Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Distributed computing, machine learning”

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 685 records · Page 38

Universal Polarization Transformations: Spatial Programming of Polarization Scattering Matrices Using a Deep Learning‐Designed Diffractive Polarization Transformer

Abstract Controlled synthesis of optical fields having nonuniform polarization distributions presents a challenging task. Here, a universal polarization transformer is demonstrated that can synthesize a large set of arbitrarily‐selected, complex‐valued polarization scattering matrices between the polarization states at different positions within its input and output field‐of‐views (FOVs). This framework comprises 2D arrays of linear polarizers positioned between isotropic diffractive layers, each containing tens of thousands of diffractive features with optimizable transmission coefficients. After its deep learning‐based training, this diffractive polarization transformer can successfully implement N i N o = 10 000 different spatially‐encoded polarization scattering matrices with negligible error, where N i and N o represent the number of pixels in the input and output FOVs, respectively. This universal polarization transformation framework is experimentally validated in the terahertz spectrum by fabricating wire‐grid polarizers and integrating them with 3D‐printed diffractive layers to form a physical polarization transformer. Through this set‐up, an all‐optical polarization permutation operation of spatially‐varying polarization fields is demonstrated, and distinct spatially‐encoded polarization scattering matrices are simultaneously implemented between the input and output FOVs of a compact diffractive processor. This framework opens up new avenues for developing novel devices for universal polarization control and may find applications in, e.g., remote sensing, medical imaging, security, material inspection, and machine vision.

Optical neural networks↗

Bayesian inverse modeling of the atmospheric transport and emissions of a controlled tracer release from a nuclear power plant

Probability distribution functions (PDFs) of model inputs that affect the transport and dispersion of a trace gas released from a coastal California nuclear power plant are quantified using ensemble simulations, machine-learning algorithms, and Bayesian inversion. The PDFs are constrained by observations of tracer concentrations and account for uncertainty in meteorology, transport, diffusion, and emissions. Meteorological uncertainty is calculated using an ensemble of simulations of the Weather Research and Forecasting (WRF) model that samples five categories of model inputs (initialization time, boundary layer physics, land surface model, nudging options, and reanalysis data). The WRF output is used to drive tens of thousands of FLEXPART dispersion simulations that sample a uniform distribution of six emissions inputs. Machine-learning algorithms are trained on the ensemble data and used to quantify the sources of ensemble variability and to infer, via inverse modeling, the values of the 11 model inputs most consistent with tracer measurements. We find a substantial ensemble spread in tracer concentrations (factors of 10 to 10 3 ), most of which is due to changing emissions inputs (about 80%), though the cumulative effects of meteorological variations are not negligible. The performance of the inverse method is verified using synthetic observations generated from arbitrarily selected simulations. When applied to measurements from a controlled tracer release experiment, the inverse method satisfactorily determines the location, start time, duration and amount. In a 2 km × 2 km area of possible locations, the actual location is determined to within 200m. The start time is determined to within 5min out of 2h, and the duration to within 50min out of 4h. Over a range of release amounts of 10 to 1000kg, the estimated amount exceeds the actual amount of 146kg by only 32kg. The inversion also estimates probabilities of different WRF configurations. In conclusion, to best match the tracer observations, the highest-probability cases in WRF are associated with using a late initialization time and specific reanalysis data products.

54 ENVIRONMENTAL SCIENCES↗

Combining In-situ Diagnostics and Data Analytics for Discovery of Process-Structure-Property Relationships in AM parts – A Step Toward Digital Twins

