ABSTRACT. Reacting to cybersecurity reports is exhausting. Recently, cybersecurity has become a more prominent issue for open source developers at ORNL. This can lead to hurried responses from development teams due to short deadlines to resolve issues discovered by the tools ORNL employs. The problem is exacerbated by virtual environments (e.g. conda, pipenv, uv), which are not updated as often as they should and frequently have similar dependencies. What if you could proactively discover security issues before cybersecurity reported them to you? This presentation will introduce tools and techniques for early discovery of cybersecurity issues and remediating them.

ABSTRACT. Dynamic power management techniques can dynamically adjust power settings for reducing energy and power consumption. Dynamic Voltage and Frequency Scaling (DVFS) offers a set of frequency-voltage points that affects the number of instructions executed by the computing unit. Power-Capping offers a number of power settings that serve as an upper power consumption limit threshold for the associated device. Utilizing these techniques in the application-level can help identify the most suitable power settings based on application workload types, considering fine-grained execution patterns to achieve energy efficiency. In this talk, we will focus on the way dynamic power management techniques --CPU DVFS and superchip Power-Capping-- impact the energy consumption and runtime performance for web requests and GPU kernels/tasks, respectively. We will also present methodologies for identifying the most suitable configuration settings and their energy impact in every case, that are the first step for developing application-level real-time power tuning strategies.

ABSTRACT. Fine-tuning existing LLMs for specialized tasks has become a very attractive alternative due to its low cost and quick development cycle. With many pre-trained LLMs available, it is an increasingly complex task to choose the correct model as the starting point or base model. In this work we discuss ChatPORT - a specialized fine-tuned LLM geared towards providing correctly translated codes from one programming model to another. We evaluate a number of base models and compare and contrast their features and characteristics that make them a viable starting point. Our objective is, through systematic evaluation, choose a base model that is efficient in terms of tuning time and memory requirements while also producing better transpiled codes. In this paper, we focus on the OpenMP offload porting capabilities of ChatPORT. We build our training data using kernels from the Heterogeneous Computing Benchmarks (HeCBench) [14] and the OpenMP Validation and Verification suite [6] to fine-tune the base models. We then test the model using unseen kernels extracted from the HeCBench benchmark suite. Our results show that: (1) not all open LLMs geared towards HPC are aware of programming models like OpenMP, (2) although all base models benefit from fine-tuning they learn differently and produce different correctness rates, (3) depending on the memory size and compute resource available, different base models can be used for fine-tuning without significantly affecting the quality of transpiled code they can generate, (4) fine-tuning improved the correctness rate of the LLM by an average of 43.2%, and (5)feedback-based training data further increased the correctness rate by an average of 6% over the LLMs tested.

ABSTRACT. PRESTO is an open-source tool developed at ORNL that automates the selection and tuning of differential privacy algorithms for machine learning workflows. Using Bayesian optimization, it recommends privacy configurations based on dataset characteristics and user preferences, balancing data utility and privacy. PRESTO supports integration with libraries like Opacus and includes tools for visualizing algorithm performance and reliability, making privacy-preserving ML more accessible and efficient.

ABSTRACT. This work focuses on visualizing uncertainty of local divergence of 2-dimensional (2D) vector fields. Divergence is one of the fundamental attributes of fluid flows, as it can help the domain scientists to analyze potential positions of sources (positive divergence) and sinks (negative divergence) in the flow. However, uncertainty inherent in vector field data can lead to erroneous divergence computations, adversely impacting downstream analysis. Uncertainty in vector field data can stem from various sources, such as fixed-bit representations, simulation model approximations, and spatiotemporal discretization. To mitigate data misrepresentation and poor decision-making, these uncertainties must be carefully considered in visualization. While Monte Carlo sampling is a classical approach for estimating divergence uncertainty, it suffers from slow convergence and limited scalability with increasing data size and sample counts. To tackle the aforementioned challenges regarding slow convergence and limited scalability, we present a two-fold contribution. (1) We derive a closed-form analytical expression for quantifying uncertainty in local divergence, assuming independently Gaussian-distributed vector uncertainties. This formulation enables efficient and accurate probabilistic visualization through contour-based techniques. (2) We integrate this approach into Viskores, a platform-portable parallel library, to accelerate the visualization of divergence uncertainty. In our results, we demonstrate enhanced accuracy of our analytical probabilistic visualization of divergence over the classical Monte Carlo visualization. Furthermore, we demonstrate significantly enhanced efficiency of our serial analytical (speedup up to 1125x) and parallel analytical Viskores (speedup up to 105056x) algorithms over the classical serial Monte Carlo approach. Finally, we demonstrate qualitative improvements of our probabilistic divergence visualizations over traditional deterministic mean-field approach, which disregards uncertainty due to the limitations of Monte Carlo sampling. We validate the accuracy and scalability of our methods on wind forecast and ocean simulation datasets.

ABSTRACT. The purpose of this poster is to provide an overview of various components of the software quality assurance (SQA) program at ORNL. This includes level of rigor applied to software applications as part of the QA process, the categories of software, and the types of software. Additionally, the 10 SQA work activities are discussed, along with factoids about the SQA program's co-led structure and the benefits this provides.

ABSTRACT. DIALED is and INTERSECT Initiative project to develop, demonstrate, and deploy an active learning microservice to the INTERSECT ecosystem to steer scientific workflows and that provides access to both new and off-the-shelf active learning algorithms. Active learning is an area of artificial intelligence and machine learning where the algorithm chooses which additional training data should be added to reach an objective. We are particularly interested in active learning methods for experimental design – methods that suggest the next experiments (or simulations) to perform given existing data. Our project is structured as a three-part co-design effort to: (i) create an active learning INTERSECT microservice that uses existing active learning methods and packages, (ii) develop new active learning algorithms with increased trustworthiness and implement them in the microservice, and (iii) support the use of the active learning microservice by partner application teams. The trustworthy active learning capability that this project delivers to the INTERSECT ecosystem will drive the next generation of autonomous experiments across ORNL, further cementing the laboratory’s role as a leader in science and technology research across a broad range of disciplines. The active learning microservice is general and flexible and as part of the INTERSECT ecosystem can be used to direct autonomous scientific workflows anywhere from ORNL’s major user facilities to smaller laboratories around ORNL.

ABSTRACT. The ORNL AI Forecasting Tool represents a novel approach to safety enhancement at Department of Energy (DOE) facilities through predictive artificial intelligence and automated safety documentation. This comprehensive framework addresses critical safety challenges from multiple angles: analyzing historical incident patterns to predict and prevent future risks, and automating the creation of Activity-Based Work Control (ABWC) documents to eliminate human oversight errors.

The system implements two primary use cases under a unified AI architecture. Use Case 1 focuses on event forecasting and risk analysis through a systematic workflow that analyzes current work plans, retrieves relevant historical events from DOE’s extensive dataset, extracts hazards and lessons learned, identifies missing safety gaps, and performs comprehensive risk analysis. This module integrates four key components: Work Planner (WP) for input processing, Information Retrieval (IR) for historical event analysis, Risk Analysis (RA) utilizing Large Language Models, and Safety Standards Integration (SI) connecting with ORNL's Standards Based Management System (SBMS). The integration with SBMS ensures cross-referencing with official policies, regulatory compliance, and adherence to institutional best practices.

Use Case 2 addresses the automation of Activity-Based Work Control document creation for Chemical Sciences operations. The current manual ABWC process is time-consuming, resource-intensive, and prone to human error, potentially leading to overlooked critical hazards.

