Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “model identification”

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 577 records · Page 32
All sourcesNASA NTRSDOE OSTI
DOE OSTI2015

Use of 16S rRNA gene for identification of a broad range of clinically relevant bacterial pathogens

According to World Health Organization statistics of 2011, infectious diseases remain in the top five causes of mortality worldwide. However, despite sophisticated research tools for microbial detection, rapid and accurate molecular diagnostics for identification of infection in humans have not been extensively adopted. Time-consuming culture-based methods remain to the forefront of clinical microbial detection. The 16S rRNA gene, a molecular marker for identification of bacterial species, is ubiquitous to members of this domain and, thanks to ever-expanding databases of sequence information, a useful tool for bacterial identification. In this study, we assembled an extensive repository of clinical isolates (n = 617), representing 30 medically important pathogenic species and originally identified using traditional culture-based or non-16S molecular methods. This strain repository was used to systematically evaluate the ability of 16S rRNA for species level identification. To enable the most accurate species level classification based on the paucity of sequence data accumulated in public databases, we built a Naïve Bayes classifier representing a diverse set of high-quality sequences from medically important bacterial organisms. We show that for species identification, a model-based approach is superior to an alignment based method. Overall, between 16S gene based and clinical identities, our study shows a genus-level concordance rate of 96% and a species-level concordance rate of 87.5%. We point to multiple cases of probable clinical misidentification with traditional culture based identification across a wide range of gram-negative rods and gram-positive cocci as well as common gram-negative cocci.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2025

Toward Drilling the Perfect Geothermal Well: An International Research Coordination Network for Geothermal Drilling Optimization Supported by Deep Machine Learning and Cloud Based Data Aggregation

The EDGE project, supported by the U.S. Department of Energy Geothermal Technologies Office under award DE-EE0008793, established a data-driven framework for improving the efficiency, cost-effectiveness, and reliability of geothermal well drilling. The project focused on developing scalable data infrastructure, advanced machine learning and probabilistic models, and integrated analytics tools to support continuous drilling optimization. A central objective was to reduce geothermal drilling costs by up to seventy percent while minimizing the risk of well failure through predictive diagnostics and adaptive planning. Over the project period, a comprehensive data repository was designed and deployed, incorporating records from over one hundred geothermal wells across varied geological settings. This repository supported both structured and unstructured data and adhered to FAIR data principles, enabling provenance tracking, quality control, and standardized metadata. The project introduced automated ingestion pipelines and a cloud-hosted platform that facilitated access to raw, processed, and derived datasets. This infrastructure served as the foundation for model development and analysis. Machine learning workflows were developed to predict key drilling metrics including rate of penetration, non-productive time, and total drilling costs. Self-organizing maps and dimensionality reduction methods were used to uncover operational patterns and outliers, while supervised learning algorithms such as random forests and deep neural networks were applied to forecast performance outcomes. The models were validated on heterogeneous datasets from both U.S. and Icelandic fields, demonstrating variable but significant predictive accuracy. The results indicated that finer temporal resolution, inclusion of lithological data, and consistency in operational annotations could substantially improve model performance. The project also implemented process mining techniques to reconstruct state-transition models from drilling event logs. These models enabled the identification of deviations from optimal workflows and provided insights into recurring failure modes. Analysis of non-productive time highlighted the impact of equipment failures, geological challenges, and human factors, offering opportunities for targeted mitigation strategies. The EDGE Dashboard was developed as a web-based expert system integrating data visualization, model outputs, and user-driven queries. It provided an accessible interface for operators to explore historical data, evaluate predicted outcomes, and compare drilling scenarios. Initial feedback from project partners suggested that the dashboard could serve as a foundation for more advanced advisory and optimization tools. Overall, the EDGE project demonstrated the feasibility and value of applying modern data science techniques to geothermal drilling. It delivered a set of interoperable tools and models that can support more efficient, lower-risk well development. The findings point toward a viable path for transitioning from advisory analytics to semi-autonomous drilling systems, contingent on continued collaboration, expanded datasets, and field validation. The project results have immediate relevance for drilling operations, data management practices, and future geothermal R&D efforts aimed at achieving reliable, cost-competitive geothermal energy at scale.

