SEARCH · Engineering Papers

Results for “data normalization”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3

All sources NASA NTRS DOE OSTI

DOE OSTI2026

Performance Modeling of Tandem Photovoltaics: A Yearlong Outdoor Degradation Analysis of a Ga⁢As//Si Minimodule

We present a performance modeling and degradation analysis framework for tandem photovoltaic modules, building upon established procedures for crystalline silicon devices and adapting them to account for the spectral sensitivity of multijunction technologies. The methodology employs filter criteria to select outdoor measurements close to standard test conditions (STC) under stable spectral and ambient conditions, followed by normalization of power production data with corrections for temperature, irradiance, and precipitable water vapor. We demonstrate this framework using a mechanically stacked four-terminal gallium arsenide (Ga⁢As) // silicon (Si) tandem solar minimodule deployed outdoors from October 2019 to January 2021 in Golden, Colorado, USA. We determined degradation rates of −4.1 ±0.2%/year for the Ga⁢As subcell and −2.5 ±0.9%/year for the Si subcell, with analysis of individual performance metrics indicating that packaging degradation, particularly delamination, was the dominant failure mode. Simulations using PVcircuit, an open-source equivalent-circuit solver, confirmed these findings. The presented methodology provides a reproducible foundation for performance modeling and degradation analysis of emerging tandem technologies.

14 SOLAR ENERGY↗

DOE OSTI2025

Deep Factorization Machine Learning for Disaggregation of Transmission Load Profiles with High Penetration of Behind-The-Meter Solar

The ever-growing integration of distributed energy resources (DERs), especially behind-the-meter (BTM) solar generations, poses imperative operational challenges to system operators such as regional transmission organizations (RTOs). It is important for RTOs to effectively and accurately extract actual load profiles at the transmission level for a single node with significant BTM solar injection. This paper first illustrates the necessity of disaggregating the daily actual load profile of a single node. Furthermore, by segmenting nodes with selected timeseries features, nodes with significant BTM solar generation are identified. Lastly, a bi-level framework is proposed, comprising reference node disaggregation and DeepFM nodal disaggregation, aimed at disaggregating the nodal load profiles from which system operators require more information. By adopting a hybrid Deep Factorization Machine (DeepFM) model, the model achieve accurate results by extracting both linear and nonlinear relations between nodes in the same region and the zonal load and nodal load profile. To overcome the lack of ground truth, this paper segments the load profile into daytime, nighttime, and zero-crossing points and utilizes the latter two for evaluation purposes. The proposed disaggregation procedure is validated using real world, minute-level, normalized, and anonymized nodal data in the PJM service territory.

42 ENGINEERING↗

DOE OSTI2024

Elliptically-Contoured Tensor-variate Distributions with Application to Image Learning

Statistical analysis of tensor-valued data has largely used the tensor-variate normal (TVN) distribution that may be inadequate for data arising from distributions with heavier or lighter tails. We study a general family of elliptically contoured (EC) TV distributions and derive its characterizations, moments, marginal, and conditional distributions. We describe procedures for maximum likelihood estimation from data that are (1) uncorrelated draws from an EC distribution, (2) from a scale mixture of the TVN distribution, and (3) from an underlying but unknown EC distribution, for which we extend Tyler’s robust estimator. A detailed simulation study highlights the benefits of choosing an EC distribution over the TVN for heavier-tailed data. We develop TV classification rules using discriminant analysis and EC errors and show that they better predict cats and dogs from images in the Animal Faces-HQ dataset than the TVN-based rules. A novel tensor-on-tensor regression and TV analysis of variance (TANOVA) framework under EC errors is also demonstrated to better characterize gender, age, and ethnic origin than the usual TVN-based TANOVA in the celebrated labeled faces of the wild dataset.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2026

OC6 Phase III: Validation of Wind Turbine Aerodynamic Loading During Surge/Pitch Motion