ABSTRACT. The Spallation Neutron Source (SNS) at Oak Ridge National Laboratory (ORNL) provides pulsed neutrons with energies varying from epithermal to cold. In preparation for VENUS, the neutron imaging beamline to be located at beam port 10, we have performed a series of experiments focused on wavelength-dependent radiography and computed tomography for a broad range of applications, from materials science to biological tissues. One of the time-of-flight (TOF) techniques that is of interest to the scientific community is the 2-dimensional mapping of phases and average crystalline plane orientation in samples both ex-situ and during applied stresses such as tensile loading and heating. This technique is known as Bragg edge imaging and relies on the identification of changes of transmission values, fitting of the edge to measure its displacement, and thus identify the shift in lattice parameter due to stresses. One of the challenges of TOF imaging measurements is the amount of data and the inability to observe Bragg edge shifts in real time during an experiment. Thus, we have been focusing on creating a Python-based interface that allows fast data processing and instantaneous mapping and fitting of the Bragg edges, and their evolution through time. This interface permits visualization, quick Bragg edge selection to fit as well as region of interest, and ultimately back-projects the fitting results onto the radiographs to display a strain map. Assuming data collection has sufficient statistics, the strain mapping analysis can be performed on a pixel-by-pixel basis. The user interface creates a config file that can be used outside via either a command line or a notebook to quickly repeat the same analysis on many other samples or projections sharing the same configuration (peak position, material of interest). This set of tools enables fast strain state extraction for each projection, first step needed to obtain the full 3D strain (tensorial) field of the sample. The poster will present the workflow as well as all the novelties implemented in this new version.

ABSTRACT. The process to go from projections to CT reconstructed data sets requires many steps. The detector used add another complexity to that workflow. We Implemented those steps using a set of jupyter notebooks that allow to use or not optional steps and visualize along the way the results of the various algorithms. The last step creates a script that will run the reconstruction as batch jobs

ABSTRACT. This project creates a resilient INTERSECT ecosystem architecture using resilience design patterns, a resilient system of systems (SoS) architecture, and a resilient microservices architecture. Its proof-of concept prototype implements a resilient federated ecosystem using the INTERSECT software development kit (SDK) and demonstrates resilience capabilities for the INTERSECT AAM cross-facility experiment between the Manufacturing Demonstration Facility (MDF) and the Oak Ridge Leadership Computing Facility (OLCF) Advanced Computing Ecosystem (ACE) testbed. The outcome of this project facilitates the proper development and deployment of resilience for federated ecosystems. It creates, implements and demonstrates a consistent design methodology that allows scientists to pick and choose the right solution for the resilience problem at hand and deploy it with ease. It enables a resilient federated ecosystem that facilitates the US Department of Energy’s (DOE’s) Integrated Research Infrastructure (IRI) vision.

ABSTRACT. The recent work for commissioning autonomous neutron powder diffraction experiments performed at the NOMAD beamline of the Spallation Neutron Source (SNS) at Oak Ridge National Laboratory will be presented. Maximizing scientific productivity through automation and steering is an important goal, according to the DOE BES report from 2019 titled “Roundtable on Producing and Managing Large Scientific Data with Artificial Intelligence and Machine Learning”.

Presentation of these recent experiments mark progress for ORNL’s neutron powder diffraction beamlines to prepare for more automation and steering in neutron scattering. Iron(III) oxide, specifically hematite (alpha-Fe2O3) was used to help commission autonomous navigation of measurement parameter space in a spin-flop magnetic phase transition, also known as the Morin temperature, via machine learning to commission experiment steering capabilities. This is based on previous work on WAND^2 beamline at the High Flux Isotope Reactor (HFIR) at ORNL led by a team from the National Institute of Standards and Technology (NIST) performing similar experimental steering by measuring MnO. These experiments helped bring together multiple software infrastructures including External Instrument Control, Interconnected Science Ecosystem (INTERSECT), Distributed INTERSECT Active Learning for Experiment Design (DIALED), and National Science Data Fabric. Together, this infrastructure enabled analysis to run on compute resources from outside the beamline network and stream data outside ORNL to the National Science Data Fabric.

This accomplishment will drive progress towards both large-scale compute resources being used to guide experiments at SNS and the HFIR as well as promote and mature this experimental steering capability for the General User Program. The future scientific impact from this study will be significant reduction in experimental time required for neutron diffraction experiments and better exploration of parameter space with the constraint of finite beamtime for Users.

ABSTRACT. The software training workshops offered by the ORNL Carpentries team will be presented.

The goal of these workshops is to teach our workforce how to use computers more effectively to make their research easier. We avoid a lot of the theory that is taught in introductory computer science classes in favor of covering more of the practical side of programming and data analysis that is necessary for conducting research.

All of our workshops actively integrate with The Carpentries.

Carpentry-style workshops rely on live coding together with the instructor. This helps us ensure that everyone in the workshop is able to follow along with, and get the most out of, the material.

The ORNL Carpentries team is a collaborative group of data and software experts from across the lab, including the Research Library, CCSD, ITSD, and ESTD. Together, we provide hands-on training to help researchers build foundational software skills for their work.

ABSTRACT. DataFed is a federated scientific data management system, supporting cross-facility research activities including experimentation, simulation, and/or analytics. DataFed provides the software service infrastructure needed to build a loosely-couple data network between geographically distributed and heterogeneous facilities. Within this DataFed network, access to data is simple and uniform regardless of physical storage location, local environment type, or security policies.

ABSTRACT. The Kokkos C++ Performance Portability Ecosystem is a versatile solution for writing modern C++ applications that run efficiently across diverse hardware architectures. Originally developed by the U.S. Department of Energy National Labs as part of the Exascale Computing Project, Kokkos is now a Linux Foundation project under the High Performance Software Foundation (HPSF), supported by organizations like Sandia National Laboratories, Oak Ridge National Laboratory, and the French Alternative Energies and Atomic Energy Commission.

The ecosystem includes three main components: the Kokkos Core Programming Model, the Kokkos Kernels Math Libraries, and the Kokkos Profiling and Debugging Tools. Kokkos Core allows developers to write portable, hardware-agnostic parallel algorithms using abstract patterns, policies, and execution/memory spaces tailored for many-core processors. It ensures high performance by mapping these abstractions onto architecture-specific optimizations. The Kokkos Kernels library provides foundational algorithms for linear algebra and graphs, delivering robust portability and performance through architecture-specific adaptations. Kokkos Tools offers profiling and debugging capabilities, enabling developers to analyze and optimize their applications’ behavior within the Kokkos framework.

Beyond its core components, Kokkos continues to expand its offerings, including supporting Fast Fourier Transforms (Kokkos-FFT) and inter-node communication (Kokkos-Comm). The team also contributes to modernizing ISO C++ standards, such as std::mdspan and std::linalg, and ensures compatibility with various toolchains.

To support new users, Kokkos offers tutorials with step-by-step lectures and exercises that teach developers how to write performance-portable applications. Overall, Kokkos is a comprehensive and evolving solution for ensuring HPC applications achieve peak performance across all modern supercomputing architectures.

ABSTRACT. A user-friendly software for scientific experimental planning

ABSTRACT. GPUs power most of the world’s fastest supercomputers and are central to modern AI breakthroughs. However, developing GPU-accelerated code remains challenging for many domain scientists. We evaluate Mojo, a new Python-like programming language, designed to bridge the gap between Python’s high productivity and the performance demands of modern computing. Mojo builds on MLIR, a modern compiler infrastructure adopted by all major vendors, enabling code portability between NVIDIA and AMD GPUs. Our work benchmarks Mojo against vendor-specific C++ implementations (CUDA and HIP) on four representative scientific workloads: a 7-point stencil, BabelStream, Hartree-Fock, and miniBUDE. These span memory-bound, compute-bound, and atomics-heavy kernels. We conducted tests on NVIDIA H100 and AMD MI300A GPUs. Our results show that Mojo delivers GPU-portable performance comparable to vendor-specific C++ code. Although Mojo is still evolving and requires familiarity with some low-level concepts, it shows strong potential for portable high-performance computing.

ABSTRACT. The INTERSECT framework for self driving labs has been applied to a variety of inter-related use cases in collaboration with the Biological and Environmental Systems project. These include AI driven modeling with the CENTURY model, data reduction through the ORNL-VOC open source library, and the data streaming from the APPL facility to provide real time updates to the SORTED dashboard.

