ABSTRACT. We propose a IPT immersive display system built on ultra-short-throw projectors, a general-purpose PC, and NVIDIA Mosaic, requiring no proprietary libraries. By treating the combined projection area as a single logical desktop, any application that can open multiple windows and synchronize rendering across them can be used without modification. Off-axis frustum parameters for stereoscopic projection are derived analytically, and because the camera model is expressed as a text-based specification, rendering code can be generated automatically by a large language model. Case studies confirm that the system accommodates diverse visualization workflows.

ABSTRACT. hree-dimensional visualization is useful for understanding scalar fields in physics and scientific visualization. In this study, we present an interactive volume rendering approach for visualizing quantum-mechanical scalar fields using hydrogen atomic orbital densities as representative data. The probability density of each orbital is calculated on a structured grid and visualized by a Unity-based application developed using the volume-rendering framework implemented in VOIR. Compared with conventional isosurface visualization,volume rendering is found to provide a continuous representation of the probability density distribution and to enable internal and low-density structures to be observed, particularly in highly excited orbitals with spatially extended density distributions. By examining the effect of opacity control through the transfer function, we also show that the visibility of internal and surface-like structures can be adjusted interactively. Our results suggest that volume rendering is a useful technique for exploring three-dimensional quantum-mechanical scalar fields and other volumetric data in scientificvisualization.

ABSTRACT. Previous studies have reported that virtual objects tend to be perceived as smaller than their actual size in virtual reality environments. However, size perception in augmented reality (AR) environments using optical see-through head-mounted displays has not been sufficiently investigated. This study examined the characteristics of size perception for virtual objects in an AR environment. A 3D virtual model of a 350 mL beverage can was presented using the XREAL Air 2 Ultra. Participants compared the virtual models of different sizes with a 3D-printed physical model and judged whether it appeared "larger," "the same," or "smaller." Based on these responses, the point of subjective equality was calculated. The results indi-cated that virtual objects in the AR environment tended to be perceived as approximately 5% smaller than their actual size.

ABSTRACT. Driven by advances in computing, cost-effective head-mounted displays (HMDs) are replacing conventional two-dimensional displays and expensive CAVE systems for the visualization of large-scale three-dimensional simulation. Previously, we developed VOIR, a general-purpose visualization software for HMDs, but virtual reality environments posed safety risks due to blocked users’ vision of the real world and required avatars for interaction. In this study, VOIR was adapted for mixed reality using the capabilities of the Meta Quest 3/3S. By integrating the Depth API, dynamic occlusion by real-world objects and human bodies was achieved, facilitating intuitive communication via natural physical gestures. Furthermore, implementing the Colocation feature enables multiple users to observe and interact with synchronized virtual objects from their respective viewpoints in a shared physical space, providing a safe and interactive collaborative visualization platform.

ABSTRACT. Hiroshima University’s compact synchrotron radiation source, HiSOR, has been operated more than 30 years, and its source performance has been gradually declining compared with that of modern synchrotron radiation facilities. Therefore, the university is considering the prompt realization of a new light source accelerator with advanced and competitive performance. To achieve a high-brightness, low-energy synchrotron radiation source within an appropriate budget, radiation protection is an important con-sideration in addition to the accelerator itself. In this study, we conduct a simulation study of radiation generation and transport in a compact electron storage ring using PHITS. As a simulation model, a straight section and a bending magnet section filled with residual gas were constructed. Through interactions with the residual gas, electrons lose energy due to bremsstrahlung gamma-ray emission and are deflected from their trajectories by Rutherford scattering. The electrons then collide with accelerator com-ponents, such as the beam duct and electromagnets, inducing electromagnetic showers. Simulation results indicated that radiation tends to concentrate in specific directions, influenced by beam loss locations and accelerator structures.

ABSTRACT. In vascular shape simulation and computational fluid dynamics (CFD) analysis for coronary artery bypass grafting (CABG), investigating the effect of bypass graft diameter on hemodynamics is of great clinical and engineering importance. In this study, we propose a shape deformation method that takes a vascular shape model obtained from postoperative CT images as input and flexibly and robustly modifies the graft diameter. Experimental results show that the proposed method is robust against geometric failures such as self-intersections.