In-situ additive manufacturing (AM) diagnostic tools (e.g., optical/infrared imaging, acoustic, etc.) already exist to correlate process anomalies to printed part defects. This current work aimed to augment existing capabilities by: 1) Incorporating in-situ imaging w/ machine learning (ML) image processing software (ORNL- developed "Peregrine") for AM process anomaly detection 2) Synchronizing multiple in-situ sensors for simultaneous analysis of AM build events 3) Correlating in-situ AM process data, generated part defects and part mechanical properties The key R&D question investigated was to determine if these new combined hardware/software tools could be used to successfully quantify defect distributions for parts build via SNL laser powder bed fusion (LPBF) machines, aiming to better understand data-driven process-structure-property- performance relationships. High resolution optical cameras and acoustic microphones were successfully integrated in two LPBF machines and linked to the Peregrine ML software. The software was successfully calibrated on both machines and used to image hundreds of layers of multiple builds to train the ML software in identifying printed part vs powder. The software's validation accuracy to identify this aspect increased from 56% to 98.8% over three builds. Lighting conditions inside the chamber were found to significantly impact ML algorithm predictions from in-situ sensors, so these were tailored to each machine's internal framework. Finally, 3D part reconstructions were successfully generated for a build from the compressed stack of layer-wise images. Resolution differences nearest and furthest from the optical camera were discussed. Future work aims to improve optical resolution, increase process anomalies identified, and integrate more sensor modalities.

36 MATERIALS SCIENCE↗

An adaptive approach to machine learning for compact particle accelerators

Abstract Machine learning (ML) tools are able to learn relationships between the inputs and outputs of large complex systems directly from data. However, for time-varying systems, the predictive capabilities of ML tools degrade if the systems are no longer accurately represented by the data with which the ML models were trained. For complex systems, re-training is only possible if the changes are slow relative to the rate at which large numbers of new input-output training data can be non-invasively recorded. In this work, we present an approach to deep learning for time-varying systems that does not require re-training, but uses instead an adaptive feedback in the architecture of deep convolutional neural networks (CNN). The feedback is based only on available system output measurements and is applied in the encoded low-dimensional dense layers of the encoder-decoder CNNs. First, we develop an inverse model of a complex accelerator system to map output beam measurements to input beam distributions, while both the accelerator components and the unknown input beam distribution vary rapidly with time. We then demonstrate our method on experimental measurements of the input and output beam distributions of the HiRES ultra-fast electron diffraction (UED) beam line at Lawrence Berkeley National Laboratory, and showcase its ability for automatic tracking of the time varying photocathode quantum efficiency map. Our method can be successfully used to aid both physics and ML-based surrogate online models to provide non-invasive beam diagnostics.

97 MATHEMATICS AND COMPUTING↗

Advancing spatiotemporal forecasts of CO 2 plume migration using deep learning networks with transfer learning and interpretation analysis

Accurate and timely forecasts of CO 2 plume distribution throughout the injection and post-injection phases are crucial for detecting plume migration, assessing leakage risks, and supporting operational decisions in geologic carbon storage (GCS). Current convolutional neural network-based approaches primarily focus on spatial information and overlook temporal dependencies in plume distributions, thus limiting their ability to capture dynamic movement effects and provide accurate predictions of plume migration. In this work, we propose two deep learning models, Auto-Encoder (AE)-LSTM and Encoder-Decoder (ED)-ConvLSTM, each uniquely designed to capture both spatial and temporal features. We apply the proposed methods to forecast the dynamic distribution of CO 2 plumes based on 108 reservoir simulations over a 30-year injection and a 30-year post-injection period. The results indicate that the ED-ConvLSTM model outperforms the AE-LSTM model in accurately predicting the spatiotemporal dynamics of CO 2 plume migration, achieving R 2 values above 0.99. To provide a deeper understanding of these model predictions, we employ a gradient-based explanation method on the trained models. This approach provides insights into the influence of input variables on plume migration forecasts and uncovers the underlying prediction mechanisms of the proposed models. Furthermore, we introduce a transfer learning technique, enabling fast and accurate plume migration forecasting in the post-injection phase by leveraging the trained model during the injection phase. This reduces the necessity for extensive data collection or re-training. In conclusion, the methods proposed in our work enhances the performance and interpretability of CO 2 plume migration forecasts, thereby facilitating informed decision-making throughout the entire lifecycle of GCS applications.

