The design of physics-augmented neural networks (PANNs) for the purposes of constitutive modeling has received considerable attention as of late for a variety of material behaviors. Here, we revisit the classical framework of isotropic incompressible hyperelasticity in light of recent advances in the study of constitutive inequalities. We show that polyconvexity implies true-stress-true-strain monotonicity for a large class of incompressible strain-energy functions. The resulting elastic law obeys the physically reasonable Legendre-Hadamard (or ellipticity) condition as well as the notion of increasing stress with increasing strain. These results then inform the architecture of four distinct PANNs which are subsequently calibrated to three different sets of experimental data each. We show that different PANN parametrizations - satisfying the same constitutive constraints a priori - have varying approximation power for the description of material behavior. Moreover, even when distinct parametrizations perform comparatively well within the calibration regime, they show pronounced differences in extrapolation. This observation motivates a critical discussion about the predictive power of PANNs which also has implications for the modeling of more complex material behavior by virtue of neural networks.
During alkaline water electrolysis, gas-evolving reactions generate bubbles that contribute to ohmic losses and an increase in activation and concentration overpotentials affect cell efficiency. Building on prior studies of bubble behavior, this work introduces two new aspects: spatially resolved void fraction near the electrode surface and locally generated gas volume flow as an indicator of surface activity. Both quantities are correlated with measured bubble dynamics. To achieve this, the oxygen evolution reaction is studied on a vertical wire electrode using Laser-Marked Shadowgraphy over a wide current-density range and multiple heights. Statistical distributions of bubble position, size, and velocity are obtained, together with local void fraction and gas volume flow. Their dependence on current density and vertical position is analyzed. Results show that bubble size primarily governs void fraction and, thus, bubble-induced resistance. A feedback mechanism is observed in which bubble-induced electrolyte flow enhances bubble transport, while bubble-bubble interactions cause a slowdown at high current densities. Spatial variations in generated gas volume flow reveal inhomogeneous apparent surface activity linked to convective transport. Overall, the results demonstrate the interplay of bubble dynamics and the electrochemical process and underscore the need for quantitative, spatially resolved, and statistically converged data for bubble dynamics.
The accurate measurement of the motion of rigid solid fuel particles is important for the study of many appli cations in energy science. These dynamics influence the particle-turbulence interaction as well as the heat and mass transfer processes governing reactor performance and chemical conversion. While many methods for the tracking of the translational velocity of such particles are available, methods for the robust measurement of angu lar velocity are scarce, especially for nearly circular particles and particles with complex shapes. To address the issue, this study presents a novel methodology for computing the angular velocity of dispersed particles in twophase flows using a wavelet-based optical flow velocimetry (wOFV) algorithm. First, the applicability of wOFV for particle velocimetry was validated against a synthetic ground-truth dataset of walnut shell fragments as a representative biomass and the influence of the regularization parameter lambda was investigated. It was found that for 0.01 <= lambda <= 5, particle rotation can be accurately determined using the presented wOFV method. Additionally, findings reveal that particle shape significantly influences the performance of both wOFV and a commonly used ellipse fit tracking method, which is used as a benchmark. While the accuracy of ellipse fit tracking correlates strongly with the particle aspect ratio, the accuracy of wOFV is more strongly determined by the ellipticity of fuel particles. The implementation of a hybrid approach that integrates wOFV and ellipse fit tracking has been demonstrated to leverage the strengths of both algorithms. This is achieved through a shape-dependent selection of the most suitable algorithm for each specific particle based on a support vector machine classification. Overall, this hybrid approach yields a reduction in the median relative error of angular velocity predictions from 4.2 % (benchmark method ellipse fit tracking) to 3.7% (hybrid method) for the employed test set. The proposed frame work demonstrates considerable potential for enhancing the precision of rotational measurements in the field of particle dynamics.
This paper examines defending the power grid against load-altering attacks using electric vehicle charging. It proposes to preventively segment the cyber infrastructure that charging station operators (CSOs) use to communicate with and control their charging stations, thereby limiting the impact of successful cyber-attacks. Using real German charging station data and a reconstructed transmission grid model, a threat analysis shows that without segmentation, the successful hack of just two CSOs can overload two transmission grid branches, exceeding the N-1 security margin and necessitating defense measures. A novel defense design problem is then formulated that minimizes the number of imposed segmentations while bounding the number of branch overloads under worst-case attacks. The resulting IP-MILP bi-level problem can be solved with an exact column and constraint generation algorithm and with heuristics for fast computation on large-scale instances. For the near-real-world Germany case, the applicability of the heuristics is demonstrated and validated under relevant load and dispatch scenarios. It is found that the simple scheme of segmenting CSOs evenly by their installed capacity leads to only 23% more segments compared to the heuristic optimization result, suggesting potential relevance as a regulatory measure.
Recent advancements in eXtended Reality (XR) technologies have opened new opportunities for integrating virtual and physical environments, enabling natural immersive user experiences. This paper addresses the challenge of developing socially acceptable XR agents that can engage in complex, meaningful interactions across various social and public settings. Specifically, we propose a modular architecture for enhancing user-XR agent interaction, integrated into the framework of the Horizon Europe project “Socially-acceptable Extended Reality Models And Systems (SERMAS)”. The architecture is designed to ensure adaptability, scalability, and the integration of diverse communication modalities, including verbal and non-verbal cues. Its main components work together to enable seamless user detection and communication. In particular, the Detection module plays a key role in managing the agent’s awareness of the surrounding environment. Its functions include identifying user intentions, monitoring group dynamics, and inferring emotional states, thereby increasing responsiveness and enabling more contextually grounded behavior. In contrast, the Communication module employs Large Language Models for personalized verbal responses, and recognizes gestures and body language for enriched non-verbal communication. Together, these approaches guarantee contextually relevant, personalized, and emotionally intelligent interactions. The architecture supports scalability and the flexible integration or replacement of modules, fostering socially acceptable user-XR agent interactions. Its real-world applicability is demonstrated across three deployment scenarios, including a digital receptionist integrated with a physical robot, an XR-based security training system for journalists practicing safety-critical procedures, and a digital assistant supporting customers in a post-office environment. The functionality of the proposed system is presented through the implementation of the XR Agent interacting with users.
