Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “open source model”

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 901 records · Page 50
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Deep Learning for Multigroup Cross-Section Representation in Two-Step Core Calculations

Here we investigate using deep learning, a type of machine-learning algorithm employing multiple layers of artificial neurons, for the mathematical representation of multigroup cross sections for use in the Griffin reactor multiphysics code for two-step deterministic neutronics calculations. A three-dimensional fuel element typical of a high-temperature gas reactor as well as a two-dimensional sodium-cooled fast reactor lattice are modeled using the Serpent Monte Carlo code, and multigroup macroscopic cross sections are generated for various state parameters to produce a training data set and a separate validation data set. A fully connected, feedforward neural network is trained using the open-source PyTorch machine-learning framework, and its accuracy is compared against the standard piecewise linear interpolation model. Additionally, we provide in this work a generic technique for propagating the cross-section model errors up to the k eff using sensitivity coefficients with the first-order uncertainty propagation rule. Quantifying the eigenvalue error due to the cross-section regression errors is especially practical for appropriately selecting the mathematical representation of the cross sections. We demonstrate that the artificial neural network model produces lower errors and therefore enables better accuracy relative to the piecewise linear model when the cross sections exhibit nonlinear dependencies; especially when a coarse grid is employed, where the errors can be halved by the artificial neural network. However, for linearly dependent multigroup cross sections as found for the sodium-cooled fast reactor case, a simpler linear regression outperforms deeper networks.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

MLOps for Beam Controls

Machine learning operations (MLOps) is the standardization and streamlining of the ML development lifecycle to address the challenges associated with large-scale machine learning applications. The full MLOps pipeline consists of open-source tools: DataHub, MinIO and MLflow. It is being used for dataset management and model development to handle changing data dependencies, varying business needs, reproducibility, and diverse teams working with differing tools and skills. To demonstrate the completion of an MLOps pipeline for particle accelerator operations, we are deploying a simple script that computes settings for the Booster’s gradient magnet power supply. Once the demonstration is complete, we will develop and deploy ML-based optimization algorithms to improve Booster’s overall efficiency. This MLOps pipeline opens the gate to systematically develop and deploy ML applications for accelerator controls and diagnostics.

43 PARTICLE ACCELERATORS↗
DOE OSTI2024

MLOps for Beam Controls

Machine learning operations (MLOps) is the standardization and streamlining of the ML development lifecycle to address the challenges associated with large-scale machine learning applications. The full MLOps pipeline consists of open-source tools: DataHub, MinIO and MLflow. It is being used for dataset management and model development to handle changing data dependencies, varying business needs, reproducibility, and diverse teams working with differing tools and skills. To demonstrate the completion of an MLOps pipeline for particle accelerator operations, we are deploying a simple script that computes settings for the Booster’s gradient magnet power supply. Once the demonstration is complete, we will develop and deploy ML-based optimization algorithms to improve Booster’s overall efficiency. This MLOps pipeline opens the gate to systematically develop and deploy ML applications for accelerator controls and diagnostics.

43 PARTICLE ACCELERATORS↗
DOE OSTI2024

MLOps for Beam Controls

Machine learning operations (MLOps) is the standardization and streamlining of the ML development lifecycle to address the challenges associated with large-scale machine learning applications. The full MLOps pipeline consists of open-source tools: DataHub, MinIO and MLflow. It is being used for dataset management and model development to handle changing data dependencies, varying business needs, reproducibility, and diverse teams working with differing tools and skills. To demonstrate the completion of an MLOps pipeline for particle accelerator operations, we are deploying a simple script that computes settings for the Booster’s gradient magnet power supply. Once the demonstration is complete, we will develop and deploy ML-based optimization algorithms to improve Booster’s overall efficiency. This MLOps pipeline opens the gate to systematically develop and deploy ML applications for accelerator controls and diagnostics.