15 GEOTHERMAL ENERGY↗
DOE OSTI2019

A quick guide for student-driven community genome annotation

High quality gene models are necessary to expand the molecular and genetic tools available for a target organism, but these are available for only a handful of model organisms that have undergone extensive curation and experimental validation over the course of many years. The majority of gene models present in biological databases today have been identified in draft genome assemblies using automated annotation pipelines that are frequently based on orthologs from distantly related model organisms and usually have minor or major errors. Manual curation is time consuming and often requires substantial expertise, but is instrumental in improving gene model structure and identification. Manual annotation may seem to be a daunting and cost-prohibitive task for small research communities but involving undergraduates in community genome annotation consortiums can be mutually beneficial for both education and improved genomic resources. We outline a workflow for efficient manual annotation driven by a team of primarily undergraduate annotators. This model can be scaled to large teams and includes quality control processes through incremental evaluation. Moreover, it gives students an opportunity to increase their understanding of genome biology and to participate in scientific research in collaboration with peers and senior researchers at multiple institutions.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS1988

A knowledge-based approach to identification and adaptation in dynamical systems control

Artificial intelligence techniques are applied to the problems of model form and parameter identification of large-scale dynamic systems. The object-oriented knowledge representation is discussed in the context of causal modeling and qualitative reasoning. Structured sets of rules are used for implementing qualitative component simulations, for catching qualitative discrepancies and quantitative bound violations, and for making reconfiguration and control decisions that affect the physical system. These decisions are executed by backward-chaining through a knowledge base of control action tasks. This approach was implemented for two examples: a triple quadrupole mass spectrometer and a two-phase thermal testbed. Results of tests with both of these systems demonstrate that the software replicates some or most of the functionality of a human operator, thereby reducing the need for a human-in-the-loop in the lower levels of control of these complex systems.

Glass, B. J.↗
DOE OSTI2022

Accelerated screening of functional atomic impurities in halide perovskites using high-throughput computations and machine learning

The pressing need for novel materials that can serve rising demands in solar cell and optoelectronic technologies makes the nexus of halide perovskites, high-throughput computations, and machine learning, very promising. Ever increasing amounts of data on the structure, fundamental properties, and device performance of halide perovskites provide opportunities for learning chemical rules and design principles that make these materials attractive, and applying them across wide chemical spaces. In this work, we show that impurity properties of halide perovskites computed using density functional theory (DFT) can be combined with machine learning (ML) to deliver predictive models and quick identification of optoelectronically active impurity atoms. Our computation lead to the largest reported dataset of the formation energies and charge transition levels of Pb-site impurities in methylammonium lead halide (MAPbX 3 ) perovskites. Descriptors are defined to uniquely represent any impurity atom in any MAPbX 3 compound and mapped to the computed impurity properties using regression techniques such as Gaussian process regression, neural networks, and random forests. We use the best optimized predictive models to make predictions for hundreds of impurities across 9 MAPbX 3 compounds and create lists of dominating impurities, that is, impurities that can shift the equilibrium Fermi level in the perovskite as determined by native point defects. Finally, this accelerated screening powered by computations and machine learning can guide the identification of problematic impurities that may cause undesired recombination of charge carriers, as well as impurities that can be deliberately introduced to tune the perovskite conductivity and resulting photovoltaic absorption.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Biosystems Design by Machine Learning

Biosystems such as enzymes, pathways, and whole cells have been increasingly explored for biotechnological applications. Yet, the intricate connectivity and complexity of biosystems pose a major hurdle in designing biosystems with desired features. As -omics and other high throughput technologies have been rapidly developed, the promise of applying machine learning (ML) techniques in biosystems design has started to become a reality. ML models enable the identification of patterns within complicated biological data across multiple scales of analysis and can augment biosystems design applications by predicting new candidates for optimized performance. ML is being used at every stage of biosystems design to help find non-obvious engineering solutions with fewer design iterations. In this review, we first describe commonly used models and modeling paradigms within ML. We then discuss some applications of these models that have already shown success in biotechnological applications. Moreover, we discuss successful applications at all scales of biosystems design, including nucleic acids, genetic circuits, proteins, pathways, genomes, and bioprocess. Lastly, we discuss some limitations of these methods and potential solutions as well as prospects of the combination of ML and biosystems design.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS2001

