Search NASA⌕ Search

SEARCH · Search NASA

Results for “network optimization”

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 1,441 records · Page 80
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Iterative Linearization for Phasor-Defined Optimal Power Dispatch

Optimal power flow (OPF) problems, which dispatch power targets to controllable generating units across a network, must generally account for non-convex constraints on power flow. Furthermore, adapting those problems so as to make them solvable with convex optimization techniques is an area of much academic and operational interest. In this paper, we present a method for solving OPF as a quadratic program by iteratively refining and re-initializing a linearized model of power flow based on the outputs of an associated nonlinear solver. The linear model on which we demonstrate this method is an adapted version of an approximation designed for use with unbalanced distribution networks. As an important benefit, the model allows for the explicit inclusion of nodal voltage phasor values in both the OPF problem's objective and its constraints, which opens the door to the idea of phasor-based control (PBC) design. We show in simulations on the IEEE 13-node test feeder that our method quickly converges to a set of phasor targets that are sufficiently precise for use in operations at the distribution level.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2024

Advanced Data Science Model for Detecting Intelligent Malware

This study focused on developing a robust artificial intelligence (AI) model capable of detecting and characterizing advanced malware in Internet of Things (IoT) devices using network data. By analyzing network traffic with various machine learning (ML) models, our AI model can identify and characterize malicious activities to significantly improve malware detection accuracy and reliability as compared to traditional methods. The developed AI/ML model was trained using network data from IoT devices, leveraging classifiers such as Random Forest, Gradient Boosting, AdaBoost, and others to optimize detection performance. This project demonstrates a scalable framework for real-time malware detection and characterization in IoT networks, capable of identifying infected devices and facilitating the necessary steps to remove or isolate them, thereby preventing further infections. Although digital twin (DT) integration is not yet implemented in the current model, it represents a promising future enhancement. By creating a virtual replica of physical IoT devices, DT technology would allow for real-time monitoring and analysis without directly accessing operational technology, thus reducing the risk of compromising or reducing the performance of actual devices. This integration would further enhance the security of IoT ecosystems, combining AI technology to better flag and detect indications of malware-infected devices within a nuclear system environment.

42 ENGINEERING↗
NASA NTRS1998

Preliminary Work for Examining the Scalability of Reinforcement Learning

Researchers began studying automated agents that learn to perform multiple-step tasks early in the history of artificial intelligence (Samuel, 1963; Samuel, 1967; Waterman, 1970; Fikes, Hart & Nilsonn, 1972). Multiple-step tasks are tasks that can only be solved via a sequence of decisions, such as control problems, robotics problems, classic problem-solving, and game-playing. The objective of agents attempting to learn such tasks is to use the resources they have available in order to become more proficient at the tasks. In particular, each agent attempts to develop a good policy, a mapping from states to actions, that allows it to select actions that optimize a measure of its performance on the task; for example, reducing the number of steps necessary to complete the task successfully. Our study focuses on reinforcement learning, a set of learning techniques where the learner performs trial-and-error experiments in the task and adapts its policy based on the outcome of those experiments. Much of the work in reinforcement learning has focused on a particular, simple representation, where every problem state is represented explicitly in a table, and associated with each state are the actions that can be chosen in that state. A major advantage of this table lookup representation is that one can prove that certain reinforcement learning techniques will develop an optimal policy for the current task. The drawback is that the representation limits the application of reinforcement learning to multiple-step tasks with relatively small state-spaces. There has been a little theoretical work that proves that convergence to optimal solutions can be obtained when using generalization structures, but the structures are quite simple. The theory says little about complex structures, such as multi-layer, feedforward artificial neural networks (Rumelhart & McClelland, 1986), but empirical results indicate that the use of reinforcement learning with such structures is promising. These empirical results make no theoretical claims, nor compare the policies produced to optimal policies. A goal of our work is to be able to make the comparison between an optimal policy and one stored in an artificial neural network. A difficulty of performing such a study is finding a multiple-step task that is small enough that one can find an optimal policy using table lookup, yet large enough that, for practical purposes, an artificial neural network is really required. We have identified a limited form of the game OTHELLO as satisfying these requirements. The work we report here is in the very preliminary stages of research, but this paper provides background for the problem being studied and a description of our initial approach to examining the problem. In the remainder of this paper, we first describe reinforcement learning in more detail. Next, we present the game OTHELLO. Finally we argue that a restricted form of the game meets the requirements of our study, and describe our preliminary approach to finding an optimal solution to the problem.

