Ryo Yokoyama
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ryo Yokoyama has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
Koji Okamoto
5 shared publications- Shuichiro Miwa
The University of Tokyo
4 shared publications - Avadhesh Kumar Sharma
Vellore Institute of Technology
3 shared publications Kai Wang
2 shared publications- Ravinder Kumar
The University of Tokyo
2 shared publications - Ruicong Xu
The University of Tokyo
2 shared publications
Latest Publications
- Open Access
- 26 May 2026
- Published by Elsevier in SSRN Electronic Journal
Discretized Momentum Equation-Constrained Physics-Informed Graph Neural Networks for Meshfree Simulation of Highly Viscous Free-Surface Flows
Lagrangian meshfree particle methods are well suited for simulating free-surface flowsinvolving large deformation and complex interface motion. However, their computationalcost becomes significant for highly viscous fluids because explicit time integration suffersfrom diffusion-number restrictions, whereas implicit schemes require repeated solutions oflarge-scale linear systems and pressure equations. To address this issue, this study proposesa discretized momentum-equation-constrained physics-informed Graph Network-basedSimulator (PI-GNS) framework for accelerating meshfree particle simulations.The training data are generated using theMoving Particle Hydrodynamics (MPH) method,and the particle system is represented as a graph, where particles correspond to nodes andneighboring particle interactions correspond to edges. Unlike the original purely data-drivenGNS, PI-GNS incorporates a physics-informed loss based on the discretized momentum equation.Specifically, the network predicts particle acceleration and pressure increments, whilegraph-based differential operators are used to evaluate the pressure-gradient and viscousdiffusionterms on the particle graph. The discrepancy between the GNS-predicted accelerationand the acceleration evaluated from the discretized governing equation is penalizedduring training.The method is applied to fundamental free-surface flow problems, including dam-breakand box-drop simulations, under different fluid viscosities. The results show that PI-GNSimproves the prediction accuracy of particle displacement and pressure compared with theoriginal GNS, regardless of the geometrical complexity and fluid viscosity. The results alsoshow that the pressure field predicted by the PI-GNS is smoother than that obtained fromthe original particle simulation.The proposed PI-GNS also substantially accelerates the simulations, achieving up to a530-fold speed-up over GPU-based computation in the highest-viscosity cases. To assess theextrapolation capability, an untrained flow condition, namely a dam-break problem with a small obstacle and the dam-break problem in a computational domain twice as large as thetraining dataset, is predicted using the trained model. Although small discrepancies remain,PI-GNS demonstrates significantly improved extrapolation capability compared with theoriginal GNS, whose rollout prediction becomes extremely unstable. These results indicatethat the momentum-equation-constrained PI-GNS provides an accurate, stable, and efficientsurrogate framework for highly viscous meshfree particle simulations.
- 26 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessDiscretized Momentum Equation-Constrained Physics-Informed Graph Neural Networks for Meshfree Simulation of Highly Viscous Free-Surface Flows
Lagrangian meshfree particle methods are well suited for simulating free-surface flowsinvolving large deformation and complex interface motion. However, their computationalcost becomes significant for highly viscous fluids because explicit time integration suffersfrom diffusion-number restrictions, whereas implicit schemes require repeated solutions oflarge-scale linear systems and pressure equations. To address this issue, this study proposesa discretized momentum-equation-constrained physics-informed Graph Network-basedSimulator (PI-GNS) framework for accelerating meshfree particle simulations.The training data are generated using theMoving Particle Hydrodynamics (MPH) method,and the particle system is represented as a graph, where particles correspond to nodes andneighboring particle interactions correspond to edges. Unlike the original purely data-drivenGNS, PI-GNS incorporates a physics-informed loss based on the discretized momentum equation.Specifically, the network predicts particle acceleration and pressure increments, whilegraph-based differential operators are used to evaluate the pressure-gradient and viscousdiffusionterms on the particle graph. The discrepancy between the GNS-predicted accelerationand the acceleration evaluated from the discretized governing equation is penalizedduring training.The method is applied to fundamental free-surface flow problems, including dam-breakand box-drop simulations, under different fluid viscosities. The results show that PI-GNSimproves the prediction accuracy of particle displacement and pressure compared with theoriginal GNS, regardless of the geometrical complexity and fluid viscosity. The results alsoshow that the pressure field predicted by the PI-GNS is smoother than that obtained fromthe original particle simulation.The proposed PI-GNS also substantially accelerates the simulations, achieving up to a530-fold speed-up over GPU-based computation in the highest-viscosity cases. To assess theextrapolation capability, an untrained flow condition, namely a dam-break problem with a small obstacle and the dam-break problem in a computational domain twice as large as thetraining dataset, is predicted using the trained model. Although small discrepancies remain,PI-GNS demonstrates significantly improved extrapolation capability compared with theoriginal GNS, whose rollout prediction becomes extremely unstable. These results indicatethat the momentum-equation-constrained PI-GNS provides an accurate, stable, and efficientsurrogate framework for highly viscous meshfree particle simulations.
