Serving Machine Learning Models in a Production Environment
A presentation on serving machine learning models in a production environment
97 MATHEMATICS AND COMPUTING↗
SEARCH · Engineering Papers
Results for “Operator 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.
Biomethanation to Upgrade Biogas to Pipeline Grade Methane
NREL is working closely with DOE, Electrochaea GmbH, and Southern California Gas Company (SoCalGas) to reduce costs of a biomethanation process capable of megawatt-scale deployment that upgrades organic biogas waste streams to produce pipeline quality renewable natural gas (RNG). Biomethanation is a two-step process using a single-celled methanogenic archaea that converts low-carbon low-cost hydrogen (H2) and waste carbon dioxide (CO2) to produce renewable methane (CH4). The process upgrades the biogenic CO2 - while allowing the CH4 to pass through - from biogas sources like dairies, wastewater treatment plants, and landfills. The CH4 produced is a drop-in direct replacement fuel and producers can participate in the growing number of carbon markets; like California's Low Carbon Fuel Standard and the Federal Renewable Fuel Standard. NREL and Argonne National Laboratory have completed a life cycle analysis using the GREET model to show that the biomethanation process produces RNG that is carbon negative even when H2 production via low-temperature water electrolysis is driven by the existing carbon intensity of California's electricity grid. And of course, even further carbon negative (-233 kg CO2e/kWh) when the electricity is produced from low-carbon sources like wind and solar. Leveraging lessons learned from operating SoCalGas' 700L 18-bar bioreactor system, NREL is designing and building a flexible RD&D platform that will enable field trials at biogas and other CO2 sources. A custom 16' long trailer will house a 20L 18-bar bioreactor, 3 - 25 kW proton exchange membrane electrolyzer, and dosing, thermal, and controls systems to support operations with only power, biogas, and water feedstocks required by the field locations. The end-of-project goal is to demonstrate pipeline quality RNG production (> 95% CH4, < 4% H2, <1% CO2, < 0.2% O2 and < 4 parts per million H2 sulfide) using real biogas feedstocks - thereby recycling both greenhouse gases for injection into the natural gas network or to be used onsite.
biogas upgrading↗
Aging Infrastructure: Application of System Health Monitoring at the Defense Waste Processing Facility - 16069
The disposition of high level waste (HLW) into borosilicate glass canistered waste forms began at SRS in 1996. Since that time, the Defense Waste Processing Facility (DWPF) has produced nearly four thousand HLW canisters. The DWPF is now approaching 20 years of radioactive service and 25 years of operation since the commencement of nonradioactive testing in 1990. Many facility systems are challenged by equipment or component obsolescence, which places a premium on proper work planning, engineering, and maintenance resource allocation to ensure facility reliability. To this end, Savannah River Remediation (SRR) implements programs and procedures for performance monitoring of structures, systems, and components (SSCs) within all of its processing facilities. As these programs have matured, they have helped to identify vulnerabilities within facility systems and provided a framework for ranking system assets to ensure correct priority is established and maintained. This paper provides an overview of how SRR structures and implements system performance monitoring with specific focus on the recent replacement of the DWPF Safety Grade Purge Flowmeters. This case study will discuss: 1) How the performance monitoring process identified an availability risk due to the minimal vendor support for the in-use flow transmitters, 2) The design changes developed as part of the project to transition to a different measurement technology, which includes operational lessons learned from the original installation, 3) The execution of the replacement effort, which was complicated by the fact that the purge system is a Vital Safety System (VSS) and not easily taken out of service for extended maintenance, and 4) The need to consider future maintainability, particularly given that continued acceleration in technical innovation will likely increase the frequency of equipment obsolescence. (authors)
12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
Emergency vacuum repairs in an aging accelerator: Case studies and lessons learned
Jefferson Lab operates the CEBAF electron accelerator at energies to 12 GeV for the Department of Energy Nuclear Physics program. The CEBAF injector beamline was designed and built in the early 1990s. Although we’ve upgraded and replaced many of the vacuum systems, we still have unique original components installed which operate daily. Over the past 3 years, we have had several vacuum leaks in ageing components leading to emergency repairs on a tight timeline. I’ll discuss the nature of these vacuum component failures, the difficulties in repair due to their ages, the lessons we’ve learned, and how we hope to minimize similar failures going forward.