Clouse, Jeff↗
DOE OSTI2018

Removal of Congo Red from Aqueous Solution by Hydroxyapatite Nanoparticles Loaded on Zein as an Efficient and Green Adsorbent: Response Surface Methodology and Artificial Neural Network-Genetic Algorithm

This study is based on the application of hydroxyapatite nanoparticles loaded on Zein (Zein/nHAp) as an efficient adsorbent for the removal of Congo red from aqueous solutions. The properties of the adsorbent were characterized using various techniques including FT-IR, XRD, FE-SEM, and BET. The influence of five parameters such as pH, temperature, contact time, initial dye concentration, and adsorbent dosage on the removal percentage of the dye was investigated. The optimum conditions of 5.83, 34.32 °C, 5.20 min, 392.10 ppm, and 0.007 g were achieved for pH, temperature, contact time, initial dye concentration, and adsorbent dosage, respectively. The maximum removal percentage of 99.48% was obtained under the optimum condition that exhibited high adsorption potential of the used adsorbent. Central composite design (CCD) under response surface methodology and artificial neural network-genetic algorithm (ANN-GA) were utilized for optimization of parameters. Comparison of the results of the two models in terms of coefficient of determination (R{sup 2}) and mean absolute percentage error confirmed the prediction potential of CCD and ANN-GA. Higher ability and accuracy of ANN-GA in prediction was found based on given results. The experimental equilibrium data were studied by Langmuir, Freundlich, Temkin and Dubinin-Radushkevic isotherm models and explored that the data well presented by Langmuir model with maximum adsorption capacity of 416.66 mg/g. The adsorption kinetic was well-fitted by the pseudo-second-order model. The thermodynamics of the adsorption displayed spontaneous and endothermic nature of the process. Regeneration investigation showed that Zein/nHAp can impressively be reused, indicating that the adsorbent was a promising one for the removal of Congo red from aqueous solution. Graphical Abstract: .

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2026

Radiation Damage Mechanisms in Oxide-Based Solid Electrolytes

Ion implantation is a promising approach to mitigate dendrite formation in metal anode solid-state batteries by modifying the chemical and mechanical properties of oxide-based electrolyte surfaces such as Li 7 La 3 Zr 2 O 12 (LLZO). Yet, the detailed atomistic mechanisms and potentially adverse side effects of ion implantation remain poorly understood. In this study, we used molecular dynamics simulations to elucidate the evolution of radiation-induced damage in LLZO. The results reveal that radiation damage is primarily driven by recoils from heavy ions in the lattice, producing antisite defect clusters whose density decreases as recoil energy increases, a counterintuitive finding that highlights complex cascade fragmentation. Radiation-induced disruption to the crystal sublattice connectivity, especially in the La–O network critical for Li-ion transport, leads to degradation in ionic pathways at low energies, while higher energy cascades can form new conduction pathways via enhanced Zr–O network connectivity. These findings provide a mechanistic foundation for optimizing ion implantation strategies that balance disruption of the lattice for surface modification for dendrite resistance with preservation of essential ionic conductivity.

Defects↗
DOE OSTI2024

Comparative genomic analysis of thermophilic fungi reveals convergent evolutionary adaptations and gene losses

Thermophily is a trait scattered across the fungal tree of life, with its highest prevalence within three fungal families (Chaetomiaceae, Thermoascaceae, and Trichocomaceae), as well as some members of the phylum Mucoromycota. We examined 37 thermophilic and thermotolerant species and 42 mesophilic species for this study and identified thermophily as the ancestral state of all three prominent families of thermophilic fungi. Thermophilic fungal genomes were found to encode various thermostable enzymes, including carbohydrate-active enzymes such as endoxylanases, which are useful for many industrial applications. At the same time, the overall gene counts, especially in gene families responsible for microbial defense such as secondary metabolism, are reduced in thermophiles compared to mesophiles. We also found a reduction in the core genome size of thermophiles in both the Chaetomiaceae family and the Eurotiomycetes class. The Gene Ontology terms lost in thermophilic fungi include primary metabolism, transporters, UV response, and O-methyltransferases. Comparative genomics analysis also revealed higher GC content in the third base of codons (GC3) and a lower effective number of codons in fungal thermophiles than in both thermotolerant and mesophilic fungi. Furthermore, using the Support Vector Machine classifier, we identified several Pfam domains capable of discriminating between genomes of thermophiles and mesophiles with 94% accuracy. Using AlphaFold2 to predict protein structures of endoxylanases (GH10), we built a similarity network based on the structures. We found that the number of disulfide bonds appears important for protein structure, and the network clusters based on protein structures correlate with the optimal activity temperature. Thus, comparative genomics offers new insights into the biology, adaptation, and evolutionary history of thermophilic fungi while providing a parts list for bioengineering applications.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2019