ABSTRACT. This poster will briefly cover the software development efforts of SNAPRed and where we will be leaving it off. SNAPRed is a Data Reduction Software used by the SNAP Instrument Team to process data produced by their neutron instrument at the SNS. The SNAP instrument is a highly configurable diffractometer, and as such proposes several data management challenges. It produces large datafiles, requires strict data mapping of its many state dependent parameters, and operates in a highly specialized domain. This poster will outline the variety of software systems implemented as well as the practices and approaches that worked best when developing a project like this.

ABSTRACT. DOE leadership computing facilities like Frontier, El Capitan,​ Aurora, Perlmutter, and the forthcoming Doudna are GPU-accelerated​. Therefore performance-portable abstraction layers are a priority for the US DOE. Julia is a just-in-time (JIT) compiled language built on LLVM, providing C-like performance with a user-friendly syntax optimized for science. We present JACC.jl, a performance portability layer built for Julia, allowing users to program parallel constructs and automatically take advantage of vendor-specific backends (Threads, CUDA, AMDGPU, oneAPI) without depending directly on one of them. This poster shows some simple code examples and the low performance overhead of using JACC compared to vendor packages. It also showcases two features: 1) JACC.shared, which makes using on-chip GPU shared memory easier, and 2) JACC.Multi, which enables distributed kernels across multiple GPUs on a node (without MPI).

ABSTRACT. The INTERconnect Science ECosysTem (INTERSECT) MULTIlaB, MultiUser CampaignS (MULTIBUS) project aims to run campaigns that span facilities across multiple DOE laboratories and handle security requirements and different protocols between facilities. INTERSECT MULTIBUS will use the current Autonomous Electron Microscopy and Experimental Steering for Powder Diffraction INTERSECT projects for use cases, along with others that benefit from this project. Both will benefit from using leadership computing facilities like ORNL Leadership Compute Facility (OLCF) and National Energy Research Scientific Computing Center (NERSC) and, together, use three different experimental DOE User Facilities: Center for Nanophase Materials Science (CNMS), Spallation Neutron Source (SNS), and the High Flux Isotope Reactor (HFIR). INTERSECT MULTIBUS will develop new INTERSCT capabilities such as handling of authentication and authorization required to span multilab and multifacility resources and also extend the domain science capabilities as needed to utilize the leadership compute facilities. Finally, where no existing security infrastructure exists already such as individual lab instruments, additional stand-alone security implementations will be added to the existing INTERSECT SDK to help provide end-to-end security for these multilab, multifacility INTERSECT campaigns.

ABSTRACT. The implementation of the new BIOSANS detector in the drtsans data reduction toolkit for Small Angle Neutron Scattering (SANS) at Oak Ridge National Laboratory (ORNL) represents a significant advancement in Q-resolution for data collected at the BIOSANS beamline. This "midrange" detector is designed to generate intensity profiles that overlap with those of the "main" and "wing" detectors. The Data Reduction Toolkit for SANS (drtsans) has been extended to include calibration and reduction for the "midrange" detector in conjunction with the "main" and "wing" detectors. Bar-scan and tube-width calibration determine the effective position, dimension, and width of each detector pixel, ensuring precise measurements by correcting for spatial distortions and occlusions. In the data reduction workflow, the toolkit converts Time-of-Flight (TOF) data collected at the "midrange" detector to momentum transfer (Q) space, applying corrections for transmission, sample thickness, and background noise. These advancements in the BIOSANS detector and the drtsans toolkit significantly enhance the capabilities of SANS experiments, providing researchers with more accurate reduced data.

ABSTRACT. PLEIADES (Python Libraries Extensions for Isotopic Analysis via Detailed Examinations of SAMMY results) democratizes R-Matrix analysis by transforming the complex, expert-only domain of neutron resonance data evaluation into an accessible, user-friendly framework. This cross-facility collaboration between Oak Ridge National Laboratory, Los Alamos National Laboratory, and international partners addresses a critical gap: while neutron resonance imaging (NRI) emerges as an essential technique for materials characterization, the R-Matrix analysis required for quantitative results traditionally demands specialized nuclear physics expertise and months of manual effort.

PLEIADES provides intuitive Python APIs that seamlessly interface with SAMMY, the comprehensive R-Matrix solver, enabling researchers from diverse backgrounds—material scientists, engineers, biologists—to perform sophisticated nuclear data analysis without deep knowledge of R-Matrix theory. The framework introduces three complexity levels (Easy/Expert/Custom), allowing users to simply select isotopes while PLEIADES automates ENDF data retrieval, parameter file generation, and complex R-Matrix calculations. Modern data validation through Pydantic models ensures physics-aware constraints, eliminating the silent failures and character-counting errors common in traditional SAMMY workflows.

By abstracting the computational complexities of R-Matrix formalism while preserving SAMMY's full analytical power, PLEIADES reduces analysis time from months to days. The software has successfully analyzed irradiated nuclear fuel samples at LANSCE and enabled isotope mapping at SNS VENUS, demonstrating its capability to handle complex materials and high-radiation environments. The upcoming v2.0 release features enhanced automation, expanded facility integration, and web API connectivity.

This open-source initiative exemplifies ORNL's commitment to making advanced nuclear data analysis tools accessible to the broader scientific community. By democratizing R-Matrix analysis, PLEIADES empowers non-experts to leverage sophisticated neutron resonance techniques, accelerating scientific discovery across materials science, nuclear engineering, and beyond while ensuring reproducible, trusted results through community-driven development.

ABSTRACT. ChatHPC democratizes large language models for the high-performance computing (HPC) community by providing the infrastructure, ecosystem, and knowledge needed to apply modern generative AI technologies to rapidly create specific capabilities for critical HPC components while using relatively modest compu- tational resources. Our divide-and-conquer approach focuses on creating a collection of reliable, highly specialized, and optimized AI assistants for HPC based on the cost-effective and fast Code Llama fine-tuning processes and expert supervision. We target major components of the HPC software stack, including programming models, runtimes, I/O, tooling, and math libraries. Thanks to AI, ChatHPC provides a more productive HPC ecosystem by boosting important tasks related to portability, parallelization, optimization, scalability, and instrumentation, among others. With relatively small datasets (on the order of KB), the AI assistants, which are created in a few minutes by using one node with two NVIDIA H100 GPUs and the ChatHPC library, can create new capabilities with Meta’s 7-billion parameter Code Llama base model to produce high-quality software with a level of trustworthiness of up to 90% higher than the 1.8-trillion parameter OpenAI ChatGPT-4o model for critical programming tasks in the HPC software stack.

ABSTRACT. RESolution is an enterprise-level application developed at ORNL to unify and simplify access to essential business functions that support research and development. Initiated in response to a 2012 whitepaper, the platform was designed to address the inefficiencies of navigating multiple legacy systems by offering a single, intuitive interface tailored to the needs of researchers and operations staff. Today, it supports thousands of users each month across the lab.

The primary goal of RESolution is to embed critical workflows—such as procurement, project tracking, publication processing, and proposal development—into a cohesive user experience. Built on a modern technology stack (Microsoft SQL Server, .NET, Vue.js), the platform uses a modular architecture that allows for scalable development and seamless integration with existing systems.

Each module—such as Action List, Projects, Publications, and Proposals—was developed to solve specific workflow challenges while maintaining a consistent and user-friendly interface. The development process emphasizes usability and quality through dedicated UX design, automated testing, and continuous integration practices.

RESolution’s key innovations include its modular design, unified interface, real-time workflow integration, and adaptability to evolving lab needs. Its user-centered approach ensures accessibility for both technical and non-technical users. The platform has also gained recognition beyond ORNL, with modules adopted by other DOE National Laboratories and national accolades including an Excellence.Gov Award.

The impact of RESolution is evident in improved operational efficiency, reduced administrative burden, and enhanced user satisfaction. Its development has been shaped by close collaboration with the research community, ensuring that the platform evolves in alignment with user needs and scientific priorities. This partnership-driven approach has helped RESolution remain responsive, relevant, and deeply integrated into the lab’s research culture.

Looking forward, RESolution will expand its modular capabilities, enabling faster updates and more agile development. Continued investment will ensure it remains a strategic asset, supporting ORNL’s mission while maximizing the return on past development efforts.

