Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “input data”

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 1,117 records · Page 62
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Evaluating the feasibility of using downwind methods to quantify point source oil and gas emissions using continuously monitoring fence-line sensors

The dependable reporting of methane (CH 4 ) emissions from point sources, such as fugitive leaks from oil and gas infrastructure, is important for profit maximization (retaining more hydrocarbons), evaluating climate impacts, assessing CH 4 fees for regulatory programs, and validating CH 4 intensity in differentiated gas programs. Currently, there are disagreements between emissions reported by different quantification techniques for the same sources. It has been suggested that downwind CH 4 quantification methods using CH 4 measurements on the fence line of production facilities could be used to generate emission estimates from oil and gas operations at the site level, but it is currently unclear how accurate the quantified emissions are. To investigate the accuracy of downwind methods, this study uses fence-line simulated data collected during controlled-release experiments as input for a non-standard closed-path eddy covariance (EC), the Gaussian plume inverse model (GPIM), and the backward Lagrangian stochastic (bLs) model in a range of atmospheric conditions. This study's EC attempt was unsuccessful due to data collection and instrumentation issues, resulting in invalid results characterized by underestimated emissions, large negative fluxes, and cospectra/ogives that deviated from their ideal shapes. Consequently, the EC results could not be compared with the GPIM and bLS model. The bLs model demonstrated the highest accuracy for single-release single-point emissions, though it exhibited greater uncertainty than GPIM under multi-release conditions. Across the GPIM and bLs model, the most reliable quantification was achieved with 15 min averaging and a narrow 5° wind sector range. Although EC was limited in this context, future studies should consider employing a standard EC system and further optimizing GPIM and bLs approaches – particularly for complex multi-source scenarios – to enhance quantification accuracy and reduce uncertainty.

03 NATURAL GAS↗
DOE OSTI2026

Estimating Flexibility Envelopes for Residential Customers From Utility Smart Meter Data: Preprint

Demand response from residential customers has significant potential to support power system operations, but accurate flexibility estimation is challenging due to the limited resolution of advanced metering infrastructure (AMI) data. Most utility AMI measurements are recorded at hourly intervals, with only a small portion at higher resolutions, and even fewer households have appliance-level energy usage data. To address this issue, this paper proposes a two-stage long short-term memory (LSTM) framework for estimating household flexibility envelopes from low-resolution AMI data. In the first stage, the heating, ventilating, and air-conditioning (HVAC) load and non-HVAC loads are estimated by using a model trained on a small set of households with appliance-level profiles. These estimated data are then used to compute the upper- and lower-flexibility bounds, which are subsequently down-sampled to lower-resolution data. In the second stage, these flexibility bounds serve as training inputs for another LSTM model, enabling direct prediction of flexibility envelopes for households with only hourly AMI data. This method is validated using Pecan Street data from two different areas, and the results demonstrate its applicability and effectiveness.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2023

Cosmogenic 137 Xe Analysis of Alternatives for nEXO

Neutrinoless double-beta decay is a hypothetical nuclear decay in which two matter particles are produced without corresponding antimatter particles. Observation of neutrinoless double-beta decay would shed light on a potential explanation of matter-antimatter observed in the universe and also indicate the Majorana nature of neutrinos, meaning that the neutrino can act as its own antiparticle. The nEXO experiment is a proposed 5-ton liquid xenon time projection chamber detector with a projected 1.35 × 10 28 yr half-life sensitivity to neutrinoless double-beta decay in 136 Xe [J. Phys. G 49, 015104 (2022)]. While nEXO has a baseline location of SNOLAB, there are several other underground facilities in the world capable of hosting the experiment. The primary change at each site will be cosmogenic 137 Xe backgrounds, caused by varied cosmogenic muon activity at each site. This study finds that while there is a negligible increase in nEXO’s half-life sensitivity at 90% C.L. from improving the cosmogenic 137 Xe veto rejection procedure for SNOLAB, much higher daily muon rates at three other sites will have an adversely negative effect on nEXO’s sensitivity and would likely require further modifications to nEXO’s baseline design to reach nEXO’s scientific goals. Monte Carlo simulations of muons inducing 137 Xe backgrounds were conducted at four underground sites, SNOLAB, SURF, LNGS, and WIPP, each with roughly a factor of 10 times higher daily muon activity than the previous site. There were two primary inputs to these simulations. The first was sample data of cosmogenic muon flight through nEXO’s outer detector, a 12.8m tall and 12.3m wide cylindrical water tank lined with photomultiplier tubes to detect Cerenkov radiation, and all of the secondary interactions produced in the detector. The second input was a parametrization of 137 Xe* production tagging when 137 Xe de-excites and release prompt gammas, as well as accidental tagging caused by neutron capture on 63 Cu, 65 Cu, 19 F, and 1 H throughout the detector. Together, the number of backgrounds relative to the previous 2021 Sensitivity Paper [J. Phys. G 49, 015104 (2022)] was found at each of these four sites across three energy windows, from 400, 1400, and 2600 keV all up to 4600 keV, just above the end point of the prompt gamma spectrum for 137 Xe*. Using level curves for the sensitivity as 137 Xe backgrounds scale relative to baseline calculations, the half-life sensitivity at 90% C.L. for each of trial was found. Even with reduced 137 Xe backgrounds, the results for SNOLAB yielded neglible improvements in nEXO’s predicted sensitivity. On the contrary, the increased backgrounds for the three other considered sites yielded sensitivity estimates below nEXO’s previous projections. This study demonstrates that the shielding from cosmogenic muons at SNOLAB makes it highly preferred even amongst other world leading underground laboratories.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2021