58 GEOSCIENCES↗

DiffLense: a conditional diffusion model for super-resolution of gravitational lensing data

Abstract Gravitational lensing data is frequently collected at low resolution due to instrumental limitations and observing conditions. Machine learning-based super-resolution techniques offer a method to enhance the resolution of these images, enabling more precise measurements of lensing effects and a better understanding of the matter distribution in the lensing system. This enhancement can significantly improve our knowledge of the distribution of mass within the lensing galaxy and its environment, as well as the properties of the background source being lensed. Traditional super-resolution techniques typically learn a mapping function from lower-resolution to higher-resolution samples. However, these methods are often constrained by their dependence on optimizing a fixed distance function, which can result in the loss of intricate details crucial for astrophysical analysis. In this work, we introduce DiffLense , a novel super-resolution pipeline based on a conditional diffusion model specifically designed to enhance the resolution of gravitational lensing images obtained from the Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP). Our approach adopts a generative model, leveraging the detailed structural information present in Hubble space telescope (HST) counterparts. The diffusion model, trained to generate HST data, is conditioned on HSC data pre-processed with denoising techniques and thresholding to significantly reduce noise and background interference. This process leads to a more distinct and less overlapping conditional distribution during the model’s training phase. We demonstrate that DiffLense outperforms existing state-of-the-art single-image super-resolution techniques, particularly in retaining the fine details necessary for astrophysical analyses.

Computer Science↗

Energy Optimization of Light and Heavy-Duty Vehicle Cohorts of Mixed Connectivity, Automation and Propulsion System Capabilities via Meshed V2V-V2I and Expanded Data Sharing (Final Scientific and Technical Report)

Vehicle connectivity and automated driving technologies individually have the potential to decrease energy consumption and/or increase safety on light, medium or heavy duty vehicles to varying degrees depending on the traffic infrastructure and specific driving scenarios. Due to advances in sensing, perception and computing power, research and development emphasis in the mobility sector has shifted away from connectivity. Prior research has shown that driving automation with the absence of connectivity can in certain circumstances increase energy consumption. The effectiveness of synergizing connectivity and driving automation technologies is the focus of this work, specifically applied to vehicle cohorts of mixed composition, light and heavy duty, and powertrains ranging from all electric to conventional internal combustion engine. The project team is led by Michigan Technological University (MTU) and partnered with AVL Mobility Technologies Inc. (AVL), Borg Warner (BW), Traffic Technology Services (TTS), American Center for Mobility (ACM) and Navistar (NAV). The main thrusts for the team are to develop a micro-traffic simulation environment with specific VD&PT system attributes and CAV capabilities, 2) field a vehicle test fleet of mixed classification, propulsion and CAV capacity, 3) develop artificial intelligence (AI) and machine learning (ML) based multi-agent optimization methods for various traffic infrastructures, 4) integrate the virtual environment and the optimization methods then deploy the system as a CAV hardware in the loop (HiL) for the vehicle test fleet and 5) conduct closed track and public road testing to validate simulation and demonstrated energy and mobility improvements at multiple scales. For a cohort of mixed vehicles, the team will demonstrate a reduction of energy consumption of 10-50% at intersection, arterial roadway and limited access highway scenarios through connectivity and automation in simulation and at a closed test track. The energy reduction objectives of the project are summarized in Table 1, indicating the infrastructure and over what distances are relevant considered. Single scenario energy reductions are not relevant and thus, the research team took the approach to vary parameters associated with the infrastructure, vehicle cohort composition and dynamic behavior to generate energy consumption distributions for both unconnected and connected scenarios.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Physics-informed heterogeneous graph neural networks for DC blocker placement

The threat of geomagnetic disturbances (GMDs) to the reliable operation of the bulk energy system has spurred the development of effective strategies for mitigating their impacts. One such approach involves placing transformer neutral blocking devices, which interrupt the path of geomagnetically induced currents (GICs) to limit their impact. The high cost of these devices and the sparsity of transformers that experience high GICs during GMD events, however, calls for a sparse placement strategy that involves high computational cost. To address this challenge, we developed a physics-informed heterogeneous graph neural network (PIHGNN) for solving the graph-based dc-blocker placement problem. Our approach combines a heterogeneous graph neural network (HGNN) with a physics-informed neural network (PINN) to capture the diverse types of nodes and edges in ac/dc networks and incorporates the physical laws of the power grid. We train the PIHGNN model using a surrogate power flow model and validate it using case studies. Results demonstrate that PIHGNN can effectively and efficiently support the deployment of GIC dc-current blockers, ensuring the continued supply of electricity to meet societal demands. Furthermore, our approach has the potential to contribute to the development of more reliable and resilient power grids capable of withstanding the growing threat that GMDs pose.