43 PARTICLE ACCELERATORS↗
DOE OSTI2026

Modeling the contributions to acoustic nonlinearity from complex dislocation networks using 3D dislocation dynamics

Nonlinear ultrasonic parameters are highly sensitive to microstructural features that affect macroscale material behavior, providing a nondestructive means to characterize their evolution. Although dislocations are known to be a strong source of acoustic nonlinearity, establishing quantitative links between the acoustic nonlinearity parameter (β), measured via Second Harmonic Generation, and dislocation morphology—such as dislocation length and density—remains an open challenge. This work advances the numerical modeling of dislocation–β relationships using 3D dislocation dynamics (DD) simulations in two approaches: a “static” method computing strain and stress fields from dislocation configurations in the absence of external loading, and a “quasi-static” method to estimate β from the curvature of dislocation lines under applied load. First, the static method is combined with finite element analysis to investigate a recent assertion that heterogeneous initial strain fields can induce higher harmonic generation in a linear elastic medium; the present results do not corroborate this outcome. Then, the quasi-static method is applied to multiple-dislocation scenarios through parametric studies, revealing behaviors not predicted by analytical models, such as the competing interactions of edge and screw dislocations and the significant influence of applied stress on β. Finally, the simulations are used to model SHG experimental results and validate the hypothesis that β can decrease during plastic deformation, despite increasing dislocation density. As the DD code used here is open-source, it provides a practical platform for future investigation into microstructure–β relationships important to the interpretation of SHG results.

Materials science↗
DOE OSTI2021

Progress on Optimizing Wind Farms and Rotor Designs Using Adjoints

Modern wind plants are increasingly tasked with multiple performance objectives. In addition to designing plants that maximize power output and minimize the levelized cost of energy (LCOE), the design and operation of wind plants is increasingly influenced by challenges regarding grid integration of variable generation renewables. This places a growing emphasis on making wind plants more controllable and predictable. WindSE is a Reynolds-averaged Navier-Stokes (RANS) model designed around analytical gradient and adjoint methods, with the ability to capture terrain-induced effects, as shown in Figure 1. The recent addition of an unsteady solver with an actuator line method (ALM) and ongoing work to enable massively parallel optimizations gives it a unique niche to explore coupled plant-level controls and design problems. This code is an open source python package built on the FEniCS framework that utilizes fast, parallel PETSc solvers to model fluid flow throughout wind-farm scale domains. Two recent studies performed using WindSE demonstrate the capability to optimize under a wide variety of flow conditions and objective functions. In the first, we present an optimization focused on modifying the layout of a wind farm with a fixed number of turbines for maximum total power output [1]. This study highlights the ability to quickly perform simulations using the steady Navier-Stokes solver combined with rotors represented as actuator disks while also stressing the importance of capturing terrain-induced effects. Gradient-based optimization using the RANS equations is viable due to the inclusion of efficiently computed adjoint derivatives. We interpret the physical results of the optimal layout and also discuss the computational cost of scaling to larger problems. In the second study, we present the capabilities of the unsteady Navier-Stokes solver, where rotor-blade profiles represented by actuator lines are optimized to enhance wake steering effects and overall power production [2]. We quantify the wind plant performance gains obtained from this type of simultaneous control co-design optimization as compared to optimizing the blade design and yaw independently. Figure 2 shows the differences between a baseline two-turbine system and an optimized system where we fine-tune the blade chord profile. Results and challenges from each study are quickly summarized and used to motivate the current development efforts within WindSE. Current and future work is focused on enabling higher-resolution studies with more degrees of freedom through parallelization of both the simulation and optimization algorithms. We present benchmarking results to show that WindSE performs well in both weak- and strong-scaling tests and further demonstrate that the optimizer obtains the same convergence rates in both shared- and distributed-memory environments. Using larger wind farms, we can study deep-array effects within an optimization context, allowing the use of objective functions that have been previously unstudied. As an example, we present ongoing work on a blockage metric which characterizes the loss of available kinetic energy due to wake effects from multiple upstream turbines.