Stutzman, Marcy [Thomas Jefferson National Acceler↗
Machine Learning-Based Security Assessment and Control for Bulk Electric System Operation
This presentation provides an overview of machine learning-based stability assessment and control for bulk electric system operation.
41 EE - Solar Energy Technologies Office (EE-4S)↗
Operator inference with roll outs for learning reduced models from scarce and low-quality data
Data-driven modeling has become a key building block in computational science and engineering. However, data that are available in science and engineering are typically scarce, often polluted with noise and affected by measurement errors and other perturbations, which makes learning the dynamics of systems challenging. Here, in this work, we propose to combine data-driven modeling via operator inference with the dynamic training via roll outs of neural ordinary differential equations. Operator inference with roll outs inherits interpretability, scalability, and structure preservation of traditional operator inference while leveraging the dynamic training via roll outs over multiple time steps to increase stability and robustness for learning from low-quality and noisy data. Numerical experiments with data describing shallow water waves and surface quasi-geostrophic dynamics demonstrate that operator inference with roll outs provides predictive models from training trajectories even if data are sampled sparsely in time and polluted with noise of up to 10%.
97 MATHEMATICS AND COMPUTING↗
Characterization and Quantification of Radiation-Induced Clusters/Precipitates in RPV Steels Using STEM-EDS and Machine Learning
Over the operational lifespan of a nuclear reactor, reactor pressure vessel (RPV) steels are subjected to significant neutron irradiation, resulting in complex microstructural changes and the consequent degradation of mechanical properties. Various physically motivated correlation models have been developed to predict neutron irradiation-induced embrittlement of RPVs under different irradiation conditions. However, the efficient and accurate characterizations and quantification of radiation-induced clusters in RPVs are still challenging, which will affect the precision of the predictive models for embrittlement of RPV components. In the DOE Visiting Faculty Program (VFP) research work at Oak Ridge National Lab (ORNL), I integrate machine learning to aid Scanning Transmission Electron Microscopy – Energy Dispersive X-ray Spectroscopy (STEM-EDS) analyses, which improve the characterization and quantification of radiation-induced clusters in RPV steels, thereby enabling more accurate predictions of material behavior under irradiation. The surveillance base- and welded- RPV steels were annealed at various temperatures of 340 °C, 450 °C and 500 °C for up to 168 hours, respectively. Afterwards, I have characterized radiation-induced clusters using advanced STEM-EDS techniques and subsequently applying machine learning algorithms to analyze and refine STEM-EDS datasets, enhancing the quantification of clusters compositions and distributions. In the end, an efficient workflow for integrating STEM-EDS data analysis with machine learning to address challenges including noise reduction has been developed. The completion of this VFP work will support bridge critical gaps in the accurate quantification of radiation-induced clusters in RPV steels using STEM-EDS and support the development of more precise models for predicting RPV embrittlement in the Light Water Reactor Sustainability program supported by Department of Energy and enhancing the collaboration between ORNL and Alred University. The outcome of the VFP project will leverage a few research papers submission to peer-reviewed journals in the relevant scientific field and a few oral presentations at national and international conferences.
