Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Statistical simulation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Evaluation of data driven low-rank matrix factorization for accelerated solutions of the Vlasov equation

Low-rank methods have shown success in accelerating simulations of a collisionless plasma described by the Vlasov equation, but still rely on computationally costly linear algebra every time step. We propose a data-driven factorization method using artificial neural networks, specifically with convolutional layer architecture, that trains on existing simulation data. At inference time, the model outputs a low-rank decomposition of the distribution field of the charged particles, and we demonstrate that this step is faster than the standard linear algebra technique. Numerical experiments show that the method achieves comparable reconstruction accuracy for interpolation tasks, generalizing to unseen test data in a manner beyond just memorizing training data; patterns in factorization also inherently followed the same numerical trend as those within algebraic methods (e.g., truncated singular-value decomposition). However, when training on the first 70% of a time-series data and testing on the remaining 30%, the method fails to meaningfully extrapolate. Despite this limiting result, the technique may have benefits for simulations in a statistical steady-state or otherwise showing temporal stability. These results suggest that while the model offers a computationally efficient alternative for datasets with temporal stability, its current formulation is best suited for interpolation rather than for predicting future states in time-evolving systems. This study thus lays the groundwork for further refinement of neural network-based approaches to low-rank matrix factorization in high-dimensional plasma simulations.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Unraveling the implications of finite specimen size on the interpretation of dynamic experiments for polycrystalline aluminum through direct numerical simulations

Normal and Pressure-shear plate impact (NPI and PSPI) experiments are popular experimental techniques for studying the mean-field macroscopic behavior of polycrystalline metals under high-rate dynamic loading. However, since both configurations rely upon geometry for subjecting the specimen to high strain rates, these experiments often involve a limited specimen size. Moreover, because of the inherent heterogeneities present within polycrystalline metals, it is difficult to ascertain if the size of the specimen and/or regions where measurements are made are sufficiently large for making representative inferences about the mean-field macroscopic properties from single-point velocity measurements. In the present study, we quantify the expected measurement variability on observable point measurements in NPI and PSPI experiments by carrying out direct numerical simulations (DNS) of statistically representative polycrystalline microstructures subjected to dynamic compression and compression-shear loading. In particular, we consider the role of specific material heterogeneities (e.g. the grain-to-grain difference in size, crystallographic orientation) on dispersion in the normal and transverse particle velocity records and on local fluctuations in key state variables (e.g. velocity, accumulated plastic strain) by incorporating these effects directly into a representative synthetic microstructure geometry and crystalline description of pure polycrystalline aluminum. The form of the present study is a large parametric investigation, consisting of ten ensembles of one hundred simulations. Each of the thousand simulations reflects a randomly realized synthetic microstructure in one of five cases of decreasing average grain size for the two loading configurations. Our analysis of the DNS results demonstrates that for both of these experimental configurations, the grain size directly correlates with the coefficient of variation (CV) in simulated point measurements, showing a convergent decrease in CV to zero (i.e. particle velocity record approaches the mean-field value) with decreasing grain size. Remarkably, the magnitude of variations in the particle velocity record is shown to be largest where the deviatoric stresses are most significant. In the case of NPI, this occurs at the elastic and plastic wavefront, whereas, in the case of PSPI, the magnitude of fluctuations are approximately constant throughout the experimental window time. The reasoning for the scatter in particle velocity due to the heterogeneous microstructure is demonstrated to be dependent on the mechanisms for accommodating deformation and on the interaction of reflection waves generated at sites of heterogeneities occurring at the scale of grains. Lastly, we develop a power-law description for the magnitude of scattering versus characteristic length, which provides a statistical framework for assessing the required number of grains per characteristic specimen dimension for minimizing scatter within these two experimental configurations (NPI, PSPI).