The objective of Phase III of the Offshore Code Comparison Collaboration, Continued, with Correlation and unCertainty (OC6) project was to validate the accuracy of aerodynamic load predictions by offshore wind modeling tools for a floating offshore wind turbine as it experiences large surge-translational and pitch-rotational motion, as would occur during normal operation. The test data considered were generated at Politecnico di Milano – a wind tunnel with a robotic excitation system to emulate wave loading on the wind turbine. Testing was performed using a scaled version of the DTU 10-MW reference model, and motion was prescribed as harmonic oscillations in the surge and pitch directions, independently. A variety of models were examined in the project, performing a three-way validation between engineering-level tools, higher-fidelity tools, and measurement data from two wind tunnel experimental campaigns. The Load Cases (LC) considered in this testing are as follows: LC 1.X - Steady Wind LC 2.X - Unsteady Wind - Surge Motion LC 3.X - Unsteady Wind - Pitch Motion Details on the results of the OC6 Phase III campaign can be found in the following reference: Bergua R., et al. OC6 project Phase III: validation of the aerodynamic loading on a wind turbine rotor undergoing large motion caused by a floating support structure, Wind Energ. Sci., 8, 465–485, https://doi.org/10.5194/wes-8-465-2023, 2023.

17 WIND ENERGY↗

DOE OSTI2026

Fracture Intersections under Stress: Laboratory Data and Code [Data set]

The connectivity of natural and induced fractures governs the injection and withdrawal of fluids from subsurface reservoirs. Connectivity depends on intersections that control how fluids mix and move through the entire system. Here, we present data sets from 3D X-ray microscopy measurements of simple fracture networks under stress. 3D printing was used to create prismatic blocks that formed fracture networks composed of 2 orthogonal fractures. The network orientation was either "x" or "+" relative to an applied vertical stress. 3D data sets were collected for normal loads of 25, 100 and 200 Newtons for samples with fracture surfaces with either correlated or uncorrelated asperity distributions. The file contains data from the 12 samples analyzed along with an example code used to extract the intersection geometry. Additional experimental details can be found in the manuscript "Geologic Stress Modulates Fluid Mixing at Fracture Intersections" (10.1038/s43247-026-03525-9)and supplemental information to appear in Communications Earth & Environment in 2026.

02 PETROLEUM↗

DOE OSTI2026

Grid Operator Analytics and Assessment Tools for Inverter- Based Resources Dominated Grid (GOAAT-IBR) Project Update

This presentation provides an update on the OPTIMA GOAAT project, with emphasis on the cloud-native data platform developed in-house to ingest, manage, and operationalize high-resolution power system data. Since our last NASPI presentation, accessible via OSTI ID #2671437, the project team advanced the design and deployment of a scalable architecture capable of handling both synchronized and non-synchronized streams, including PMU, point-on-wave (POW), COMTRADE, and SCADA data. These materials review the project status, recent progress, and key lessons learned. The core of the presentation examines the architecture and engineering of our cloud-native ingestion and data management platform. We then explain how pipelines were designed to collect, normalize, time-align, store, and serve heterogeneous data at scale. We will discuss design choices such as data models, streaming versus batch ingestion, storage tiers, and interoperability with analytics applications. Practical experiences with cloud-native technologies were shared during the event, including benefits, limitations, and integration challenges in a utility environment, along with methods used to improve performance, reduce latency, and optimize resource usage. The presentation also showcases user interface designs and visualization tools that convert raw measurements and analytics results into intuitive, actionable insights for operators and engineers. During the presentation examples were provided demonstrating how visualization, event views, and summarized analytics enhance situational awareness and support operational decision-making. These use cases illustrate how a well-designed data infrastructure can bridge the gap between high-volume measurements and practical grid operations.

