arXiv subjects

Marc Bernacki

Publications and source records attributed to Marc Bernacki.

At least 19 recordsLinked to original sources

arXiv2026

Generation of representative powder particle packing in 2D/3D: which tool for which application?

Although dense sphere packings serve as the initial state for simulations in powder metallurgy, additive manufacturing and granular physics, the choice of a packing generator is rarely guided by a systematic benchmark. A representative packing must be (i)-(ii) physically admissible (non-overlapping particles in gravitational equilibrium); (iii) faithful to the target particle size distribution (PSD); (iv) representative in relative density Phi; and (v) computationally affordable. Four open-source tools have been benchmarked, meeting (i)-(ii) by construction: the sequential DR (dropping-and-rolling) and its densified variant DR-ME, and the discrete element method (DEM) codes LAMMPS (gravity) and dp3D (isostatic compression). Across four configurations (2D/3D lognormal, 3D bimodal, and a 3D domain-size study), they are compared against an industrial MIM-grade powder, with PSD fidelity measured by the bin-width-independent Hellinger distance and Phi against the feedstock solid loading (phi_exp = 0.62, by Archimedes' method). In 3D, the DEM codes reach the densest packings but run more than three orders of magnitude slower: for approximatively 20 000 particles, DR shows a 9% phi shortfall relative to dp3D while running 1800x faster. These idealised model packings yield application-driven tool-selection guidelines.

cond-mat.mtrl-sci↗

arXiv2026

Physics-Informed Attention Mechanism and Generalization Capability of Deep Learning-Based Grain Growth Evolution Prediction

Machine Learning (ML) models for grain growth prediction are typically trained on idealized synthetic data, yet practical applications require generalization to conditions outside the training distribution. This study evaluated the Out-Of-Distribution (OOD) generalization capability of the trained model from our previous study across three test cases, including experimental microstructures, microstructures characterized by a bimodal grain size distribution, and abnormal grain growth. To further probe whether physics-informed architectural design could improve robustness under these different conditions, a boundary-masked attention mechanism was proposed specifically for grain growth, constraining attention to grain boundary pixels. Both the baseline and the proposed physics-informed attention model were evaluated without retraining or fine-tuning on the OOD data. Both models successfully generalized to all three test cases, yet the boundary-masked attention mechanism provided substantial improvements, with the most notable gains for microstructures characterized by a bimodal grain size distribution, where Structural Similarity Index Measure (SSIM) improved from \num{0.6221} to \num{0.7609} and mean grain size ($\overline{R}$) error decreased from \SI{8.75}{\percent} to \SI{3.57}{\percent}. The attention heatmap analysis revealed that the boundary-masked attention model learned to concentrate attention on large grain boundaries in a manner consistent with curvature-driven grain growth physics, emerging from training without being explicitly encoded into the architecture. These results indicate that models trained on synthetic data can generalize to diverse OOD conditions without retraining, and that physics-informed attention may improve accuracy when the boundary morphology matches the training domain.

cond-mat.mtrl-sci↗

arXiv2026

Predicting Grain Growth Evolution Under Complex Thermal Profiles with Deep Learning through Thermal Descriptor Modulation

Predicting microstructure evolution during thermomechanical treatment is essential for determining the final mechanical properties of a material, yet conventional simulations based on Partial Differential Equations (PDEs) remain computationally expensive. Our prior Deep Learning (DL) framework using Convolutional Long Short-Term Memory (ConvLSTM) has proven effective in accelerating grain growth prediction, though its applicability was limited to constant-temperature or single-rate thermal profiles. As the model was trained exclusively under constant thermal conditions, it cannot account for the thermal history dependence of grain boundary kinetics, fundamentally limiting its applicability to the time-varying thermal profiles characteristic of industrial heat treatment processes. This study extends the previous framework by incorporating Feature-wise Linear Modulation (FiLM) for thermal conditioning to predict grain growth under complex, time-varying thermal profiles. The model was trained on a large dataset of grain growth evolution under thermal profiles with heating and cooling rates ranging from 0.01 kelvin per second to 10 kelvin per second. The results demonstrate that the proposed thermal conditioning mechanism enables the model to capture the influence of variable thermal profiles on grain boundary migration kinetics. Across the three test scenarios of increasing complexity, the model achieved a Structural Similarity Index Measure (SSIM) of up to 0.93 and mean grain size error below 3.2%. Despite the architectural extensions, inference time remains on the order of seconds per prediction sequence, preserving the computational advantage over PDE-based simulations.