Machine Learning for Gearbox Fault Prediction by Using Both Scada and Modeled Data

This presentation outlines the work in the paper titled "Prognostics of Wind Turbine Gearbox Bearing Failures Using SCADA and Modeled Data" published by the PHM Society and presented at its 2020 annual conference. It is accessible at https://papers.phmsociety.org/index.php/phmconf/article/download/1292/862. The technical work is on machine learning approaches for prognostics for gearbox faults. The methodology combines SCADA time series data and physics domain modeling data, derived from the models developed by the NREL team, as inputs to machine learning models to predict gearbox bearing failures with one month lead time. Based on SCADA data, modeled data, and bearing failure log data from an actual wind plant, the performances of different machine-learning models on unseen data are then evaluated using industry-standard metrics such as precision, recall, and F1 score, and AUC (area under receiver operating characteristic curve). Results show the overall system performance enhancement in predicting bearing failure when modeled data are included with SCADA data. The reduction in terms of false alarms is about 50%, and improvement in terms of precision, and F1 score, and AUC is about 33%, and 12%, and 6% respectively, based on the best modeling case in this study.

49 EE - Wind and Water Power Program - Wind (EE-4W↗
DOE OSTI2022

A multi-stage stochastic programming model for adaptive biomass processing operation under uncertainty

Variations of physical and chemical characteristics of biomass reduce equipment utilization and increase operational costs of biomass processing. Biomass processing facilities use sensors to monitor the changes in biomass characteristics. Integrating sensory data into the operational decisions in biomass processing will increase its flexibility to the changing biomass conditions. In this paper, we propose a multi-stage stochastic programming model that minimizes the expected operational costs by identifying the initial inventory level and creating an operational decision policy for equipment speed settings. These policies take the sensory information data and the current biomass inventory level as inputs to dynamically adjust inventory levels and equipment settings according to the changes in the biomass' characteristics. We ensure that a prescribed target reactor utilization is consistently achieved by penalizing the violation of the target reactor feeding rate. A case study is developed using real-world data collected at Idaho National Laboratory's biomass processing facility. We show the value of multi-stage stochastic programming from an extensive computational experiment. Our sensitivity analysis indicates that updating the infeed rate of the system, the processing speed of equipment, and bale sequencing based on the moisture level of biomass improves the processing rate of the reactor and reduces operating costs.

09 BIOMASS FUELS↗
DOE OSTI2026

Conditional distribution estimation of building characteristics with diffusion models for urban energy modeling

Understanding current energy consumption behavior in communities is critical for informing future energy use decisions and enabling efficient energy management. Urban energy models, which are used to simulate these energy use patterns, require large datasets with detailed building characteristics for accurate outcomes. However, such detailed characteristics at the individual building level are often unknown and costly to acquire, or unavailable. Through this work, we propose using a generative modeling approach to generate realistic building attributes to fill in the data gaps and finally provide complete characteristics as inputs to energy models. Our model learns complex, building-level patterns from training on a large-scale residential building stock model containing 2.2 million buildings. We employ a tabular diffusion-based framework that is designed to handle heterogeneous (discrete and continuous) features in tabular building data, such as occupancy, floor area, heating, cooling, and other equipment details. We develop a capability for conditional diffusion, enabling the imputation of missing building characteristics conditioned on known attributes. We conduct a comprehensive validation of our conditional diffusion model, firstly by comparing the generated conditional distributions against the underlying data distribution, and secondly, by performing a case study for a Baltimore residential region, showing the practical utility of our approach. Our work is one of the first to demonstrate the potential of generative modeling to accelerate building energy modeling workflows.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2020