ABSTRACT. In the standard ICCG method, the numerical solution cannot be obtained for the large-scale EFG-type saddle-point (EFG-SP) problem. In order to resolve the above demerit, the improved Variable Reduction Method (iVRM) has been proposed. In general, the computational cost for the iVRM is generally higher than that for the standard ICCG method. In this study, the standard ICCG method is improved so that it can stably solve the large-scale EFG-SP problem. In addition, its performance is investigated numerically. The numerical computations show that the proposed method can stably obtain the numerical solution for the large-scale EFG-SP problem in contrast to the standard ICCG method. Therefore, not only the iVRM but also the ICCG($\tau$) method might be useful as the solver of the large-scale EFG-SP problem.

ABSTRACT. We have proposed a hierarchical parallel solver for saddle point problems that exploits their block structure, obtaining the solution via an intermediate linear system with multiple right-hand sides. To further improve the accuracy of the solution, we apply iterative refinement. This paper presents a theoretical convergence analysis of the hierarchical parallel solver with iterative refinement and validates the results through numerical experiments.

ABSTRACT. To reduce the adoption barrier of in-house solvers, we present an integrated CFD workflow platform that unifies geometry preparation, mesh generation, solver execution, and result visualization in a single environment without modifying the solver core, eliminating manual operations typical of in-house solver workflows. Development used an AI-assisted approach with domain experts retaining full architectural and validation responsibility. Applied to in-house CFD solver UNICONES, the platform reduced required tools from five to one and workflow steps from twenty to five, with onboarding time estimated at two days versus two weeks. The design is generalizable to other in-house solvers.

ABSTRACT. Strongly coupled methods can be classified into two categories; the monolithic method and the partitioned iterative method. In this study, we evaluated and compared their computational costs in piezoelectric analysis using the number of floating-point operations or FLOPs through a numerical example of piezoelectric bimorph. The results showed that the partitioned itera-tive method is clearly superior when solving systems of linear equations using direct solvers, while the monolithic method may be superior using iterative solvers.

ABSTRACT. In this study, adaptive mesh control methods consisting of three types of mesh control tech-niques are extended and applied to fluid-structure-control interaction (FSCI) analysis of insect flight. This method uses adaptive mesh control to deform and move the mesh surrounding the insect in response to wing motion and rigid-body motion of the body. Furthermore, by com-bining this with the expansion and contraction of distant mesh elements, the proposed method enables the analysis of insect free flight. By simulating flight altitude control under conditions where the insect flies vertically, this method demonstrates its effectiveness for analyzing in-sect free flight.

ABSTRACT. When we enjoy music, it’s usually in indoor spaces like live music clubs. An indoor environment like the human body gives sound affect. Listeners cannot listen to sounds in a uniform way. We aim to study numerical wave analysis methods using the finite element method (FEM) for musical instrument acoustics.

ABSTRACT. How do magnetically coupled inflows and outflows shape the birth of Sun-like stars? Addressing this question requires simulations that connect physical processes across scales and translate their predictions into observable signatures. This talk presents an integrative approach linking science-driven code development, high-performance astrophysical simulations, and synthetic observations, building on ASIAA’s CompAS program and the Coordinated Hydrodynamic and Astrophysical Research, Modeling, and Synthesis (CHARMS) project.

I first outline how astrophysical questions guide the development of numerical tools for modeling magnetized flows in protostellar systems. I then explore the interplay of jets, winds, and outflows with accreting material in magnetically collapsing envelopes. ALMA and JWST have revealed intricate, nested morphological and kinematic structures that offer new opportunities to test this framework. Magnetized bubbles driven by winds from the innermost regions, together with flattened inflow structures known as pseudodisks, provide a physical basis for interpreting these observations within a coupled jet–outflow–envelope system.

Tomographic projections and synthetic observations bridge three-dimensional simulations and telescope measurements, allowing the underlying dynamics to be examined through their projected spatial and velocity structures. I discuss ongoing efforts to connect these dynamical models with microphysics, chemical diversity, and dust complexity, alongside prospects for incorporating machine learning and artificial intelligence into future computational workflows. Finally, I introduce complementary explorations in laboratory astrophysics and cosmochemistry through studies of primitive extraterrestrial materials. Together, these directions illustrate how computation, observation, and laboratory investigation can inform an increasingly integrated understanding of stellar birth and the origins of our Solar System.