System Identification and POD Method Applied to Unsteady Aerodynamics

The representation of unsteady aerodynamic flow fields in terms of global aerodynamic modes has proven to be a useful method for reducing the size of the aerodynamic model over those representations that use local variables at discrete grid points in the flow field. Eigenmodes and Proper Orthogonal Decomposition (POD) modes have been used for this purpose with good effect. This suggests that system identification models may also be used to represent the aerodynamic flow field. Implicit in the use of a systems identification technique is the notion that a relative small state space model can be useful in describing a dynamical system. The POD model is first used to show that indeed a reduced order model can be obtained from a much larger numerical aerodynamical model (the vortex lattice method is used for illustrative purposes) and the results from the POD and the system identification methods are then compared. For the example considered, the two methods are shown to give comparable results in terms of accuracy and reduced model size. The advantages and limitations of each approach are briefly discussed. Both appear promising and complementary in their characteristics.

Tang, Deman↗
NASA NTRS1983

System identification of large flexible structures by using simple continuum models

Simple continuum models for the representation of large truss beams and truss platforms are constructed and employed in the solution of system identification problems permitting the significant reduction of the number of structural parameters. Two simple and easily computer-implemented algorithms are presented for the estimation of a simple model's equivalent structural parameters, and analytical and numerical solutions are given for the truss beam and truss platform cases in order to illustrate the basic concepts. It is found that model simplicity, in conjunction with the principle of least square erors, yields a feasible integration of structural and identification problems.

Juang, J. N.↗
DOE OSTI2014

MO-E-17A-08: Attenuation-Based Size Adjusted, Scanner-Independent Organ Dose Estimates for Head CT Exams: TG 204 for Head CT

Purpose: AAPM Task Group 204 described size specific dose estimates (SSDE) for body scans. The purpose of this work is to use a similar approach to develop patient-specific, scanner-independent organ dose estimates for head CT exams using an attenuation-based size metric. Methods: For eight patient models from the GSF family of voxelized phantoms, dose to brain and lens of the eye was estimated using Monte Carlo simulations of contiguous axial scans for 64-slice MDCT scanners from four major manufacturers. Organ doses were normalized by scannerspecific 16 cm CTDIvol values and averaged across all scanners to obtain scanner-independent CTDIvol-to-organ-dose conversion coefficients for each patient model. Head size was measured at the first slice superior to the eyes; patient perimeter and effective diameter (ED) were measured directly from the GSF data. Because the GSF models use organ identification codes instead of Hounsfield units, water equivalent diameter (WED) was estimated indirectly. Using the image data from 42 patients ranging from 2 weeks old to adult, the perimeter, ED and WED size metrics were obtained and correlations between each metric were established. Applying these correlations to the GSF perimeter and ED measurements, WED was calculated for each model. The relationship between the various patient size metrics and CTDIvol-to-organ-dose conversion coefficients was then described. Results: The analysis of patient images demonstrated the correlation between WED and ED across a wide range of patient sizes. When applied to the GSF patient models, an exponential relationship between CTDIvol-to-organ-dose conversion coefficients and the WED size metric was observed with correlation coefficients of 0.93 and 0.77 for the brain and lens of the eye, respectively. Conclusion: Strong correlation exists between CTDIvol normalized brain dose and WED. For the lens of the eye, a lower correlation is observed, primarily due to surface dose variations. Funding Support: Siemens-UCLA Radiology Master Research Agreement; Disclosures - Michael McNitt-Gray: Institutional Research Agreement, Siemens AG; Research Support, Siemens AG; Consultant, Flaherty Sensabaugh Bonasso PLLC; Consultant, Fulbright and Jaworski.