SAM Plug-in Development (Phase I Final Report)

The DOE Office of Nuclear Energy (NE) has created an extensive set of advanced modeling and simulation tools for nuclear engineering analysis. The advanced capabilities of these newer analysis codes require more in-depth training, skills, and knowledge in order to effectively utilize them for the design, analysis, and licensing of advanced nuclear systems and experiments. A high learning curve for inexperienced users may deter organizations from incorporating these tools into their internal processes. This project involved development of a plug-in to the Symbolic Nuclear Analysis Package (SNAP) for the System Analysis Module (SAM) tool. SAM is an advanced system analysis tool for reactor transient analyses being developed at Argonne National Laboratory under the U.S. DOE Office of Nuclear Energy’s Nuclear Energy Advanced Modeling and Simulation (NEAMS) program. SAM utilizes an object-oriented application framework (MOOSE), and its underlying meshing and finite-element library (libMesh) and linear and non-linear solvers (PETSc), to leverage modern advanced software environments and numerical methods. SNAP provides a highly flexible framework for creating, modifying and documenting input for engineering analysis codes such as SAM as well as extensive functionality for submitting, monitoring, and interacting with the codes through an intuitive graphical user interface (GUI). The common user interface provided by SNAP minimizes the learning curve for engineers starting with a new analysis code and provides an intuitive framework for transitioning between different analysis codes. SNAP provides a powerful but intuitive interface to facilitate access to advanced modeling and simulation tools for inexperienced users. Unlike many “form based” GUI’s, SNAP maps each engineering code’s component input to an internal database which manages all component input parameters along with component interconnections. This level of abstraction permits SNAP to support several advanced capabilities such as renodalization, model validation and consistency checks, embedded documentation, model notebook generation, data ownership and reviewer tracking, and variable assignment for inputs to name a few. SNAP includes a built-in Python interpreter and is interfaced to several commercial and open source packages including CPython, MATLAB/OCTAVE, Microsoft Office, Open Office, and SANDIA’s DAKOTA package which provides Uncertainty Quantification analysis through the SNAP plug-ins. Phase I of this project involved development a fully functional basic SAM plug-in to SNAP. This plug-in provides the ability to import existing models, graphically construct, edit and submit models using SNAP’s extensive functionality.

99 GENERAL AND MISCELLANEOUS↗
DOE OSTI2023

2023 Data Science Challenge Notebooks

THE DATA SCIENCE INSTITUTE WILL BE INVITING STUDENTS FROM UC MERCED AND UC RIVERSIDE TO PARTICIPATE IN A DATA SCIENCE CHALLENGE AT LLNL. IN THE CHALLENGE, THE STUDENTS WILL EXPLORE A DATA-DRIVEN APPROACH TO RECONSTRUCTING ELECTROANATOMICAL MAPS OF THE HEART AT CLINICALLY RELEVANT RESOLUTIONS, COMBINING INPUT FROM THE STANDARD 12-LEAD ECG WITH ADVANCED MACHINE LEARNING TECHNIQUES. THE DATA THE STUDENTS WILL HAVE ACCESS TO HAS BEEN IM REVIEWED FOR EXTERNAL RELEASE. (LLNL-MI-835833) THESE NOTEBOOKS WILL PROVIDE THE STUDENTS WITH INSTRUCTIONS OF HOW TO LOAD THE DATA AND WORK WITH THE DATA, ALONG WITH TUTORIALS TO DO BASIC MACHINE LEARNING TASKS LIKE LOGISTIC REGRESSION AND IMAGE CLASSIFICATION.