- Open Access
- 15 April 2026
- Published by Taylor & Francis in Nuclear Technology
- p. 1-25
Study on Aerosol Deposition Characteristics of an Obstacle Structure During Ventilation for Fukushima Daiichi Decommissioning
I. In a severe nuclear accident scenario, reactor core disruption following serious structural damages may lead to fuel debris formation and spreading, causing radioactive material release inside t...
- 15 April 2026
- Published by Taylor & Francis in Nuclear Technology
- p. 1-25
Open AccessStudy on Aerosol Deposition Characteristics of an Obstacle Structure During Ventilation for Fukushima Daiichi Decommissioning
I. In a severe nuclear accident scenario, reactor core disruption following serious structural damages may lead to fuel debris formation and spreading, causing radioactive material release inside t...
- Open Access
- 16 February 2026
- Published by Elsevier in SSRN Electronic Journal
Experimental Study on 3D Large-Scale Debris Bed Formation Using Sub-Millimeter Particles for Core Catcher Design Optimization
In a core disruptive accident of a sodium-cooled fast reactor, molten fuel is rapidly quenched by liquid sodium and fragmented into fine debris particles, which subsequently settle and form a debris bed. The debris bed shape plays a crucial role in determining coolability; however, the physical mechanisms governing debris-bed formation under three-dimensional conditions remain insufficiently understood.In this study, large-scale three-dimensional water-pool experiments were conducted using sub-millimeter to millimeter-sized particles to investigate debris-bed formation during underwater sedimentation. The debris bed shape was characterized using several geometric indicators, including the inner cavity diameter, outer diameter, maximum bed height, and outer slope angle.The results show that debris bed formation can be interpreted as the cumulative outcome of multiple physical processes occurring during particle sedimentation, namely momentum conversion associated with a granular jump near the bottom surface, particle transport toward the outer region driven by fluid motion and dispersion, and particle rearrangement near the bed periphery. By comparing the geometric indicators with relevant dimensionless parameters, a clear transition in the dominant physical behavior was identified around a Stokes number of unity, separating particle–fluid interaction–dominated and particle–particle interaction–dominated regimes.Based on this regime distinction, empirical correlations for the outer slope angle were formulated using physically motivated dimensionless parameters. The proposed correlations reproduce the experimental data with prediction errors comparable to the experimental repeatability, demonstrating that the dominant features of debris-bed formation can be quantitatively captured through dimensionless scaling grounded in experimentally observed physical mechanisms.
- 16 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExperimental Study on 3D Large-Scale Debris Bed Formation Using Sub-Millimeter Particles for Core Catcher Design Optimization
In a core disruptive accident of a sodium-cooled fast reactor, molten fuel is rapidly quenched by liquid sodium and fragmented into fine debris particles, which subsequently settle and form a debris bed. The debris bed shape plays a crucial role in determining coolability; however, the physical mechanisms governing debris-bed formation under three-dimensional conditions remain insufficiently understood.In this study, large-scale three-dimensional water-pool experiments were conducted using sub-millimeter to millimeter-sized particles to investigate debris-bed formation during underwater sedimentation. The debris bed shape was characterized using several geometric indicators, including the inner cavity diameter, outer diameter, maximum bed height, and outer slope angle.The results show that debris bed formation can be interpreted as the cumulative outcome of multiple physical processes occurring during particle sedimentation, namely momentum conversion associated with a granular jump near the bottom surface, particle transport toward the outer region driven by fluid motion and dispersion, and particle rearrangement near the bed periphery. By comparing the geometric indicators with relevant dimensionless parameters, a clear transition in the dominant physical behavior was identified around a Stokes number of unity, separating particle–fluid interaction–dominated and particle–particle interaction–dominated regimes.Based on this regime distinction, empirical correlations for the outer slope angle were formulated using physically motivated dimensionless parameters. The proposed correlations reproduce the experimental data with prediction errors comparable to the experimental repeatability, demonstrating that the dominant features of debris-bed formation can be quantitatively captured through dimensionless scaling grounded in experimentally observed physical mechanisms.
- 23 April 2025
- Published by Taylor & Francis in Journal of Nuclear Science and Technology
- Vol. 62 (12) , 1156-1179
Efficient radioactive aerosol removal with the development of water mist pre-injection methods and advanced spray charging system for safe Fukushima Daiichi decommissioning: insights from recent experimental studies in UTARTS facility
- 23 April 2025
- Published by Taylor & Francis in Journal of Nuclear Science and Technology
- Vol. 62 (12) , 1156-1179
Efficient radioactive aerosol removal with the development of water mist pre-injection methods and advanced spray charging system for safe Fukushima Daiichi decommissioning: insights from recent experimental studies in UTARTS facility