24 POWER TRANSMISSION AND DISTRIBUTION↗

SU-E-J-254: Evaluating the Role of Mid-Treatment and Post-Treatment FDG-PET/CT in Predicting Progression-Free Survival and Distant Metastasis of Anal Cancer Patients Treated with Chemoradiotherapy

Purpose: To evaluate the role of mid-treatment and post-treatment FDG-PET/CT in predicting progression-free survival (PFS) and distant metastasis (DM) of anal cancer patients treated with chemoradiotherapy (CRT). Methods: 17 anal cancer patients treated with CRT were retrospectively studied. The median prescription dose was 56 Gy (range, 50–62.5 Gy). All patients underwent FDG-PET/CT scans before and after CRT. 16 of the 17 patients had an additional FDG-PET/CT image at 3–5 weeks into the treatment (denoted as mid-treatment FDG-PET/CT). 750 features were extracted from these three sets of scans, which included both traditional PET/CT measures (SUVmax, SUVpeak, tumor diameters, etc.) and spatialtemporal PET/CT features (comprehensively quantify a tumor’s FDG uptake intensity and distribution, spatial variation (texture), geometric property and their temporal changes relative to baseline). 26 clinical parameters (age, gender, TNM stage, histology, GTV dose, etc.) were also analyzed. Advanced analytics including methods to select an optimal set of predictors and a model selection engine, which identifies the most accurate machine learning algorithm for predictive analysis was developed. Results: Comparing baseline + mid-treatment PET/CT set to baseline + posttreatment PET/CT set, 14 predictors were selected from each feature group. Same three clinical parameters (tumor size, T stage and whether 5-FU was held during any cycle of chemotherapy) and two traditional measures (pre- CRT SUVmin and SUVmedian) were selected by both predictor groups. Different mix of spatial-temporal PET/CT features was selected. Using the 14 predictors and Naive Bayes, mid-treatment PET/CT set achieved 87.5% accuracy (2 PFS patients misclassified, all local recurrence and DM patients correctly classified). Post-treatment PET/CT set achieved 94.0% accuracy (all PFS and DM patients correctly predicted, 1 local recurrence patient misclassified) with logistic regression, neural network or support vector machine model. Conclusion: Applying radiomics approach to either midtreatment or post-treatment PET/CT could achieve high accuracy in predicting anal cancer treatment outcomes. This work was supported in part by the National Cancer Institute Grant R01CA172638.

60 APPLIED LIFE SCIENCES↗

Dynamic Model Agnostic Reliability Evaluation of Machine-Learning Models Integrated in Instrumentation & Control Systems

In recent years, the field of data-driven neural network-based machine learning (ML) algorithms has grown significantly and spurred research in its applicability to instrumentation and control systems. While they are promising in operational contexts, the trustworthiness of such algorithms is not adequately assessed. Failures of ML-integrated systems are poorly understood; the lack of comprehensive risk modeling can degrade the trustworthiness of these systems. In recent reports by the National Institute for Standards and Technology, trustworthiness in ML is a critical barrier to adoption and will play a vital role in intelligent systems' safe and accountable operation. Thus, in this work, we demonstrate a real-time model-agnostic method to evaluate the relative reliability of ML predictions by incorporating out-of-distribution detection on the training dataset. It is well documented that ML algorithms excel at interpolation (or near-interpolation) tasks but significantly degrade at extrapolation. This occurs when new samples are "far" from training samples. The method, referred to as the Laplacian distributed decay for reliability (LADDR), determines the difference between the operational and training datasets, which is used to calculate a prediction's relative reliability. LADDR is demonstrated on a feedforward neural network-based model used to predict safety significant factors during different loss-of-flow transients. LADDR is intended as a "data supervisor" and determines the appropriateness of well-trained ML models in the context of operational conditions. Ultimately, LADDR illustrates how training data can be used as evidence to support the trustworthiness of ML predictions when utilized for conventional interpolation tasks.