adjoint optimization↗
NASA NTRSDate not supplied

Advancing Open Source Science Initiatives Through Public-Private Partnerships

Collaboration is fundamental to advancing open science within the science community. With the recent developments in technology and research, the establishment of formal partnerships between the private sector and government agencies are needed to bridge the knowledge gaps and expedite the time to actionable science. NASA’s Interagency Implementation and Advanced Concepts Team (IMPACT) seeks to address this challenge by establishing non-reimbursable Space Act Agreements with industry leaders in cloud computing, artificial intelligence (AI) and machine learning. The purpose of these agreements is to advance open source science initiatives in the areas of data discovery, access and use of high value NASA science data sets on the cloud. As well as, jointly work on common research problems to accelerate the development and adoption of new AI technologies. Current success stories include co-locating NASA datasets from multiple science disciplines on one platform using Amazon Web Services Open Data Registry, developing AI Foundation Models for Science with IBM and co-hosting training workshops and tutorials for the science community aimed at providing hands-on experience with using NASA data and AI models on the cloud. In summary, we will present an overview of our partnerships supporting open source science initiatives, describe current activities and lessons learned that may be useful to others considering similar partnerships with the private sector.

Elizabeth Fancher↗
DOE OSTI2025

Damage modeling of power tower receiver tubes using the SRLIFE tool

Concentrating Solar Power (CSP) molten-salt central receivers are subject to high, transient incident flux during daily operation. The resulting creep-fatigue damage impacts the receiver’s reliability and restricts the permissible incident flux distribution for a given receiver. This paper aims to reduce CSP plants’ levelized cost of electricity by developing a methodology to predict lifetime and identifies the primary damage mechanism (creep vs fatigue) for any given fluid temperature and temperature gradient. Results are presented in the form of a damage map that serves as a valuable operation guide and design tool. Damage maps can be used to reduce maintenance costs by improving reliability and reduce receiver capital costs by better utilizing the receiver area. FEA simulation and damage modeling of tubes subject to asymmetrical flux conditions is performed in the open-source receiver design tool srlife. Parametric studies are performed over a range of inner tube temperatures and thermal gradients for A230, 316H, 740H, A282, A617, and 800H high temperature alloys. Damage maps are presented for each alloy. A parametric, FEA-based methodology is presented for comparison of fatigue-creep ratios and prediction of tube lifetime based on the critical thermal operating conditions. Fatigue is found to be negligible compared to creep for almost every case. Here, this finding suggests that fatigue effects associated with cloud events are insignificant compared to creep at these high temperature operating conditions. Additionally, lifetime predictions identify thermal conditions where small changes in operating conditions can result in large changes in predicted lifetime.