ABSTRACT. This poster showcases a scientific dataset: road elevation model (REM) for the State of Texas. We created this REM dataset to enable state-wide high-resolution road inundation mapping and forecast. We addressed computational challenges in processing massive lidar point cloud data with a parallel computing workflow on GPU. 50TB Texas lidar data provided by TxGIO, spanning from year 2006 to 2024, was processed on the ORNL Research Cloud to produce 3D road shape, road surface, and road lidar output on 758,276 kilometers of Texas roads. The 340GB output REM dataset was published as an open public dataset on June 19, 2025 for building the Texas road flood forecast system on all 25 TxDOT districts. We have registered this dataset using OLCF's DOI service (DOI: 10.13139/ORNLNCCS/2574440). The dataset can be downloaded from https://web.corral.tacc.utexas.edu/nfiedata/road3d/ .

The production of this dataset is in part supported by a Strategic Partnership Project with UT Austin and an NSF ACCESS computing allocation award that enables fast massive data movement between TACC Corral and ORNL CADES/OLCF using Globus.

ABSTRACT. The ASGarD project has the goal of building a solver specifically targeting high-dimensional PDEs where the "curse-of-dimensionality" has previously precluded useful continuum / Eularian (grid or mesh based as opposed to Monte-Carlo sampling) simulation. Our approach is based on a Discontinuous-Galerkin finite-element solver build atop an adaptive hierarchical sparse-grid (note this is different from the "combination technique" when applied to sparse-grids).

ABSTRACT. Advances in health informatics increasingly rely on multimodal data to improve prediction in health services research, yet limited guidance exists on systematically integrating diverse data sources. We present a multimodal prediction framework that processes synthetic clinical, imaging, text, wearable, and environmental features through dedicated sub-networks, including dense layers for numerical inputs and an LSTM for text embeddings. Encoded representations were fused for binary prediction of state-level variation in veteran facility admissions (NSDUH 2018–2019, question B8). Trained on 5,000 samples with 26 features, the model achieved 75% accuracy and an ROC AUC of 0.87, with precision of 0.72 and recall of 0.80 for the positive class. These findings highlight both the promise and challenges of leveraging multimodal data for community-level prediction. This scalable framework demonstrates how synthetic multimodal datasets can support predictive modeling in veteran and broader health service analysis, informing decision-making and guiding future methodological refinement.

ABSTRACT. Accurately estimating nonhighway fuel consumption is crucial for the fair distribution of the Highway Trust Fund, which allocates billions of dollars annually to state highway programs. Since states do not uniformly report nonhighway fuel usage, the Federal Highway Administration (FHWA) needed to rely on estimation models to fill data gaps and ensure consistent fuel tax attributions. However, the current nonhighway fuel consumption models, last updated in 2013-2014, have become outdated due to reliance on older datasets and model parameters. To this end, we developed an Excel-based tool for the FHWA to estimate annual gasoline and diesel consumption at the state-level in US nonhighway sectors, including aviation, agriculture, industrial, commercial, construction, and boating. This tool incorporates updated methodologies and integrates recent datasets such as the 2021 Vehicle Inventory and Use Survey, the 2022 Census of Agriculture, and the latest industry and state-level studies. The tool is designed for ease of use and reproducibility, featuring built-in data validation, transparent parameter tables, automated quality checks, and export-ready outputs. Users can update a few input data to get fuel consumption estimates by sectors automatically every year. This tool makes it easier to estimate fuel use in a reliable and repeatable way, helping ensure that Highway Trust Fund dollars are fairly allocated across states and sectors.

ABSTRACT. By applying data governance standards and curatorial best practices, ORNL is providing a federated data solution that increases the discoverability of energy related datasets while limiting risk to the energy grid itself. This poster will highlight our use of the National Institute of Standards and Technology’s (NIST) Research Data Framework (RDaF) to help establish the overarching policies for the project along with the roles and responsibilities of our data stewards; our abstraction of the Data Curation Network's CURATE(d) steps through Jupyter Notebooks to rigorously surface anomalies and identify potentially problematic datasets; and our application of Role-Based Access Controls (RBACs) to ensure that the right user communities have access to the right datasets. Over the past year we have found that data governance, management and curation are complementary activities that when deployed at the beginning of a project help ensure long-term success.

ABSTRACT. Verification, validation, and uncertainty quantification (VVUQ) is the practice of making sure scientific computing results are credible for decision makers. However, VVUQ analysis is often time-consuming and difficult to perform. This is especially true for computational fluid dynamics (CFD), where practitioners often do not have the time to develop the skillsets needed for effective VVUQ on top of the demands of CFD. However, this has slowed the adoption of CFD in many key scientific and engineering disciplines because not all CFD results can be deemed credible. A key inhibitor has been the lack of software tools to aid CFD practitioners with VVUQ analysis. CFDverify is an open-source Python library designed to aid analysts in conducting solution verification of their results. This poster presents the statement of need for CFDverify, how it was designed, how it fits into the greater VVUQ analysis, and key results from using it.

ABSTRACT. The need to develop affordable and clean energy is of paramount importance as US energy demand grows due to shifts in manufacturing, population, consumer behavior, technology development, and other drivers. Yet, the task of energy planners and decision-makers has become increasingly more challenging in a world of rapid change and considering the multitude of stakeholder inputs that are crucial to solutions being robust, resilient, durable, and lasting. All energy infrastructure scenarios come with their own unique trade-offs: both benefits and risks/costs across multiple dimensions (social, economic, and environmental). ORNL’s ARMADA (Action Relevant Modeling and Decision Analysis) effort aims to quantify those trade-offs through the development of a one-of-a-kind, high-resolution, integrated modeling framework that links a heterogeneous suite of research capabilities. We present initial findings and insights focusing on demonstration scenarios for deploying nuclear SMR (small modular reactor) and solar PV (photovoltaic; utility-scale and rooftop) infrastructure to 2038 in the service area of the Tennessee Valley Authority (all of Tennessee plus portions of surrounding states). The modeling exercise is conducted at a minimum granularity of 4x4 kilometers and maximum of 3x3 meters. Our results evidence varied implications for energy affordability, security, and resilience, as assessed across a variety of quantitative metrics related to land use change, energy and consumer product costs/prices, job growth potential, flood risk, and environmental impacts, among others. In short, we aim to illuminate potential trade-offs across different engineered, environmental, and social systems, both geospatially and temporally.

ABSTRACT. Developing interactive geospatial dashboards for complex environmental data is often hindered by visualization challenges, implementation complexity, and limited automation. We present a novel generative AI framework leveraging Large Language Models to automate geospatial dashboard creation from user inputs, such as UI wireframes, requirements, and data sources. Central to our approach is a structured knowledge graph that embeds domain expertise. Equally important is the use of a Context-Aware Visual Prompting technique that extracts spatial and interface semantics to guide precise code generation. The framework features an interactive validation loop driven by an AI agent, which iteratively evaluates and refines generated dashboards using semantic metrics and agent-based reasoning to assure accuracy and reliability. With finely tuned dashboard code snippets and pairing with visualization knowledge bases, the system produces scalable, multi-page React-based dashboards ready for deployment. This new approach enhances accuracy, reproducibility, and deployment readiness for environmental risk analysis tools. We demonstrate improved performance over baseline methods and discuss applications in supporting timely and informed risk analysis of big environmental data and decision-making.* Corresponding author email: yuxiaoying@ornl.govResearch efforts were supported by the Nuclear Safety Research and Development (NSR&D) program sponsored by National Nuclear Security Administration (NNSA) Office of Environment, Health, Safety and Security (EHSS) of the U.S. Department of Energy (DOE). Oak Ridge National Laboratory is managed by UT-Battelle, LLC, under contract DE-AC05-00OR22725 for the US DOE.