cond-mat.mtrl-sci↗

arXiv2026

High-fidelity level-set modeling of polycrystalline grain growth

Accurate modeling of polycrystalline microstructure evolution under strong crystallographic heterogeneities remains a major challenge for full-field numerical methods at the mesoscopic scale. In this work, we present a high-fidelity level-set framework for capillarity-driven grain growth in polycrystals with highly-heterogeneous, disorientation-dependent grain boundary energies. The novel framework represents a polycrystalline extension of our level-set formulation, previously developed and validated using a single triple junction benchmark case. In-depth comparisons with three established level-set models demonstrate that the proposed method yields the most energetically-consistent evolution of grain statistics, disorientation distribution function, and triple junction dihedral angles. Accuracy and robustness are maintained across the entire heterogeneity spectrum. To the best of our knowledge, this approach delivers the highest-fidelity front-capturing level-set modeling of grain growth based on Mullins' mean curvature flow theory, paving the way for state-of-the-art digital twins for annealing applications.

cond-mat.mtrl-sci↗

arXiv2025

InfinityEBSD : Metrics-Guided Infinite-Size EBSD Map Generation With Diffusion Models

Materials performance is deeply linked to their microstructures, which govern key properties such as strength, durability, and fatigue resistance. EBSD is a major technique for characterizing these microstructures, but acquiring large and statistically representative EBSD maps remains slow, costly, and often limited to small regions. In this work, we introduce InfinityEBSD, a diffusion-based method for generating monophase realistic EBSD maps of arbitrary size, conditioned on physically meaningful microstructural metrics. This approach supports two primary use cases: extending small experimental EBSD maps to arbitrary sizes, and generating entirely new maps directly from statistical descriptors, without any input map. Conditioning is achieved through eight microstructural descriptors, including grain size, grain perimeter, grain inertia ratio, coordination number and disorientation angle distribution, allowing the model to generate maps that are both visually realistic and physically interpretable. A patch-wise geometric extension strategy ensures spatial continuity across grains, enabling the model to produce large-scale EBSD maps while maintaining coherent grain boundaries and orientation transitions. The generated maps can also be exported as valid Channel Text Files (CTF) for immediate post-processing and analysis in software such as MTEX or simulation environments like DIGIMU. We quantitatively validate our results by comparing distributions of the guiding metrics before and after generation, showing that the model respects the statistical targets while introducing morphological diversity. InfinityEBSD demonstrates that diffusion models, guided by physical metrics, can bridge the gap between synthetic and realistic materials representation, paving the way for future developments such as 3D realistic microstructure generation from 2D data.

cond-mat.mtrl-sci↗

arXiv2025

Predicting Grain Growth in Polycrystalline Materials Using Deep Learning Time Series Models

Grain Growth strongly influences the mechanical behavior of materials, making its prediction a key objective in microstructural engineering. In this study, several deep learning approaches were evaluated, including recurrent neural networks (RNN), long short-term memory (LSTM), temporal convolutional networks (TCN), and transformers, to forecast grain size distributions during grain growth. Unlike full-field simulations, which are computationally demanding, the present work relies on mean-field statistical descriptors extracted from high-fidelity simulations. A dataset of 120 grain growth sequences was processed into normalized grain size distributions as a function of time. The models were trained to predict future distributions from a short temporal history using a recursive forecasting strategy. Among the tested models, the LSTM network achieved the highest accuracy (above 90\%) and the most stable performance, maintaining physically consistent predictions over extended horizons while reducing computation time from about 20 minutes per sequence to only a few seconds, whereas the other architectures tended to diverge when forecasting further in time. These results highlight the potential of low-dimensional descriptors and LSTM-based forecasting for efficient and accurate microstructure prediction, with direct implications for digital twin development and process optimization.

