Minisymposia

Modeling and simulation of polycrystalline materials

Organizer: Matthias Neumann

Abstract: Polycrystalline materials, including e.g. metals, ceramics, and semiconductors, exhibit complex microstructures characterized by grains, grain boundaries, and crystallographic orientations that critically influence their macroscopic behavior. Understanding and predicting the evolution and properties of these materials require an integrated approach combining advanced modeling, simulation, and quantitative microstructural analysis. This minisymposium explores recent developments in stochastic and numerical modeling of polycrystalline materials, with a focus on understanding and predicting microstructure-driven mechanical behavior. Contributions span a range of multiscale approaches that combine physical modeling, image-based analysis, and data-driven techniques to capture the complexity of polycrystalline systems.

Image analysis and data-driven stochastic modeling with applications in materials science

Organizers: Benedikt Prifling and Orkun Furat

Abstract: The 3D structure of many functional materials has a strong impact on effective macroscopic properties such as diffusivity, permeability or mechanical properties. Thus, a deeper understanding of quantitative structure-property relationships is crucial to further optimize the performance of materials. For this purpose, data-driven stochastic modeling can be leveraged, by considering models from stochastic geometry (such as random tessellations, random fields, point processes) and/or tools from machine learning (such as generative adversarial networks, stable diffusion). Once calibrated, these models can generate virtual structures that can be used as geometry input for numerical simulations of macroscopic material properties—enabling the systematic, computer-based investigation of structure-property relationships. Usually, the basis for model calibration is experimentally acquired 2D, 3D or even 4D image data, which has to be appropriately segmented into different phases and/or individual objects such as particles or cells. For these purposes, tools from image processing, mathematical morphology as well as methods from machine learning such as convolutional neural networks can be used. Frequently, image segmentation is followed by image analysis to quantify complex structures in terms of suitable geometrical descriptors. Often, the latter are computed from experimentally acquired data that is used as reference for model calibration. After model calibration and validation, a systematic variation of model parameters to generate a broad range of virtual, but still realistic structures just at the cost of computer simulations, which then serve as geometry input for spatially resolved numerical simulation of effective macroscopic properties. The focus of the present minisymposium is on recent methodological advances in image analysis or data-driven stochastic microstructure modeling with applications in materials science.

Testing in spatial statistics

Organizer: Tomáš Mrkvička

Abstract: Testing the statistical significance brings many challenges in spatial statistics. The parametric spatial model is constructed from more ingredients than the regression model with iid residuals; therefore, the nonparametric method can be useful in spatial statistics even more than in regression analysis with iid residuals. In this minisymposium, new nonparametric methods for variable selection will be presented both in spatial regression and point pattern analysis. Connected to the variable selection in point pattern is the common assumption of second-order intensity-reweighted stationarity, which will also be discussed. Also, nonparametric methods for goodness-of-fit testing will be presented in the new context, i.e., for hot-spot detection in a point pattern or multiple point patterns.