ABSTRACT. Monitoring states in harsh environments through sensors presents significant challenges. Not only is it difficult for sensors to endure extreme conditions, but it is also often impractical, both physically and economically, to deploy sensors at every location of interest. These challenges raise crucial questions regarding the feasibility of estimating states at specific locations by strategically placing appropriate sensors, whether at those locations or elsewhere. Compounding these difficulties is the prevalence of noise in nearly all physical quantity measurements, where sensor data can be plagued by noise and intermittency due to temporary data loss resulting from factors such as data transmission over wireless networks. To tackle challenges such as measurement uncertainty, temporal and spatial measurement loss, this work explores the implementation of a state estimator to monitor corrosion in a coal-fired power plant. State estimators like the Kalman filter and its variants are useful tools that can estimate the ‘true’ value of a state using sensor measurements and process models, even in the face of missing sensors and uncertain measurements. A model of the hot corrosion mechanism was developed and validated using the data obtained from an industrial boiler. The corrosion process is represented by a differential-algebraic equation (DAE) system. For state estimation, a modified version of the Unscented Kalman Filter (UKF) is applied to the DAE system. The impact of variation in the sensor network was studied by simulating a number of scenarios including corrosion characteristics and mechanisms. It was observed that with only a few sensors placed at optimal locations, an accurate estimate of the corrosion profile can be obtained.

ABSTRACT. The GRAVITAS project (GPU-based Resolution And Visualization Interface for Triple-Axis Spectrometers) is designed to replace current SPICE based data manipulation, and to provide resolution calculation and model fitting for the triple axis spectrometers (TAS). Since multi-dimensional convolution of models with instrumental resolution is a very computational intensive process, we believe a parallelization using GPUs is going to allow this to happen in real time. We show the progress we achieved, including accelerating phonon scattering calculations.

ABSTRACT. PETINA (Privacy prEservaTIoN Algorithms) supports privacy both at the data level and throughout the machine learning process. It is a flexible tool that helps make machine learning safer by protecting private information. PETINA works for both centralized and federated training and supports several privacy methods, including Laplace, Gaussian, and CountSketch mechanisms. It has built-in tools to track how much privacy is used, process data efficiently, and add noise in a way that balances privacy and accuracy. Tests show PETINA can keep accuracy close to non-private models while keeping extra runtime under control. With NVFLARE, PETINA makes it possible for teams to work together on sensitive data, such as medical records, without revealing personal details. Its modular design and simple interface make it useful for both research and real-world applications that need strong privacy.

ABSTRACT. thornado is an open-source library of modules for solving the equations of neutrino-radiation hydrodynamics in relativistic astrophysics, with applications including core-collapse supernovae. Its development emphasizes specialized solvers based on discontinuous Galerkin methods, optimized for node-level performance and portability across heterogeneous computing systems. The core kernels update solution representations stored in logically Cartesian data structures, enabling straightforward integration into large-scale simulation frameworks—demonstrated by its deployment within the multi-physics code Flash-X. Beyond its native modules, thornado supports large-scale simulations through a dedicated AMReX-based layer, providing adaptive mesh refinement and distributed parallelism. This poster will present thornado’s current capabilities and outline planned developments for advancing relativistic neutrino-radiation hydrodynamics.

ABSTRACT. Big data challenges are commonly encountered when conducting radiological and chemical hazard analysis for nuclear facilities in the Department of Energy (DOE). In the context of nuclear safety, extreme weather significantly influences environmental conditions and site operations, underscoring the need to incorporate accurate weather characterization into hazard assessments and safety planning. We applied EDA to six years (2017–2022) of high-resolution meteorological observations collected from six towers across the ORNL campus, with sensors mounted at multiple heights at 15-minute intervals. Conventional statistical methods were used to characterize the overall distribution, while EDA examined daily, weekly, monthly, and seasonal trends using the Mann–Kendall test and Sen’s slope estimation. The results reveal asymmetric seasonal trends—warming during summer weeks and cooling during winter weeks. Clustering analysis was employed to interpret underlying patterns, such as the frequent co-occurrence of high temperature and high humidity during summer. Extreme weather events were further defined using feature-specific thresholds (e.g., temperature–moisture hazards, wind chill events, high-wind conditions), informed by both regulatory guidelines and clustering outcomes. This study demonstrates exploratory data analysis (EDA) as a recommended step prior to applying Machine Learning for extreme event classification, as it provides deeper insight into data quality and underlying patterns. Our results show that EDA can effectively assess big, long-term meteorological datasets and extract actionable information for site operations, particularly in relation to potentially hazardous extreme events.

ABSTRACT. SAGESim (Scalable Agent-Based GPU-Enabled Simulator) is the first scalable, pure-Python, general-purpose agent-based modeling framework that supports both distributed computing and GPU (graphics processing unit) acceleration for use on the latest generation of high-performance computing systems. SAGESim can be ubiquitously installed on both your off-the-shelf laptop computer with a single GPU device or deployed on premier supercomputing systems such as the Frontier supercomputer. It allows researchers to study complex adaptive systems at unprecedented scales, offering a general-purpose framework that runs efficiently on modern high-performance infrastructure and a Python API for easy integration with widely used machine learning and data analytics libraries.

ABSTRACT. Streamlines are widely used to visualize vector fields in fluid simulations, but they often ignore uncertainty in the underlying data. In computational fluid dynamics (CFD), such uncertainty commonly arises from truncation errors in numerical integration. Understanding these errors is essential for making informed decisions — for example, in Formula One racing, engineers must know how accurate their aerodynamic models are before adjusting car designs. We present a method to quantize and visualize streamline uncertainty for clearer, more informative flow analysis. Using the fourth-order Runge–Kutta (RK4) method, we compute both high-resolution and low-resolution/compressed streamlines. At each RK4 step, we calculate the Euclidean distance between the high-resolution and low-resolution paths, then highlight the streamline according to these distances. This stepwise coloring reveals where the streamline deviates most from the true solution. To handle cases where streamlines have different numbers of RK4 steps, we fit cubic splines between points, enabling direct comparison by measuring spline-to-spline distances or computing Euclidean distances between spline points. This approach provides a compact, interpretable representation of uncertainty, supporting more informed decision-making in CFD and related fields.

ABSTRACT. The Knowledge Discovery Infrastructure (KDI) at Oak Ridge National Laboratory (ORNL) is a research and development (R&D) platform for protected data. The system offers a secure, scalable, multi-tenant environment, for tailored Artificial Intelligence (AI), traditional High-Performance Computing (HPC), and Private Cloud operations. Datasets include any category of Controlled Unclassified Information (CUI) and/or Protected Health Information (PHI). KDI supports dozens of research projects, including the Veterans Administration’s Million Veteran Program (MVP) and a growing slate of cyber, infrastructure, and biometrics efforts, providing custom hardware, software, and user support solutions for multiple ORNL Research Directorates and their external research partners.

ABSTRACT. This poster presents an overview of a high-fidelity compressible flow solver that utilizes the continuous Galerkin (CG) method with entropy viscosity for stabilization to solve a variety of steady and unsteady benchmark inviscid flow problems. Discretizing the Euler equations with the CG approach produces a more cost-effective stencil compared to other finite-element discretizations, as well as simpler well-posed boundary conditions. We demonstrate that the reduced stencil of CG, combined with the low amount of artificial diffusion required when using the entropy viscosity method outlined in this work, leads to stable and highly accurate results. When combined with the adaptive mesh refinement approach, our results show that the aforementioned flow solver achieves even more accurate results. A variety of inviscid flow cases are presented including transient 2D cases with complex shock structures, several steady 3D airfoil configurations, and a transient 3D forward step.

ABSTRACT. This project introduces a proxy service to bridge two innovative platforms: the Interconnected Science Ecosystem (INTERSECT) and the Neutron Data Interpretation Platform (NDIP). INTERSECT enables seamless communication between distributed components in scientific ecosystems through standardized messaging, while NDIP provides an integrated environment for managing data analysis workflows, tools, and datasets within neutron scattering research.

Our proxy service facilitates interoperability between these platforms by transforming INTERSECT messages into NDIP-compatible calls, enabling the initiation of workflows, tools, and dataset registrations directly through INTERSECT. By seamlessly connecting these two ecosystems, the proxy reduces barriers to resource integration, enhances automation of scientific workflows, and streamlines tool usability for end users.