Aminifar, Farrokh↗

DOE OSTI2025

FY2025 Status Report: Model 9975 O Ring Fixture Long-Term Leak Performance

Leak testing experiments to monitor the aging performance of Viton® GLT and GLT-S O-rings used in the model 9975 shipping package have been ongoing since 2004 at Savannah River National Laboratory. Seventy tests using mock-up 9975 primary containment vessels (PCVs) with GLT O-rings were assembled and heated to temperatures ranging from 200 to 450 °F. Due to material substitution, fourteen tests with GLT-S O-rings were initiated in 2008 and heated to temperatures ranging from 200 to 400 °F. The conditioning temperatures are elevated compared to the calculated maximum O-ring temperature in a 9975 package in storage, 158 °F, to accelerate aging and enable observations of O-ring failures in a reasonable time frame. The mock-up PCV fixtures are leak tested periodically, and all GLT O-ring fixtures aged at 350 °F or above have failed to maintain a leak-tight seal. Eight GLT O-ring fixtures aged at 300 °F have failed after 2.8 to 5.7 years at temperature while the remaining fixtures at 300 °F were retired from testing following more than five years of aging without failure. Two of these retired fixtures were returned to testing and heated to 350 °F to evaluate the impact of additional heating at higher temperature for aged O-rings. Fixture #20 failed after 3 months while fixture #18 failed after 9 months at 350 °F. These O-rings demonstrated that aged and in-service O-rings can continue to be used, even at higher temperatures, after being in storage, and their leak performance are consistent with other samples at 350 °F. There has been one GLT O-ring fixture which failed after 13.4 years of aging at 200 °F. However, 20 other GLT O-rings aging at 200 °F have remained leak-tight for over 16.9 years and remain in test. There are two GLT O-ring fixtures at 270 °F; one fixture has failed after 12.9 years while the other fixture remains in test after 12.5 years. All GLT-S O-ring fixtures aged at 300 °F or above have failed their leak test. No failures have yet been observed in GLT-S O-ring fixtures aging at 250 °F for 14.9 years, while one GLT-S O-ring fixture failed after 12.4 years at 200 °F. The leak testing data to date suggest the GLT and GLT-S O-rings aging in the K-Area Complex (KAC) storage at temperatures of 158 °F might maintain a leak-tight seal for up to 59 years. Data from the O-ring fixtures are generally consistent with results from compression stress-relaxation testing and provide confidence in the predictive models based on those results. However, uncertainty exists in extrapolating these elevated temperature results to the lower temperatures of interest for normal storage in KAC. The collective data from these test efforts suggest the minimum O-ring service life at KAC normal storage conditions should be at least 34 years for GLT and GLT-S O-rings. Measurement of compression set in O-rings removed from failed fixtures, compared to that from KAC surveillance O-rings, indicate significant margin remains for O-rings still in service in 9975 packages in KAC. Aging and periodic leak testing will continue for the remaining 24 mock-up PCV fixtures.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

DOE OSTI2025

Welch Method and Bootstrapping Applied to Subcritical Gamma Noise

We measured the prompt neutron decay constant 𝛼 of the CROCUS zero-power reactor at the Swiss Federal Institute of Technology Lausanne using cross-power spectral density (CPSD) analysis of gamma-gamma correlations from two trans-stilbene organic scintillators positioned near the reactor core. We measured critical and subcritical states, with water levels ranging from 960 mm (critical) to 800 mm (𝜌=−1.4 $ subcritical). Our analysis used the Welch method, dividing signal segments for fast Fourier transform (FFT) frequency analysis and applying bootstrapping uncertainty quantification that uses Welch-defined segments. Results demonstrated a clear increase in the measured 𝛼 as reactor reactivity decreased, distinguishing critical from subcritical conditions. At the 960-mm critical level, 𝛼 was estimated at 155.9 ± 0.7 s −1 , and for the 800-mm subcritical level, 𝛼 increased significantly to 367.3 ± 6.9 s –1 . A linear regression of subcritical states yielded a critical estimate of 154.0 ± 3.1 s –1 , aligning with the static 𝛼 estimate at critical. The bootstrapping method produced normally distributed 𝛼 estimates, confirming data consistency. The gamma CPSD 𝛼 estimates clearly distinguish reactor states and improve monitoring of zero-power reactors. The future deployment of modular and microreactors as potential candidates for noise analysis is demonstrated in CROCUS, particularly zero-power mock-ups of new designs. The improvement of noise analysis in the subcritical domain from this work will support experimental data for reactor deployment and procedure.

CROCUS↗

DOE OSTI2025

Results from the T2K Experiment on Neutrino Mixing Including a New Far Detector 𝜇-like Sample

We have made improved measurements of three-flavor neutrino mixing with 19.7⁢(16.3) × 10 20 protons on target in (anti-)neutrino-enhanced beam modes. A new sample of muon-neutrino events with tagged pions has been added at the far detector, as well as new proton and photon-tagged samples at the near detector. Significant improvements have been made to the flux and neutrino interaction modeling. T2K data continue to prefer the normal mass ordering and upper octant of sin 2 ⁡𝜃 23 with a near-maximal value of the charge-parity violating phase with best-fit values in the normal ordering of 𝛿 CP = −2.18$^{+1.22}_{−0.47}$, sin 2 ⁡𝜃 23 = 0.559$^{+0.018}_{−0.078}$ and Δ⁢𝑚$^{2}_{32}$ = (+2.506$^{+0.039}_{−0.052}$) × 10 −3 eV 2 .