22 GENERAL STUDIES OF NUCLEAR REACTORS↗
Machine learning enabling high-throughput and remote operations at large-scale user facilities
Code used for generating results in the publication: "Machine learning enabling high-throughput and remote operations at large-scale user facilities"
Maffettone, PhillipM. [Brookhaven National Lab. (B↗
Enhancing Security and Resiliency in Operational Technology Environments Through Network Slicing and Federated Learning
The growing convergence of Information Technology (IT) and Operational Technology (OT) within Industry 4.0 environments has introduced new demands on industrial network infrastructure. As cyber-physical systems become increasingly interconnected, ensuring the secure, timely, and efficient exchange of critical data is essential. This thesis explores how network slicing, a method of creating isolated virtual network segments, can be applied within OT environments to address challenges such as latency, security, and resource allocation. The first research question addressed in this thesis is: How can OT networks take advantage of NFV and SDN technology to become cyber resilient? This study examines the operational, security, and architectural implications of introducing network slicing into traditionally static OT infrastructures such as Industrial Control Systems (ICS) and SCADA. Through simulated deployments and case studies, the research demonstrates how slicing enables better isolation between critical and non-critical services, thereby improving response time, throughput, and security in sensitive environments. The second question considers: How to dynamically implement network slicing and take advantage of network resources towards integrating decentralized machine learning? In response, this thesis proposes a framework that combines Software-Defined Networking (SDN), Network Function Virtualization (NFV), and Federated Learning (FL) to enable real-time analytics while maintaining data locality. The proposed approach reduces the burden on centralized infrastructure and minimizes privacy risks by supporting on-site training of models across distributed OT nodes, coordinated through dynamically allocated network slices. The third focus explores: How slicing helps to increase the resiliency of OT networks through the orchestration of a dynamic DMZ? To answer this, the thesis presents a method for creating and managing Dynamic Demilitarized Zones (DMZs) using network slicing. This enables flexible and automated isolation of sensitive subsystems during threat scenarios or high-risk operations. Coupled with intelligent orchestration and containerized security services, the dynamic DMZ significantly enhances the system's ability to respond to cyber incidents without halting production. Ultimately, this thesis contributes a comprehensive architecture that blends network slicing with machine learning, secure segmentation, and automation, paving the way for resilient, adaptive, and intelligent OT environments. Performance evaluations across multiple scenarios show improvements in system reliability, threat response time, model accuracy, and resource utilization, providing a strong foundation for future industrial automation systems.
Rodiles Delgado, Brian G↗
Data-Driven Operator Theoretic Methods for Global Phase Space Learning
In this work, we propose to apply the recently developed Koopman operator techniques to explore the global phase space of a nonlinear system from time-series data. In particular, we address the problem of identifying various invariant subsets of a phase space from the spectral properties of the associated Koopman operator constructed from time-series data. Moreover, in the case when the system evolution is known locally in various invariant subspaces, then a phase space stitching result is proposed that can be applied to identify a global Koopman operator. A biological system, bistable toggle switch and a second-order nonlinear system example are considered to illustrate the proposed results. The construction of this global Koopman operator is very helpful in experimental works as multiple experiments can't be run at the same time starting from several initial conditions.
Nandanoori, Sai Pushpak↗
Mixed Precision Fermi-Operator Expansion on Tensor Cores from a Machine Learning Perspective
Here we present a second-order recursive Fermi-operator expansion scheme using mixed precision floating point operations to perform electronic structure calculations using tensor core units. A performance of over 100 teraFLOPs is achieved for half-precision floating point operations on Nvidia’s A100 tensor core units. The second-order recursive Fermi-operator scheme is formulated in terms of a generalized, differentiable deep neural network structure, which solves the quantum mechanical electronic structure problem. We demonstrate how this network can be accelerated by optimizing the weight and bias values to substantially reduce the number of layers required for convergence. We also show how this machine learning approach can be used to optimize the coefficients of the recursive Fermi-operator expansion to accurately represent the fractional occupation numbers of the electronic states at finite temperatures.
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
Learning Only on Boundaries: A Physics-Informed Neural Operator for Solving Parametric Partial Differential Equations in Complex Geometries
Recently, deep learning surrogates and neural operators have shown promise in solving partial differential equations (PDEs). However, they often require a large amount of training data and are limited to bounded domains. In this work, we present a novel physics-informed neural operator method to solve parameterized boundary value problems without labeled data. By reformulating the PDEs into boundary integral equations (BIEs), we can train the operator network solely on the boundary of the domain. This approach reduces the number of required sample points from $O(N^d)$ to $O(N^{d-1}$), where $d$ is the domain’s dimension, leading to a significant acceleration of the training process. Additionally, our method can handle unbounded problems, which are unattainable for existing physics-informed neural networks (PINNs) and neural operators. Finally, our numerical experiments show the effectiveness of parameterized complex geometries and unbounded problems.