The resulting framework demonstrates how distributed scientific platforms can be effectively integrated to empower researchers with richer, more cohesive workflows. This work represents a significant step towards fostering a resilient and interoperable ecosystem for large-scale, collaborative scientific research.

ABSTRACT. High-risk property classification in Department of Energy environments demands transparent, auditable decisions while regulations—such as the International Traffic in Arms Regulations, rules of the Nuclear Regulatory Commission, and the Commerce Control List—change frequently. This work specifies a traceability-first software architecture with three coordinated layers.

The data layer combines keyword and meaning-based search with a modern language model to produce policy-grounded determinations, confidence estimates, and linked evidence passages. The control layer manages versioned assets—policy snapshots, search indexes, model and instruction versions, and routing thresholds—so outcomes are reproducible and can be safely rolled back. The observability layer records a unique trace identifier for every request, component timings, and evidence lineage in audit-ready logs.

A stable application programming interface supports both interactive decisions and large-scale processing, while expert feedback is captured and reused to improve instructions and retrieval settings. The result is a maintainable, transparent approach to regulated classification that keeps pace with policy change and preserves human oversight.

ABSTRACT. Federated Learning (FL) is critical for edge and High Performance Computing (HPC) where data is not centralized and privacy is crucial. We present OmniFed, a modular framework designed around decoupling and clear separation of concerns for configuration, orchestration, communication, and training logic. Its architecture supports configuration-driven prototyping and code-level override-what-you-need customization. We also support different topologies, mixed communication protocols within a single deployment, and popular training algorithms. It also offers optional privacy mechanisms including Differential Privacy (DP), Homomorphic Encryption (HE), and Secure Aggregation (SA), as well as compression strategies. These capabilities are exposed through well-defined extension points, allowing users to customize topology and orchestration, learning logic, and privacy/compression plugins, all while preserving the integrity of the core system. We evaluate multiple models and algorithms to measure various performance metrics. By unifying topology configuration, mixed-protocol communication, and pluggable modules in one stack, OmniFed streamlines FL experimentation and deployment across heterogeneous environments.

ABSTRACT. Neutron Scattering workflows are a core part of daily operations for SNS and HFIR scientists at Oak Ridge National Laboratory (ORNL). However, there can be challenges relating to the ease of access, use, and reproducibility of these workflows. We have developed the Neutron Data Interpretation Platform (NDIP) and the Neutrons Open Visualization and Analysis (NOVA) framework to tackle these challenges. NDIP and NOVA provide scientists with the tools to create web based user interfaces (UIs) and take advantage of computational resources, advanced visualization tools, and reproducible analysis pipelines.

NDIP is built around the open-source Galaxy project, a web-based scientific workflow engine and UI, built to enable scientists to easily manage data, create workflows, and simplify the reproduction of past results. Our efforts have integrated our own Galaxy instance, Calvera, with various computational resources including Frontier at ORNL and Perlmutter at NERSC. Calvera handles the authentication, the submission, and the reporting of results from these resources.

The NOVA framework was designed to enable scientists to develop custom UIs that take advantage of NDIP's capabilities to run workflows and tools. NOVA provides tools to create Model-View-ViewModel applications utilizing a variety of front–end technologies including Trame, Panel, and QT. Scientists can also use NOVA to launch jobs through NDIP, run workflows, retrieve their results, and display them in their interfaces.

Our poster will highlight the work done on both NDIP and NOVA, describe their capabilities, and show examples of how NDIP and NOVA have facilitated the development of Neutron Scattering workflows and interfaces.

ABSTRACT. Accurate prediction of metallurgical phase transformations is essential for understanding the microstructural evolution and resulting mechanical properties in metal alloys. During processes such as additive manufacturing, welding, and heat treatment, metal alloys experience thermal cycle(s), which induce physical state changes (solid to liquid and vice versa) and metallurgical solid-state phase transformations (for instance, austenite phase to martensite phase transformation). Capturing these phase transformations is key to linking the process generated thermal histories to the final material properties. PyJMAK is a stand-alone Python package for simulating temperature-induced metallurgical phase transformations. Designed for researchers and engineers in AM and heat treatment, it is a lightweight, flexible framework for predicting time-resolved metallurgical phase fractions from temperature histories. It has been currently implemented for typically used alloys in metal AM: Ti-6Al-4V and steel. The framework models phase transformation from the raw material (such as powder) via melting/solidification followed by metallurgical solid-state transformations induced by thermal cycles in processes such as AM and heat treatment. Diffusional transformations are captured using modified Johnson-Mehl-Avrami-Kolmogorov (JMAK) model, while martensitic transformations are modelled using Koistinen-Marburger (KM) equations. Non-isothermal conditions, typical in AM, are incorporated via the additivity rule. Users define possible transformations via a parent-child paradigm, specify the thermal conditions when the transformations can occur, and indicate whether the transformation is reversible, diffusional, or non-diffusional. PyJMAK models transformation kinetics over arbitrary temperature–time profiles, and outputs phase fraction histories. The implementation been validated against both experimental results and predictions from commercial software.

ABSTRACT. The Rust programming language is mainly used for systems programming and relies on calling C functions through its foreign function interface to interact with GPUs. With native GPU support, Rust could use its built-in memory and concurrency safety to run code without memory errors and increase researcher productivity without sacrificing performance. This poster explores the current capabilities of building and running Rust on GPUs in NVIDIA and AMD environments.

ABSTRACT. * The Autonomous Chemistry Lab (ACL) has been outfitted with automated chemical synthesis equipment and assessment hardware * Custom software controls automation, yet requires custom tooling for external integration and analysis * INTERSECT integration allows for flexible connectivity with multiple clients for data storage and analysis * Initial demonstration processes data using Neutron Data Interpretation Platform (NDIP) and ML optimization in collaboration with the DIAL INTERSECT project

ABSTRACT. With the rise of diversity in the architecture available in high-performance computing systems like supercomputers, there is a huge demand of performance portable software that can help domain scientists develop applications with minimum effort made towards porting/tuning them to these different architectures while getting the best performance. Typically, for achieving the best possible performance, the optimization effort starts with identifying the compact set of performance-critical mathematical abstractions called motifs and use them to rewrite the code to different composition of operations resulting in improved performance. There are many domain specific libraries that have been/are being developed with a focus of providing performance and programming productivity to the user for specific set of motifs. However, it is difficult to provide optimization support for all the motifs like dense linear algebra, sparse linear algebra, particles, fast Fourier transform (FFT), and many more under one single library. One must often use multiple libraries that focus on separate motifs to write an optimized code for a single scientific application. Moreover, not all libraries specialize on providing both performance and programming productivity, leaving the domain scientist struggling to find the right combination to use for their application development. To find such a combination for the domain of particle methods, we present a case study where we do a performance study of the vortex method called the Method of Local Corrections (MLC), a particle-particle particle-mesh method (PPPM), which is used to study vortex blobs to simulate incompressible fluid flow. The implementation is developed by combining two motif-based libraries namely Cabana and FFTX. Cabana is a performance portable library written in C++ for particle simulations and FFTX is a high-performance fast Fourier transform (FFT) library backed by the SPIRAL code generation system. The velocity induced by the vorticities is computed in two parts: the short-range (particle-particle method) and the long-range force (particle mesh method). Cabana provides the short-range particle computations but doesn’t implement the FFT directly which is needed for computing the long-range force or the FFT-based Poisson solver. Meanwhile, FFTX provides optimized FFT code that can interface with Cabana providing the developer with performance portability and programming productivity. Initial results for single device/GPU on Perlmutter indicate an approximate 10X speedup compared to the base CPU implementation.

ABSTRACT. Building upon the Neutron Data Interpretation Platform (NDIP) and the Neutrons Open Visualization and Analysis (NOVA) framework developed at Oak Ridge National Laboratory (ORNL), we have developed a VTK-based interactive software tool, CT Scan Visualizer, for performing rapid visualization of data produced by imaging instruments such as VENUS at the Spallation Neutron Source (SNS).

This data commonly takes the form of a stack of 2D Tagged Image File Format (TIFF) images, so the stack represents a 3-dimensional volume. Our tool can take several hundred gigabytes of this type of data and display an interactive volume rendering and slices of the data in a few seconds.