14 SOLAR ENERGY↗
DOE OSTI2023

Gas-Liquid Flow Modeling for Renewable Fuels Production

Aerobic/anaerobic and gas fermentation pathways have emerged as promising new technologies for the generation of renewable fuels/chemicals from biomass derived sugars, and mixtures of greenhouse/energy rich gas streams (CO2/CH4/H2/CO) via microbial action. Example pathways include sugars-to-ethanol conversion, biomethanation (CO2/H2 to CH4), biogas upgrading, CO fermentation and wet-waste conversion. Gas and liquid phase transport, mass-transfer, and mixing physics at large length scales can significantly affect microbial conversion rates, particularly when the microbial reaction requires a narrow set of conditions. These phenomena are difficult to study in small-scale bench-top reactors that are typically well-mixed. Predictive computational fluid dynamics (CFD) based simulations can therefore aid in the scale-up, design and optimization of these reactors. This work presents multiphase Euler-Euler CFD simulations of at-scale (~500 m3) bioreactors. Our mathematical model treats the gas and liquid as interpenetrating phases. This approach reduces the computational complexity of tracking individual gas bubbles that are several orders of magnitude smaller than reactor dimensions. We solve the Reynolds averaged Navier-Stokes (RANS) multiphase equations that account for phase and chemical species transport, interphase mass and momentum transfer and uses a phenomenological model for gas uptake by microbes. We use a customized solver derived from open-source CFD toolbox, OpenFOAM [1], to perform these simulations, which has been validated against small-scale reactors in our previous work [2]. There is currently a knowledge-gap regarding bubble-size distributions when using gas mixtures with vastly different properties, which can have a significant impact overall mass-transfer. For example, hydrogen bubbles are more buoyant compared to other relatively heavier gases (CO2/CH4/CO), resulting in a large distribution of residence times and bubble sizes. This work therefore develops a deeper understanding of bubble dynamics and interphase mass transfer in such heterogenous gas mixtures through well-resolved computational models. We use a population balance model (PBM) for bubble-size-distribution modeling that is validated against small-scale experiments in our solver with an uncertainty quantification study for bubble coalescence and break-up model parameters. Results pertaining to multiple simulations of gas-fermentation reactors are presented where gas mixtures with varying compositions of CO2/CH4/CO/H2 are imposed at the sparger boundaries. The spatio-temporal variations in bubble-size distribution and mass transfer coefficient are analyzed for varying superficial velocities and gas-compositions for varying sizes of bubble-column and airlift reactors. This work will also examine the performance of different reactor designs, viz. bubble column reactor, airlift reactor with an internal draft tube, and a stirred-tank reactor with Rushton impellers. Reactor mass-transfer coefficient, gas hold-up, and dissolved gas distribution are critically analyzed among reactors, and sensitivity studies pertaining to gas flow rates and reactor geometry will be presented. [1] Weller, H., Tabor, G., Jasak, H. and Fureby, C., A tensorial approach to computational continuum mechanics using object-oriented techniques, Computers in physics, 12, 6, 620--631, 1998. [2] Rahimi, M., Sitaraman, H., Humbird, D. and Stickel, J., Computational fluid dynamics study of full-scale aerobic bioreactors: Evaluation of gas-liquid mass transfer, oxygen uptake, and dynamic oxygen distribution, Chemical Engineering Research and Design, 139: 283-295.

BIOMASS FUELS↗
DOE OSTI2026

Phenomena-based graph representations and applications to chemical process simulation

Rapid and robust simulation of chemical processes is critical to conduct process design, optimization, techno-economic analysis, and sustainability analysis. Yet, efficiently solving simulation models remains a challenge due to the highly coupled and nonlinear nature of the underlying algebraic equations that capture the physical phenomena taking place in the process (e.g., material and energy conservation, phase equilibrium, reactions). In this work, we show that graph-theoretic representations of the physical phenomena within unit operations can help navigate and decompose equations to systematically identify alternative approaches for fast and robust numerical solutions. Specifically, we present a graph-theoretic abstraction that captures the connectivity between the model variables/equations and use this abstraction to group variables/equations into fundamental phenomena. We show that phenomena-based decomposition of the underlying equations can help decouple nonlinearities and enforce material/energy conservation at the process level to accelerate convergence. The proposed decomposition approach differs from the more traditional sequential modular simulation approach, in which equations are grouped and decomposed by unit operations. We implemented the phenomena-based decomposition in BioSTEAM—an open-source process simulation platform in Python—and demonstrated that this approach can converge a variety of separation process models. Compared to sequential modular simulation, the phenomena-based approach can converge idealized systems faster, but it can be slower for (or even fail to converge) highly coupled and nonideal process systems.

Convergence↗
NASA NTRS2014

Static Aeroelastic Analysis with an Inviscid Cartesian Method

An embedded-boundary, Cartesian-mesh flow solver is coupled with a three degree-of-freedom structural model to perform static, aeroelastic analysis of complex aircraft geometries. The approach solves a nonlinear, aerostructural system of equations using a loosely-coupled strategy. An open-source, 3-D discrete-geometry engine is utilized to deform a triangulated surface geometry according to the shape predicted by the structural model under the computed aerodynamic loads. The deformation scheme is capable of modeling large deflections and is applicable to the design of modern, very-flexible transport wings. The coupling interface is modular so that aerodynamic or structural analysis methods can be easily swapped or enhanced. After verifying the structural model with comparisons to Euler beam theory, two applications of the analysis method are presented as validation. The first is a relatively stiff, transport wing model which was a subject of a recent workshop on aeroelasticity. The second is a very flexible model recently tested in a low speed wind tunnel. Both cases show that the aeroelastic analysis method produces results in excellent agreement with experimental data.

