Science journal in April 2024 (https://doi.org/10.1016/j.pmatsci.2023.101224).
.Invitation to present the developments related to the RealIMotion chair as a plenary conference at the ReX&GG2023 conference.
.Finalization of DIGIMU 5.0 code with numerous additions (application of CDRX to zirconium alloys, heterogeneous intragranular energy field, germ size distribution).

. In the next months, new mechanisms will be deployed on the ToRealMotion formalism.

. The databases currently under construction are unique on this theme; they are already being tested to produce the first reduced-order models usable in computational metallurgy for the prediction of recrystallization mechanisms, grain growth, and solid-state phase transformation in the solid state.

. These models will then be coupled with macroscopic scale simulations and integrated into DIGIMU6.0.

. This new version of the software will then be intensively used by industrial partners to validate/improve it.

• Invitation to present the developments related to the RealIMotion chair as a plenary lecture at the reference conference dedicated to recrystallization mechanisms and grain growth (ReX&GG2023).
• The prestigious journal Progress in Materials Science commissioned an article dedicated to the numerical methods developed within the framework of the RealIMotion chair; this article was published in April 2024 (https://doi.org/10.1016/j.pmatsci.2023.101224).
• Release of DIGIMU5.0.
• Recruitment of a faculty member (end of 2022 at the start of the chair) specializing in solid-state phase changes, actively participating in the project.
• Over the period: 10 A-rank articles, 9 international conferences, 5 national conferences, and multiple partnerships for various items.
• 1 PhD thesis award at the ESAFORM 2024 conference (Bronze Medal for the best presentation by P. Hahn).
• Awarded to the PI, the Albert Portevin Medal by the French Society of Metallurgy and Materials in 2024.

One of the European Union’s objectives in climate change consists of reaching net-zero greenhouse gas emissions by 2050. Such perspective puts the metallic materials industry, as a large contributor to carbon emissions, under tremendous pressure for change and requires the existence of robust and qualitative computational materials strategies to design, to enhance, to calibrate, with a very high degree of confidence, new metallic materials technologies with a limited environmental impact. From a more general perspective, the in-use properties and durability of metallic materials are strongly related to their microstructures, which are themselves inherited from the thermomechanical treatments.

Hence, understanding and predicting microstructure evolutions are nowadays a key to the competitiveness of industrial companies, with direct economic and societal benefits in all major economic sectors.

Multiscale materials modeling, and more precisely, simulations at the mesoscopic scale, constitute the more promising numerical framework for the next decades of industrial simulations as it compromises between the versatility and robustness of physically-based models, computation times, and accuracy.

In this context, a breakthrough numerical strategy to describe the microstructure evolutions of metallic materials during complex industrial thermomechanical treatments has been developed through the ANR Industrial Chair DIGIMU (Oct.2016-Mar.2021). The outcoming DIGIMU® software is now available for the industry, and able for quantitative predictions of microstructure evolutions on material volumes in the range of one mm3, with typical computation times of a few days when performed on a simple laptop. Such simulations and computational efficiency were a dream ten years ago, a reality now with the DIGIMU developments.

The purpose of the RealIMotion project is to push the limits of numerical metallurgy further and develop a promising new numerical framework coupled with a machine learning physically-based strategy to aim for massive computations, consideration of much larger material volumes, in connection with macroscopic simulations and still with reasonable computation times to be compatible with industrial daily uses. Such a leap in the models will open the door for industrial partners to tune numerically thermomechanical routes, build microstructure-targeted industrial processing maps and automatically propose new enhanced homogenized models. RealIMotion project brings the cutting-edge and exploding strategies of data science, physically-based models, and machine learning at the service of industrial metallurgy. Major advances regarding the concept of digital twins in metallurgy and a worldwide leading position of the RealIMotion partners concerning Integrated Computational Materials Engineering (ICME) developments are expected outcomes of the Chair program.

The RealIMotion PI is a pioneer and a world leader in mesoscopic scale modeling of microstructure during hot metal forming. The French industrial consortium supporting the developments of digital materials in the context of hot metal forming has grown in the RealIMotion proposal. Constellium and Aperam are new partners. Framatome, Aubert&Duval, ArcelorMittal, CEA, and Safran brought new targeted applications on zirconium alloys, aluminum alloys, dual-phase steels, and new generation nickel based superalloys.

The students recruited in the RealIMotion Chair will enjoy a perfect environment to become experts in computational metallurgy, digital twins, and IA and meet the metallurgical industry needs for the future. The expected benefits will be job-creating for all the partners involved. The RealIMotion Chair will contribute to the materials science teaching effort by offering to the universities concerned free access to DIGIMU® software for pedagogic purposes, and a turnkey tutorial set adapted for practicum.