ABSTRACT. Spherical shell magnetohydrodynamics (MHD) dynamo simulations were performed using a model in which the dynamo is driven by externally imposed plasma flows. Simulations were conducted by varying the number of regions in which the flow-driving force was applied. The results showed that magnetic fields were less readily generated when the number of forcing regions was small. In contrast, the dipole magnetic field component was most dominant when four forcing regions were used. These results suggest that, under the present conditions, the use of four forcing regions is the most appropriate configuration for investigating dipole magnetic field generation.

ABSTRACT. We incorporated the VISMO in-situ visualization library into a Yin–Yang-grid MHD dy-namo simulation to investigate the mechanism of geomagnetic dipole polarity reversals. In-situ visualization enables efficient analysis on HPC systems while reducing storage and data-transfer costs. Streamlines of the prescribed plasma flow and isosurfaces of the magnetic energy equation terms were visualized during the simulation. Although the current visualization successfully captures the spatial structures of individual physical quantities, it is insufficient for identifying their correlations or temporal evolution. Fu-ture work will focus on advanced visualization techniques for analyzing multiphysics relationships during dipole polarity reversals.

ABSTRACT. Pumice rafts from submarine eruptions can drift over wide areas and damage fisheries, shipping, and coastal facilities, as seen after the 2021 Fukutoku-Oka-no-Ba eruption. Simulating it therefore depends on the ocean model driving the particle tracking, and ocean simulations of the seas around Japan are now available at 10 km, 2 km, and prospectively 500 m. Differences between them are usually summarized by broad-target arrival statistics, which compress an evolving field into a few numbers and leave the shape of the difference unseen. This study renders tracked particles as accumulated luminous trajectories and juxtaposes two simulations under identical particles, tracking, and rendering, so no visible difference can come from the setup. Applied to a release from Fukutoku-Oka-no-Ba, the two are visually alike for 60 days, then diverge in structure; descriptors computed from the same trajectories separate two kinds of reading. Brightness and overall spread prove hostage to how many particles are still moving—the runs ground very differently, the finer grid resolving islands the coarser one does not—while the shape of the structures and the reach of the western tail survive controls for both. The comparison thus shows where the models differ, which readings of such an image can be trusted, and what a subsequent quantitative analysis should measure.

ABSTRACT. We have been developing a cross-reference simulation code by Code-To-Code Adapter (CoToCoA) library for the study of cross-scale coupling processes occurring in planetary magnetospheres. The cross-reference simulation is a simulation model combining multi-ple simulation codes, which enables us to carry out multi-scale and multi-physics simula-tions. Recently we developed a Python interface to the CoToCoA library to enhance its usability. The developed Python interface was applied to visualize MHD simulation re-sults in real-time. In this paper, we present examples of the cross-reference simulation and results obtained by the Python interface of the CoToCoA library.

ABSTRACT. Mobile additive manufacturing (mobile AM) is capable of fabricating structures larger than the machine itself. In this study, to investigate the effect of control methods on fab-rication errors in mobile AM using an inchworm robot, we performed simulations of toolpaths during fabrication and conducted actual fabrication experiments. By control-ling the centre of gravity during operation, the machine was able to fabricate without falling over. Furthermore, we found that positioning errors in the planar direction occur due to the vertical positioning movements of the nozzle during fabrication.

ABSTRACT. The SLIM project aimed to achieve high-precision landing of a small unmanned lunar probe and successfully landed in January 2024. This study proposes a crater-based feature point detection method . This method extracts crater template images using PCA and achieves high-precision crater detection. However, the method lacks flexibility in handling crater shading directions. Additionally, it can only be applied to regions with a high density of craters. This study proposes a crater_based feature point detection method that is not dependent on shading direction and an efficient local feature extraction method, thereby aiming to con-struct a more practical self-localization system. The proposed method is evaluated in terms of rotation and scale invariance , as well as robustness against external disturbances .

ABSTRACT. Surface electromyography (sEMG) is a biosignal that noninvasively measures the elec- trical activity associated with muscle contraction using electrodes placed on the skin surface. An sEMG signal is composed of the superposition of motor unit action potentials (MUAPs) generated by multiple motor units. A model that represents the delay time as a variable function is the nonliner delayed ARX (NDARX) model. The objective of this study is to improve the predic- tion accuracy of elbow joint angle by using pseudo-motor unit action potential trains (MUAPTs) obtained from sEMG signals as inputs to an NDARX model that accounts for Electromechanical Delay (EMD). Pseudo-MUAPTs were generated from sEMG signals using independent component analysis based on a linear mixing model of MUAP components. To account for electromechanical delay, an NDARX model with output-dependent coefficients and a variable dead-time element was employed in this study. In this study, elbow joint flexion and extension movements were measured to estimate the elbow joint angle from sEMG signals. To generate the identification inputs based on the measured sEMG signals, LPF processing and MUAPT processing were applied. The estimation accuracy was evaluated using the root mean squared error. The pseudo-MUAPT input improved the estimation accuracy of the NDARX model compared with LPF-processed sEMG, reducing both the average error and standard deviation for 10 test datasets.

