SEARCH · Engineering Papers

Results for “regression”

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.

29 records · Page 2

All sources NASA NTRS DOE OSTI

DOE OSTI2026

Conditional diffusion machine-learning framework for mapping valence electron distribution from convergent beam electron diffraction

Quantitative convergent beam electron diffraction (CBED) enables determination of aspherical valence electron distributions through refinement of low-order structure factors, which are highly sensitive to chemical bonding and charge density variations. However, conventional quantitative CBED (QCBED) requires solving a highly nonlinear inverse problem with many coupled parameters, and computationally intensive dynamical diffraction calculations, making it time-consuming and difficult to apply to complex systems. More broadly, reconstructing charge density and orbital electron distribution from diffraction data has long been a central challenge in both x-ray and electron crystallography. Here, in this study, we introduce an artificial-intelligence (AI)-based framework that replaces traditional refinement with a data-driven inverse solver. Using a large synthetic CBED dataset generated by Bloch-wave simulations, we train a conditional diffusion model to directly infer crystal structural parameters and multipole density formalism parameters, and hence valence electron distributions, from CBED patterns alone. By learning from forward simulations across realistic parameter space, the model effectively solves the inverse problem. Compared with direct regression approaches, the diffusion-based framework provides posterior parameter distributions for rigorous uncertainty quantification while preserving quantitative fidelity and reducing analysis time by orders of magnitude. By eliminating the need for external single-crystal x-ray diffraction data and complex nonlinear refinement, this approach enables practical, high-throughput, and in situ quantitative CBED, enabling real-time mapping of valence electron distributions and their correlation with functional responses in quantum and energy materials.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

DOE OSTI2026

Multitask graph neural networks for elastoplastic response prediction in dual-phase polycrystals

Microstructure-sensitive prediction of elastoplastic response remains a recurring bottleneck in multiscale damage and fatigue modeling, where large ensembles of statistically distinct polycrystals are required to quantify variability and extreme-value behavior. In this work, we develop a multitask graph neural network (GNN) surrogate that maps dual-phase ferrite–martensite polycrystal microstructures to Statistical Volume Element (SVE)-level elastoplastic Quantities of Interest (QoIs). Each SVE is represented as a grain-adjacency graph, with node features encoding phase, geometry, and crystallographic orientation, and edge features encoding relative misorientation. A message-passing graph convolution generates node embeddings, which are pooled into a graph representation and passed to a multitask regression head that jointly predicts 10 scalar QoIs and vector-valued stress–strain responses in orthogonal loading directions across multiple martensite volume fractions and SVE sizes. Results show high accuracy for scalar QoIs and strong agreement for full stress–strain trajectories, with population envelopes reproducing both median behavior and finite-SVE variability across compositions and partition scales. A unified model trained on pooled volume-fraction data preserves most within-regime accuracy relative to regime-specific models while also capturing the broader cross-regime variation reflected in the pooled test set. Distributional comparisons further demonstrate that the surrogate preserves heterogeneity under SVE partitioning, enabling statistically consistent block-wise random-field construction for mesoscale analyses. Overall, the proposed grain-graph surrogate provides a practical pathway to accelerate ensemble-based studies of SVE-level constitutive variability in dual-phase polycrystals.

Crystal plasticity↗

NASA NTRSDate not supplied

BEAST: Expanding Sustainable Data Infrastructure for High-Enthalpy Facilities

Reproducible, data-driven thermal protection system (TPS) research requires that experimental records from high-enthalpy testing be consistently structured, traceable, and accessible across campaigns and institutions. In practice, however, arcjet and plasma facilities data remain largely fragmented: raw diagnostics are stored in ad hoc formats, material sample histories are disconnected from test conditions, and metadata standards are absent, precluding systematic cross-campaign analysis and long-term reuse. BEAST (Backend for Experiment Analysis, Storage, and Traceability) is an open-source, web-based platform that addresses these limitations by providing a unified, queryable infrastructure for high-enthalpy ground-test data [1]. First presented at the 15th Ablation Workshop [2], BEAST has since undergone significant development. The platform ingests and structures multi-channel time-series diagnostics, facility configurations, and material property records within a common provenance model, ensuring end-to-end traceability from raw sensor acquisition to reduced experimental quantities. A versioned material library links specimen identity and processing history to the specific runs in which each sample was tested. An integrated modeling workbench enables training and evaluation of regression models directly on archived experimental data, supporting condition interpolation and the construction of empirical material response databases. Beyond its original deployment at NASA Ames Research Center, BEAST has been designed to be facility-agnostic, with ongoing efforts to extend its adoption to other facilities. Its modular architecture accommodates heterogeneous diagnostic setups and facility types, and its future open-source distribution allows institutions to build on a common data standard rather than maintaining isolated, bespoke solutions. BEAST is further integrated within a broader ecosystem of companion tools: arcjetCV [3] extracts recession rates and shock standoff distances from high-speed video using computer vision, and miniSTARscan [4] provides sub-minute, portable photogrammetric surface reconstruction of test articles before and after exposure. All tools share a common data schema, enabling seamless ingestion of surface geometry, imagery, and time-series data into a single, coherent experimental record.

