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

Sterley Labady, Youssef Mesri, Daniel Pino Munoz, Baptiste Flipon, Marc Bernacki · 2025

Read InfinityEBSD : Metrics-Guided Infinite-Size EBSD Map Generation With Diffusion Models by Sterley Labady, Youssef Mesri, Daniel Pino Munoz, Baptiste Flipon, Marc Bernacki (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-17859.txt.

Introduction

Sterley Labady and four coauthors work with EBSD, a lab method that maps the tiny crystal grains inside a material. Such maps guide judgments about strength and wear, yet they are slow and costly to collect over large areas. The team built InfinityEBSD to grow maps of any size with a diffusion model.

The model takes eight physical measures, such as grain size and shape, as its guide. It can stretch a small measured map or create a fresh one from numbers alone, keeping grain borders smooth across patches. The output feeds straight into standard analysis tools, which helps engineers test designs on realistic structures.

Catalog record

CREATORSterley Labady, Youssef Mesri, Daniel Pino Munoz, Baptiste Flipon, Marc Bernacki
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-17859

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

InfinityEBSD : Metrics-Guided Infinite-Size EBSD Map Generation With Diffusion Models Sterley Labady, Youssef Mesri, Daniel Pino Munoz, Baptiste Flipon, Marc Bernacki Abstract 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. K eywords Polycrystalline Microstructures, Electron Backscatter Diffraction (EBSD), Diffusion Models, Attention, Transformers, Residual Learning, Deep Learning 1 Introduction Materials play a critical role in modern industries, particularly in sectors such as aerospace, automotive, and renewable energy. The design and optimization of materials are intrinsically linked to their microstructures, which directly influence their performance and durability. Grain orientation and size, boundary connectivity (number of neighbors, disorientation) shape the macroscopic behavior of materials. Despite their importance, the acquisition of high-quality microstructures data, remains a significant challenge. Even with the increasing availability of advanced imaging techniques, including Electron Backscatter Diffraction (EBSD) [ 52 ] and Focused Ion Beam - Scanning Electron Microscopy (FIB-SEM) [ 16 ] , most microstructure datasets remain expensive to obtain, and constrained in size. These constraints are particularly visible in 2D EBSD maps acquisitions [ 46 ] , where they are often limited to small regions, capturing just a few dozen to a few hundred grains. While these are sufficient for qualitative inspection, they are rarely enough to provide statistical and physical representativeness. This limitation can lead to biased analysis, especially in materials with heterogeneous microstructures. Moreover acquiring larger EBSD maps, which offer a better statistical representation, is slow and expensive, both in terms of microscope time and post-processing effort [ 13 ] . These constraints have increased interest in generating EBSD maps of arbitrary size through data-driven methods that maintain realistic microstructural characteristics. Given the practical constraints of experimental microstructure acquisition, computational generation has emerged as a valuable and complementary strategy. By simulating microstructures [ 55 , 57 , 39 ] , instead of capturing them physically, researchers gain access to a flexible and scalable approach that can significantly reduce cost and experimentation time. Synthetic microstructures make it possible to explore conditions that would be too time-consuming or difficult to reproduce experimentally. In addition, they provide a controlled baseline for assessing and refining experimental techniques. Synthetic microstructure reconstruction is not new. Traditional approaches to microstructure reconstruction often rely on statistical and physical models. Techniques such as correlation functions [ 56 ] and geometric approximations using tessellation methods (e.g., Voronoi [ 42 ] and Saltykov [ 51 ] ) have been widely used to generate polycrystalline microstructures [ 24 ] . However, they tend to produce oversimplified grain morphologies with simplistic shapes and are rarely able to capture real-world microstructure details and boundary networks. More recently, machine learning models such as Convolutional neural networks (CNN) [ 32 ] , Generative Adversarial Networks (GAN) [ 19 ] , and Denoising Diffusion Probabilistic Models (DDPM) [ 25 , 54 ] have been applied with success to microstructure generation [ 33 , 29 , 7 , 26 , 15 , 45 ] . However, most of these models treat microstructures as grayscale images [ 15 , 33 , 29 , 45 ] . As a result, they may produce plausible-looking textures that lack physical meaning. In this work, we adopt the view that generative models should not create random microstructures. Instead, these models need to be guided or constrained by physically meaningful information such as grain size, shape, and orientation distribution. To date, there is no established method that can generate EBSD maps of arbitrary size while ensuring consistency with a complete set of microstructural metrics. These include grain size, grain perimeter, grain inertia ratio, coordination number, grain boundary disorientation and the three Euler angles that describe crystal orientation. The ability to generate EBSD maps that match given metrics is crucial for producing synthetic data that is not only visually realistic but also physically meaningful for simulations or analysis. Figure 1: Overview of the proposed method for extending EBSD maps from small ones (512×512 cells) to large generated maps (2048×2048 cells). The generated microstructures follow statistical and physical characteristics of the input microstructures. Map (a) corresponds to Inconel 718 (In718), a nickel-based superalloy and Map (b) corresponds to an austenitic stainless steel (316L). In this work, we present InfinityEBSD, as shown in Figure 1 , a new method to generate large EBSD maps of arbitrary size, conditioned on microstructural metrics, using DDPM trained in the latent space of a Variational AutoEncoder (VAE) [ 31 ] . We extend small EBSD maps of arbitrary size input, into large ones of arbitrary size output, preserving both the morphological realism and the statistical properties of the input map. Additionally, the model can generate entirely new EBSD maps from scratch by conditioning on a set of target microstructural metrics, without requiring any input map. When dealing with high-resolution data such as EBSD maps, a common problem with DDPM is their computational consumption [ 47 , 28 , 61 , 62 ] . That is why we address this issue by shifting the generation process to a compressed latent space. Rather than relying on raw experimental Euler angles values, we first encode EBSD maps into a compact 4D latent space learned by a dedicated VAE [ 47 ] . This separation allows the model to focus its generative capacity on meaningful map content while discarding imperceptible noise or repetitive information. This results in substantial gains in efficiency without a noticeable loss in detail or structure. A conditional UNet [ 48 ] then learns to denoise these latents, conditioned on microstructural metrics encoded as continuous embeddings. As it will be illustrated, that guided DDPM actually are capable of generating high quality EBSD maps. InfinityEBSD also introduces a geometric extension mechanism for generating maps of arbitrary size. Unlike standard inpainting and outpainting [ 44 ] methods that work within fixed-size contexts, our method generates a series of

Read the complete text