ABSTRACT. The JAXA/ISAS SLIM project achieves pinpoint lunar landing by matching craters detected in onboard images against a pre-built map. The target craters span only 8 to 13 pixels, and detection must run under strict power constraints. Conventional CNN detectors are accurate but power-hungry, whereas spiking neural networks (SNNs), which compute with sparse binary spikes and addition-only operations, are a natural low-power alternative. We adapt a spiking VGG16-SSD detector to lunar crater detection and, under identical architecture and data, compare it against a conventional CNN (ANN), adding Temporal Efficient Training and an L1 firing-rate regularization to im-prove spike sparsity. Evaluating four models on both detection and spiking efficiency, our full model attains an F1 of 0.862 and precision of 0.883 while requiring an estimated 3.5-5.0× less compute energy, demonstrating that spiking detectors can match conventional accuracy at substantially lower power for onboard crater detection.

ABSTRACT. This research proposes an autonomous optimization system for motion pictogram gen-eration using video generation AI, in which an Agentic Multimodal Large Language Model (MLLM) is integrated as an evaluation agent to control the quality variability in-herent in probabilistic generative AI outputs. In the proposed system, an MLLM as-signed the role of an expert in semiotics and human-centered design evaluates each mo-tion pictogram generated via a Text-guided Image-to-Video (TI2V) approach across four criteria: semantic clarity, semantic consistency, object consistency, and motion smooth-ness. When the evaluation falls below a defined threshold, the MLLM verbalizes the identified failure factors and feeds them back into the next generation prompt, forming an Agentic AI optimization loop. Experiments on five pictogram types demonstrated that MLLM numerical scores showed no statistically significant correlation with human evaluations; however, the three-tier overall judgment (Usable/Acceptable/Unusable) ex-hibited partial effectiveness as a filter for rejecting low-quality outputs. The effective-ness of the optimization loop varied by pictogram type: pictograms involving locally describable visual defects showed significant improvement through iterative refinement, while cases where MLLM hallucinations led to erroneous rejection of acceptable outputs resulted in a significant decline in final quality. These findings highlight both the prom-ise and the current limitations of Agentic AI-based autonomous quality control for AI-generated visual content.

ABSTRACT. Recent advances in speech synthesis technology have enabled the generation of high-quality synthetic speech that is difficult to distinguish from natural human speech. Although conventional synthetic speech detection methods often achieve high performance, they generally require large amounts of training data and rely on black-box models, making it difficult to interpret their decisions. To address these issues, this study focuses on the nonlinear characteristics of speech signals and proposes a classification method combining the Lyapunov Exponent (LE) and the Higuchi Fractal Dimension (HFD). Features based on LE and HFD were extracted from speech frames, and classification thresholds were determined using ROC analysis. Experimental results using the JVS corpus and VOICEVOX demonstrated that the proposed method achieved an Ac-curacy of 0.932, an F1-Score of 0.930, and an AUC of 0.960, outperforming methods using LE or HFD alone. These results demonstrate the effectiveness of combining LE and HFD is effective for synthetic speech detection.

ABSTRACT. With the advancement of AI technology, the misuse of high-quality synthetic speech has become a growing concern, making it urgent to develop technology capable of distin-guishing it from natural speech. Conventional methods using deep learning face chal-lenges such as being black box in nature and requiring large amounts of training data. Chaos analysis using the maximum Lyapunov exponent (MLE), which captures the nonlinear dynamics of speech, is effective; however, while static analysis efficiently captures the basic properties of these metrics, further clarification is needed where their distributions exhibit structural overlap. Therefore, this study proposes a discrimination method that superimposes minute Gaussian noise onto speech and utilizes the differ-ences in response characteristics to disturbances. This approach is based on the hypoth-esis that the two signals, which have different chaotic structures, exhibit distinct dy-namic responses to noise disturbances. The experimental results showed that the overlap in the MLE distributions of the two signals was effectively eliminated. Evaluation using test data revealed that the proposed method achieved an accuracy of 0.8859, an F1 score of 0.89342, and an AUC of 0.94189, demonstrating a significant improvement in classification accuracy compared to conventional methods.