60 APPLIED LIFE SCIENCES↗
NASA NTRS2001

Integration of the QMSFRG Database into the HZETRN Code

Accurate nuclear interaction data bases are needed for describing the transport of space radiation in matter including space craft structures, atmospheres, and tissues. Transport models support the identification and development of new material concepts for human and electronic part protection. Quantum effects are manifested in nuclear reactions in several ways including interference effects between terms in the multiple scattering series, the many-body nuclear wave functions (for e.g. the roles of shell structure and Fermi momentum) and nuclear clustering. The quantum multiple scattering fragmentation model (QMSFRG) is a comprehensive model for generating nuclear interaction databases for galactic cosmic ray (GCR) transport. Other nuclear databases including the NUCFRG model and Monte-Carlo simulation codes such as FLUKA, LAHET, HETC, and GEANT ignore quantum effects. These codes fail to describe many important features of nuclear reactions and are thus inaccurate for the evaluation of materials for radiation protection. Previously we have shown that quantum effects are manifested through constructive interference in forward production spectra, the effects of Fermi momentum on production spectra, cluster nuclei knockout, and the nuclear response function. Quantum effects are especially important for heavy ions with mass numbers less than 20 that dominate radiation transport in human tissues and for the materials that are expected to be superior in space radiation protection. We describe the integration of the QMSFRG model into the HZETRN transport code. Integration milestones include proper treatment of odd-even charge-mass effects in nuclear fragmentation and the momentum distribution of nucleon production from GCR primary heavy ions. We have also modified the two-body amplitudes in the model to include nuclear medium effects. In order to include a comprehensive description of the GCR isotopic composition in materials, we have described the isotopic composition of the GCR by extending the 59-isotope version of HZETRN to an 120-isotope version. The isotopic composition of most primary GCR elements (including H, He, C, N, O, Ne, Mg, Si, Ar, Ca, Cr, and Fe) are included in the extended model. We discuss results for the high-energy neutron composition inside materials, and the charge and mass distribution for benchmark GCR problems.

Cucinotta, F. A.↗
DOE OSTI2018

Decay Heat Surrogate Modeling for High Temperature Reactors

The accuracy of the safety analysis of Nuclear Power Plants, in general, and the High Temperature Reactors, in particular, is driven by several factors, among which the accuracy of decay heat curves or calculation. This paper is aimed to present a methodology based on the usage of the RAVEN code for the construction of a multi-dimensional decay heat curve for High Temperature Reactors, via calculations performed by ORIGEN. The main goal of this activity has been the identification of a model able to surrogate the phenomenology of the decay heat after shutdown, satisfying the following requirements: 1) Reasonable prediction accuracy till 1 month after the shutdown, 2) Ability to capture the main deviation effects determined by field conditions (e.g. burn-up, fuel temperature, fissile enrichment, etc.), 3) As easy as possible formalism in order to possibly allow the implementation of the curve into system codes (e.g. RELAP5-3D) inputs exploiting, for example, control variable capabilities. Three surrogate models (also known as decay heat curves in the safety analysis nomenclature) have been identified: 1) Spline Exponential surrogate model, 2) A Support Vector Regression model and, 3) A Dynamic Mode Decomposition model. The aim of this paper is to present the mathematical background of the identified surrogate models and their performance in the prediction of decay heat evolution for High Temperature Reactors. A preliminary comparison of the surrogate model performances is finally reported.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2023

Using Ultrasound Image Augmentation and Ensemble Predictions to Prevent Machine-Learning Model Overfitting

Deep learning predictive models have the potential to simplify and automate medical imaging diagnostics by lowering the skill threshold for image interpretation. However, this requires predictive models that are generalized to handle subject variability as seen clinically. Here, we highlight methods to improve test accuracy of an image classifier model for shrapnel identification using tissue phantom image sets. Using a previously developed image classifier neural network—termed ShrapML—blind test accuracy was less than 70% and was variable depending on the training/test data setup, as determined by a leave one subject out (LOSO) holdout methodology. Introduction of affine transformations for image augmentation or MixUp methodologies to generate additional training sets improved model performance and overall accuracy improved to 75%. Further improvements were made by aggregating predictions across five LOSO holdouts. This was done by bagging confidences or predictions from all LOSOs or the top-3 LOSO confidence models for each image prediction. Top-3 LOSO confidence bagging performed best, with test accuracy improved to greater than 85% accuracy for two different blind tissue phantoms. This was confirmed by gradient-weighted class activation mapping to highlight that the image classifier was tracking shrapnel in the image sets. Overall, data augmentation and ensemble prediction approaches were suitable for creating more generalized predictive models for ultrasound image analysis, a critical step for real-time diagnostic deployment.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