97 MATHEMATICS AND COMPUTING↗

Multiscale modeling of packed-bed microwave reactors and estimation of intrinsic materials' permittivity

Modeling of packed-bed microwave reactors relies on an accurate representation of particle size, shape, and distribution within the bed, as well as the particles' dielectric properties. The measured permittivity of microwave susceptors (powders or structured materials) depends on the geometric features of the particles and the porosity of the bed, as well as the specific form factor of a structured material. These are effective properties and cannot be used to analyze other reactor configurations unless the geometric effects are removed. Therefore, we introduce a methodology for extracting the intrinsic particle permittivity from experimentally measured effective permittivity by combining cavity-based measurements with multiscale simulations and machine learning. Further, we develop the first multiscale model of packed-bed microwave reactors that incorporate particle effects (geometric features, random packing, and particle contact). This approach bridges macroscopic observables with mesoscopic physics, enabling analysis of local hotspots, arcing, and contact effects that control reactor performance. Using polymer-based spherical activated carbon (PBSAC) and silicon carbide (SiC) as examples, we demonstrate that the inferred particle permittivity is consistent with independent experimental heating profiles we collect from microwave reactors without adjustable parameters. Finally, this methodology establishes a foundation for predictive, multiscale design of microwave packed-bed reactors that explicitly accounts for particle-scale effects, enabling the estimation of intrinsic permittivity for the first time.

97 MATHEMATICS AND COMPUTING↗

Situational Awareness of Grid Anomalies (SAGA) for Visual Analytics—Near-Real-Time Cyber-Physical Resiliency Through Machine Learning

The Situational Awareness of Grid Anomalies (SAGA) project built upon foundational power system tools developed at the National Renewable Energy Laboratory (NREL) integrated with an ever-increasing set of Gridmetrics data extracted from the cable television (CATV) broadband network infrastructure while assimilating other time-series geospatial data and information, such as weather and cyber-physical phenomena, to demonstrate a disruptive technology for power system data analytics relying on existing infrastructure. Three research thrusts supported (1) visual analytics, (2) cyber-physical power system simulation, and (3) anomaly detection. SAGA created technology that leverages, couples, and fortifies two vastly different realms - power and broadband - to increase the resiliency of the power grid in the face of increasing cyberattacks and operational challenges related to integrating DERs. The exploration of potential synergies of broadband-enabled grids resulted in identifying a mutually beneficial symbiosis that can increase the resiliency of both power and broadband services. Broadband networks perform better with reliable power and are good at providing real-time measurements that identify where the grid is under attack, is failing, or is weak. Likewise, sensor-starved distribution grids perform better and can be more reliable when their operation is buttressed with observations of broadband-detected anomalies. Future research can explore broadband's contribution to continuing to improve grid resiliency, reliability, and cost-effective operation.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Resolving turbulent magnetohydrodynamics: a hybrid operator-diffusion framework

We present a hybrid machine learning framework that combines physics-informed neural operators (PINOs) with score-based generative diffusion models to simulate the full spatio-temporal evolution of two-dimensional, incompressible, resistive magnetohydrodynamic turbulence across a broad range of Reynolds numbers (Re). The framework leverages the equation-constrained generalization capabilities of PINOs to predict coherent, low-frequency dynamics, while a conditional diffusion model stochastically corrects high-frequency residuals, enabling accurate modeling of fully developed turbulence. Trained on a comprehensive ensemble of high-fidelity simulations with Re ϵ {100, 250, 500, 750, 1000, 3000, 10000}, the approach achieves state-of-the-art accuracy in regimes previously inaccessible to deterministic surrogates. At Re = 1000 and 3000, the model faithfully reconstructs the full spectral energy distributions of both velocity and magnetic fields late into the simulation, capturing non-Gaussian statistics, intermittent structures, and cross-field correlations with high fidelity. At extreme turbulence levels (Re = 10 000), it remains the first surrogate capable of recovering the high-wavenumber evolution of the magnetic field, preserving large-scale morphology and enabling statistically meaningful predictions.