CP violation↗

DOE OSTI2025

Deep-learning-based domain adaptation for cavity fault prediction at Jefferson Laboratory

Superconducting radio-frequency (SRF) cavities are the core components of the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab, providing high-power electron beams for nuclear physics experiments. The facility comprises 418 SRF cavities, and any fault in these cavities can lead to interruptions in the electron beam supply. Cavity faults are the leading cause of beam trips in CEBAF. Predicting and mitigating those faults before onset can help maintain normal operation. Existing models face challenges in distinguishing between normal and fault signals when changes occur in the underlying time-series data, from changes in control software, operational parameters, or the environment. This work proposes a deep learning domain adaptation model that leverages transfer learning to address fault prediction challenges by improving accuracy. The model is trained and fine-tuned using a dataset collected for faulty and normal operation using a data acquisition system in CEBAF. Our deep learning-based domain adaptation model achieves a prediction accuracy of 89.61% of the fault and normal signals. The developed model effectively predicts normal running signals compared to the baseline approach without domain adaptation. This capacity is essential for the fault prediction task in the CEBAF because of heavily imbalanced data containing vast amounts of normal signals. The model performs well for predicting faults several hundred milliseconds before the fault onset compared to other models where no adaptation is applied. Incorporating deep learning-based domain adaptation techniques will significantly improve the fault prediction performance.

Rahman, Md Monibor [Old Dominion Univ., Norfolk, V↗

DOE OSTI2026

Daily, 30 m Resolution NDSI Data for the East River Watershed, CO for 2000-2020

This dataset contains daily Normalized Difference Snow Index (NDSI) values at 30 m spatial resolution for the East River watershed in Colorado, USA. The temporal range of these data includes water years 2001-2020. These data were created using the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM). This model fuses low spatial and high temporal resolution data from MODIS (500 m, daily) with high spatial and low temporal resolution data from Landsat (30 m, 16 days) to create a 30m synthetic daily snow product. This product allows for the analysis of historical snow covered area trends in the East River Watershed at fine spatiotemporal resolutions where it was not available previously. This research was performed as a part of the Department of Energy’s Subsurface Biogeochemical Research Program with the primary intent of better understanding the timing and spatial patterns of water delivery to the Critical Zone in mountain watersheds. Each .zip file contains one "water year" of data (October 1 - September 30; i.e., water year 2010 starts October 1, 2010 and ends September 30, 2011). Each zip file contains the following: STARFM daily Normalized Difference Snow Index (NDSI) fusion data files in GeoTiff format with one layer for each day between Landsat data acquisition dates (i.e., for dates of Landsat acquisition, the Landsat image is included for that date). The study area is located in an area of Landsat path overlap, so Landsat dates acquisitions are every 7-9 days. Landsat NDSI files containing the high spatial (30m), low temporal (7-9 days due to Landsat path overlap) resolution data used as input to STARFM in GeoTiff format with one layer for each day. Dates for which no Landsat data were obtained are included as NoData layers. MODIS NDSI files containing the high temporal (daily), low spatial (500m) resolution data used as input to STARFM in GeoTiff format with one layer for each day. Please note the MODIS data were resampled to 30m pixels for input into the STARFM model. The data have a scale factor of 10,000 and a no data value of -32767. The projection of all datasets is WGS 84 (EPSG: 4326), which has a latitude/longitude based degree resolution of 0.0002694946 X 0.0002694946, and approximates to the 30 m spatial resolution mentioned above. The Layer Index files in .csv format. They contain information for each layer in the above GeoTiff files regarding the corresponding date for each layer, the fraction of pixels in the image that contain valid data (missing data is due to either cloud cover or poor data quality; these values are not percent snow cover). Dates of Landsat overpass are indicated in these files. If no Landsat data were able to be obtained due to cloud cover or lack of Landsat Tier 1 data available on Google Earth Engine, this is also noted.