36 MATERIALS SCIENCE↗
DOE OSTI2021

Impact of a New Cloud Microphysics Parameterization on the Simulations of Mesoscale Convective Systems in E3SM

Abstract Mesoscale convective systems (MCSs) are one of the most climatically significant forms of convection because of their large role in water and energy cycles. The mesoscale features associated with MCS are difficult to represent in climate models because the relevant dynamics and physics are absent or poorly represented with coarse model resolution (∼100 km). Using a regionally refined model (RRM) with 0.25° grid spacing embedded in the Energy Exascale Earth System Model (E3SM), we explore the impact of cloud microphysics parameterizations on the simulation of precipitation, particularly MCS precipitation over the contiguous United States. The Predicted Particle Properties (P3) cloud microphysics scheme has been modified and implemented into E3SM to overcome the limitations of the default Morrison and Gettelman (MG2) scheme in which rimed precipitating ice particles (graupel/hail) are absent and frozen particles are artificially partitioned into cloud ice and snow. We show that P3 improves the simulation of precipitation statistics including frequency distribution compared with MG2 with a limited effect on the diurnal cycle. P3 predicts higher hourly rain rates, resulting in 20% more MCSs and a higher total MCS precipitation (4.4%) compared to MG2, agreeing better with observations. The improvements with P3 mainly result from improved representations of ice microphysics, which not only produces higher rain rates through melting but also leads to a stronger large‐scale ascending motion by releasing more latent heating. This study suggests that improving microphysics parameterization is important for simulating MCS precipitation as future climate model resolutions continue to increase.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

Microphysics of shock-grain interaction for inertial confinement fusion ablators in a fluid approach

Ablator materials used for inertial confinement fusion, such as high-density carbon (HDC) and beryllium, have grain structure which may lead to small-scale density nonuniformity and the generation of perturbations when the materials are shocked and compressed. Here, we use a combination of a linear theory of shock interaction with density nonuniformity [Velikovich et al., Phys. Plasmas 14, 072706 (2007)] and numerical simulations to study shock interaction with a model representation of HDC grains. While the shock-grain interaction is nonlinear, the linear theory shows some key features of the shock-grain interaction, which also hold for the (nonlinear) simulations. The postshock perturbations are made up of sonic reflections off of grain boundaries and vorticity deposition along them, with the latter dominating the perturbed energy content. The mean (per mass) postshock perturbed kinetic energy decreases with increasing grain size, but energy will be deposited at increasing spatial scale. From the perspective of the postshock perturbed energy, the detailed linear theory largely supports a proposed method [S. Davidovits et al., Phys. Plasmas 29, 112708 (2022)] for deresolving the grains (in a similar grains model) that treats the grains statistically. Finally, our simulation results highlight the influence of thermal conduction on the perturbation dynamics at grain scales.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Deep learning closure models for large-eddy simulation of flows around bluff bodies

Near-wall flow simulation remains a central challenge in aerodynamics modelling: Reynolds-averaged Navier–Stokes predictions of separated flows are often inaccurate, and large-eddy simulation (LES) can require prohibitively small near-wall mesh sizes. A deep learning (DL) closure model for LES is developed by introducing untrained neural networks into the governing equations and training in situ for incompressible flows around rectangular prisms at moderate Reynolds numbers. The DL-LES models are trained using adjoint partial differential equation (PDE) optimization methods to match, as closely as possible, direct numerical simulation (DNS) data. They are then evaluated out-of-sample – for aspect ratios, Reynolds numbers and bluff-body geometries not included in the training data – and compared with standard LES models. The DL-LES models outperform these models and are able to achieve accurate LES predictions on a relatively coarse mesh (downsampled from the DNS mesh by factors of four or eight in each Cartesian direction). We study the accuracy of the DL-LES model for predicting the drag coefficient, near-wall and far-field mean flow, and resolved Reynolds stress. A crucial challenge is that the LES quantities of interest are the steady-state flow statistics; for example, a time-averaged velocity component $\langle {u}_i\rangle (x) = \lim _{t \rightarrow \infty } ({1}/{t}) \int _0^t u_i(s,x)\, {\rm d}s$ . Calculating the steady-state flow statistics therefore requires simulating the DL-LES equations over a large number of flow times through the domain. It is a non-trivial question whether an unsteady PDE model with a functional form defined by a deep neural network can remain stable and accurate on $t \in [0, \infty )$ , especially when trained over comparatively short time intervals. Our results demonstrate that the DL-LES models are accurate and stable over long time horizons, which enables the estimation of the steady-state mean velocity, fluctuations and drag coefficient of turbulent flows around bluff bodies relevant to aerodynamics applications.