Landajuela Larma, Mikel↗
DOE OSTI2021

Brady Geodatabase for Geothermal Exploration Artificial Intelligence

These files contain the geodatabases related to Brady's Geothermal Field. It includes all input and output files for the Geothermal Exploration Artificial Intelligence. Input and output files are sorted into three categories: raw data, pre-processed data, and analysis (post-processed data). In each of these categories there are six additional types of raster catalogs which are titled Radar, SWIR, Thermal, Geophysics, Geology, and Wells. These inputs and outputs were used with the Geothermal Exploration Artificial Intelligence to identify indicators of blind geothermal systems at the Brady Hot Springs Geothermal Site. The included zip file is a geodatabase to be used with ArcGIS and the tar file is an inclusive database that encompasses the inputs and outputs for the Brady Hot Springs Geothermal Site.

15 GEOTHERMAL ENERGY↗
DOE OSTI2024

Interpretable Machine Learning Models for Practical Antimonate Electrocatalyst Performance

Computationally predicting the performance of catalysts under reaction conditions is a challenging task due to the complexity of catalytic surfaces and their evolution in situ, different reaction paths, and the presence of solid-liquid interfaces in the case of electrochemistry. We demonstrate here how relatively simple machine learning models can be found that enable prediction of experimentally observed onset potentials. Inputs to our model are comprised of data from the oxygen reduction reaction on non-precious transition-metal antimony oxide nanoparticulate catalysts with a combination of experimental conditions and computationally affordable bulk atomic and electronic structural descriptors from density functional theory simulations. From human-interpretable genetic programming models, we identify key experimental descriptors and key supplemental bulk electronic and atomic structural descriptors that govern trends in onset potentials for these oxides and deduce how these descriptors should be tuned to increase onset potentials. Here, we finally validate these machine learning predictions by experimentally confirming that scandium as a dopant in nickel antimony oxide leads to a desired onset potential increase. Macroscopic experimental factors are found to be crucially important descriptors to be considered for models of catalytic performance, highlighting the important role machine learning can play here even in the presence of small datasets.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Effects of experimental and seasonal drying on soil microbial biomass and nutrient cycling in four lowland tropical forests

Changes in precipitation represent a major effect of climate change on tropical forests, which contain some of the earth’s largest terrestrial carbon (C) stocks. Such changes are expected to influence microbes, nutrients, and the fate of C in tropical forest soils. To explore this, we assessed soil microbial biomass, potential extracellular enzyme activities, and nutrient availability in a partial throughfall exclusion experiment in four seasonal lowland tropical humid forests in Panama with wide variation in precipitation and soil fertility. We hypothesized that throughfall exclusion would reduce microbial biomass and activity and accentuate dry season soil nutrient accumulation, with larger effects in wetter, less droughtresistant forests. We observed a baseline seasonal pattern of decreased microbial biomass and increased extractable dissolved organic C (DOC), total dissolved nitrogen (TDN), nitrate (NO 3 – ), and resinextractable phosphorus (P) in the dry season, with the strongest patterns for nitrogen (N). However, potential enzyme activities showed no consistent seasonality. In line with seasonal drying, throughfall exclusion decreased soil microbial biomass in the wet season and increased TDN and NO 3 – , especially in the dry season. In contrast to seasonal drying, throughfall exclusion decreased DOC and did not affect resinextractable P, but slightly decreased potential phosphatase activities. Potential enzyme activities varied among sites and sampling times, but did not explain much variation in microbial biomass or substrate availability. We conclude that reduced rainfall in tropical forests might accentuate some dry season patterns, like reductions in microbial biomass and accumulation of extractable nutrients. However, our data also suggest new patterns, like reduced inputs of DOC to soils with drying, which could have cascading effects on soil ecological function and C storage.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Latent map Gaussian processes for mixed variable metamodeling

Gaussian processes (GPs) are ubiquitously used in sciences and engineering as metamodels. Standard GPs, however, can only handle numerical or quantitative variables. Here we introduce latent map Gaussian processes (LMGPs) that inherit the attractive properties of GPs and are also applicable to mixed data which have both quantitative and qualitative inputs. The core idea behind LMGPs is to learn a continuous, low-dimensional latent space or manifold which encodes all qualitative inputs. To learn this manifold, we first assign a unique prior vector representation to each combination of qualitative inputs. We then use a low-rank linear map to project these priors on a manifold that characterizes the posterior representations. As the posteriors are quantitative, they can be directly used in any standard correlation function such as the Gaussian or Matern. Hence, the optimal map and the corresponding manifold, along with other hyperparameters of the correlation function, can be systematically learned via maximum likelihood estimation. Through a wide range of analytic and real-world examples, we demonstrate the advantages of LMGPs over state-of-the-art methods in terms of accuracy and versatility. In particular, we show that LMGPs can handle variable-length inputs, have an explainable neural network interpretation, and provide insights into how qualitative inputs affect the response or interact with each other. We also employ LMGPs in Bayesian optimization and illustrate that they can discover optimal compound compositions more efficiently than conventional methods that convert compositions to qualitative variables via manual featurization.