ABSTRACT. The shielding effect of divertor leg structures against boron dust particle injection was investigated in the Large Helical Device (LHD) using edge plasma and dust particle transport simulation codes. The results show that the boron ion density and the radiation power induced by the boron dust particles in the plasma periphery are lower in an inward magnetic-axis-shift configuration than those in outward-shift configurations. This difference is attributed to the deflection of boron dust particle trajectories injected by an impurity powder dropper. These findings demonstrate that the divertor leg structures effectively shield the main plasma confinement region against dust particles, thereby improving the robustness of plasma discharges against impurity influx originating from dust particles released from the closed helical divertor region via plasma-wall interactions during long-pulse discharges.

ABSTRACT. In fusion power generation, plasma particles such as deuterium and tritium ions collide with the reactor wall, releasing hydrogen molecules. Understanding the transport of these emitted hydrogen molecules is essential for predicting edge plasma behavior. Neutral transport simulations require accurate predictions of the vibrational and rotational state distributions of the emitted hydrogen molecules. In this study, we developed a simulation system that automatically generates machine learning models to predict these state distributions by specifying the target material, incident particle species, and other simulation conditions. The system auto-mates molecular dynamics simulations, data collection, machine learning, and model generation. It also efficiently manages jobs and multiple computational resources through a dedicated management computer, enabling fully automated construction of predictive models.

ABSTRACT. We have developed a visualization system that reconstructs the Large Helical Device (LHD) at the National Institute for Fusion Science within a virtual environment using a Head-Mounted Display (HMD). The HMD enables users to explore simulation results with a high level of immersion and interactivity, potentially facilitating new insights into complex physical phenomena. In this study, we implemented a new visualization function for simulation results describing the behavior of lithium (Li) particles in the LHD. This paper presents the design, implementation, and evaluation of the proposed visualization function.

ABSTRACT. Tungsten, a leading plasma-facing material in fusion reactors, has its atomic structure continuously modified by hydrogen and helium. Hybrid molecular dynamics (MD)–kinetic Monte Carlo (kMC) simulations can treat both particle impact and long-time diffusion, but repeatedly recalculating trapping sites from the potential-energy field is costly. Extending a previous W–H model, we trained a three-dimensional pix2pix network to map voxelized W, H, and He configurations (128×128×128 over a 12.8 Å cube) to the potential distribution. For a representative test case the model reproduced the overall field, with errors concentrated near the steep atomic contours, demonstrating the feasibility of accelerating potential-field evaluation in future MD-kMC simulations.

ABSTRACT. In this study, CFD analysis was conducted using flying snake models with two undulatory configurations to clarify the effects of a mirror-image configuration of an S shape and the S-shaped configuration, which appear alternately during the undulatory motion of flying snakes in gliding flight, on gliding direction. In addition, the effect of the model's center-of-mass position on gliding distance was investigated, and the gliding directions of the two configurations were compared at the optimal center-of-mass position. As a result, the longest gliding distance was obtained when the center of mass was located at 47.5\% of the body length from the head tip. Furthermore, it was confirmed that the two configurations glided in opposite directions. These results suggest that, during the undulatory motion of flying snakes, the mirror image of an S-shaped configuration and the S-shaped configuration appear alternately, and the bias in gliding direction generated by each configuration is canceled out, enabling straight gliding flight.

ABSTRACT. As cyberattacks become increasingly sophisticated, deep learning-based network intru-sion detection systems (NIDSs) are gaining attention; however, the lack of transparency in AI decisions can delay operators’ responses. In this study, we propose a multi-stage NIDS that integrates detection, explanation, and classification to reduce the operational burden. The proposed system performs anomaly detection using a deep LSTM autoen-coder, visualizes the basis for decisions using SHAP (a form of XAI), and classifies at-tack types using a random forest based on extracted features. The proposed method achieved an attack classification accuracy of 93.63% and filtered 81.78% of false posi-tives generated by the anomaly detector.