Characterization of large-volume Frisch grid detector fabricated from as-grown CdZnTeSe

CdZnTe (CZT) has been under continuous development as a room-temperature radiation detector for the past 2-3 decades. Due to its intrinsic defects, such as a high concentration sub-grain boundary network, it has been very challenging to consistently produce high-quality CZT detectors with high yield, particularly for detector volumes exceeding a few cm 3 . In this paper, we report characterization of a relatively large-volume Frisch-grid detector (dimensions: 11x10.8x19.4 mm 3 ) fabricated from a new material Cd 0.9 Zn 0.1 Te 0.985 Se 0.015 . Furthermore, the detector was fabricated from an as-grown ingot with a two-inch diameter grown by the Traveling Heater Method (THM). The detector sample was evaluated by infrared (IR) transmission microscopy to study the presence of Te inclusions. X-ray topography using a synchrotron light source was deployed to measure the presence of sub-grain boundaries and their networks. Finally, the detector performance was measured for an optimized Frisch-grid length to extract the best energy resolution at room temperature.

36 MATERIALS SCIENCE↗
DOE OSTI2019

Runtime Concurrency Control and Operation Scheduling for High Performance Neural Network Training

In this paper, we extend an existing runtime system (the TensorFlow runtime) to enable automatic concurrency control and scheduling of operations. We explore performance modeling to predict the performance of operations with various thread- level parallelism. Our performance model is highly accurate and lightweight. Leveraging the performance model, our runtime system employs a set of scheduling strategies that co-run opera- tions to improve hardware utilization and system throughput. Our runtime system demonstrates a significant performance benefit. Comparing with using the recommended configurations for concurrency control and operation scheduling in TensorFlow, our approach achieves 36% performance (execution time) im- provement on average (up to 49%) for four neural network models, and achieves high performance close to the optimal one manually obtained by the user.

Liu, Jiawen↗
DOE OSTI2018

Regression for Radioactive Source Detection

Detection of radioactive sources is an important capability that led to the deployment of networks for detecting and localizing low-level, hazardous radiation sources. It is generally expected that such networks outperform the individual detectors by intelligently fusing information from several dispersed sensors. In this paper, we develop a network detection method that uses the slope of a linear regression fit as a test statistic for detecting a point radioactive source within a field of detectors. In our regression model, we compute a least-squared linear fit between the average radiation counts at the detectors and the inverse-squared distances of known detector locations to an estimated source location. We show that the slope of this regression fit is an estimate of the source intensity and can be used as a threshold for source detection purposes. We compare performance of our proposed detection method with that of a fusion-based Sequential Probability Ratio Test (SPRT) method. For performance analyses, two datasets from the Domestic Nuclear Detection Ofces Intelligence Radiation Sensors Systems (IRSS) outdoor tests are used. Each of these tests consists of several runs of a single radioactive source moving in and out of a detector network. We present receiver operating characteristic (ROC) curves and optimal threshold values for the performance each detection method, and determine that our detection method using linear regression fit has slightly better overall performance than SPRT method.

Cordone, G.↗
DOE OSTI2020

Data-Driven Preemptive Voltage Monitoring and Control Using Probabilistic Voltage Sensitivities

Increased penetration levels of distributed variable renewable generation can cause random voltage fluctuations and violations at multiple nodes. Traditional methods of voltage control typically involve reactionary responses of capacitor banks, tap changers, and recently even smart inverters. But because of the lack of foresight in voltage violations, these controls are ineffective to completely mitigate the issue. Therefore, new methods of predicting voltage violations subject to random power injection changes in the distribution network are needed, which can be used to guide optimal and dynamic methods of voltage control. This work lays the foundation for such preemptive voltage monitoring and control by proposing an analytical and sensor data-driven voltage sensitivity analysis method. Driven by stochastic data and forecasts, the method can be used to develop probabilistic voltage sensitivities and consequently to predict system nodes with high likelihood of voltage limit violations. The effectiveness of this method is tested on IEEE 69-node distribution system integrated with distributed solar. The results demonstrate the proposed method's ability to successfully predict nodes with high probability of voltage violations for a specific time-series simulation. The results also demonstrate the ability to guide timely power injection control actions to mitigate future voltage violations.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2018