42 ENGINEERING↗
DOE OSTI2025

SparcleQC: Automated Input File Creation for QM/MM Studies of Protein:Ligand Complexes

SparcleQC is a Python package that, given a protein:ligand complex in the Protein Data Bank (PDB) file format, can create quantum mechanics/molecular mechanics (QM/MM)-like input files for the electronic structure theory packages PSI4, QChem, and NWChem. The resulting input files include quantum mechanical representations of the ligand and a small section of the protein, surrounded by point charges that represent the rest of the protein. Creation of these QM/MM input files includes cutting and capping the QM subregion, obtaining point charges for the protein, and adjusting charges at the QM/MM boundary; and each of these tasks are automated by the software. In this article, we describe the details of SparcleQC’s procedure, show examples of the Python API, and explain additional features that are helpful in protein:ligand interaction studies. Finally, we show that SparcleQC enables automated preparation of input files for QM/MM calculations, which can return can return accurate interaction energies in minutes, while a fully quantum mechanical computation on the protein:ligand complex could take days, if it is even possible.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Operating advanced scientific instruments with AI agents that learn on the job

Advanced scientific user facilities, such as next generation X-ray light sources and self-driving laboratories, are revolutionizing scientific discovery by automating routine tasks and enabling rapid experimentation and characterizations. However, these facilities must continuously evolve to support new experimental workflows, adapt to diverse user projects, and meet growing demands for more intricate instruments and experiments. This continuous development introduces significant operational complexity, necessitating a focus on usability, reproducibility, and intuitive human-instrument interaction. In this work, we explore the integration of agentic AI, powered by Large Language Models (LLMs), as a transformative tool to achieve this goal. We present our approach to developing a human-in-the-loop pipeline for operating advanced instruments including an X-ray nanoprobe beamline and an autonomous robotic station dedicated to the design and characterization of materials. Specifically, we evaluate the potential of various LLMs as trainable scientific assistants for orchestrating complex, multi-task workflows, which also include multimodal data, optimizing their performance through optional human input and iterative learning. We demonstrate the ability of AI agents to bridge the gap between advanced automation and user-friendly operation, paving the way for more adaptable and intelligent scientific facilities.

Large Language Models↗
DOE OSTI2023

O’Hare Airport Short-Term Ground Transportation Modal Demand Forecast Using Gaussian Processes

Here, the principal objective of this study is to analyze the spatial and temporal variation of ground transportation airport demand and provide demand forecast to inform planning capability and explore alternatives for investments to accommodate airport growth. Because of its good adaptability and strong generalization ability for dealing with high-dimensional input, small-sample, and nonlinear spatial data, Gaussian process (GP) regression is used to provide forecast estimates using data from transportation network company (TNC) trips and urban rail passengers at Chicago's O'Hare International Airport. TNC airport trips differ significantly, with three times more distance, more than twice the travel time, and half of the share requests compared with nonairport trips. This highlights the need for separate demand models. Hourly analysis of the rail service indicates that this is likely heavily used by airport workers, whereas TNC services focus on travelers because of variations in the peak demand hours. Heteroscedastic GP regression is implemented because of differences in trip variance between night and day hours. Estimates are given for weekdays and weekend trips, and the 95% confidence intervals are calculated. The introduction of flight schedule information into the models shows marginal improvements in their performance. However, fitting a GP regression becomes computationally expensive with increased sample size and the introduction of spatial components. Transportation planners and policymakers can use the results and methods implemented in this study to optimize transportation assets and provide long-range simulations of the current and future conditions in the area.