Mechanics↗
DOE OSTI2024

A Gaussian process autoregressive model capturing microstructure evolution paths in a Ni–Mo–Nb alloy

Additive manufacturing is increasingly being employed to produce components of complex geometries in structural alloys because of the expected energy savings associated with the near-net-shape capability and the ability to build in novel internal features that are not possible with many conventional manufacturing approaches. However, because of the extreme thermal conditions encountered, the non-equilibrium microstructures produced during powder bed-based additive manufacturing processes must be subjected to custom post-heat treatment processes to recover the target mechanical properties. Phase-field models and simulation techniques have matured to a state where the microstructure evolution paths, and the morphologies of the resulting precipitate phases can be predicted reasonably accurately, considering alloy-specific thermodynamic and kinetic aspects of the nucleation and growth processes. However, phase-field simulations are computationally intensive, which precludes the ability to apply the simulations directly to the length scale of the entire component. Therefore, it is highly desirable to develop low-computational-cost surrogate models that effectively capture the physics at the microstructural length scale, while facilitating the design of optimized processing conditions resulting in location-specific targeted microstructures at the component scale. The work presented here demonstrates the application of the materials knowledge system framework to develop a surrogate model that effectively captures the microstructural path during annealing of a Ni–Mo–Nb alloy containing different Mo and Nb compositions known to segregate during solidification under additive manufacturing conditions. Specifically, the surrogate model built in this work is based on a Gaussian process autoregressive model informed by statistical representation of simulated microstructures using two-point correlations and dimensionality reduction through principal component analysis. In conclusion, this surrogate model is shown to capture the bifurcation of the microstructural path during precipitation, which yields a microstructure dominated by the $\gamma^{\prime\prime}$ phase at high Nb concentrations and the $\delta$ phase at low Nb concentrations.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Atomistically-informed modeling of point defect clustering and evolution in irradiated ThO 2

A cluster dynamics (CD) model has been developed to investigate the nucleation and growth of point defect clusters, i.e., interstitial prismatic loops and nanoscale and sub-nanoscale voids, in ThO 2 during irradiation by energetic particles. The model considers cluster off-stoichiometry due to the asymmetry of point defect generation on the O and Th sublattices under irradiation, as well as the point defect diffusivities and the defect binding energies to clusters. The energies were established using detailed molecular dynamics simulations considering the statistical variability of cluster configuration. A high-order adaptive time-integration has been used to solve the model. The predicted loop density and their average size is in good agreement with reported experimental observations for proton irradiated ThO 2 at 600°C. In conclusion, the model did not predict void evolution due to the sluggish kinetics of cation vacancies, explaining the absence of voids in proton irradiated ThO 2 (and other oxides) at relatively low temperatures.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Towards the Accurate Thermodynamic Characterization of Enzyme Reaction Mechanisms

We employed QM/MM molecular dynamics (MD) simulations to characterize the rate-limiting step of the glycosylation reaction of pancreatic α-amylase with combined DFT/molecular dynamics methods (PBE/def2-SVP : AMBER). Upon careful choice of four starting active site conformations based on thorough reactivity criteria, Gibbs energy profiles were calculated with umbrella sampling simulations within a statistical convergence of 1 -2 kcal · mol -1 . Nevertheless, Gibbs activation barriers and reaction energies still varied from 11.0 to 16.8 kcal · mol -1 and -6.3 to +3.8 kcal · mol -1 depending on the starting conformations, showing that despite significant state-of-the-art QM/MM MD sampling (0.5 ns/profile) the result still depends on the starting structure. In conclusion, the results supported the one step dissociative mechanism of Asp197 glycosylation preceded by an acid-base reaction by the Glu233, which are qualitatively similar to those from multi-PES QM/MM studies, and thus support the use of the latter to determine enzyme reaction mechanisms.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Turbulent burning velocity of lean premixed hydrogen/air flames at engine conditions: Effects of turbulence intensity and length scale