Distribution Locational Marginal Pricing (DLMP) for Congestion Management and Voltage Support

In this paper, a day-ahead market-clearing model for smart distribution systems is proposed. Various types of distributed energy resources (DERs), such as distributed energy storage, distributed generators, microgrids, and load aggregators, can bid into the day-ahead distribution-level electricity market. Considering system Volt/VAR control, network reconfiguration, and interactions with the wholesale market, an optimization model is built to clear the day-ahead market, through which the distribution locational marginal pricing (DLMPs) for both active power and reactive power are determined. Through derivations of the Lagrangian function and sensitivity factors, DLMPs are decomposed to five components (i.e., marginal costs for active power, reactive power, congestion, voltage support, and loss), which provide price signals to motivate DERs to contribute to congestion management and voltage support. Finally, case studies demonstrate the effectiveness of the proposed method.

congestion management↗
DOE OSTI2019

Cross‐compartment metabolic coupling enables flexible photoprotective mechanisms in the diatom Phaeodactylum tricornutum

Summary Photoacclimation consists of short‐ and long‐term strategies used by photosynthetic organisms to adapt to dynamic light environments. Observable photophysiology changes resulting from these strategies have been used in coarse‐grained models to predict light‐dependent growth and photosynthetic rates. However, the contribution of the broader metabolic network, relevant to species‐specific strategies and fitness, is not accounted for in these simple models. We incorporated photophysiology experimental data with genome‐scale modeling to characterize organism‐level, light‐dependent metabolic changes in the model diatom Phaeodactylum tricornutum . Oxygen evolution and photon absorption rates were combined with condition‐specific biomass compositions to predict metabolic pathway usage for cells acclimated to four different light intensities. Photorespiration, an ornithine‐glutamine shunt, and branched‐chain amino acid metabolism were hypothesized as the primary intercompartment reductant shuttles for mediating excess light energy dissipation. Additionally, simulations suggested that carbon shunted through photorespiration is recycled back to the chloroplast as pyruvate, a mechanism distinct from known strategies in photosynthetic organisms. Our results suggest a flexible metabolic network in P. tricornutum that tunes intercompartment metabolism to optimize energy transport between the organelles, consuming excess energy as needed. Characterization of these intercompartment reductant shuttles broadens our understanding of energy partitioning strategies in this clade of ecologically important primary producers.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2021

OPFLearn.jl [SWR-21-109]

OPFLearn.jl is a Julia package for creating datasets for machine learning approaches to solving AC optimal power flow (AC OPF). It was developed to provide researchers with a standardized way to efficiently create AC OPF datasets that are representative of more of the AC OPF feasible load space compared to typical dataset creation methods. The OPFLearn dataset creation method uses a relaxed AC OPF formulation to reduce the volume of the unclassified input space throughout the dataset creation process. Over time this input space tightens around the relaxed AC OPF feasible region to increase the percentage of feasible load profiles found while uniformly sampling the input space. Load samples are processed using AC OPF formulations from PowerModels.jl. More information on the dataset creation method can be found in our publication, "OPF-Learn: An Open-Source Framework for Creating Representative AC Optimal Power Flow Datasets". To use OPFLearn.jl a PowerModels network data dictionary is required (can be loaded from Matpower ".m" files) to define the network the dataset is being created for.

Joswig-Jones, Trager↗
DOE OSTI2023

Graph Neural Networks and Applied Linear Algebra v.1.0

SAND2024-01365O The Graph Neural Networks and Applied Linear Algebra is companion software for the educational article with the same title. The software provides illustrative examples of graph neural networks in Matlab and Python. These stand-alone algorithms are for educational purposes. The software also includes graph neural network-based algorithms for a trainable Jacobi iteration as well as diffusion coefficient estimation. The software provides human-interpretable implementations of Graph Neural Networks in Matlab and Python. These implementations are not optimized for performance and instead emphasize readability. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

SciDAC↗
DOE OSTI2019

Self-powered Wireless Dual-mode Langasite Sensor for Pressure/Temperature Monitoring of Nuclear Reactors