EARTH SCIENCE > CRYOSPHERE > SNOW/ICE↗

DOE OSTI2026

Connecting ground-state properties of 6 Li to each other and to scattering data

We examine the relationship between the asymptotic normalization coefficient (ANC) of 6 Li and other low-energy observables in the α–deuteron system. Our analysis uses a set of calculations carried out within the ab initio no core shell model with continuum (NCSMC) using a variety of inter-nucleon interactions and basis sizes, and yielding 6 Li deuteron separation energies between 1.3 and 1.8 MeV (Hebborn et al 2022 Phys. Rev. Lett. 129 042503). These NCSMC calculations show that the square of the ANC is strongly correlated with the separation energy over this range. In this work, we investigate the origin of this correlation using the phenomenological R-matrix, a single-channel potential and a perturbative approach. We show that this correlation occurs because the depth of the α–deuteron central potential changes by only a small relative amount as the separation energy varies. We then investigate if the ANC can be accurately extracted from α–deuteron phase shifts in an ideal case in which low-energy data are available and there are no experimental errors. We find that both R-matrix and Coulomb-modified effective-range theory (CM-ERE) yield extracted ANCs close to, although not exactly equal to, the NCSMC value, provided the extrapolation is constrained by the known position of the bound-state pole and at least three terms are included in the fit function. The R-matrix approach converges faster than the CM-ERE as the number of parameters increases and is also more robust against the inclusion of low-energy and high-energy phase shift data. Finally, our study also shows that a naive quantification of uncertainties by comparing different truncations used in both theories is not accurate, and suggests the accuracy of ANCs extracted from phase shift data needs further investigation.

R-matrix↗

DOE OSTI2025

Development of Machine Learning Algorithm for Pebble Bed Modular Reactor Misuse Detection

The objective of this work was to develop a machine learning ensemble that could assist pebble bed reactor verification by evaluating whether a given pebble circulating through a PBR was normal or anomalous using gamma spectroscopy measurements from a notional PBR burnup measurement system. Using a PBR reference design, data sets of synthetic gamma spectra representative of BUMS measurements of normal and anomalous pebbles that may be used to produce special fissile material were generated to train and test an ML anomaly detection ensemble on two reference scenarios – substitution of normal pebbles with target pebbles for production of Pu or 233 U. The ML ensemble correctly identified all anomalous pebbles in the testing data set, and while perfect ensemble performance is normally indicative of overfitting, it was concluded that significantly lower photon intensity of target pebbles produced distinctly less intense photon spectra to where perfect ensemble performance was expected.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOE OSTI2026

NeuNorm

NeuNorm is a scipp-based Python library for neutron imaging normalization and time-of-flight (TOF) data processing at Oak Ridge National Laboratory imaging facilities (MARS at HFIR and VENUS at SNS). NeuNorm 2.0 is a complete, scipp-based rewrite of the original NeuNorm normalization library, adding HDF5 output, automatic uncertainty propagation, and TOF/event-mode processing.

Zhang, Chen [Oak Ridge National Laboratory (ORNL),↗

DOE OSTI2025

Graphene SHDMC Data

The data used to produce all of the figures and tables in the manuscript titled "Highly Accurate Many-Body Theory Reaches 2D Materials" can be found here. This data set includes: -SHDMC results for graphene -selected CI with and without re-normalized second-order perturbation (rPT2) theory corrections for graphene -data demonstrating that SHDMC displays an exponential rate of convergence -data used for sCI + rPT2 complete basis set extrapolation -data used to extrapolate SHDMC energies to the infinite basis limit -data used to demonstrate compactness of SHDMC wavefunction

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

DOE OSTI2025

Oral microbiome and mycobiome dynamics in cancer therapy-induced oral mucositis

Cancer therapy-induced oral mucositis is a frequent major oncological problem, secondary to cytotoxicity of chemo-radiation treatment. Oral mucositis commonly occurs 7–10 days after initiation of therapy; it is a dose-limiting side effect causing significant pain, eating difficulty, need for parenteral nutrition and a rise of infections. The pathobiology derives from complex interactions between the epithelial component, inflammation, and the oral microbiome. Our longitudinal study analysed the dynamics of the oral microbiome (bacteria and fungi) in nineteen patients undergoing chemo-radiation therapy for oral and oropharyngeal squamous cell carcinoma as compared to healthy volunteers. The microbiome was characterized in multiple oral sample types using rRNA and ITS sequence amplicons and followed the treatment regimens. Microbial taxonomic diversity and relative abundance may be correlated with disease state, type of treatment and responses. Identification of microbial-host interactions could lead to further therapeutic interventions of mucositis to re-establish normal flora and promote patients’ health. Data presented here could enhance, complement and diversify other studies that link microbiomes to oral disease, prophylactics, treatments, and outcome.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2024