ABSTRACT. The purpose of this paper is to develop a path interpolation algorithm to improve the effectiveness of the path-following control system for the Lizard-Inspired Single Actuated Robot (LISA). In a conventional control system, effectiveness was compromised when the initial error between LISA and the path was large. In this paper, we develop an algorithm to generate a path that interpolates between LISA's initial position and the reference path by combining two circular paths and one straight-line path. To this end, we formulate the conditions that these three types of paths must satisfy. We verify through numerical simulations that the developed algorithm can successfully generate interpolated paths.

ABSTRACT. LiDAR-based forest surveys are affected by occlusion caused by surrounding trees. In our previous study, we proposed a machine-learning-based method for determining scan positions to reduce occlusion and verified its effectiveness through simulation. In this study, we evaluate the method in an actual forest environment. It generates a forest map from point clouds acquired using a 2D LiDAR and sequentially determines the next scan position from the map. In the field experiment, 14 trees were detected at the position determined by the method, compared with 10 at the uniformly spaced baseline position, resulting in a 25\% improvement in measurement coverage. These results confirm its effectiveness under field conditions.

ABSTRACT. To solve time-harmonic problems with complex symmetric matrices in electromagnetic field computations more robustly and efficiently, ten new product-type Krylov subspace (PT) methods have been proposed. They have been derived from the COCG or COCR methods and have been evaluated through finite element analyses. As a result, in several test cases, iteration counts and the computational time have been reduced. In this paper, PT methods based on the COCG or COCR methods are applied to the hierarchical do-main decomposition method, and are evaluated in the parallel environment.

ABSTRACT. Microwave scalpels, which are energy devices used in surgical procedures, emit high-frequency electromagnetic waves at 2.45 GHz, raising concerns about electromagnetic interference with other equipment. Ensuring electromagnetic compatibility (EMC) is essential for mitigating electromagnetic noise and maintaining a safe surgical environment. The objective of this study is to establish a large-scale high-frequency electromagnetic field analysis method for evaluating EMC during the use of microwave scalpels in operating rooms. We perform analyses of a simplified model using the finite element electromagnetic field analysis software ADVEN-TURE_FullWave and a PC cluster, and conduct studies aimed at accelerating and stabilizing the simulations. Furthermore, to calibrate the parameters set during the analysis, we measure the intensity of the electric field emitted by the microwave scalpel and incorporate this data into the simulations.

ABSTRACT. The medical microwave scalpel is a surgical device that generates minimal smoke and reduces intraoperative blood loss, making it beneficial for both patients and surgeons, and has been widely adopted in laparoscopic procedures in gynecology. However, the dielectric heating induced by microwaves is extremely rapid, raising concerns that thermal damage to adjacent structures such as the ureter, ovaries, and intestines may lead to postoperative complications and a deterioration in patients' postoperative quality of life (QOL). Our research group aims to advance preoperative thermal impact assessment for microwave scalpel use and to establish safer surgical practices. In this study, we conducted heat conduction simulations based on experimentally measured data obtained using a microwave scalpel and chicken tissue, and explored the future direction of our research.

ABSTRACT. We present a simulation model for investigating multiscale plasma physics in the high- energy regime through the formation of artificial-scale magnetospheres. The model combines a particle-in-cell description of collisionless plasma dynamics with phase-space transport of high- energy photons. Applying the model to the interaction between a high-energy particle beam and a magnetized plasma demonstrates the formation of magnetosphere-like structures governed by electron-scale plasma dynamics, enabling investigation of the effects of relativistic processes and high-energy photon reactions on plasma dynamics. Ongoing code development, including the im- plementation of a nested hierarchical mesh, aims to extend the model to larger-scale magnetosphere formation involving ion dynamics.

ABSTRACT. We present an end-to-end differentiable surrogate for fluid–structure interaction (FSI) coupling a Graph Immersed-boundary FSI Transformer Network (GIFTN) with an analytic modal ROM. GIFTN embeds the moving boundary as an adaptive volume-fraction field that gates graph attention weights via a learnable support radius, preserving geometry-awareness throughout the closed-loop rollout. The analytic modal ROM covers a range of Young’s moduli through an exact frequency scaling law, enabling parametric stiffness sweeps without retraining. Unlike conventional ALE/IBM solvers which are accurate but slow and gradient-free, both components remain active at training and inference, eliminating the sim-to-sim gap and delivering a design-loop-ready framework for near-real-time FSI prediction and gradient-based inverse design.