97 MATHEMATICS AND COMPUTING↗
Lessons from commissioning of the cryogenic system for the Short-Baseline Neutrino Detector at Fermilab
Results from commissioning and first year of operations of the cryogenic system of the Short-Baseline Neutrino Detector (SBND) and its membrane cryostat installed at the Fermi National Accelerator Laboratory are described. The SBND detector is installed in a 200 m$^3$ membrane cryostat filled with liquid argon, which serves both as target and as active media. For the correct operation of the detector, the liquid argon must be kept in very stable thermal conditions while the contamination of electronegative impurities must be consistently kept at the level of small fractions of parts per billion. The detector is operated in Booster Neutrino Beams (BNB) at Fermilab for the search of sterile neutrinos and measurements of neutrino-argon cross sections. The cryostat and the cryogenic systems also serve as prototypes for the much larger equipment to be used for the LBNF/DUNE experiment. Since its installation in 2018-2023 and cooldown in spring of 2024, the cryostat and the cryogenic system have been commissioned to support the detector operations. The lessons learned through installation, testing, commissioning, cooldown, and initial operations are described.
43 PARTICLE ACCELERATORS↗
Lessons from commissioning of the cryogenic system for the Short-Baseline Neutrino Detector at Fermilab
Results from commissioning and first year of operations of the cryogenic system of the Short-Baseline Neutrino Detector (SBND) and its membrane cryostat installed at the Fermi National Accelerator Laboratory are described. The SBND detector is installed in a 200 m3 membrane cryostat filled with liquid argon, which serves both as target and as active media. For the correct operation of the detector, the liquid argon must be kept in very stable thermal conditions while the contamination of electronegative impurities must be consistently kept at the level of small fractions of parts per billion. The detector is operated in Booster Neutrino Beams (BNB) at Fermilab for the search of sterile neutrinos and measurements of neutrino-argon cross sections. The cryostat and the cryogenic systems also serve as prototypes for the much larger equipment to be used for the LBNF/DUNE experiment. Since its installation in 2018-2023 and cooldown in spring of 2024, the cryostat and the cryogenic system have been commissioned to support the detector operations. The lessons learned through installation, testing, commissioning, cooldown, and initial operations are described.
Geynisman, Michael [Fermilab]↗
Scalable Wind Turbine Generator Bearing Fault Prediction Using Machine Learning: A Case Study
Operation and maintenance (O&M) costs for wind turbines pose a risk to competitiveness and asset owners. With machine-learning technologies and digitalization rapidly maturing, the wind industry is actively investigating these new technologies to optimize O&M practices and reduce costs. This paper reviews recent work on machine-learning approaches to generator bearing failure prediction and presents a relevant real-world case study through a collaboration between the National Renewable Energy Laboratory and Envision Digital Corporation. In the case study, we evaluate the performance of representative machine-learning algorithms for predicting wind turbine generator bearing failures. Operational supervisory control and data acquisition data from one wind power plant was used to train and test the machine-learning models. The investigated data channels are chosen based on whether physically they reflect the failed generator bearing conditions and the component historical usage, including both environmental and operational conditions. Benefits and drawbacks of different methods are identified.
generator bearing failures↗
An Edge-Cloud Integrated Solution for Buildings Demand Response Using Reinforcement Learning
Buildings, as major energy consumers, can provide great untapped demand response (DR) resources for grid services. However, their participation remains low in real-life. One major impediment for popularizing DR in buildings is the lack of cost-effective automation systems that can be widely adopted. Existing optimization-based smart building control algorithms suffer from high costs on both building-specific modeling and on-demand computing resources. To tackle these issues, this paper proposes a cost-effective edge-cloud integrated solution using reinforcement learning (RL). Beside RL’s ability to solve sequential optimal decision-making problems, its adaptability to easy-to-obtain building models and the off-line learning feature are likely to reduce the controller’s implementation cost. Using a surrogate building model learned automatically from building operation data, an RL agent learns an optimal control policy on cloud infrastructure, and the policy is then distributed to edge devices for execution. Simulation results demonstrate the control efficacy and the learning efficiency in buildings of different sizes. A preliminary cost analysis on a 4-zone commercial building shows the annual cost for optimal policy training is only 2.25% of the DR incentive received. Results of this study show a possible approach with higher return on investment for buildings to participate in DR programs.