Safety issues in nuclear power plants are of extreme importance. Numerous sensors are used within nuclear power plants for sensing and monitoring of its critical operation parameters. As currently implemented, during massive losses of power across the plant in accidents, the backup batteries for the monitoring sensors will ultimately be drained and can cease to operate. The overall objective of this NEUP project is to develop a novel self-powered wireless hybrid sensor which can accurately monitor both pressure and temperature using a single device, even in the extreme harsh environments of severe nuclear accidents, without the requirement of external electricity. At Virginia Tech side, a heat pipe assisted thermoelectric generator (TEG) energy harvester and associated power management and data transmission circuits were developed. The energy harvester can produce electrical power continuously, even in loss of plant and external power situations, providing power for important data transmission for accident diagnosis. The self-powered sensing system could be integrated directly onto key nuclear components including pipes, pump housings, heat exchangers, reactor vessels, and shielding structures inside and outside of the reactor core. Two TEG based energy harvesters which can work in relatively low (50-250 ºC) and high temperature ranges (250-350 ºC) were established and tested. In the prototypes, heat pipes were introduced to reduce the heat resistances of the device at the hot and cold ends. The power output of the heat pipe assisted TEG module is 2.22 W, which is 6 times more than one using an aluminum rod. In-lab tests showed that these devices canmeet the requirement to power the sensors and circuit during normal and off-normal conditions.Amodel was then built to analyze the thermal network in the energy harvester, and was used to optimize its performance and reduce its geometry size. The design and testing of the vibrational energy harvester were completed. The harvester, has shown to effectively increase the vibrational potential of the main coolant pump, though it has been shown that thermal energy has a much higher potential for the powering of sensors. In this project, a self-start charger circuit for the energy harvester was designed to regulate and manage the power. An energy storage element was incorporated into the energy harvesting module for sensors and wireless transceivers in an off-normal situation. The signal conditioning and wireless communication electronic circuits were developed and tested in the lab. An important consideration for an energy harvester within a nuclear power plant is the level of radiation it will encounter during operation. The two main sources of radiation in a nuclear power plant are gamma and neutron radiation. Aradiation experiment was conducted in the Westinghouse Electric Company during which the prototypes were placed in a 106 rad radiation environment to test their ability to operate in the extreme environment. The results showed that this level of radiation had little effect on the performance of the TEGmaterial and heat pipe heat exchanging system. Its influence on the circuit was still under research, however the need for more development on radiation hardened electronics is needed. Pressure measurement based on piezoelectric crystal resonators has advantages over the commercially available methods like capacitive, electromagnetic, optical, potentiometric, and piezo resistive strain gauges. Since the output of piezoelectric based sensors are digital in format, it has a very high resolution, high accuracy, and long-term stability. Quartz is one of the commonly used piezoelectric material in digital pressure sensors design. This material is elastic in nature, providing high stability, high repeatability, high elasticity, repeatable mechanical behavior for many cycles of operation, and is free of hysteresis. The quartz resonator has a high Q, which means that very high vibration can be driven with the very low electric power. Langasite (LGS) is a new promising piezoelectric material that combines the advantageous properties of quartz with a better performance at high temperatures. LGS will not undergo any phase transitions up to its melting temperature of 1473 C, and hence, the application of Langasite in measuring pressure at high temperatures is promising. In this project, the design and fabrication of the piezo oscillator crystal for temperature and pressure measurement has been completed. The entire sensor has been fabricated and tested to see its sensitivity to temperature and pressure changes. The design and optimization of the radiation shielding systems that can effective protect the duel-mode Langasite sensor inside reactor core were demonstrated. To maximize the radiation shielding effectiveness, different types of materials are investigated based upon recent developments of radiation shielding systems including aluminum + Boron carbide and bismuth borosilicate glass B2O3 in addition to tungsten/B4C. New materials based upon metatheses is included as potential shielding to mitigate neutron fluencies. The radiation shielding effectiveness of selected composites are evaluated within a broader neutron spectrum from thermal, epithermal, fast (1 MeV) and 15 MeV neutrons. Performance analysis was also conducted to evaluate the effects of selected composite systems.Aconceptual design was also proposed for the packaging and integration of the radiation shielding system with the sensor, and its applicability within the reactor core was discussed. At end of the project, all components necessary for a complete sensing and monitoring system, including a langasite sensor, electrodes, the TEG harvester, an energy storage device, and power management and transmission electronics were integrated to demonstrate feasibility in powering the sensor. Finally, integration of each component into a full package, and a full demonstration were done.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Initial Systems-Level Assessment of a Distributed Direct Air Capture System Concept at the Urban-scale (UrbanDAC)

