Numerical simulations of contaminant dispersion after a gas leakage incident at a chemical plant can provide valuable insights for both emergency response and preparedness. High-fidelity simulation approaches combine incompressible Navier-Stokes (INS) equations with advection-diffusion processes to model wind and concentration field. However, their computational cost increases rapidly for complex geometries and extended domains like urban environments. This renders them unfeasible in time-critical or multi-query "what-if'' scenarios, making model order reduction (MOR) techniques particularly valuable for enabling fast, accurate predictions. This study describes a MOR-based, application-driven workflow for uncertainty-aware contaminant concentration prediction, demonstrated for a two-dimensional benchmark system. Starting with the model selection, relevant comparison criteria and trade-offs are discussed for well-established proper orthogonal decomposition based (non-)intrusive methods, applied to the computationally more demanding parametric INS problem. They include accuracy, computational efficiency, data requirements, and extrapolation capability. Based on these insights, a non-intrusive parametric reduced-order model is constructed that enables accelerated Monte Carlo simulations for estimating the influence of potential wind measurement uncertainties on the spatio-temporal concentration field. This enables extracting valuable information for local consequence analyses and evacuation planning. Simulation results are furthermore interactively visualized in a dashboard and can serve as a building block within broader decision-support systems.
This paper presents modified augmented Lagrangian block preconditioners for the mixed-dimensional coupling of three-dimensional solid bodies with embedded one-dimensional torsion-free Kirchhoff-Love beams using Lagrange multipliers for constraint enforcement. The finite element discretization of this mixed formulation leads to an indefinite saddle-point system. An augmented Lagrangian formulation is employed to regularize the linear system while maintaining exact enforcement of the coupling constraints. Starting from the corresponding ideal augmented Lagrangian block preconditioner, more practical block-triangular variants are derived in which the solid, beam, and Schur complement blocks can be treated independently. In addition, different variants of Schur complement approximations are introduced. Numerical experiments investigate the dependence on model parameters and mesh refinement, as well as weak and strong scalability for short-fiber composite representative volume elements. For the tested parameter regime, a suitable choice of the Schur complement approximation yields favorable results, indicating the suitability of the proposed approach for large-scale simulations of mixed-dimensional models in solid and structural mechanics, as illustrated by an engineering example involving a composite sandwich plate.
Advancements in artificial intelligence (AI), particularly in generative AI and agentic AI, have intensified challenges related to transparency and explainability. While explainable AI (XAI) research has evolved to facilitate human interaction with such complex, black-box systems, research and practice lack clarity on the causal chain linking explanations, user perceptions, and real-world outcomes, a relationship that remains conceptually fragmented. To address this gap, we conducted a systematic literature review of 107 experimental user studies on XAI. We developed a conceptual framework guided by the stimulus-organism-response-consequences (S-O-R-C) model to systematize current human-XAI research and examine how users respond to explanations. Our study contributes to the literature by clarifying how explanations shape user interactions and downstream effects in real-world settings. We propose five research directions to help navigate the challenges of emerging AI systems (e.g., LLMs, AI agents) and evolving human-AI delegation.
The Taylor-Green vortex (TGV) serves as a canonical benchmark for studying the transition from laminar to turbulent flow in the absence of solid boundaries. Despite its widespread use in turbulence model validation, the degree to which the TGV exhibits true isotropy and homogeneity, particularly at late stages of decay, remains insufficiently examined. This study employs high-order numerical simulations to investigate these properties for both the standard and isotropic variants of the TGV. Statistical measures, including Reynolds stress anisotropy, coherent structure functions, homogeneity indices and integral length scales, are used to assess flow behaviour over time. Results show that the standard TGV remains anisotropic and inhomogeneous even during late decay stages, with unequal longitudinal length scales and directionally dependent homogeneity indices. The isotropic TGV maintains isotropy by design but still deviates from the characteristics of ideal homogeneous isotropic turbulence, exhibiting larger transverse than longitudinal length scales. Both configurations reveal persistent spatial inhomogeneities manifested as fixed peaks in turbulent kinetic energy and the coherent structure function of plane-averaged statistics. The findings highlight that while the isotropic TGV provides a more balanced and symmetric configuration, neither flow achieves fully homogeneous isotropic turbulence.
For autonomous vehicles, safe navigation in complex environments depends on handling a broad range of diverse and rare driving scenarios. Simulation- and scenario-based testing have emerged as key approaches to development and validation of autonomous driving systems. Traditional scenario generation relies on rule-based systems, knowledge-driven models, and data-driven synthesis, often producing limited diversity and unrealistic safety-critical cases. With the emergence of foundation models, which represent a new generation of pre-trained, general-purpose AI models, developers can process heterogeneous inputs (e.g., natural language, sensor data, HD maps, and control actions), enabling the synthesis and interpretation of complex driving scenarios. In this paper, we conduct a survey about the application of foundation models for scenario generation and scenario analysis in autonomous driving (as of May 2025). Our survey presents a unified taxonomy that includes large language models, vision-language models, multimodal large language models, diffusion models, and world models for the generation and analysis of autonomous driving scenarios. In addition, we review the methodologies, open-source datasets, simulation platforms, and benchmark challenges, and we examine the evaluation metrics tailored explicitly to scenario generation and analysis. Finally, the survey concludes by highlighting the open challenges and research questions, and outlining promising future research directions. All reviewed papers are listed in a continuously maintained repository, which contains supplementary materials and is available at https://github.com/TUM-AVS/FM-for-Scenario-Generation-Analysis.