For turbulent lean premixed hydrogen flames with strong thermodiffusively instabilities, most previous studies have focused on the influence of turbulence intensity, whereas the role of turbulence length scale is less well understood. Here, this study addresses this gap by conducting direct numerical simulations (DNS) of statistically planar turbulent premixed flames for a lean (ϕ=0.35) hydrogen/air mixture under independently varied turbulence intensity (u') and length scale (l T ) at engine-relevant thermodynamics conditions. Results show that as u' increases, the flame front becomes increasingly wrinkled, forming smaller cellular structures. In contrast, l T variations do not significantly alter the size of these structures. For the turbulent burning velocity (S T ), the normalized S T (i.e., S T /S L , where S L is the laminar flame speed) increases linearly with u', driven by both enhanced flame surface wrinkling (i.e., increased A T /A L ) and enhanced local burning rate (i.e., increased I 0 ). However, increasing l T reduces I 0 , despite a continued increase in A T /A L , resulting in only a marginal increase in S T /S L . To reveal the underlying mechanisms, especially the decreasing trend of I 0 with l T , local flame dynamics analyses are performed. It is found that as l T increases, the interaction between thermodiffusive effects and turbulence weakens due to the reduced tangential strain rate, while the flame curvature remains largely unchanged. This suppresses local reactivity enhancement and thus decreases I 0 , In contrast, an increase in u' enhances the interaction by amplifying both curvature fluctuation and tangential strain rate, leading to increased local reactivity (increased I 0 ). Finally, based on the DNS data, several new scaling models are proposed for the three global properties, S T /S L , A T /A L , and I 0 , and showed improvements compared to existing models. These findings provide new insights into the flame-turbulence interactions in thermodiffusively unstable hydrogen flames. The DNS dataset is also useful for the development of turbulent combustion models applicable to practical engine simulations.

Engine-relevant condition↗
DOE OSTI2026

Lattice-scale variations in viscosity are correlated with solution structure at mineral-water interfaces

At solid-liquid interfaces, the viscosity increases markedly from the bulk due to the collective interactions of ions and water molecules, influencing phenomena relevant to nanofluidics, colloidal dynamics, and electrochemistry. Here, in this study, we investigated dissipative forces at the boehmite-water interface using 3D atomic force microscopy. We observed an increase in interfacial solution viscosity, η, by 10-100-fold as the nanoprobe approached the surface in normal direction, with up to four oscillatory features showing average peaks of η/η bulk = 44-71. Moreover, the viscosity showed sub-nanometer variations within 0.5 nm from the interface, templated by the underlying crystal lattice and correlated with the interfacial solution structure. Beyond a near-wall region of approximately 1.2 nm, the dissipative response was comparable to that in bulk solution. Molecular dynamics simulations, along with statistical mechanical analyses, provided details on hydrodynamic structures near the interface. Specifically, the lattice-dependent dissipative responses are correlated with extensive hydrogen bonding by interfacial water molecules, which increased friction, particularly along the [001] direction. These results demonstrate how solution viscosity at mineral-water interfaces is anisotropic and correlated with the local solution structure, providing insights into the dynamics of nanocrystal attachment.

Viscosity↗
DOE OSTI2026

Ripening of Rh Nanoparticle Catalysts in Reverse Water–Gas Shift via a Data-Driven Model Combining Physics, Theory, and Experiment

Degradation via sintering is an ongoing challenge that impedes the broad commercial success of supported metallic nanoparticle catalysts. To mitigate degradation via informed catalyst design and process operations, here we aim to disambiguate the underlying mechanisms of sintering by combining theory and experiment in a quantitative framework. While mechanistic sintering models exist, they only model a single sintering pathway, even though multiple sintering mechanisms can occur simultaneously or dominate at different stages of the process. Data-driven machine learning models have emerged as a means to represent complex processes through data regression. However, machine learning models have very large data needs and lack mechanistic insights due to their black-box encoding. To develop an interpretive model of catalyst degradation via sintering, we constructed a hybrid model combining mechanistic “physics-based” models and data-driven methods to obtain both reliable predictions and mechanistic insights regarding experimentally observed sintering phenomena. Focusing on nanoparticle sintering in the Rh–TiO 2 catalyst for the reverse water–gas shift (RWGS) reaction, the hybrid model couples a mechanistic term for Ostwald ripening with energy values calculated via density functional theory (DFT) with a parametric, data-driven discrepancy function term for unmodeled mechanisms. The hybrid model is trained using Bayesian inference with data collected from small-angle X-ray scattering (SAXS) in situ experiments wherein average nanoparticle diameter versus time was measured at three relevant operating temperatures. The calibrated hybrid model results show that an Ostwald ripening-only model parameterized with fixed DFT energies does not fully capture the time and temperature dependence of the SAXS-observed sintering kinetics, and that an additional functional contribution, or DFT energy calibration, is required to reconcile simulation and experiment. Analysis of the hybrid-model error confirms that the hybrid model outperforms both the purely mechanistic and purely data-driven alternatives in terms of expected predictive accuracy for time-evolving average particle sizes. Furthermore, the results support the hypothesis that the Ostwald ripening mechanism is less important for explaining the sintering phenomena as operating temperature increases under an assumed fixed DFT parameterization. This could be explained in one of two ways: either latent, unmodeled sintering mechanisms dominate at higher temperatures, or the DFT uncertainty increases with temperature. The proposed modeling approach directly links theory to experiments and simulations via a statistical hybrid modeling framework and can be extended to other catalytic systems to improve predictive models and mechanistic understanding.

