In fields such as material design or biomedicine, fiber materials play an important role. Fiber simulations, also called digital twins, provide a basis for testing and optimizing the material's physical behavior digitally. Inter-fiber contacts can influence the thermal and mechanical behavior of a fiber system; to our knowledge, however, there exist no parametric fiber models allowing for explicit modeling of the number of inter-fiber contacts. Therefore, this paper proposes an extension of the iterative force-biased fiber packing by Altendorf & Jeulin. In this extension, we model the inter-fiber contacts explicitly and add another force to the force-biased packing to increase the number of contacts. We successfully validate the packing with respect to its parameter accuracy. Moreover, we show that the extension indeed increases the number of contacts, even exceeding theoretical values. Hence, this packing scheme has the potential to achieve higher accuracy in physical simulations.

Spallation in ductile metals involves complex void nucleation and growth mechanisms, but the interactions between voids and the resulting statistical structure of fracture surfaces remain a persistent challenge for both experimental and theoretical modeling. This study develops a generative model to capture the statistical features of spall-induced fracture surfaces in high-purity aluminum. Aluminum samples were subjected to nanosecond laser-induced spallation, and the resulting fracture surfaces were imaged via scanning electron microscopy (SEM) and reconstructed in 3D. Individual dimples were segmented and analyzed to extract void size distributions and the spatial arrangement of nucleation sites. Boolean models and Gaussian random fields were then used to generate synthetic surfaces and compared against the experimental data using one- and two-point statistics. The analysis revealed a Poisson distribution of nucleation centers within the spall plane but significant out-of-plane spatial correlations in nucleation depth. The extended generative model successfully reproduces both the surface height distribution and the spatial covariance observed experimentally. These results emphasize the need to incorporate large-scale spatial correlations in predictive models of dynamic ductile damage. The proposed framework provides a basis for future studies of collective void growth and spall surface formation in dynamic ductile fracture.

We develop a physics-informed neural network pipeline for solving linear elastic micromechanics in three dimensions, on a statistical volume element (SVE) of a polycrystalline material periodic geometry. The presented approach combines a convolutional neural network containing residual connections with physics-informed non-trainable layers. The latter are introduced enforce the strain field admissibility and the constitutive law in a way consistent with so-called fast Fourier transform (FFT) algorithms. More precisely, differential operators are discretized finite differences in accordance with the Green operator used in FFT computations and treated convolutions with fixed kernels. The deterministic relationship between crystalline orientations and stiffness tensors is transferred to the network by an additional non-trainable layer. A function dependent on the divergence of the predicted stress field allows for updating the neural network's parameters without further supervision from ground truth data. The surrogate model is trained on untextured synthetic polycrystalline SVEs with periodic boundary conditions, realized from a stochastic 3D microstructure model based on random tessellations. Once trained, the network is able to predict the periodic part of the displacement field from the crystalline orientation field (represented as unit quaternions) of an SVE. The proposed self-supervised pipeline is compared to a similar one trained with a data-driven loss function instead. Further, the accuracy of both models is analyzed by applying them to microstructures larger than training inputs, as well as to SVEs generated by the stochastic 3D microstructure model, utilizing various different parameters. We find that the self-supervised pipeline yields more accurate predictions than the data-driven one, at the expense of a longer training. Finally, we discuss the trained surrogate model can be used to solve certain inverse problems on polycrystalline domains by gradient descent.

The properties of porous glasses and their field of application strongly depend on the characteristics of the void space. Understanding the relationship between their porous structure and failure behaviour can contribute to the development of porous glasses with long-term reliability optimized for specific applications. In the present work, we used X-ray computed tomography with nanometric resolution (nano-CT) to image a controlled pore glass (CPG) with 400 nm-sized pores whilst undergoing uniaxial compression in-situ to emulate a stress process. Our results show that in-situ nano-CT provides an ideal platform for identifying the mechanisms of damage within glass with pores of 400 nm, as it allowed the tracking of the pores and struts change of shape during compression until specimen failure. We have also applied computational tools to quantify the microstructural changes within the CPG sample by mapping the displacements and strain fields, and to numerically simulate the behaviour of the CPG using a Fast Fourier Transform/phase-field method. Both experimental and numerical data show local shear deformation, organized along bands, consistent with the appearance and propagation of +/- 45 degrees cracks.

A stochastic 3D microstructure model for polycrystals is introduced which incorporates two types of twin grains, namely neighboring and inclusion twins. They mimic the presence of crystal twins in γ-TiAl polycrystalline microstructures as observed by 3D imaging techniques. The polycrystal grain morphology is modeled by means of Voronoi and –more generally– Laguerre tessellations. The crystallographic orientation of each grain is either sampled uniformly on the space of orientations or chosen to be in a twinning relation with another grain. The model is used to quantitatively study relationships between morphology and mechanical properties of polycrystalline materials. For this purpose, full-field Fourier-based computations are performed to investigate the combined effect of grain morphology and twinning on the overall elastic response. For γ-TiAl polycrystallines, the presence of twins is associated with a softer response compared to polycrystalline aggregates without twins. However, when comparing the influence on the elastic response, a statistically different polycrystalline morphology has a much smaller effect than the presence of twin grains. Notably, the bulk modulus is almost insensitive to the grain morphology and exhibits much less sensitivity to the presence of twins compared to the shear modulus. The numerical results are consistent with a two-scale homogenization estimate that utilizes laminate materials to model the interactions of twins.