24 POWER TRANSMISSION AND DISTRIBUTION↗
Structure Learning in Power Distribution Networks
Traditionally power distribution networks are either not observable or only partially observable. This complicates development and implementation of new smart grid technologies, such as these related to demand response, outage detection and management, and improved load-monitoring. Here, inspired by proliferation of the metering technology, we discuss statistical estimation problems in structurally loopy but operationally radial distribution grids consisting in learning operational layout of the network from measurements, e.g. voltage data, which are either already available or can be made available with a relatively minor investment. Our newly suggested algorithms apply to a wide range of realistic scenarios. The algorithms are also computationally efficient – polynomial in time – which is proven theoretically and illustrated computationally on a number of test cases. The technique developed can be applied to detect line failures in real time as well as to understand the scope of possible adversarial attacks on the grid.
24 POWER TRANSMISSION AND DISTRIBUTION↗
Dilute Source CO 2 Capture: Management of Atmospheric Coal-Produced Legacy Emissions (Final Technical Report)
This report presents the scientific and technical findings by Carbon Engineering Ltd (CE) resulting from the DOE sponsored research and development project on Direct Air Capture (DAC) process. The work was performed by CE under the Award DE-FE0026861 from the US Department of Energy’s (DOE) National Energy Technology Laboratory (NETL). The work plan was centered on advancing the Recipient’s DAC technology, developing a better understanding of its performance and cost. The report outlines the results from two key activities that were performed under this award: Studying the performance characteristics of the DAC plant at both pilot and lab scale Techno-economic analysis (TEA) to study two implementations of CE’s proprietary DAC process: i) An early commercial DAC process that captures ~1 MT CO 2 /year of atmospheric CO 2 (Baseline case), and ii) A modified early commercial DAC process that captures 1 MT CO 2 /year of flue gas CO 2 downstream of a subcritical PC post-CCS power plant (Case 1 Option C). Within the first key activity of the project, significant operational improvements and learnings were realized, including for key equipment in the DAC process in addition to the overall system. Learnings included impact of equipment design on the capture, calcination and slaking processes, and operational parameter modifications for optimizing the pelletization process, along with the complex interactions and associated considerations of these on overall system performance. In parallel, benchtop scale equipment was successfully utilized to validate pilot scale data, such as, retention in the pellet reactor, and capture performance parameters of the air contactor. CE continues to conduct core research and development activities; the learnings from this DOE project will contribute to the design, testing and operation activities of CE’s fully integrated DAC validation plant, as well as the first commercial DAC plant. The second key activity incorporated the above learnings, as well as input from vendors and EPC experts, to produce the TEA report. More specifically, the TEA results were based on measured pilot plant performance data for all major unit operations, and on an energy and material balance for the process at commercial scale computed using an Aspen One® process simulation. The cost estimation methodology follows AACE International recommended practices; with an uncertainty range of -15%/+30%. Depending on financial assumptions, energy costs, and the specific choice of inputs and outputs, the levelized cost per ton CO 2 captured from the atmosphere for an early commercial plant for the baseline case ranges from 227 to 304 $/t-CO 2 . And, the levelized cost to capture 98% of the flue gas CO 2 emissions in a subcritical PC power plant (Case 1 Option C) through a combination of Cansolv carbon capture system and a modified DAC system was estimated to range from 109 to 160 $/MWh of net power generated. Process improvement opportunities identified during this work, in addition to cost reduction opportunities identified during the Value Engineering exercise associated with the Baseline case indicate that significant reductions in capital costs are available as CE scales up and proceeds through deployment of initial facilities.
20 FOSSIL-FUELED POWER PLANTS↗