Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “data-driven method”

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 145 records · Page 8
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Beyond interpolation: Physics-inspired gating transformers for extrapolating irradiation conditions to novel nuclear fuels

The qualification of advanced nuclear fuels relies on irradiation experiments in test reactors that emulate commercial conditions. Designing these tests requires accurate prediction of key irradiation quantities, particularly heat generation rate and burnup, yet obtaining them typically involves computationally expensive multi-step simulation workflows. We propose a physics-inspired gating transformer (PIGT) that integrates an inverse-square, distance-based attenuation into the encoder representation to bias attention toward physically relevant spatial relationships while retaining data-driven flexibility. Using MiniFuel irradiation data from the High Flux Isotope Reactor at Oak Ridge National Laboratory, we benchmark against ensemble methods, feedforward and recurrent networks, convolutional models, and standard transformers. While baseline models perform well under interpolation, they exhibit a pronounced generalization gap when evaluated on fuels not included in the training set. The proposed model consistently improves extrapolative accuracy and stability, yielding the strongest performance on unseen fuel configurations. These results indicate that a lightweight physics structure embedded within attention mechanisms can substantially improve robustness, enabling more reliable surrogate predictions to accelerate the design of nuclear fuel irradiation experiments.

Fuel qualification↗
DOE OSTI2026

Machine-Learning-Driven Discovery of Water Splitting BaFe 2 O 4 and Human-in-the-Loop Improvement via Al-Substitution for Increased Thermal Stability

Thermochemical hydrogen (TCH) production offers a promising method for converting thermal energy into hydrogen fuel through heat-driven redox cycles of metal oxides. Here, in this work a defect graph neural network (dGNN) was used to predict oxygen vacancy formation energies ΔH V O combined with Materials Project predictions of oxygen chemical potential stability to screen candidate oxides via high-throughput database analysis. BaFe 2 O 4 was identified as a promising material for experimental validation based on its predicted ΔH V O , oxygen chemical potential stability range, and potential for tunable substitutions to improve thermal properties. Experimental validation using thermogravimetric analysis (TGA), stagnation flow reactor (SFR), X-ray diffraction (XRD), and electron microscopy confirmed positive water-splitting behavior but also revealed limitations in thermal stability under aggressive reduction conditions. To address this, a human-in-the-loop modification strategy was employed introducing Al substitution in BaFe 2–x Al x O 4 ; this modification improves thermal stability, alters the crystal structure and enhances overall performance. These results demonstrate a combined computational and experimental workflow in which machine learning accelerates identification of promising candidates, while targeted experimental design enables optimization of functional performance. This approach advances the development of robust, cost-effective TCH materials and highlights the importance of integrating data-driven discovery with human-guided materials design in paving the way for scalable hydrogen production technologies.

organic↗
DOE OSTI2026

Automated and High-Throughput Phase Separation Control for Supramolecular Polymer Blends Enabled by Machine Learning

Supramolecular polymer blends (SPBs) offer tunable morphologies that dictate their macroscopic properties, yet their rational design is limited by the absence of predictive structure−morphology models. Here, we introduce a data-driven highthroughput workflow that integrates modular polymer synthesis, robotic formulation, automated morphology characterization, and machine learning (ML) for accelerated SPB discovery. Using a plug-and-play synthetic strategy, 33 hydrogen-bonding endfunctional homopolymers were prepared and orthogonally combined to generate 260 SPBs in 1 day. A fully automated atomic force microscopy (AFM) pipeline enabled systematic imaging, producing 2340 morphology data sets with minimal human intervention. Domain spacings were extracted through complementary imageprocessing methods and used to train ML models. A support vector regression (SVR) model accurately predicted target phase-separation sizes (50, 100, and 150 nm), which were experimentally validated. This work demonstrates the power of coupling high-throughput experimentation with ML to accelerate morphology discovery and provides one of the first large-scale experimental data sets for supramolecular polymer systems.