Bayesian hybrid modeling↗
DOE OSTI2022

A Machine‐Learning‐Assisted Stochastic Cloud Population Model as a Parameterization of Cumulus Convection

Abstract A machine‐learning‐assisted stochastic cloud population model is coupled with the Advanced Research Weather Research and Forecasting (WRF) model to represent fluctuations in the cloud‐base mass flux associated with the life cycles and interactions among cumulus convection cells. In this cloud population model, the size distribution and the associated cloud‐base mass flux of the convective cells are related to their previous state and to the change in the total convective area via a transition function. The convective area tendency in turn is assumed to depend on the cloud‐base mass flux that is resolved by the host WRF model. The transition function is represented by a single hidden‐layer neural network trained by the evolution of convective cell size distributions in a 1‐km grid‐spacing WRF simulation run over the Australian Monsoon region. At every grid point of the host model, the cloud population model predicts the cell size and cloud‐base mass flux distributions from which a random sample of cells is fed to an entraining parcel model that calculates precipitation as well as the associated liquid water potential temperature and total moisture tendencies. These tendencies are averaged over the cells and provided to the host model. Several regional simulations are performed over tropical and midlatitude domains to test this as a potential approach to scale‐aware parameterization. It is shown that such an approach could be a new promising path to simulating realistic precipitation statistics and propagation of precipitation associated with the Madden‐Julian Oscillation while maintaining realistic depictions of the diurnal cycle over both land and ocean.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

The local wavenumber model for computation of turbulent mixing

We present an overview of the current status in the development of a two-point spectral closure model for turbulent flows, known as the local wavenumber (LWN) model. The model is envisioned as a practical option for applications requiring multi-physics simulations in which statistical hydrodynamics quantities such as Reynolds stresses, turbulent kinetic energy, and measures of mixing such as density-correlations and mix-width evolution, need to be captured with relatively high fidelity. In this review, we present the capabilities of the LWN model since it was first formulated in the early 1990s, for computations of increasing levels of complexity ranging from homogeneous isotropic turbulence, inhomogeneous and anisotropic single-fluid turbulence, to two-species mixing driven by buoyancy forces. The review concludes with a discussion of some of the more theoretical considerations that remain in the development of this model.

42 ENGINEERING↗
DOE OSTI2025

Interferometric binary phase estimations

We propose an interferometric scheme where each photon returns one bit of the binary expansion of an unknown phase. It sets up a method for estimating the phase value at arbitrary uncertainty. This strategy is global, since it requires no prior information, and it achieves the Heisenberg bound independently of the output statistics. We provide simulations and a characterization of this architecture.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

The Information Length Concept Applied to Plasma Turbulence

A methodology to study statistical properties of anomalous transport in fusion plasma is investigated. Three time traces generated by the full-f gyrokinetic code GKNET are analyzed for this purpose. The time traces consist of heat flux as a function of the radial position, which is studied in a novel manner using statistical methods. The simulation data exhibit transport processes with both medium and long correlation length along the radius. A typical example of a phenomenon with long correlation length is avalanches. In order to investigate the evolution of the turbulent state, two basic configurations are studied, one flux-driven and one gradient-driven with decaying turbulence. The information length concept in tandem with Boltzmann–Gibbs and Tsallis entropy is used in the investigation. It is found that the dynamical states in both flux-driven and gradient-driven cases are surprisingly similar, but the Tsallis entropy reveals differences between them. This indicates that the types of probability distribution function are nevertheless quite different since the higher moments are significantly different.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2023