Aeroelastic↗
DOE OSTI2020

MASTODON: An Open-Source Software for Seismic Analysis and Risk Assessment of Critical Infrastructure

Seismic analysis and risk assessment of safety-critical infrastructure like hospitals, nuclear power plants, dams, and facilities handling radioactive materials involve computationally intensive numerical models and coupled multiphysics scenarios. They are also performed in a strict regulatory environment that requires high software quality assurance standards, and in the case of safety-related nuclear facilities, a conformance to the American Society of Mechanical Engineers Nuclear Quality Assurance (NQA-1) standard. This paper introduces the open-source finite-element software, MASTODON (Multi-hazard Analysis of Stochastic Time-Domain Phenomena), which implements state-of-the-art seismic analysis and risk assessment tools in a quality-controlled environment. MASTODON is built on MOOSE (Multi-physics Object-Oriented Simulation Environment), which is a highly parallelizable, NQA-1 conforming, coupled multiphysics, finite-element framework developed at Idaho National Laboratory. MASTODON is capable of fault rupture and source-to-site wave propagation using the domain reduction method, nonlinear site response, and soil-structure interaction analysis, implicit and explicit time integration, automated stochastic simulations, and seismic probabilistic risk assessment. When coupled with other MOOSE applications, MASTODON can also solve strongly and weakly coupled multiphysics problems. This paper presents a summary of the capabilities of MASTODON and some demonstrative examples.

42 ENGINEERING↗
DOE OSTI2020

Modeling a ring magnet in ALEGRA

We show here that Sandia's ALEGRA software can be used to model a permanent magnet in 2D and 3D, with accuracy matching that of the open-source commercial software FEMM. This is done by conducting simulations and experimental measurements for a commercial-grade N42 neodymium alloy ring magnet with a measured magnetic field strength of approximately 0.4 T in its immediate vicinity. Transient simulations using ALEGRA and static simulations using FEMM are conducted. Comparisons are made between simulations and measurements, and amongst the simulations, for sample locations in the steady-state magnetic field. The comparisons show that all models capture the data to within 7%. The FEMM and ALEGRA results agree to within approximately 2%. The most accurate solutions in ALEGRA are obtained using quadrilateral or hexahedral elements. In the case where iron shielding disks are included in the magnetized space, ALEGRA simulations are considerably more expensive because of the increased magnetic diffusion time, but FEMM and ALEGRA results are still in agreement. The magnetic field data are portable to other software interfaces using the Exodus file format.

36 MATERIALS SCIENCE↗
DOE OSTI2024

PVDeg: Development of a Streamlined Tool for PV Degradation Modeling

The photovoltaic (PV) industry constantly aims for lower costs, higher-efficiency cells, and improved module designs. These trends lead to using new materials, designs, and manufacturing processes, resulting in a continually changing technological landscape. These changes can potentially introduce new, unknown degradation mechanisms and failure modes that are difficult to diagnose, analyze, test, and model. This introduces uncertainty into the expected lifetime of PV modules of 25 to 50 years. research efforts aim to achieve this while keeping performance degradation at a minimum for decades. This puts considerable pressure on improving the accuracy of long-term durability and reliability assessments. There is a need to organize the existing degradation data into an accessible format and to provide industry relevant tools for extrapolation from laboratory to field conditions. Because the core of this type of analysis involves calculations that are complicated but ubiquitous for many degradation processes, an enhanced predictive modeling framework will facilitate the analysis to help researchers keep up with the rapid pace of technological changes. In this work, we present an online tool that can be used to search for and analyze degradation information and extrapolate PV module performance and durability to field exposure. The tool will simplify many of the routine computational operations that are common to many degradation studies. The prediction tool will be built modular and published as open source, enabling users to expand on the existing framework. This repository will contain various degradation models and material parameters suitable for the reliability and durability assessment of materials and components deployed outdoors.

degradation↗
DOE OSTI2024

MIXv2: a long-term mosaic emission inventory for Asia (2010–2017)