In this study, we revisit the Choquet capacity in the framework of convolutional neural networks, in (max,+)-algebra. By incorporating a discrete and learnable Choquet capacity model, we enhance the ability to represent the spatial arrangement and density variations in random point processes of convolutional neural networks. To validate the effectiveness of our approach, numerical experiments are conducted on synthetic datasets simulating diverse spatial point patterns of the Neyman-Scott process. When compared to classical convolutional neural networks, the proposed approach exhibits comparable or improved performances in terms of classification. Superior results are also observed in regression problems involving the Neyman-Scott parameter that monitors the point patterns spatial dispersion.

We investigate the influence of crystallographic twins on the elastoplastic response of γ-TiAl intermetallics via full-field FFT-based computations. We first introduce a hierarchical stochastic model, which is used to simulate synthetic polycrystalline microstructures containing twin grains with certain morphologies, and apply it to generate representative volume elements. Second, we develop a Fourier-based method with regularization for solving the effective and local mechanical response of polycrystalline media using the Méric-Cailletaud crystal plasticity constitutive law. Numerical results show that, across configurations of twinning, the corresponding average effective response is similar. Although differences were quantified, the effect of twins on the yield stress is negligible in practice (less than 1%).

The Deep Soil Mixing (DSM) material is a cementitious composite consisting of a well-mixed soil/cement matrix and may contain unmixed soil inclusions and possibly gravel. To obtain a 3D reconstruction of a DSM specimen structure, a destructive method was investigated, as an alternative to X-Ray Computed Tomography scans. The proposed method is based on image analysis and photographs of few millimeters thickness slices of a specimen. A detailed morphological description of inclusions, including volume fraction, shape, size distribution and spatial arrangement using common image analysis and CAD softwares preceded the validation of the 3D destructive method by the non-destructive X-ray CT based on 3 criteria (volume fraction, size distribution and shadow zone area). Compared to 1D and 2D basic methods found in the literature, the 3D developed method prevented biases linked to the external surface estimation of inclusions’ volume fraction. Next, to improve the experimental procedure, an extended methodology was established for estimating the inclusions volume fractions. Such method relies on a database of 3D reconstructed inclusion shapes and on the optimization of the sawing thickness. The aim consisted in the obtaining of unbiased estimates of the volume fractions with fewer cuts, between 4 and 18 instead of 50–60, which made such method as a low-cost, practical and accurate procedure. Finally, the 3D mesh generated by the 3D destructive method was used in numerical approach to propose, as an example, a successful hydromechanical numerical simulation of the DSM material response. Furthermore, a machine learning framework using the 2D slices of specimens produced during the application of the 3D destructive method, was proposed to generate various realistic 3D shapes instead of the usual spherical shape for inclusions, which will allow testing high number of meso-structures in numerical simulation.

to the reviewed document: hal-03110877v1

Vacuum chambers used in high-energy particle accelerator experiments are conventionally made of bulk beryllium, which shows significant drawbacks due to cost and toxicity. An alternative solution could be to develop chambers made of polymer-based composites. Since these materials exhibit high outgassing not compatible with an ultra-high vacuum environment, a suitable gas-tight coating is required. Cold spray deposition of aluminum can be a solution, provided that the coating behaves as a perfect vacuum barrier. Porosity, especially percolating porous networks, is key to coating gas tightness issues. This work addresses the relationship between porosity and gas-tightness in cold spray coatings. To do so, coatings with different porosity were achieved playing with powder morphology, composition, and process parameters. Their gas tightness was evaluated by helium leak tests. Classical microscopy, being essentially a 2D analysis, is strongly limited when dealing with 3D properties as porosity percolation. For this reason, 3D X-ray microtomography images of coatings were obtained and treated by image analysis methods: pores were compared in terms of size and shape. Overall porosity properties, including percolation and a homogeneity criterion, were also investigated. Percolating porosity was highlighted for several samples which showed poor gas-tightness properties. The permeability of percolating pore structures was then numerically computed by a fast Fourier transform-based method, to quantify the mass flow through the coating. Results of those computations were finally compared to experimental coating leak rate measurements, in an effort to elucidate the link between gas tightness and morphology of the pore space.

We develop a machine-learning image segmentation pipeline that detects ductile (as opposed to brittle) fracture in fractography images. To demonstrate the validity of our approach, use is made of a set of fractography images representing fracture surfaces from cold-spray deposits. The coatings have been subjected to varying heat treatments in an effort to improve their mechanical properties. These treatments yield markedly different microstructures and result in a wide range of mechanical properties that combine brittle and ductile fracture once the materials undergo rupture. To detect regions of ductile fracture, we propose a simple machine learning network based on a 32-layers U-Net framework and trained on a set of small image patches. These regions most often contain small dimples and differ by the surface roughness. Overall, the machine-learning method shows good predictive capabilities when compared to segmentation by a human expert. Finally, we highlight other possible applications and improvements of the proposed method.