Domain Adaptive Graph Neural Networks for Constraining Cosmological Parameters Across Multiple Data Sets

State of the art astronomical simulations have provided datasets which enabled the training of novel deep learning techniques for constraining cosmological parameters. However, differences in subgrid physics implementation and numerical approximations among simulation suites lead to differences in simulated datasets, which pose a hard challenge when trying to generalize across diverse data domains and ultimately when applying models to observational data. Recent work reveals deep learning algorithms are able to extract more information from complex cosmological simulations than summary statistics like power spectra. We introduce Domain Adaptive Graph Neural Networks (DA-GNNs), trained on CAMELS data, inspired by CosmoGraphNet (Villanueva-Domingo et al 2023). By utilizing GNNs, we can capitalize on their capacity to capture both astrophysical and topological features of galaxy distributions. Mixing these capabilities with domain adaptation techniques such as Maximum Mean Discrepancy (MMD), which enable extraction of domain-invariant features, our framework demonstrates enhanced accuracy and robustness. We present experimental results, including the alignment of distributions across domains through data visualization. These findings suggest that DA-GNNs are an efficient way of extracting domain independent cosmological information, a vital step toward robust deep learning for real cosmic survey data.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2024

Data assimilation for combustion ignition delay time simulation using schlieren image velocimetry

This study sought to improve the accuracy of simulating spray penetration and combustion ignition delay by means of data assimilation (DA). The simulations were conducted using the Reynolds-averaged Navier–Stokes (RANS) equations and assimilating the schlieren image data. In DA, an ensemble square root filter (EnSRF) was used to build the statistical model, making the simulation results more accurate without any change in the governing equations. Recognizing that the spray-cone injection angle has a large effect on penetration, we created ensemble members with different injection angles. And we applied the two-component velocity distribution calculated via SIV and updated both velocity and temperature by using a DA statistical model derived from RANS ensemble simulations. The ignition delay time is generally known to vary even under the same experimental conditions because it is influenced by many factors. In this study, we attempted the transient DA-assisted RANS simulation to predict the ignition delay time even when the temporal resolution and accuracy of the observation data ware insufficient. Our trials offer an example of how a combination of techniques can be effectively used to assimilate experimental data obtained under restricted conditions.

Combustion simulation↗
DOE OSTI2024

GrainGNN: A dynamic graph neural network for predicting 3D grain microstructure

We propose GrainGNN, a surrogate model for the evolution of polycrystalline grain structure under rapid solidification conditions in metal additive manufacturing. High fidelity simulations of solidification microstructures are typically performed using multicomponent partial differential equations (PDEs) with moving interfaces. The inherent randomness of the PDE initial conditions (grain seeds) necessitates ensemble simulations to predict microstructure statistics, e.g., grain size, aspect ratio, and crystallographic orientation. Here, currently such ensemble simulations are prohibitively expensive and surrogates are necessary.In GrainGNN, we use a dynamic graph to represent interface motion and topological changes due to grain coarsening. We use a reduced representation of the microstructure using hand-crafted features; we combine pattern finding and altering graph algorithms with two neural networks, a classifier (for topological changes) and a regressor (for interface motion). Both networks have an encoder-decoder architecture; the encoder has a multi-layer transformer long-short-term-memory architecture; the decoder is a single layer perceptron.We evaluate GrainGNN by comparing it to high-fidelity phase field simulations for in-distribution and out-of-distribution grain configurations for solidification under laser power bed fusion conditions. GrainGNN results in 80%–90% pointwise accuracy; and nearly identical distributions of scalar quantities of interest (QoI) between phase field and GrainGNN simulations compared using Kolmogorov-Smirnov test. GrainGNN's inference speedup (PyTorch on single x86 CPU) over a high-fidelity phase field simulation (CUDA on a single NVIDIA A100 GPU) is 150×–2000× for 100-initial grain problem. Further, using GrainGNN, we model the formation of 11,600 grains in 220 seconds on a single CPU core.

36 MATERIALS SCIENCE↗