cs.LG↗

arXiv2025

Vertex and front-tracking methods for the modeling of microstructure evolution at the solid state: a brief review

In mesoscopic scale microstructure evolution modeling, two primary numerical frameworks are used: Front-Capturing (FC) and Front-Tracking (FT) ones. FC models, like phase-field or level-set methods, indirectly define interfaces by tracking field variable changes. On the contrary, FT models explicitly define interfaces using interconnected segments or surfaces. In historical FT methodologies, Vertex models were first developed and consider the description of the evolution of polygonal structures in terms of the motion of points where multiple boundaries meet. Globally, FT-type approaches, often associated with Lagrangian movement, enhance spatial resolution in 3D surfacic and 2D lineic problems using techniques derived from finite element meshing and remeshing algorithms. These efficient approaches, by nature, are well adapted to physical mechanisms correlated to interface properties and geometries. They also face challenges in managing complex topological events, especially in 3D. However, recent advances highlight their potential in computational efficiency and analysis of mobility and energy properties, with possible applications in intragranular phenomena.

cond-mat.mtrl-sci↗

arXiv2025

High-fidelity modeling of interface crossing in the diffusion welding process at the polycrystalline scale

Controlling the microstructure of a diffusion welded interface is a critical point to ensure optimum mechanical properties and the homogeneity of the joint. Beyond the intimate contact formation between bonded parts studied in the literature, this article focuses on the grain boundary crossing of the interface during this process and its measurement. Following this perspective, a Level-Set method has been used for full-field microstructure simulations in 2D with various interface parameters. Two crossing measurement models have been formulated, tested and discussed over the simulations.

cond-mat.mtrl-sci↗

arXiv2025

High-fidelity Grain Growth Modeling: Leveraging Deep Learning for Fast Computations

Grain growth simulation is crucial for predicting metallic material microstructure evolution during annealing and resulting final mechanical properties, but traditional partial differential equation-based methods are computationally expensive, creating bottlenecks in materials design and manufacturing. In this work, we introduce a machine learning framework that combines a Convolutional Long Short-Term Memory networks with an Autoencoder to efficiently predict grain growth evolution. Our approach captures both spatial and temporal aspects of grain evolution while encoding high-dimensional grain structure data into a compact latent space for pattern learning, enhanced by a novel composite loss function combining Mean Squared Error, Structural Similarity Index Measurement, and Boundary Preservation to maintain structural integrity of grain boundary topology of the prediction. Results demonstrated that our machine learning approach accelerates grain growth prediction by up to \SI{89}{\times} faster, reducing computation time from \SI{10}{\minute} to approximately \SI{10}{\second} while maintaining high-fidelity predictions. The best model (S-30-30) achieving a structural similarity score of \SI{86.71}{\percent} and mean grain size error of just \SI{0.07}{\percent}. All models accurately captured grain boundary topology, morphology, and size distributions. This approach enables rapid microstructural prediction for applications where conventional simulations are prohibitively time-consuming, potentially accelerating innovation in materials science and manufacturing.

cond-mat.mtrl-sci↗

arXiv2025

An accurate and robust level-set formulation for multiple junction kinetics

The front-capturing Level-Set (LS) method is widely employed in academia and industry to model grain boundary (GB) migration during the microstructure evolution of polycrystalline materials under thermo-mechanical treatments. During capillarity-driven grain growth, the conventional mean curvature flow equation, $\vec{v} = - \mu \gamma \kappa \vec{n}$, is used to compute the GB normal migration velocity. Over recent decades, extensive efforts have been made to incorporate polycrystalline heterogeneity into this framework. However, despite increased complexity and computational costs, these approaches have yet to achieve fully satisfactory performance. This paper introduces a simple yet robust LS formulation that accurately captures multiple junction kinetics, even with extreme GB energy ratios. Validation against existing analytical solutions highlights the method's accuracy and efficiency. This novel approach offers significant potential for advancing the study of highly heterogeneous interface systems.