While existing tools can provide detailed volume rendering and slicing, this achieves a level of responsiveness and integration with our facilities to open the possibility to perform on-the-fly volume rendering as instruments produce data.

The tool allows the user to browse through their experiments and select the data they want to display. It additionally provides controls for swapping between rendering modes and thresholding the data to be visualized.

ABSTRACT. The SDL provides an integrated, ontology-driven ecosystem that connects scientific platforms, workflows, and data management services into a coherent whole. Built on a system-of-systems architecture and grounded in Linked Data Platform (LDP) principles, the SDL enables modular integration of diverse services while preserving interoperability across scientific domains. Core ontologies such as SSN/SOSA for sensor and observation modeling, DCAT for resource cataloging, and PROV-O for provenance tracking provide a semantic backbone that ensures all entities—data, instruments, workflows, and results—are described in a machine-actionable, reusable way.

ABSTRACT. Retrieval augmented generation (RAG) has become a powerful tool for LLMs to ingesting and synthesizing new information, which was not part of the original (and costly) training process, into a form easily interpreted by humans. AI-powered applications are a burgeoning field of development with several competing frameworks aimed at alleviating developer burden. As such, it is quite often necessary to adapt components from different frameworks to create effective software. We demonstrate how "data centric" programming paradigms, as are promoted by Python's `pydantic` library, can be used to claify application flow logic. For example, the framework `langchain` provides a simple DSL for defining a basic RAG workflow, but extending that workflow to return sources used or their similarity scores requires accurate modeling of `State`, a typed dictionary. As much as possible, we subclass the `Pydantic` `BaseModel` to achieve data validation, LSP type hints, and other benefits. Here we construct a custom RAG application using Llama-4 open source LLM hosted on Forerunner using publicly available PDF and Docx datasets. We show that even when presented with textual data extracted from PDF maps, RAG using Llama-4 is able to answer positional queries correctly.

ABSTRACT. Power systems software was never optimized for large scale models. Throughout the 20th century, limited computational resources forced the industry to rely on overcapacity and safety margins to handle demand fluctuations. With shrinking investment in new equipment and rising electrification of the U.S. economy, those margins are nearly depleted. Legacy software is now too slow to process the growing volume of data or deliver timely responses. SCADA systems once limited to a few thousand major substations per system operator must now extend to hundreds of thousands of distribution feeders. Rising fuel costs demand more efficient system operation, while renewables multiply the scenarios that must be optimized. At the same time, new cybersecurity requirements add further complexity to operations and contingency planning.

ABSTRACT. The Spallation Neutron Source (SNS) database contains legacy documents with inconsistent labeling from various national labs and industry partners, making efficient retrieval difficult. The archive includes both textual documents (e.g., policies, procedures) and engineering diagrams. We propose a two-stage solution to increase the efficiency of legacy archive searching. First, we apply semantic search and retrieval to classify each file as either a document or a diagram. Documents are further categorized, metadata is extracted, and results are ranked by relevance. Excess or obsolete data may be filtered out to ensure only current, useful content is shown to the user. Second, we use Vision Transformers (ViTs) to extract structured information from engineering drawings/diagrams to build parent child relationships and link pages belonging to the same series. This pipeline allows users to query the database, retrieve only relevant results, and seamlessly explore multi-page engineering drawings in a unified interface.

ABSTRACT. Long-range, non-bonded interactions—such as electrostatics and dispersion—are critical to many molecular properties but are difficult for standard graph neural networks (GNNs) to capture without resorting to deep stacks of message-passing layers that suffer from oversmoothing. We present a general and scalable approach that additively fuses local message passing with a global attention mechanism within the GPS framework. Positional and structural encoders, together with chemically informed atomic descriptors, enable the model to mix short-range topology with informative long-range signals. We integrate this design with HydraGNN to support multi-task learning over both graph- and node-level targets, and train efficiently on large-scale datasets using distributed data parallelism; hyperparameter optimization is automated with DeepHyper. Experiments on a benchmarked suite of molecular prediction tasks demonstrate improved accuracy—especially for properties influenced by long-range physics—while maintaining practical scalability. We also outline limitations arising from the quadratic complexity of global attention on large graphs and propose subgraph sampling strategies and post-hoc explainability analyses to quantify the complementary roles of attention and message passing. This work underscores the value of combining global attention with lightweight message passing for trustworthy, energy-efficient molecular property prediction at scale.

ABSTRACT. The framework  allowing AI/ML closure model integration into VERTEX was developed. The interface utilizes the TensorFlow Light library to perform fast model interface at runtime within VERTEX CFD. The AI/ML closure models developed in VERTEX Closures were implemented in VERTEX-CFD. An Additional source terms were added to the Chien Low-Reynolds model to account for the MHD effects on the Reynolds averaged turbulence model.

ABSTRACT. VERTEX-CFD is a free, open-source computational fluid dynamics (CFD) and multiphysics code released by Oak Ridge National Laboratory. It is based upon Trilinos, an open-source library released by Sandia National Laboratory.  VERTEX-CFD was developed with performance portability as the primary goal, and as such is compatible with a variety of CPU and GPU computing architectures. VERTEX-CFD currently supports single phase, incompressible flow, with options to include RANS And LES turbulence modeling, heat transfer, magnetohydrodynamics, conjugate heat transfer (CET) and conjugate electromagnetic (CEM).

ABSTRACT. The Atmospheric Radiation Measurement (ARM) Data Studio is part of the ARM Data Workbench Ecosystem, a suite of applications and services designed to provide users with a streamlined, no-code interface for querying and visualizing ARM data. Users have the ability to select from a continuously growing list of nearly 70 data products and hundreds of measurements for generating various types of plots. The application also empowers users to save and share configurations as projects or download them as CSV files, which they can optionally stage to ARM's Workbench JupyterHub for in-situ data analyses.

ABSTRACT. Magnetohydrodynamics (MHD) combines the principles of fluid dynamics and electromagnetism to model the behavior of electrically conducting fluids, such as plasmas, liquid metals, and saltwater. These equations play a crucial role in understanding phenomena in astrophysics, fusion energy, and industrial applications. Hence, it is crucial to have a robust and fast software to simulate the complex interactions between electromagnetism and fluid flow. Vertex-CFD is a performance portable GPU enabled multiphysics solver that has MHD and conjugate electromagnetic (CEM) capabilities. In this poster, we demonstrate the Vertex-CFD MHD and CEM capabilities on benchmark problems with analytical solutions. Vertex-CFD shows an excellent agreements compared to analytical solutions.

ABSTRACT. VERTEX-CFD must have general capabilities for the solution of incompressible turbulent flow problems to support simulation of transport processes for fusion applications. The gold standard of turbulent flow simulation remains direct numerical simulation, in which the flow is fully resolved, but this technique remains computationally expensive (often prohibitively). An intermediate approach that models the smaller scales of motion is the Reynolds-averaged Navier-Stokes (RANS) approach, which solves a modified form of the Navier-Stokes equations with additional modeled variables. These turbulence models must be modified to account for the finite element method used by VERTEX-CFD and tested to verify that the models work correctly.

ABSTRACT. VERTEX-RAD provides scalable, high performance solvers for stiff, many-species, Reaction-Advection-Diffusion (RAD) systems within the VERTEX framework. The VERTEX-RAD package utilizes newly develop exponential time integration methods implemented within the Trilinos/Tempus software library to solve multi-species reactive transport problems. This poster showcases VERTEX-RAD results for a transient molten salt reactor multi-species transport application. Additionally, an open source prototype package called ORMATEX (https://www.osti.gov/doecode/biblio/150070) was created to accelerate advanced exponential time integration methods research activities. Results from the VERTEX-RAD solver are compared against equivalent solvers in ORMATEX to check the correctness of the exponential time integration implementation in Trilinos.