42 ENGINEERING↗
DOE OSTI2021

Rapid 3D nanoscale coherent imaging via physics-aware deep learning

Phase retrieval, the problem of recovering lost phase information from measured intensity alone, is an inverse problem that is widely faced in various imaging modalities ranging from astronomy to nanoscale imaging. The current process of phase recovery is iterative in nature. As a result, the image formation is time consuming and computationally expensive, precluding real-time imaging. Here, we use 3D nanoscale X-ray imaging as a representative example to develop a deep learning model to address this phase retrieval problem. We introduce 3D-CDI-NN, a deep convolutional neural network and differential programing framework trained to predict 3D structure and strain, solely from input 3D X-ray coherent scattering data. Our networks are designed to be “physics-aware” in multiple aspects; in that the physics of the X-ray scattering process is explicitly enforced in the training of the network, and the training data are drawn from atomistic simulations that are representative of the physics of the material. We further refine the neural network prediction through a physics-based optimization procedure to enable maximum accuracy at lowest computational cost. 3D-CDI-NN can invert a 3D coherent diffraction pattern to real-space structure and strain hundreds of times faster than traditional iterative phase retrieval methods. Our integrated machine learning and differential programing solution to the phase retrieval problem is broadly applicable across inverse problems in other application areas.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
DOE OSTI2022

Shock initiation of low density polymer bonded explosive LX-14: A study of two morphologies

A series of six shock initiation experiments have been carried out on low density LX-14 powder in order to simulate a shock insult on the heavily damage polymer bonded explosive, LX-14. Two distinct morphologies were studied (tap density molding powder and machined swarf), both at the same density of 0.942 g/cm 3 , or 50.1% theoretical maximum density. The purpose of these experiments was to provide data to help make an assessment of the effects that damage has on the material sensitivity to a planar shock. This was achieved primarily by providing shock sensitivity data in the form of a Pop plot, and also reactants equation of state data, which aids in the determination of input conditions for both the experiments performed in this work and also future experiments of this material type. The experiments were of a cut-back format, consisting of four sample heights on each shot and diagnosed with optical velocimetry. The experiments were carried out at the Technical Area 40 Chamber 9 gas gun facility at the Los Alamos National Laboratory, where the LX-14 targets were subjected to Al 6061 and Oxygen Free High Conductivity copper impactors launched to velocities up to 2.14 km/s. Time corrected reactive growth wave profiles are presented in this paper along with the derivation of the following Hugoniot parameters for this explosive, where the molding powder and machine swarf are represented by the following linear equations, respectively, U s = 1.64 (±0.64)u p + 2.25 (±0.49) and U s = 0.87 (±.0.60)u p + 4.08 (±0.49). The results show that the steady increase in shock sensitivity with increasing void fraction reaches an inflection point beyond which the shock sensitivity begins to decease. This inflection point lies between 65% and 50% of the theoretical maximum density of the LX-14. In both cases, the damaged LX-14 was not as sensitive as expected, with a relative shock sensitivity of the molding powder being less than the pressed LX-14, and the machined swarf having a shock sensitivity that is comparable to pressed LX-14.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2024

Investigating resource-efficient neutron/gamma classification ML models targeting eFPGAs

There has been considerable interest and resulting progress in implementing machine learning (ML) models in hardware over the last several years from the particle and nuclear physics communities. A big driver has been the release of the Python package, hls4ml, which has enabled porting models specified and trained using Python ML libraries to register transfer level (RTL) code. So far, the primary end targets have been commercial field-programmable gate arrays (FPGAs) or synthesized custom blocks on application specific integrated circuits (ASICs). However, recent developments in open-source embedded FPGA (eFPGA) frameworks now provide an alternate, more flexible pathway for implementing ML models in hardware. These customized eFPGA fabrics can be integrated as part of an overall chip design. In general, the decision between a fully custom, eFPGA, or commercial FPGA ML implementation will depend on the details of the end-use application. In this work, we explored the parameter space for eFPGA implementations of fully-connected neural network (fcNN) and boosted decision tree (BDT) models using the task of neutron/gamma classification with a specific focus on resource efficiency. We used data collected using an AmBe sealed source incident on Stilbene, which was optically coupled to an OnSemi J-series silicon photomultiplier (SiPM) to generate training and test data for this study. We investigated relevant input features and the effects of bit-resolution and sampling rate as well as trade-offs in hyperparameters for both ML architectures while tracking total resource usage. The performance metric used to track model performance was the calculated neutron efficiency at a gamma leakage of 10 -3 . The results of the study will be used to aid the specification of an eFPGA fabric, which will be integrated as part of a test chip.

47 OTHER INSTRUMENTATION↗