cs.CE↗

arXiv2025

Efficient and accurate simulation of the Smith-Zener pinning mechanism during grain growth using a front-tracking numerical framework

This study proposes a new full-field approach for modeling grain boundary pinning by second phase particles in two-dimensional polycrystals. These particles are of great importance during thermomechanical treatments, as they produce deviations from the microstructural evolution that the alloy produces in the absence of particles. This phenomenon, well-known as Smith-Zener pinning, is widely used by metallurgists to control the grain size during the metal forming process of many alloys. Predictive tools are then needed to accurately model this phenomenon. This article introduces a new methodology for the simulation of microstructural evolutions subjected to the presence of second phase particles. The methodology employs a Lagrangian 2D front-tracking methodology, while the particles are modeled using discretized circular shapes or pinning nodes. The evolution of the particles can be considered and modeled using a constant velocity of particle shrinking. This approach has the advantages of improving the limited description made of the phenomenon in vertex approaches, to be usable for a wide range of second-phase particle sizes and to improve calculation times compared to front-capturing type approaches.

cs.CE↗

arXiv2024

High-fidelity level-set modeling of diffusive solid-state phase transformations for polycrystalline materials

The formation of microstructures in metallic alloys during hot metal forming involves simultaneous metallurgical complex phenomena. Traditional high-fidelity numerical frameworks used on the polycrystalline scale tend to focus on single-phase microstructures or isolate phase transformations from grain boundary migration mechanisms. The level-set method is highlighted as effective in proposing a global framework for modeling multiphase polycrystalline materials and diffusive solid-state phase transformations. This framework includes novel techniques for efficient large-scale microstructural representation, strong coupling with ThermoCalc software for real-time thermodynamic data, application for ternary alloys and beyond by taking solute drag aspects, and the use of advanced nucleation models. Numerous applications are then illustrated.

cond-mat.mtrl-sci↗

arXiv2024

Particle pinning during grain growth -- A new analytical model for predicting the mean limiting grain size but also grain size heterogeneity in a 2D polycrystalline context

This study proposes a new analytical model for grain boundary pinning by second phase particles in two-dimensional polycrystals. This approach not only considers how particles impede grain growth, but also elucidates their role in preventing grain disappearance, thereby leading to stabilised microstructures characterised by heterogeneous grain size distribution comprising a mixture of small and large grains. By quantifying the number of particles intercepted by grain boundaries during grain growth or shrinkage, we are able to calculate the respective sizes and fractions of large and small grains. Furthermore, we identify ranges of particle surface fractions and particle sizes that maximise the heterogeneity in grain size. Additionally, we demonstrate the significant influence of initial grain size on the limiting grain size in pinned microstructures. Our analytical model's results are compared with those obtained from full-field level-set simulations conducted in this study and from phase-field calculations reported in the literature, revealing very good agreement. Finally, the differences between the proposed model and existing ones in the literature are discussed.

cond-mat.mtrl-sci↗

arXiv2023

Comparison of grain growth mean-field models regarding predicted grain size distributions

Mean-field models have the ability to predict grain size distribution evolution occurring through thermomechanical solicitations. This article focuses on a comparison of mean-field models under grain growth conditions. Different microstructure representations are considered and discussed, especially regarding the consideration of topology in the neighborhood construction. Experimental data obtained with a heat treatment campaign on a 316L austenitic stainless steel are used for material parameters identification and as a reference for model comparisons. Mean-field models are also confronted to both mono- and bimodal initial grain size distributions to investigate the interest of introducing neighborhood topology in microstructure predictions models. This article exposes that improvements in the predictions are obtained in monomodal cases for topological models. In bimodal test, no comparison with experimental data were performed as no data were available. But relative comparisons between models indicate few differences in predictions. The interest of neighborhood topology in grain growth mean-field models gives overall small improvements compared to classical mean-field models when comparing implementation complexity.