The MIXv2 Asian emission inventory is developed under the framework of the Model Inter-Comparison Study for Asia (MICS-Asia) Phase IV and produced from a mosaic of up-to-date regional emission inventories. We estimated the emissions for anthropogenic and biomass burning sources covering 23 countries and regions in East, Southeast and South Asia and aggregated emissions to a uniform spatial and temporal resolution for seven sectors: power, industry, residential, transportation, agriculture, open biomass burning and shipping. Compared to MIXv1, we extended the dataset to 2010–2017, included emissions of open biomass burning and shipping, and provided model-ready emissions of SAPRC99, SAPRC07, and CB05. A series of unit-based point source information was incorporated covering power plants in China and India. A consistent speciation framework for non-methane volatile organic compounds (NMVOCs) was applied to develop emissions by three chemical mechanisms. The total Asian emissions for anthropogenic/open biomass sectors in 2017 are estimated as follows: 41.6/1.1 Tg NO x , 33.2/0.1 Tg SO 2 , 258.2/20.6 Tg CO, 61.8/8.2 Tg NMVOC, 28.3/0.3 Tg NH 3 , 24.0/2.6 Tg PM 10 , 16.7/2.0 Tg PM 2.5 , 2.7/0.1 Tg BC (black carbon), 5.3/0.9 Tg OC (organic carbon), and 18.0/0.4 Pg CO 2 . The contributions of India and Southeast Asia were emerging in Asia during 2010–2017, especially for SO 2 , NH 3 and particulate matter. Gridded emissions at a spatial resolution of 0.1° with monthly variations are now publicly available. This updated long-term emission mosaic inventory is ready to facilitate air quality and climate model simulations, as well as policymaking and associated analyses.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRSDate not supplied

A Partitioned - Task Parallel Implementation of the NASA Multiscale Analysis Tool for High Performance Computing

The NASA Multiscale Analysis Tool (NASMAT) is a platform for multiscale modeling of composites which can perform analysis of materials with any arbitrary number of length scales. The platform supports modularity, scalability, and interoperability using recursive procedures and data structures. A Macro solver driven parallelization scheme often limits the capability of NASMAT to scale as it has access to limited memory and number of cores (often one core/thread) and often forces to implement macro solver specific changes to the platform. In this work, a partitioned task-parallel approach is adopted, where the parallelization strategy adopted for NASMAT is independent of the macro solver and the computational resources are managed independently. The programming architecture takes into account the hierarchy of multiple scales (task-dependence) and the heterogeneous nature (dynamic load balancing) of computation through implementation of a hierarchy-informed task parallel model. The partitioned nature of the framework further extends the “plug and play” capability of NASMAT. preCICE, an open-source library for coupling multiphysics solver in a partitioned manner, is adopted to integrate NASMAT with an external macro solver by implementing a NASMAT adapter for preCICE. Speedup and scalability of the framework is studied for micromechanical models of varying size.

task-parallel↗
NASA NTRS1987

A directional array approach for the measurement of rotor noise source distributions with controlled spatial resolution

A special array system has been designed to examine noise source distributions over a helicopter rotor model. The particular measurement environment is for a rotor operating in the open jet of an anechoic wind tunnel. An out-of-flow directional microphone element array is used with a directivity pattern whose major directional lobe projects on the rotor disk. If significant contributions from extraneous tunnel noise sources in the direction of the side lobes are excluded, the dominant output from the array would be that noise emitted from the projected area on the rotor disk. The design incorporates an array element signal blending features which serves to control the spatial resolution of the size of the directional lobes. (Without blending, the resolution and side lobe size are very strong functions of frequency, which severely limits the array's usefulness).