Direct Air Capture (DAC) systems offer a promising solution for mitigating global carbon emissions by directly removing ambient carbon dioxide (CO 2 ) from the atmosphere. While future DAC facilities are typically envisioned as being large and centralized, small-scale systems present an alternative approach with advantages such as adaptability and lower uptake costs. By harnessing waste heat from the built environment, such small-scale systems become distributed DAC at the urban scale (UrbanDAC) that benefit from existing urban infrastructure, while presenting challenges such as identifying eligible buildings and sustainable transportation and storage of captured CO 2 . Collaborating with engineering experts and developers of a DAC unit that can be co-located with cooling towers of existing commercial buildings, this study explores the systems-level implications of UrbanDAC using a geographically explicit multi-decision criteria analysis (MCDA) framework. By considering various infrastructure and environmental factors, network analysis and geospatial techniques are applied to identify optimal building candidates for distributed DAC units within Knoxville, Tennessee, USA, as a representative mid-size city. The selected outputs of the MCDA are used to explore a scenario that assumes a CO 2 collection and transport route for 20 high-ranking candidate buildings; total carbon emissions, EV energy consumption, and net carbon dioxide removal (CDR) are then calculated. Results suggest that the spatial variation of optimal candidates between thriving commercial areas is an important planning consideration. Examining the feasibility of UrbanDAC at an urban planning level provides valuable insights into the barriers and enabling conditions for CDR in cities, where the vast majority of CO 2 emissions are produced, and supports decision-making processes for the implementation of decarbonization initiatives. Through this initial assessment, this research acts as a pilot study for an emerging technology that highlights the importance of distributed DAC technologies in addressing climate change and emphasizes the need for further research and exploration in this domain.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2023

Integrating Survival Analysis with Bayesian Statistics to Forecast the Remaining Useful Life of a Centrifugal Pump Conditional to Multiple Fault Types

To improve the viability of nuclear power plants, there is a need to reduce their operational costs. Operational costs account for a significant portion of a plant’s yearly budget, due to their scheduled-based maintenance approach. In order to reduce these costs, proactive methods are required that estimate and forecast the state of a machine in real time to optimize maintenance schedules. In this research, we use Bayesian networks to develop a framework that can forecast the remaining useful life of a centrifugal pump. To do so, we integrate survival analysis with Bayesian statistics to forecast the health of the pump conditional to its current state. We complete our research by successfully using the Bayesian network on a case study. This solution provides an informed probabilistic viewpoint of the pumping system for the purpose of predictive maintenance.

42 ENGINEERING↗
DOE OSTI2025

Machine Learning-Driven Conservative-to-Primitive Conversion in Hybrid Piecewise Polytropic and Tabulated Equations of State

We present a novel machine learning (ML)-based method to accelerate conservative-to-primitive inversion, focusing on hybrid piecewise polytropic and tabulated equations of state. Traditional root-finding techniques are computationally expensive, particularly for large-scale relativistic hydrodynamics simulations. To address this, we employ feedforward neural networks (NNC2PS and NNC2PL), trained in PyTorch (2.0+) and optimized for GPU inference using NVIDIA TensorRT (8.4.1), achieving significant speedups with minimal accuracy loss. The NNC2PS model achieves 𝐿 1 and 𝐿 ∞ errors of 4.54 × 10 −7 and 3.44 × 10−6, respectively, while the NNC2PL model exhibits even lower error values. TensorRT optimization with mixed-precision deployment substantially accelerates performance compared to traditional root-finding methods. Specifically, the mixed-precision TensorRT engine for NNC2PS achieves inference speeds approximately 400 times faster than a traditional single-threaded CPU implementation for a dataset size of 1,000,000 points. Ideal parallelization across an entire compute node in the Delta supercomputer (dual AMD 64-core 2.45 GHz Milan processors and 8 NVIDIA A100 GPUs with 40 GB HBM2 RAM and NVLink) predicts a 25-fold speedup for TensorRT over an optimally parallelized numerical method when processing 8 million data points. Moreover, the ML method exhibits sub-linear scaling with increasing dataset sizes. We release the scientific software developed, enabling further validation and extension of our findings. By exploiting the underlying symmetries within the equation of state, these findings highlight the potential of ML, combined with GPU optimization and model quantization, to accelerate conservative-to-primitive inversion in relativistic hydrodynamics simulations.

conservative-to-primitive conversion↗