cond-mat.mtrl-sci↗

arXiv2023

Kinetic equations and level-set approach for simulating solid-state microstructure evolutions at the mesoscopic scale: state of the art, limitations, and prospects

For over three decades, the front-capturing level-set method has demonstrated its prowess for the simulation, at the mesoscopic scale, of numerous mechanisms in the context of microstructure evolution occurring during complex thermomechanical paths. This review delves into the foundations of this numerical framework, charting its evolution concerning polycrystalline materials, examining its recent advancements, scrutinizing its current shortcomings, and exploring future possibilities. Special attention will be given to the context of hot metal forming processes. In this context, this article also aims to reintroduce, as simply as possible, the kinetic equations related to the grain boundary migration.

cs.CE↗

arXiv2023

Modeling CDRX and PDRX during hot forming of zircaloy-4

A recently developed full field level-set model of continuous dynamic recrystallization is applied to simulate zircaloy-4 recrystallization during hot compression and subsequent heat treatment. The influence of strain rate, final strain and initial microstructure is investigated, by experimental and simulation tools. The recrystallization heterogeneity is quantified. This enables to confirm that quenched microstructures display a higher extent of heterogeneity. The simulation results replicate satisfactorily experimental observations. The simulation framework is especially able to capture such recrystallization heterogeneity induced by a different initial microstructure. Finally, the role of intragranular dislocation density heterogeneities over the preferential growth of recrystallized grains is pointed out thanks to additional simulations with different numerical formulations.

cond-mat.mtrl-sci↗

arXiv2022

Simulation of continuous dynamic recrystallization using a level-set method

Dynamic recrystallization is one of the main phenomena responsible for microstructure evolutions during hot forming. Consequently, getting a better understanding of DRX mechanisms and being able to predict them is crucial. This paper proposes a full-field numerical framework to predict the evolution of subgrain structures upon grain growth, continuous dynamic and post-dynamic recrystallization. The microstructure representation into the numerical environment is presented. The developments made to improve substructure description are detailed extensively. Using these simulation tools, simulation of grain growth of a fully substructured microstructure are run. The influence of microstructure topology, of subgrain parameters and of some remaining stored energy due to plastic deformation is discussed. An analysis of the criterion for discrimination of recrystallized grains is proposed. Finally, the ability of the framework to model continuous dynamic and post-dynamic recrystallization is assessed upon a case study. The influence of grain boundary properties and of nucleation rules are studied. The representativity of the results in regards of experimental data will be discussed in an upcoming article.

cs.CE↗

arXiv2022

A level-set formulation to simulate diffusive solid/solid phase transformation in polycrystalline metallic materials -- Application to austenite decomposition in steels

Numerous full-field numerical methods exist concerning the digital description of polycrystalline materials and the modeling of their evolution during thermomechanical treatments. However, these strategies are globally dedicated to the modeling of recrystallization and grain growth for single-phase materials, or to the modeling of phase transformations without considering recrystallization and related phenomena. A generalized numerical framework capable of making predictions in a multi-phase polycrystalline context while respecting the concomitance of the different microstructural mechanisms is thus of prime interest. A novel finite element level-set based full-field numerical formulation is proposed to principally simulate diffusive solid-solid phase transformation at the mesoscopic scale in the context of two-phase metallic alloys. A global kinetic framework, capable of accounting for other concomitant mechanisms such as recrystallization and grain growth is considered in this numerical model. The proposed numerical framework is shown to be promising through a couple of illustrative 1D and 2D test cases in the context of austenite decomposition in steels and compared with ThermoCalc estimations.

cs.CE↗