Diffusion-Integrated Neural Operators↗

Explainable AI for a No-Teardown Vehicle Component Cost Estimation: A Top-Down Approach

The broader ambition of this article is to popularize an approach for the fair distribution of the quantity of a system's output to its subsystems while allowing for underlying complex subsystem level interactions. Particularly, we present the use of this framework on a very specific (but generalizable) application, interesting for a more general AI audience. Here, we detail a data-driven approach to vehicle price modeling and its component price estimation by leveraging a combination of concepts from machine learning and game theory. We show an alternative to common teardown methodologies and surveying approaches for component and vehicle price estimation at the manufacturer's suggested retail price (MSRP) level that has the advantage of bypassing uncertainties involved in gathering teardown data, the need to perform expensive and biased surveying, and the need to perform retail price equivalent or indirect cost multiplier adjustments to mark up direct manufacturing costs to MSRP. This novel exercise not only provides accurate pricing of the technologies at the customer level, but also shows the, a priori known, large gaps in pricing strategies between manufacturers, vehicle classes, market segments, etc. There is also clear interaction between the price of technologies and other specifications present in vehicles. Those results are indication that old methods of manufacturer-level component costing, aggregation, and application of flat and rigid adjustment factors should be carefully examined. The findings are based on a database developed by Argonne, which includes over 64,000 vehicles covering MY1990 to MY2020 with hundreds of vehicle specs.

97 MATHEMATICS AND COMPUTING↗

Combustion Performance and Emissions Optimization Through Integration of a Miniaturized High-Temperature Multi Process Monitoring System

The main objective of this research program is to design, manufacture, and demonstrate a miniaturized, multi-process, monitoring system (mMPMS) for boiler condition management and easy system deployment to obtain a higher spatial resolution of wall conditions in utility boilers. This system will facilitate a Condition-Based Maintenance (CBM) philosophy that actively monitors the health of assets to predict and prevent failures and maximize availability and generating capacity at a reduced cost. CBM systems can provide boiler data that the advanced process control (APC) system can utilize for plant performance optimization, which is increasingly relevant as coal power plants shift from predominantly base-load operation to predominantly transient operation involving large load swings. The mMPMS is based on an electrochemical sensor that provides a real-time indication of the risk of damage to key locations in the radiant or convective section of a coal-fired boiler such as metal loss rates, heat flux, metal surface temperature, and deposit thickness. These indications can be utilized to optimize boiler performance as well as improve boiler availability in conjunction with corresponding operating conditions. This monitoring system was developed and tested in the high-temperature regions of a coal-fired utility boiler in this project but can be applied to many other industries and applications as well. The new sensor design, leveraging the existing electrochemical noise-based monitoring system, is small enough to be installed through the webbing of the waterwalls without the need for long shut-downs to bend tubes and to make it feasible to obtain high spatial resolution in the boiler. Data can be transferred to the plant distributed control system (DCS) and any other control system. The sensor body that houses the sensor assembly was designed to ensure good conductive contact with boiler tubes to ensure the sensor is held at an identical temperature to the tube surface temperature. The data acquisition and signal conditioning modules were redesigned into a small footprint with optimized cooling of the module. System software was developed specifically for the new signal conditioning module and is compatible with plant PLCs. After the preliminary testing at a pilot-scale facility, five mMPMS were installed at a full-scale pulverized coal-fired plant, Basin Electric Power Cooperative’s Leland Olds Unit 1. The systems were demonstrated over a 6-week period during typical operation. Sensor measurements of deposit thickness were validated during the demonstration and subsequently leveraged to determine sensor-based boiler cleaning strategies. These strategies have the benefit of reduced thermal stresses on boiler tubes from over-cleaning and improved boiler water management. At the end of the project, continued development of the sensor technology was carried out at PacifiCorp’s Hunter Station. REI leveraged the permanent installation of the mMPMS in Unit 3 made possible by DOE funding on a separate program. The work at Hunter Plant focused on application of machine learning and artificial intelligence-based models for integration of sensor signals into control and optimization of Hunter Unit 3 processes.