Review of the Current State of Knowledge on the Effects of Radiation on Concrete

A review of the current state of knowledge on the effects of radiation on concrete in nuclear applications is presented. Emphasis is placed on the effects of radiation damage as reflected by changes in engineering properties of concrete in the evaluation of the long-term operation (LTO) and for Plant Life or Aging Management of nuclear power plants (NPPs) in Japan, Spain, and the United States. National issues and concerns are described for Japan and the US followed by a discussion of the fundamental understanding of the effects radiation on concrete. Specifically, the effects of temperature, moisture content, and irradiation on ordinary Portland cement paste and the role of temperature and neutron energy spectra on radiation induced volumetric expansion (RIVE) of aggregate-forming minerals are described. This is followed by a discussion of the bounding conditions for extended operation, the significance of accelerated irradiation conditions, the role of temperature, creep, and how these issues are being incorporated into numerical and meso-scale models. From these insights on radiation damage, analyses of these effects on concrete structures are reviewed and the current status of work in Japan and the US are described. Also discussed is the recent formation of a new international scientific and technical organization, the International Committee on Irradiated Concrete (ICIC), to provide a forum for timely information exchanges among organizations pursuing the identification, quantification, and modeling of the effects of radiation on concrete in commercial nuclear applications. Lastly, the paper concludes with a discussion of research gaps including: 1) interpreting test-reactor data, 2) evaluating service-irradiated concrete for aging management and to inform radiation damage models with the Zorita NPP (Spain) serving as the first comprehensive test case, 3) irradiated-assisted alkali-silica reactions, and 4) RIVE under constrained conditions.

36 MATERIALS SCIENCE↗
DOE OSTI2023

CoolPINNs: A physics-informed neural network modeling of active cooling in vascular systems

Emerging technologies like hypersonic aircraft, space exploration vehicles, and batteries avail fluid circulation in embedded microvasculatures for efficient thermal regulation. Modeling is vital during the design and operational phases of these engineered systems. However, many challenges exist in developing a modeling framework. What is lacking is an accurate framework that (i) captures sharp jumps in the thermal flux across complex vasculature layouts, (ii) deals with oblique derivatives (involving tangential and normal components), (iii) handles nonlinearity because of radiative heat transfer, (iv) provides a high-speed forecast for real-time monitoring, and (v) facilitates robust inverse modeling. Here, this paper addresses these challenges by availing the power of physics-informed neural networks (PINNs). We develop a fast, reliable, and accurate Scientific Machine Learning (SciML) framework for vascular-based thermal regulation—called CoolPINNs: a PINNs-based modeling framework for active cooling. The proposed mesh-less framework elegantly overcomes all the mentioned challenges. The significance of the reported research is multi-fold. First, the framework is valuable for real-time monitoring of thermal regulatory systems because of rapid forecasting. Second, researchers can address complex thermoregulation designs since the approach is meshless. Finally, the framework facilitates systematic parameter identification and inverse modeling studies, perhaps the most significant utility of the current framework.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Model America: Data and Models for every U.S. Building

The 5-year goal of the “Model America” concept was to generate a model of every building in the United States. This data repository delivers on that goal with "Model America v1". Oak Ridge National Laboratory (ORNL) has developed the Automatic Building Energy Modeling (AutoBEM) software suite to process multiple types of data, extract building-specific descriptors, generate building energy models, and simulate them on High Performance Computing (HPC) resources. For more information, see AutoBEM-related publications (bit.ly/AutoBEM). There were 125,715,609 buildings detected in the United States. Of this number, 122,146,671 (97.2%) buildings resulted in a successful generation and simulation of a building energy model. This dataset includes the full 125 million buildings. Future updates may include additional buildings, data improvements, or other algorithmic model enhancements in "Model America v2". This dataset contains OSM and IDF zip files for every U.S. county. Each zip file contains the generated buildings from that county. The .csv input data contains the following data fields: 1. ID - the Unique Building Identifier (UBID), generated using the Pacific Northwest National Laboratory (PNNL) BuildingID framework 2. Centroid - building center location in latitude/longitude (from Footprint2D) 3. Footprint2D - building polygon of 2D footprint (lat1/lon1_lat2/lon2_...) 4. State_abbr - state name 5. Area - estimate of total conditioned floor area (ft2) 6. Area2D - footprint area (ft2) 7. Height - building height (ft) 8. NumFloors - number of floors (above-grade) 9. WWR_surfaces - percent of each facade (pair of points from Footprint2D) covered by fenestration/windows (average 14.5% for residential, 40% for commercial buildings) 10. CZ - ASHRAE Climate Zone designation 11. BuildingType - DOE prototype building designation (IECC=residential) as implemented by OpenStudio-standards 12. Standard - building vintage This data is made free and openly available in hopes of stimulating any simulation-informed use case. Data is provided as-is with no warranties, express or implied, regarding fitness for a particular purpose. We wish to thank our sponsors which include Oak Ridge National Laboratory (ORNL) Laboratory Directed Research and Development (LDRD), U.S. Dept. of Energy’s (DOE) Building Technologies Office (BTO), Office of Electricity (OE), Biological and Environmental Research (BER), and National Nuclear Security Administration (NNSA). Update (September 23, 2025): We corrected the ID field in all state-level.csv input files to ensure one-to-one consistency with the corresponding .osm and .idf output files. The schema and file structure are unchanged; only the values in the ID column were modified. No files were added or removed, and the .zip bundles (containing .osm / .idf) are unchanged. The corrected .csv inputs were re-extracted in March 2025 from the original data generated ~ 2021 (Theta supercomputer runs), and published here to align input IDs with model outputs. Update (September 6, 2026): The Model America dataset was updated to replace the previous building ID field with the Unique Building Identifier (UBID), using the Pacific Northwest National Laboratory (PNNL) BuildingID framework. UBIDs provide standardized, location-based identifiers for individual building footprints and improve interoperability with other building and geospatial datasets. The data files containing the previous building identifiers were updated to include UBIDs. This update standardizes building identification; the underlying Model America building characteristics and energy simulation results were not recomputed as part of this update.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2016