ML-guided polymer design↗
DOE OSTI2026

Regularization by denoising diffusion models for solving inverse PDE problems with application to full waveform inversion

Partial differential equation (PDE)-governed inverse problems are fundamental across various scientific and engineering applications; yet they face significant challenges due to nonlinearity, ill-posedness, and sensitivity to noise. Here, we introduce a computational framework, regularization by denoising using diffusion models for partial differential equations (RED-DiffEq), by integrating physics-driven inversion and data-driven learning. RED-DiffEq leverages pretrained diffusion models as a regularization mechanism for PDE-governed inverse problems. We apply RED-DiffEq to solve the full waveform inversion problem in geophysics, a challenging seismic imaging technique that seeks to reconstruct high-resolution subsurface velocity models from seismic measurement data. Our method shows enhanced accuracy and robustness compared to benchmark methods. Additionally, it exhibits strong generalization and domain decomposition capacity, enabling the inversion of more complex velocity models with larger domains than those used in training the diffusion model. Our framework can also be directly applied to diverse PDE-governed inverse problems.

Shan, Siming [Yale University, New Haven, CT (Unit↗
DOE OSTI2026

Accelerating kinetic plasma simulations with machine-learning-generated initial conditions

Computational models of plasma technologies often solve for the system operating conditions by time-stepping an initial value problem to a quasi-steady solution. However, the strongly nonlinear and multi-timescale nature of plasma dynamics often necessitate millions, or even hundreds of millions, of steps to reach convergence, reducing the effectiveness of these simulations for computer-aided engineering. We consider acceleration of kinetic plasma simulations via data-driven machine-learning-generated initial conditions, which initialize the simulations close to their final quasi-steady-state, thereby reducing the number of steps to reach convergence. Three machine-learning models are developed to predict the density and ion kinetic profiles of capacitively coupled plasma discharges relevant to the microelectronics industry. The models are trained on kinetic simulations over a range of device operating frequencies and pressures. Best performance was observed when simulations were initialized with ion kinetic profiles generated by a convolutional neural network, reducing the mean number of steps to reach convergence by 17.1× when compared to initialization with a zero-dimensional global model. We also outline a workflow for continuous data-driven model improvement and simulation speedup, with the aim of generating sufficient data for full device digital twins.

Artificial neural networks↗
DOE OSTI2026

ZENN: A thermodynamics-inspired computational framework for heterogeneous data–driven modeling

Traditional entropy-based methods—such as cross-entropy loss in classification problems—have long been essential tools for representing the information uncertainty and physical disorder in data and for developing artificial intelligence algorithms. However, the rapid growth of data across various domains has introduced new challenges, particularly the integration of heterogeneous datasets with intrinsic disparities. To address this, we introduce a zentropy-enhanced neural network (ZENN), extending zentropy theory into the data science domain via intrinsic entropy, enabling more effective learning from heterogeneous data sources. ZENN simultaneously learns both energy and intrinsic entropy components, capturing the underlying structure of multisource data. To support this, we redesign the neural network architecture to better reflect the intrinsic properties and variability inherent in diverse datasets. We demonstrate the effectiveness of ZENN on classification tasks and energy landscape reconstructions, showing its superior generalization capabilities and robustness-particularly in predicting high-order derivatives. In image and text classification tasks, ZENN demonstrates superior generalization by introducing a learnable temperature variable that models latent multisource heterogeneity, allowing it to surpass state-of-the-art models on CIFAR-10/100, BBC News, and AG News. As a practical application in materials science, we employ ZENN to reconstruct the Helmholtz energy landscape of Fe3Pt using data generated from density functional theory and capture key material behaviors, including negative thermal expansion and the critical point in the temperature–pressure space. Overall, this work presents a zentropy-grounded framework for data-driven machine learning, positioning ZENN as a versatile and robust approach for scientific problems involving complex, heterogeneous datasets.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Latent Twins

Over the past decade, scientific machine learning has transformed the development of mathematical and computational frameworks for analyzing, modeling, and predicting complex systems. From inverse problems to numerical partial differential equations (PDEs), dynamical systems, and model reduction, these advances have pushed the boundaries of what can be simulated. Yet they have often progressed in parallel, with representation learning and algorithmic solution methods evolving largely as separate pipelines. With Latent Twins, we propose a unifying mathematical framework that creates a hidden surrogate in latent space for the underlying equations. Whereas digital twins mirror physical systems in the digital world, Latent Twins mirror mathematical systems in a learned latent space governed by operators. Through this lens, classical modeling, inversion, model reduction, and operator approximation all emerge as special cases of a single principle. We establish the fundamental approximation properties of Latent Twins for both ordinary differential equations (ODEs) and PDEs and demonstrate the framework across three representative settings: (i) canonical ODEs, capturing diverse dynamical regimes; (ii) a PDE benchmark using the shallow-water equations, contrasting Latent Twin simulations with deep operator network and forecasts with a four-dimensional variational method baseline; and (iii) a challenging real-data geopotential reanalysis dataset, reconstructing and forecasting from sparse, noisy observations. Latent Twins provide a compact, interpretable surrogate for solution operators that evaluate across arbitrary time gaps in a single-shot, while remaining compatible with scientific pipelines such as assimilation, control, and uncertainty quantification. Looking forward, this framework offers scalable, theory-grounded surrogates that bridge data-driven representation learning and classical scientific modeling across disciplines.