ABSTRACT. The violin is represented by the masterpieces of Antonio Stradivari (1644–1737), and numerous studies have investigated. Reliable finite element reproduction remains difficult in the Bridge-Hill region, the 2-5 kHz range linked to bridge-body transmission, bright-ness, and projection. This study develops a violin FE analysis framework that combines computed tomography based three dimensional orthotropic material mapping, part wise stiffness optimization, and local wood axis assignment. Hounsfield Unit values were mapped to element wise density and stiffness, and the model was evaluated using laser doppler vibrometer operational deflection shape data. These results suggest that CT informed material optimization may provide a practical route for improving FE calculation.

ABSTRACT. This study develops an enhanced configuration-specific physics-informed neural network surrogate for the spatiotemporal evolution of large-deformation fluid-structure interaction in a two-dimensional elastic flap benchmark. Based on an Eulerian-Lagrangian PINN framework, it introduces traction-balance interface coupling and evaluates learnable activation functions to improve predictions near moving boundaries. Results show that the traction-balance coupling reduces the global pressure error and provides a smaller but consistent improvement in the global velocity errors, based on percentage RMSE computed on normalized field variables.

ABSTRACT. In this study, an irregular wave field reproduction simulator based on observed sea state data was developed for evaluating the operation of a small wave observation buoy in waves. Currently operated wave observation systems have limited observation points, making it difficult to understand the marine environment in areas without observation points. To address this issue, a wave observation system using an autonomously mobile wave observation buoy has been proposed. Although a machine-learning-based wave estimation method has been proposed for this system, training data must be obtained through experiments using the actual buoy under various observation points and wave conditions.In the developed simulator, the water surface elevation at arbitrary times and positions is calculated by superimposing multiple linear wave components corresponding to the wave conditions in the observed data. As a result, an irregular wave field was reproduced from the observed data, and the water surface elevation and dynamic pressure components were calculated.

ABSTRACT. This study investigates a crop row-following control method for a compact weeding robot in a high-density cultivation field. To achieve stable row-following navigation, a tracking control system is implemented based on row detection from camera images and state estimation using an extended Kalman filter (EKF). The performance of the control is evaluated through integrated simulation using Gazebo. In particular, the tracking per-formance is compared under conditions without observation dropout and with 25% random observation dropout.

ABSTRACT. The computational core of finite element procedures for multiphysics including an　incompressible viscous fluid is the matrix-vector product in the iterative solution process for the fluid pressure Poisson equation. In this study, for the purpose of achieving its GPGPU-based massive scaling, we propose the matrix storage free (MSF) method with no assembly of the global matrix and compare the MSF method with the sparse matrix–vector multiplication (SpMV) on a GPGPU environment. We evaluate their data transfer amounts for linear triangular or tetrahedral elements and show that the advantage of the MSF method over the SpMV in terms of the data transfer amount depends on the element type. Furthermore, through two-dimensional cavity flow analyses, we demonstrate that the performance difference between the MSF method and the SpMV is determined by the data transfer amount in large-scale analysis while affected by the number of kernel launches in small-scale analysis, and the MSF method is superior to the SpMV.

ABSTRACT. In large scale finite element structural analysis, conventional iterative solvers with simple preconditioning suffer from poor convergence when solving practical nonlinear problems and fail to sufficiently leverage GPU performance due to low arithmetic intensity. To address these challenges, this study integrates three types of Algebraic Multigrid (AMG) preconditioning into the open source structural analysis software FrontISTR: Trilinos/ML, an in house Smoothed Aggregation AMG (SA-AMG) implementation, and NVIDIA AmgX. Standardized under optimized algorithmic configurations using W cycles and Chebyshev smoothers, their performance is evaluated using a complex 3D mold model. The in house SA-AMG solver achieves the best overall efficiency, while ensuring com-plete development autonomy without external lifecycle dependencies. These comparative findings clarify the structural and algorithmic requirements for next generation GPU solvers in FrontISTR.

ABSTRACT. In a previous work, the iterative linear solver of FrontISTR, an open-source large-scale parallel finite element method (FEM) software for nonlinear structural analysis, was ported to GPUs and evaluated on a dedicated tuning branch. This paper reports that the GPU acceleration of the solver, preconditioner, and communication has now been integrated into the main development version and is available to every user, and it examines the performance of the integrated version. Using the roofline model on an NVIDIA A100, we found that the dominant sparse matrix-vector kernel operates well below the roofs and is latency-bound. A data-structure choice made for maintainability raises the register pressure and limits the occupancy, so the integration carries a performance cost for this kernel. Recovering this cost is left as future work.