Brooks, T. F.↗
DOE OSTI2024

HydraGNN_Predictive_GFM_2024 - Ensemble of predictive graph foundation models for ground state atomistic materials modeling

We provide the ensemble of fifteen pre-trained graph foundation models (GFMs) for atomistic materials modeling applications. Each one of the fifteen GFMs has been trained on five open-source datasets that (once aggregated) amount to over 154 million atomistic structures, which cover over two-thirds of the natural elements of the periodic table and that comprises a broad set of organic and inorganic compounds. This vast set of atomistic structures comprises ground state configurations that are dynamically stable (i.e., equilibrated structures with atomic forces approximately close to zero values) as well as dynamically unstable structures (i.e., non-equilibrium structures with non-negligible non-zero values of atomic forces). The ensemble of datasets aggregated does NOT include excited states. The datasets have been curated to remove atomistic structures with spectral norm of the force tensor above 100 eV/angstrom. Moreover, a linear term of the energy was computed for each dataset using a linear regression model that uses the chemical concentration of each natural element as regressor. The linear term predicted by the linear regression model has been subtracted from each original energy value to perform a re-alignment of the energy values across different electronic structures approximation theories performed to generate the diverse multi-source, multi-fidelity datasets. The folder "ADIOS_files" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "ADIOS_files" directory contains 6 sub-directories named as follows: - ANI1x-v3.bp - MPTrj-v3.bp - OC2020-20M-v3.bp - OC2020-v3.bp - OC2022-v3.bp - qm7x-v3.bp Each sub-directory contains the pre-processed datasets converted in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used to the development, training, and performance testing of the ensemble go predictive graph foundation models. Each GFM was developed using HydraGNN (https://github.com/ORNL/HydraGNN) as underlying graph neural network (GNN) architecture. The multi-task learning (MTL) capability of HydraGNN was used to simultaneously train the GFMs on labeled values for direct predictions of energy (a total system property of an atomistic structure that measures the chemical stability) and atomic forces (an atomic level property of an atomistic structure that measures the dynamical stability). The hyper parameters of the GFM have been tuned using scalable hyperparameter optimization (HPO) algorithms implemented in the software DeepHyper (https://github.com/deephyper/deephyper). The pre-training of each HPO trial was performed using distributed data parallelism (DDP) to scale the training across 128 compute nodes of the exascale OLCF supercomputer Frontier. Each HPO trial was trained only for 10 epochs and an early stopping was performed to avoid wasting significant computational resources on GNN architectures that were clearly underperforming. For each HPO trial, the 'omnistat' tool developed by (AMD Research - Advanced Micro Device) was used to measure the total energy consumption in kWh. The ensemble of GFMs was obtained by selecting the fifteen best performing HPO trials. Four models have been selected for their clear advantage in accuracy, and these are the GFMs with IDs 229, 156, 147, 260. Additional eleven models have been selected based on judicious balance between accuracy and energy consumption needed for training, and these are the GFMs with IDs 165, 78, 137, 1, 175, 171, 181, 67, 179, 167, 351. Each selected GFM of the ensemble was continued to cumulate a total of at most 30 epochs. In some cases, the total number of epochs actually performed was les than 30 due to two combined factors: (1) the size of the GFM (i.e., the number of model parameters to train) and (2) the total wall-clock time for which the computational resources could be allocated on OLCF-Frontier. The "Ensemble_of_models" directory contains 15 sub-directories named as follows: - gfm_0.229 - gfm_0.156 - gfm_0.147 - gfm_0.260 - gfm_0.165 - gfm_0.78 - gfm_0.137 - gfm_0.1 - gfm_0.175 - gfm_0.171 - gfm_0.181 - gfm_0.67 - gfm_0.179 - gfm_0.167 - gfm_0.351 Each one of these sub-directories refers to one of the fifteen HPO trials that have been selected to continue the pre-training with at most 30 epochs. With each sub-directory associated with a specific HPO trial, the following files can be found: - config.json: file for argument parsing to develop and train an HydraGNN architecture - gfm_0.ID_epoch_N.pk: file with model parameters for HPO ID trial after N epochs of training The ensemble of fifteen GFM architectures was used for (1) ensemble averaging to stabilize the predictions of energy and atomic forces after pre-training for post-processing analysis and (2) ensemble uncertainty quantification (UQ). The code used to develop, pre-train, and load the pre-trained models for post-processing analysis is available on the ORNL-GitHub at the following link: https://github.com/ORNL/HydraGNN/tree/Predictive_GFM_2024

36 MATERIALS SCIENCE↗