ABSTRACT. The Centralized Health and Exposomic Resource (C-HER) ecosystem is an ontology-driven, metadata-first platform designed to unify fragmented health, environmental, sociodemographic, and infrastructure datasets into an interoperable, scalable framework. C-HER integrates a robust PostGIS backend, semantic alignment through the MORPH ontology tool, real-time geospatial analytics via ADDRESS services, and intuitive visualization through the MINERVA platform. By linking distributed data sources without centralizing ownership, C-HER ensures data provenance, accelerates analysis, and supports reproducible research. This poster introduces C-HER’s architecture, core components, and governance model, illustrating its potential through real-world use cases ranging from public health surveillance to disaster response. C-HER positions itself as the connective tissue enabling trusted, timely, and actionable insights across domains where integrated data is essential.

ABSTRACT. High fidelity simulations of complex turbulent flows require accurate, time-resolved resolution of the flow fields to achieve realistic predictions of drag, heat transfer, and other phenomena. Large eddy simulation (LES) techniques, both implicit (ILES) and modeled, are increasingly popular among analysts for this purpose, enabled by increases in algorithmic efficiency and computing power. This work examines the performance of two LES approaches in VERTEX-CFD: a simple algebraic LES model, and an ILES model which relies on numerical diffusion introduced by the discretization schemes to account for sub-grid dissipation. Both methods yield high-quality predictions of average and fluctuating velocity fields and integral values.

ABSTRACT. The open-source VERTEX-computational fluid dynamics (VERTEX-CFD) software focuses on modeling the incompressible Navier-Stokes equations, a temperature equation, and an electric potential equation with the objective of modeling a fusion blanket problem. When performing multiphysics simulations relating to fusion blanket problems, strong localized gradients present issues when attempting to accurately discretize the domain. VERTEX-AMR aims add an adaptive mesh refinement (AMR) routine to optimize the unstructured discretization scheme to improve stability, efficiency, and accuracy on 2D and 3D, non-periodic meshes.

ABSTRACT. VERTEX-computational fluid dynamics (VERTEX-CFD): An open-source multi-physics code with focus on modeling fusion blanket problem using the incompressible Navier-Stokes equations, a temperature equation, and an electric potential equation Computational cost and resource requirements for most of the simulations executed warrant using multiple CPUs and offloading to GPUs Trilinos provides the portability via kokkos but optimization is needed to better utilize resources

ABSTRACT. VERTEX-Particles has focused primarily on supporting plasma impurity tracking for VERTEX, in addition to general particle capabilities for multi-physics simulations. To this end, physics needed for the global impurity transport (GITR) approach has been developed within a performance portable simulation framework, banjo, for fusion plasmas and more. VERTEX-Particles has also developed generalized interfaces for particles within unstructured finite element method (FEM) multi-physics environments for more complex geometries and potential use within VERTEX-CFD and VERTEX-MAXWELL in the future.

ABSTRACT. EnergyPlus is DOE's flagship building energy modeling package, and is used by engineers, architects, and researchers to model both energy consumption and water use in buildings. Under development since 1997 as a Fortran program, the package is now developed as a multi-lab collaboration between ORNL, NREL, LBNL, and PNNL in the C++ language. Based on a description of a building (including location, geometry, construction materials, systems, occupancy, and usage), EnergyPlus calculates the heating and cooling loads necessary to maintain thermal control setpoints, the conditions throughout secondary HVAC systems and coil loads, the energy consumption of plant equipment, and many other details relating to the operation of the building. Simulations may be calibrated to the energy use of real, existing buildings or may be purely speculative models of buildings that have yet to be built. Single building models may be used to investigate the performance of a particular retrofit applied to a specific building, or many millions of simulations may be used to assess the potential impact of the retrofit on the entire building stock. Recent additions have added to the package’s ability to represent more complex situations, including a Python plug-in feature that allows the model’s behavior to be modified by user-written Python code. Finally, as artificial intelligence has become more and more important, new avenues for the development, use, and improvement of models and the package itself are opening up.

ABSTRACT. Accurate modeling of plasma edge dynamics is vital for designing plasma-facing components in fusion devices, where heat fluxes reach extreme levels. While high-fidelity codes such as SOLPS-ITER capture scrape-off layer (SOL) physics in detail, their computational cost limits parameter exploration. We present MATEY (Multiscale Adaptive foundation model for spatiotemporal physical systems), a transformer-based foundation model enabling efficient autoregressive prediction of plasma states.

Pretrained on diverse fluid systems and fine-tuned on SOLPS-ITER data, MATEY predicts density, temperature, and radiated power with high fidelity. Autoregressive training yields stable long-term rollouts, achieving <10% relative error over 1000 steps—forty times longer than training sequences—while maintaining 1–2% error for short horizons. Attention maps further reveal physically consistent relationships between core and divertor-proximal regions, offering interpretability of learned dynamics.

MATEY provides a fast, accurate surrogate for expensive plasma edge simulations, supporting rapid scenario exploration and future control applications.

ABSTRACT. Ambient loop geothermal energy networks (GEN) offer substantial benefits for reducing energy consumption and minimizing peak electric demand at both building and grid levels. However, there are currently no publicly available tools capable of effectively designing and conducting techno-economic analyses of GEN systems.

GeoWise is an online tool originally developed for the preliminary design and techno-economic evaluation of ground source heat pump (GSHP) systems for individual commercial or residential buildings. This talk presents recent upgrades to GeoWise that enable users to design and analyze GEN systems.

New features have been implemented to support the selection and specification of multiple new or existing buildings within a GEN. A comprehensive database of over 125 million U.S. buildings was created, allowing users to input building addresses and optionally modify building characteristics—such as footprint, construction vintage, primary function, number of floors, and window-to-wall ratio.

Energy simulation models are automatically generated using the Automatic Building Energy Modeling software. EnergyPlus simulations are then conducted to estimate the thermal loads of the selected buildings.

Based on these thermal loads, a simplified GEN system is sized, including the number, spacing, and depth of boreholes. The design accounts for heat loss and the thermal inertia of water in the ambient loop. A central borehole heat exchanger (BHE) is sized using the RowWise algorithm from GHEDesigner, ensuring the thermal loads are met within user-defined land constraints for BHE installation.

The upgraded GeoWise tool provides outputs including the required heat pump capacity for each building, central BHE design specifications, and projected reductions in energy consumption and associated cost savings compared to conventional HVAC systems. Additionally, the tool calculates key economic indicators such as simple payback period and return on investment.

ABSTRACT. The presentation will cover the importance of software quality assurance practices incorporation into the software development, procurement, and implementation process at ORNL. The basic software quality assurance (SQA) process at ORNL will be discussed as well as the benefits to provided by the SQA program.

ABSTRACT. The DOE/ORNL Heat Pump Design Model (HPDM) is a public-domain equipment design and modeling software. HPDM is a flagship software developed with support from the DOE to model and design building equipment in heating, ventilation, air conditioning, refrigeration, and water heating (HVAC&R, WH) sectors. It is a detailed equipment design tool, used widely by international scholars and US engineers, adopted by many equipment manufacturers and supported development of thousands of products. It supported many DOE’s initiatives to assess the feasibility, including high efficiency rooftop units, cold climate heat pumps, and low GWP refrigerants, etc. It is the most used, influential, public-domain building equipment modelling software in the industry. Some features are illustrated below: • Comprehensive knowledge-base containing state-of-the-art developments in object-oriented programming, numerical recipes, advanced solver and optimization algorithms, property handling, and fundamental thermal science at component and system levels, etc. • Hardware-based equipment design tool promotes energy saving technologies and provides technical support to transfer new technologies from ORNL to U.S. industry. • Flexible component-based platform to model extensive building equipment types and complicated configurations. • Comprehensive library simulates details of most HVAC&R and WH components. • Extensive handling of fluid properties, including air, water, glycol, hydrochlorofluorocarbon (HCFC), hydrofluorocarbon (HFC), new hydrofluoro-olefin (HFO), and natural (CO2, propane) refrigerants. • Extensive co-simulation capability integrated to EnergyPlus and Modelica to facilitate building energy co-simulations, control development, fault diagnosis and energy storage. In addition to an on-line version, HPDM is also provided as a free, downloadable desktop version with a python wrapper.