Scintillation light from cosmic-ray muons in liquid argon

This paper reports the results of an experiment to directly measure the time-resolved scintillation signal from the passage of cosmic-ray muons through liquid argon. Scintillation light from these muons is of value to studies of weakly-interacting particles in neutrino experiments and dark matter searches. The experiment was carried out at the TallBo dewar facility at Fermilab using prototype light guide detectors and electronics developed for the Deep Underground Neutrino Experiment. Two models are presented for the time structure of the scintillation light, a phenomenological model and a physically-motivated model. Both models find tT = 1:52 ms for the decay time constant of the Ar 2 triplet state. These models also show that the identification of the “early” light fraction in the phenomenological model, FE 25% of the signal, with the total light from singlet decays is an underestimate. The total fraction of singlet light is FS 36%, where the increase over FE is from singlet light emitted by the wavelength shifter through processes with long decay constants. The models were further used to compute the experimental particle identification parameter Fprompt, the fraction of light coming in a short time window after the trigger compared with the light in the total recorded waveform. The models reproduce quite well the typical experimental value 0.3 found by dark matter and double b-decay experiments, which suggests this parameter provides a robust metric for discriminating electrons and muons from more heavily ionizing particles.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
NASA NTRS1981

The use of frequency methods in rotorcraft system identification

A new approach to model structure determination is examined. Flight data from the Rotor Systems Research Aircraft (RSRA) are transformed into the frequency domain and truncated to provide band limiting. The stepwise regression technique is then used to identify a quasistatic state-space model from the transformed data. The data processing requirements for both time domain and frequency domain identification are discussed and the results of the two techniques are compared.

Duval, R. W.↗
DOE OSTI2026

Sim-to-real supervised domain adaptation for radioisotope identification

Machine learning has the potential to improve the speed and reliability of radioisotope identification using gamma spectroscopy. However, meticulously labeling an experimental dataset for training is often prohibitively expensive, while training models purely on synthetic data is risky due to the domain gap between simulated and experimental measurements. In this research, we demonstrate that supervised domain adaptation can substantially improve the performance of radioisotope identification models by transferring knowledge between synthetic and experimental data domains. We consider two domain adaptation scenarios: (1) a simulation-to-simulation adaptation, where we perform multi-label proportion estimation using simulated high-purity germanium detectors, and (2) a simulation-to-experimental adaptation, where we perform multi-class, single-label classification using measured spectra from handheld lanthanum bromide (LaBr) and sodium iodide (NaI) detectors. We begin by pretraining a spectral classifier on synthetic data using a custom transformer-based neural network. After subsequent fine-tuning on just 64 labeled experimental spectra, we achieve a test accuracy of 96% in the sim-to-real scenario with a LaBr detector, far surpassing a synthetic-only baseline model (75%) and a model trained from scratch (80%) on the same 64 spectra. Furthermore, we demonstrate that domain-adapted models learn more human-interpretable features than experiment-only baseline models. Overall, our results highlight the potential for supervised domain adaptation techniques to bridge the sim-to-real gap in radioisotope identification, enabling the development of accurate and explainable classifiers even in real-world scenarios where access to experimental data is limited.

Lalor, Peter W.↗