42 ENGINEERING↗

FM4NPP: A Scaling Foundation Model for Nuclear and Particle Physics

Large language models have revolutionized artificial intelligence by enabling large, generalizable models trained through self-supervision. This paradigm has inspired the development of scientific foundation models (FMs). However, applying this capability to experimental particle physics is challenging due to the sparse, spatially distributed nature of detector data, which differs dramatically from natural language. This work addresses if an FM for particle physics can scale and generalize across diverse tasks. We introduce a new dataset with more than 11 million particle collision events and a suite of downstream tasks and labeled data for evaluation. We propose a novel self-supervised training method for detector data and demonstrate its neural scalability with models that feature up to 188 million parameters. With frozen weights and task-specific adapters, this FM consistently outperforms baseline models across all downstream tasks. The performance also exhibits robust data-efficient adaptation. Further analysis reveals that the representations extracted by the FM are task-agnostic but can be specialized via a single linear mapping for different downstream tasks.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Flexible AI Models for Grid Resilience

The rapid growth in size and complexity of artificial intelligence (AI) and machine learning (ML) models has led to increased energy demands, posing a threat to the reliability of the existing power grid. This project addresses the challenge of highly intermittent and energy-intensive inference workloads by (1) developing fidelity-adaptive neural networks capable of dynamic response to grid conditions and (2) integrating these networks with power flow simulations to assess their impact on power grid reliability. We will explore both top-down and bottom-up approaches to create hierarchies of submodels that provide a controlled trade-off between power draw and prediction accuracy. The top-down method utilizes NN pruning to reduce a flagship model into progressively smaller, energy-efficient variants. The bottom-up approach employs geometrically principled weight setting strategies to construct depth-efficient models from the ground up. A real-time hardware-in-the-loop (HIL) platform will be developed to simulate a scaled AC power grid, integrating live AI workload power draw and enabling dynamic model switching in response to grid feedback. This work will provide a novel framework for evaluating the impact of flexible AI/ML workloads on grid performance and establish new methodologies for energy-aware computing in data centers. The outcomes will demonstrate that adaptive AI/ML can play a critical role in improving grid stability while advancing NREL's leadership in energy-efficient computing research.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Physics-Informed Machine Learning with Application to Solid Oxide Fuel Cell System Modeling and Optimization

Traditional data-driven machine learning (ML) techniques were combined with the physics-based SOFC-MP model toward improving SOFC system-level performance prediction. Four different physics-informed ML methods were proposed to demonstrate better prediction accuracy compared with traditional ML methods. First, it was preferred that the existing Kriging-based surrogate model not be entirely abandoned, so errors between Kriging regression prediction and true solution are learned using ML. The goal was to find the pattern of the error distribution as a function of model input so discrepancies between the Kriging prediction and true solution could be reduced by adding the additional ML-trained error term to the existing prediction. Second, deep neural networks (DNN) coupled with the mass balance model (MBM) have been developed to significantly decrease the reduced order model (ROM) prediction error with fewer training data compared to the traditional Kriging-based ROM and traditional DNN regression approaches. Third, DNN regression coupled with a neural networks (NN) classifier and MBM was developed. It provides superior performance on classifying the physically operational conditions for natural gas fuel cell (NGFC) than traditional classification approaches. Finally, two kinds of DNN transferring were studied and tested. One was transferring DNN for low-fidelity model data to a high-fidelity model. The second one was transferring DNN from state-of-the-art NGFC to advanced NGFC. These physic-informed ML frameworks can serve as a solid foundation for estimating U.S. Department of Energy-targeted fuel cell performance and exhibit great potential to be employed to other engineering applications.

30 DIRECT ENERGY CONVERSION↗