Latent Twins↗
DOE OSTI2026

spammR: an R package designed for analysis and integration of spatial multi-omic measurements

Spatial omics is a young and evolving field and as such shows rapid development of novel technologies and analysis methods to measure transcripts, proteins, metabolites, and post-translational modifications at high spatial resolution. These advances in technology have enabled the simultaneous generation of abundance profiles for multiple different omics types and associated microscopy imaging data, as well as their analysis in a spatial context. However, most analytical tools are designed for spatial transcriptomics platforms and are challenging to use in other contexts such as mass spectrometry-based measurements or metagenomics. To this end we present spammR (spatial analysis of multi-omics measurements in R), an R package that enables end-to-end analysis with a specific focus on mass-spectrometry derived spatial omics datasets with (1) smaller sample sizes and spatial sparsity of samples, (2) considerable missingness, and (3) no a-priori knowledge about proteins or genes of interest, relying on a fully data-driven approach.

spammR↗
DOE OSTI2026

TCAD-Machine Learning Enabled TID Compact Model Development for Commercial SiC MOSFET

We propose a TCAD (Technology Computer Aided Design)-machine learning coupled approach that combines a TCAD tool (Charon), optimization/uncertainty quantification tool (Dakota), surrogate models, and Bayesian learning capabilities. The coupling approach is used for accurate modeling and calibration of total ionizing dose (TID) induced threshold voltage (V th ) shifts in Commercial-Off-The-Shelf (COTS) semiconductor devices and to develop physics-informed TID compact models. This versatile approach is applied to model the TID effect in an exemplar COTS 3.3 kV SiC power MOSFET (Metal-Oxide-Semiconductor Field-Effect Transistor). With the Charon-Dakota coupling, we can determine key device geometry and doping values based on device physics, which are difficult to obtain or not available for COTS devices but important for TCAD simulation; additionally, we can efficiently generate thousands of simulation results in a large parameter space, which makes it possible to develop data-driven surrogate models and perform Bayesian calibration. Utilizing the full tool-coupling approach, we achieve calibrated TCAD simulation models that accurately capture the average TID-induced V th shifts behavior with total doses and V th shifts saturation at high doses as observed in experimental data. More importantly, the calibrated TCAD simulations are obtained with determined TID model parameters (e.g., hole trap density and capture cross section) values that contain well quantified uncertainties. Furthermore, we can isolate and quantify the noises that are not captured by the TCAD models but exist in the measured data due to measurements and devices variabilities. Lastly, the calibrated surrogate models are used to develop physics-informed TID compact models. The method is generalizable to other devices and/or radiation conditions with few modifications and can provide well-determined uncertainties.

COTS↗
DOE OSTI2026

DIVA/DeviceEditor v6.1.2

DIVA is an end-to-end DNA design and construction management platform that streamlines how researchers design, build, and receive sequence-verified DNA constructs. Through a web-based BioCAD interface (DeviceEditor), researchers independently design DNA constructs and submit them to a centralized queue with a single action. Designs progress transparently through standardized states which allow researchers to track status and access finished constructs via a central DNA repository. Submitted designs are reviewed by dedicated staff for feasibility and optimization, reducing costly failures and improving downstream execution. Automated DNA assembly software optimizes construction strategies by reusing existing parts where possible and sourcing synthetic DNA only when needed. Standardized, sequence-agnostic assembly methods enable many independent constructs to be built in parallel using lab automation, dramatically increasing throughput. High-throughput next-generation sequencing is used to verify construct accuracy, with flexible platforms selected based on task requirements. Throughout the process, detailed success and failure data are captured and analyzed, enabling continuous improvement of assembly protocols. Compared to traditional, manual DNA construction workflows, DIVA offers higher scalability, transparency, reproducibility, and data-driven optimization.