Photonuclear cross sections for 197 Au: An update on the gold standard

Cross sections for the 197 Au(γ, n) reaction are broadly used in nuclear physics as a standard for normalizing photonuclear reaction cross-section data at photon beam energies above approximately 8 MeV. In this paper, we report cross-section measurements for the 197 Au(γ, n) 196 Au g+m1 reaction at beam energies from 13 to 31 MeV. Our measurements provide the first cross-section data for this reaction at beam energies above 20 MeV, enabling the use of this reaction as a cross-section standard up to 30 MeV. Also, this work provides first cross-section measurements for the 197 Au(γ, n) 196 Au m2 reaction. In addition, we measured cross-section data for the 197 Au(γ, 3n) 194 Au reaction, which can be used as a cross-section standard above about 25 MeV. These measurements were performed using a new target activation method that is based on the angle-energy correlation of the laser Compton- scattered photon beams at the High Intensity Gamma-ray Source (HIγS). The technique enables measuring photonuclear reaction cross-sections at several discrete beam energies concurrently via a single irradiation on a stack of different targets. Measurements were carried out by irradiating a stack of concentric-ring targets consisting of Au, TiO 2 , Zn, Os, and Au (in order of the γ-ray beam direction). Our data for the 197 Au(γ,n) 196 Au g+m1 reaction in the energy range of 13 to 20 MeV are in good agreement with existing ones measured using monoenergetic γ-ray beams, but differ from data acquired using a bremsstrahlung γ-ray beam. Also, above 18 MeV, our data for the 197 Au(γ,n) 196 Au g+m1 and 197 Au(γ,n) 196 Au m2 reactions differ significantly from the most recent TENDL and JENDL evaluations, suggesting a need to update these data libraries. Furthermore, the TENDL evaluation and existing data are consistent with our data for the 197 Au(γ,3n) reaction, but differ significantly from the JENDL evaluation above 26 MeV.

190 ≤ A ≤ 219↗

DOE OSTI2026

Leveraging large language models to address data scarcity in machine learning for graphene synthesis

Machine learning in experimental materials science faces significant challenges due to the scarcity of data, which are costly and time-consuming to generate, particularly when relying on in-house experiments. Literature data mining offers a potential solution but introduces issues like mixed data quality, inconsistent formats, and non-uniform reporting of synthesis parameters, resulting in partially missing and heterogeneous features across the dataset. Here, we propose data imputation and feature engineering methods that employ pre-trained large language models (LLMs) to enhance machine learning performance on scarce, heterogeneous datasets, demonstrated on graphene CVD synthesis data and the ML-HydPARK hydrogen storage dataset. GPT models perform data imputation via tailored prompting and semantic normalization of inconsistently reported features through embeddings, for example, to harmonize the complex nomenclature of CVD substrates. Beyond yielding more diverse and richer feature representations than traditional methods such as K-nearest neighbors (KNN) and Multivariate Imputation by Chained Equations (MICE), LLM-based data imputation is evaluated against dataset characteristics and prompting strategies. We vary the level of autonomy granted to the LLM, from generic prompting that leverages pre-trained knowledge for autonomous data generation to data-informed prompting that constrains outputs using target-specific information, and demonstrate which level of autonomy yields superior imputation performance across datasets and feature types. The proposed data engineering methods markedly improve downstream performance; for example, in graphene layer number classification using a support vector machine (SVM), binary accuracy increases from 39% to 65% and ternary accuracy from 52% to 72%. Fine-tuning experiments on both datasets show that combining our proposed LLM-based data imputation and feature encoding methods with numerical machine learning predictors outperforms standalone fine-tuned LLM predictors in data-scarce settings. The proposed strategies emphasize data enhancement techniques rather than refining learning architectures or regularizing loss functions, offering a broadly applicable framework for improving machine learning performance on scarce, inhomogeneous datasets.

Chemical vapor deposition↗