Plahar, Hector [Lawrence Berkeley National Laborat↗
DOE OSTI2026

Application of artificial intelligence methods in the international roughness index prediction of rigid and composite pavements: a systematic review

The International Roughness Index (IRI) is a widely adopted metric for quantifying pavement roughness, directly influencing vehicle safety, ride comfort, and overall roadway performance. In recent years, the use of Machine Learning (ML) models for IRI prediction has gained momentum, with the goal of improving the allocation of maintenance and rehabilitation resources by enabling accurate assessments of pavement conditions. Most prior reviews, however, have concentrated on flexible pavements, leaving a notable gap regarding rigid and composite pavements. To address this gap, the present study conducts a systematic review of Artificial Intelligence (AI) methods applied to IRI prediction for rigid and composite pavements. Literature published between 2004 and 2025 is synthesized to highlight prevailing trends, methodological contributions, and directions for future research. Particular attention is given to the types of models employed, the datasets used for training and validation, and the role of input variables and data-processing strategies. Across the included studies, ensemble learning methods (especially gradient boosting variants such as XGBoost), artificial neural networks, and hybrid architectures frequently achieved high predictive skill, with several models reporting test-set coefficients of determination approaching 0.9–0.96, indicating strong potential for capturing the influence of traffic, pavement structure, and climatic factors. Since these results are obtained from heterogeneous datasets and evaluation protocols, they are interpreted qualitatively rather than as strict cross-study rankings. Analysis of input variables revealed that pavement age and initial IRI were included in 91% (21 of 23) and 78% (18 of 23) of studies, respectively. Climatic variables such as the freezing index appeared in 57% (13 of 23), while traffic-related factors were considered in 65% (15 of 23). The findings underscore the importance of standardized, high-quality datasets, such as those from the Long-Term Pavement Performance (LTPP) program, along with data consistency, model interpretability, computational efficiency, and replicability in enhancing IRI prediction. Future research should focus on incorporating input variable selection techniques to identify the most influential predictors, thereby improving accuracy and robustness. Integrating these approaches with advanced non-linear data-driven models, coupled with robust hyperparameter optimization, holds considerable promise for strengthening the reliability of IRI prediction and supporting resilient pavement management strategies.

42 ENGINEERING↗
DOE OSTI2025

An Open Benchmark of One Million High-Fidelity Cislunar Trajectories

Cislunar space spans from geosynchronous altitudes to beyond the Moon and will underpin future exploration, science, and security operations. We describe and release an open dataset of one million numerically propagated cislunar trajectories generated with the open-source Space Situational Awareness Python package (SSAPy). The model includes high-degree Earth/Moon gravity, solar gravity, and Earth/Sun radiation pressure; other planetary gravities are omitted by design for computational efficiency. Initial conditions uniformly sample commonly used osculating-element ranges, and each trajectory is propagated for up to six years under a single, fixed start epoch. The dataset is intended as a reusable benchmark for method development (e.g., space domain awareness, navigation, and machine-learning pipelines), a reference library for statistical studies of orbit families, and a starting point for community-driven extensions (e.g., alternative epochs). We report empirically observed stability trends (e.g., a band near ~5 GEO and persistence of some co-orbital classes including L4/L5 librators) as dataset descriptors rather than new dynamical results. The chief contribution is the scale, fidelity, organization (CSV/HDF5 with full state time series and metadata), and open availability, which together lower the barrier to comparative and data-driven studies in the cislunar regime.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2026

Toward a microscopic picture of hadronization and multi-parton processes

This project advanced the understanding of how quarks and gluons produced in high-energy collisions transform into the hadrons observed in particle detectors, a fundamental process known as quantum chromodynamics (QCD) hadronization. By combining theoretical calculations, quantum simulation methods, and modern AI techniques, the research developed new tools to study multi-parton dynamics and nonperturbative effects that are essential for interpreting data from current and future nuclear physics experiments. Key outcomes include new theoretical frameworks for jet and hadron measurements, pioneering quantum simulation algorithms for real-time dynamics in field theories, and the development of advanced machine-learning models, such as diffusion models and explainable classifiers, to simulate and analyze collider events. These results are directly relevant to experiments at Jefferson Lab, Brookhaven National Laboratory, and the future Electron-Ion Collider, and they also have a broader impact in areas such as quantum information science and data-driven modeling of complex systems. The project supported the training of graduate students and postdoctoral fellows and contributed to the broader scientific community through publications, workshops, and collaborative activities. Overall, this work provides new insights into the microscopic mechanisms of hadron formation and establishes a foundation for future studies at the intersection of nuclear physics, artificial intelligence, and quantum computing.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2026

Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events

Danovo Energy Solution's presented its paper named: Feature-Based PMU Event Classification under Variable PMU Participation and Overlapping Events at the 2026 Georgia Tech Fault & Disturbance Analysis Conference. The full paper can be found at OSTI ID# 3169150 Paper Abstract—Phasor Measurement Units (PMUs) stream time synchronized, high-resolution measurements from the grid, enabling data-driven techniques for event detection and classification. Accurate event classification improves grid reliability and stability. Events can be detected by varying numbers of PMUs and exhibit different durations depending on the event type. This variability challenges standard classifiers that require uniform input sizes. Moreover, multiple events may coincide, which increases classification complexity. Standard classifiers assign each instance to the class with the highest predicted probability, whereas overlapping events may exhibit comparable probabilities across multiple classes. In this study, to handle data size variability, we extract a wide range of time–frequency domain features from all available PMUs for each event into a fixed-length vector, facilitating the application of standard machine learning classifiers, including Random Forest, XGBoost, LightGBM, Support Vector Machine, and Multilayer Perceptron. To account for overlapping events, a probabilistic post-processing step is applied. For a given data instance, if multiple predicted class probabilities exceed 30% and the differences between them are less than 10%, the event is assigned to multiple classes. Experiments using real-world PMU data demonstrate that the Random Forest and XGBoost models achieve the highest accuracy, while the proposed post-processing method yields perfect classification performance on external unseen test sets.

Nematirad, Reza [Danova Energy Solutions]↗
DOE OSTI2026

Detecting and Characterizing Fracture Zones Using a Convolutional Neural Network

This project directly supports the Geothermal Technologies Office (GTO) objectives outlined in the Multi-Year Program Plan (MYPP) by advancing two key research areas: “Exploration and Characterization” and “Data, Modeling, and Analysis.” This project has successfully demonstrated a pre-drilling ability to image and characterize the distribution and connectivity of subsurface faults and fractures, key parameters for identifying permeable pathways that enable geothermal fluids to circulate and produce energy. Specifically, we developed and implemented innovative machine learning methodologies to enhance geothermal exploration. Large-scale faults were detected using a Convolutional Neural Network (CNN), while small-scale fractures were characterized using a novel Double-Beam Neural Network (DBNN). These tools have proven both technically effective and cost-efficient by reducing reliance on expensive exploratory drilling. Through collaboration with our geothermal industry partner, this research has significantly advanced techniques for identifying hidden geothermal systems and extending the productive lifespan of existing geothermal fields. We applied our methods to two geothermal fields—Soda Lake (Nevada) and Lightning Dock (New Mexico)—to identify shallow steam-charged fracture zones and characterize deep faults at depths of 1.5-2 km. The steam zone identified at the Soda Lake geothermal field showed excellent agreement with prior drilling data, validating the effectiveness of our approaches. In addition, the analysis revealed three new prospective drilling targets for further development and verification. The outcomes of this project improve our scientific understanding of geothermal reservoir behavior, enhance exploration efficiency, extend the economic life of existing geothermal plants. Ultimately, these advancements contribute to GTO’s goal of achieving more sustainable, affordable, and data-driven geothermal energy development across the United States.

15 GEOTHERMAL ENERGY↗
DOE OSTI2025

Comparing Outdoor to Indoor Performance for Bifacial Modules Affected by Polarization-Type Potential-Induced Degradation

Bifacial photovoltaic (PV) modules have the advantage of using light reflected off of the ground to contribute to power production. Predicting the energy gain is challenging and requires complex models to do so accurately. Often, module degradation over time is neglected in models for the sake of simplicity or is underestimated. Comparing outdoor and indoor current–voltage (I–V) performance for bifacial modules is more challenging than for monofacial modules, as there are additional variables to consider such as rear albedo non-uniformity, cell mismatch, and their effects on temperature. This challenge is compounded when heterogeneous degradation modes occur, such as polarization-type potential-induced degradation (PID-p). To examine the effects of PID-p on I–V predictions using an empirical data-driven approach, 16 bifacial PERC modules are installed outdoors on racks with different albedo conditions. A subset is exposed to high-voltage biases of −1500 V or +1500 V. Outdoor data are traced at irradiance ranges of 150–250 W/m 2 , 500–600 W/m 2 , and 900–1000 W/m 2 . These curves are corrected using control module temperature, wire resistivity, and module resistance measured indoors. We examine several methods to transform indoor I–V curves to accurately, and more simply than existing methods, approximate outdoor performance for bifacial modules without and with varying levels of PID-p degradation. This way, bifacial performance modeling can be more accessible and informed by fielded, degraded modules. Distributions of percent errors between indoor and outdoor performance parameters and Mean Absolute Percent Errors (MAPEs) are used to assess method quality. Results including low-irradiance data (150–250 W/m 2 ) are discussed but are filtered for quantifying method quality as these data introduce substantial errors. The method with the most optimal tradeoff between low MAPE and analysis simplicity involves measuring the front side of a module indoors at an irradiance equal to plane-of-array irradiance plus the product of module bifaciality and albedo irradiance. This method gives MAPE values of 1–6.5% for non-degraded and 1.6–5.9% for PID-p degraded module performance.

14 SOLAR ENERGY↗
DOE OSTI2026

Flux Cube Reconstruction from Slitless Spectroscopy

Slitless spectroscopy enables efficient, large-area surveys without target preselection, yet it faces challenges from source blending, higher noise, and lost spatial–spectral information. We present an advanced, nonparametric, data-driven algorithm that leverages multiple dispersion angles to reconstruct three-dimensional flux distributions, providing low-resolution integral field unit capabilities from slitless data. By treating each pixel as an independent element, our method naturally handles source confusion without requiring prior assumptions regarding redshifts, templates, or model libraries. We validate the algorithm using simulated Roman Space Telescope wide-field slitless spectroscopy images that are equivalent to what is expected from the High-Latitude Time-Domain Survey. First, we demonstrate that a host-galaxy model reconstructed from multiple dispersion angles can be used to accurately subtract host light from a transient, recovering a Type Ia supernova spectrum with minimal bias. Second, we showcase a high-fidelity flux-cube reconstruction of a complex galaxy, successfully measuring the redshift and recovering continuum, emission, and absorption features. This approach highlights the potential of multi-dispersion-angle slitless data to provide spatially resolved spectral information in a nonparametric way, which is traditionally accessible only with integral field spectroscopy, opening a new window into large, unbiased, and spatially resolved studies of galaxy evolution.

Griggio, M. [Space Telescope Science Institute, Ba↗
DOE OSTI2026

Leveraging machine learning to enhance aerosol classification using Single-Particle Mass Spectrometry

Advancing automated classification of atmospheric aerosols from Single-Particle Mass Spectrometry (SPMS) data remains challenging due to overlapping ion signatures, compositional diversity, and limited labeled data. This study evaluates supervised and semi-supervised learning frameworks to enhance aerosol identification by jointly leveraging labeled and unlabeled spectra. Four models were compared: a supervised Support Vector Machine (SVM), a self-training SVM, a stacked autoencoder classifier, and a stacked autoencoder trained using a temporal-ensembling Mean Teacher approach. All models achieved high and stable accuracies (90.0 %–91.1 %), surpassing previous results on the same dataset (87 %) and matching the performance of state-of-the-art deep learning methods. Despite small global metric differences (≤ 1 %), semi-supervised variants yielded up to 5 %–10 % improvements for compositionally rare particle types – such as soot (0.77 % of spectra, F1-score: 0.93–0.97) and hazelnut pollen (0.98 % of spectra, F1-score: 0.97–1.00) – equating to roughly ∼ 187 additional correctly classified spectra. These gains are scientifically significant, as such rare particles exert disproportionate influence on radiative absorption and ice nucleation processes; their improved detection reduces modeled uncertainties in aerosol absorption optical depth and mixed-phase cloud ice nucleation rates. The models' residual misclassifications (≈ 9 %) largely arise from true spectral overlap among chemically adjacent species (e.g., Na- vs. K-feldspar, coated vs. uncoated feldspars), reflecting physical compositional continuity rather than algorithmic error. Collectively, these findings demonstrate that leveraging unlabeled data to learn robust spectral representations and refine classification enhances both fidelity and interpretability, bridging data-driven analysis with aerosol–climate process understanding.

54 ENVIRONMENTAL SCIENCES↗