Database↗

NASA NTRSDate not supplied

BEAST: Expanding Sustainable Data Infrastructure for High-Enthalpy Facilities

Database↗

DOE OSTI2026

Explainable machine learning reveals that local structural motifs encode the thermodynamic state across the CuZr metallic glass-forming range

Metallic glasses derive their properties from the statistics of local atomic motifs rather than from long-range order, yet a quantitative, chemistry-specific link between motif populations and the underlying glassy state has remained elusive. In this work we combine large-scale molecular dynamics, Voronoi tessellation, deep neural networks, and SHapley Additive exPlanations (SHAP) to identify which local structural motifs define the glassy state of Cu—Zr metallic glasses. A dataset of 17,180 atomistic configurations spanning ten compositions (Cu 20 Zr 80 –Cu 80 Zr 20 ) and four quench rates (10 9 –10 12 K/s) is used to train a feed-forward neural network that regresses temperature across the 50–2000 K liquid–supercooled–glass range, achieving a mean absolute error of 19.89 K and R 2 = 0.9974, confirming that the local structural state is faithfully encoded in motif-level structure. SHAP analysis then reveals that a tightly coupled near-icosahedral family of motifs (coordination numbers (CN) 11–13, including the full icosahedron 001200 and its single-atom-perturbation sibling 10930) collectively encodes the thermodynamic state of the system across the full glass-forming range. The CN = 11–13 ordered members carry negative SHAP values at high populations, tracking the most deeply-quenched configurations, while 10930 shows the reversed signature consistent with its role as a soft-spot host whose population shrinks as the icosahedral network deepens. The analysis demonstrates that explainable machine learning can isolate the minimal motif vocabulary defining the glassy state and recovers the near-icosahedral building blocks previously identified by data-driven analyses of Cu—Zr. The approach provides a general, chemistry-specific route for characterizing the structural state of disordered materials.

36 MATERIALS SCIENCE↗

DOE OSTI2026

Heat exposure and maternal stress: evidence from the GRAPHS pregnancy cohort in Ghana

Heat exposure has been linked to psychosocial stress, an established antecedent of perinatal depression; however, evidence on heat-related stress during pregnancy in sub-Saharan Africa remains limited. We analyzed psychosocial stress scores and covariate data from the Ghana Randomized Air Pollution and Health Study, linking daily maximum and minimum shaded wet bulb globe temperature (WBGT) metrics to participants’ stress scores derived from the Crisis in Family Systems-Revised Life Events Questionnaire. We evaluated associations using ordinal logistic regression of pregnancy-average and trimester-average exposures and distributed lag non-linear models (DLNMs) to assess time-varying associations across gestation. Higher average maximum WBGT exposure across pregnancy was associated with increased odds of higher psychosocial stress; each 1 °C increase in maximum WBGT was associated with 64% higher odds of belonging to a higher stress category (OR = 1.64; 95% CI = 1.17–2.31; p = 0.0040). In trimester-average models, higher first-trimester maximum WBGT was also associated with higher stress (OR = 1.44; 95% CI = 1.15–1.81; p = 0.0014). DLNMs suggested that relatively cooler daily maximum WBGT values (25th percentile) were associated with decreased odds of stress in early pregnancy, whereas extreme daily maximum WBGT values (99th percentile) showed a pattern consistent with increased odds of stress in mid-to-late gestation. These findings highlight gestational windows in which heat exposure may influence stress, emphasizing the need for further research into underlying mechanisms and effective interventions to protect maternal mental health in heat-vulnerable settings.

White, Lewis [Columbia University] (ORCID:00090005↗

NASA NTRS2026

Updated global and regional trends of stratospheric ozone profiles

We present updated evaluation of stratospheric ozone profile trends in the 60° S–60°N latitude range using long-term ground-based and satellite climate data records, as well as simulations by chemistry-climate models. The trends are evaluated using the LOTUS (Long-term Ozone Trends and Uncertainties in the Stratosphere) regression model. Analyses of satellite data confirm the statistically significant positive ozone trends in the period 2000–2024 in the upper stratosphere of ~1–3% per decade, with larger trends at mid-latitudes compared to the tropics. The trends are slightly positive or close to zero in the middle stratosphere, and mostly negative, -1 to -2% per decade, in the lower stratosphere, but they are not statistically significant. The morphology and magnitude of ozone trends are similar to previous analyses (2000–2020 trends). Ozone trends in 2000–2024 predicted by chemistry-climate model simulations are in good agreement with combined satellite trends. In the upper stratosphere, models predict a slightly stronger ozone recovery than observations. In the lower stratosphere, both models and satellite observations report negative trends in the tropics, while modelled ozone trends are slightly positive at mid-latitudes. Ozone profile trends over several stations estimated from ground-based records capture the same overall vertical pattern of ozone trends as merged gridded satellite datasets. Analyses of regional ozone profile trends in 2003–2024 using merged satellite datasets confirmed the previous observations of a longitudinal structure in ozone trends in the NH mid-latitude stratosphere, with positive trends over Scandinavia and negative trends over Siberia. However, the magnitude of this dipole-like structure is reduced compared to previous analyses.

trends↗

NASA NTRS2026

Machine Learning for Predicting Team Functioning in HERA Missions

Team functioning is integral to success in future long term space exploration missions. Proactively detecting declines in team functioning can mitigate conflict and ensure mission success. This project developed a speech-based artificial intelligence (AI) system that unobtrusively predicts degradation in team functioning, including performance and cohesion, in the Human Exploration Research Analog (HERA) Campaigns 4 and 5. The AI system conducted automated analysis of the prosodic (tone of voice) and linguistic (language content) components of speech, modeling interpersonal dynamics at both the turn-taking and day-wide levels. We investigated team functioning via observing structured interactions (i.e., multi-mission space exploration vehicle-extra vehicular activity [MMSEV-EVA], team interaction battery [TIB]) and unstructured interactions before the MMSEV-EVA task. We developed machine learning models to predict team functioning (objective task accuracy, self reported team efficacy and self reported team cohesion) by analyzing OpenSmile acoustic features, linguistic descriptors extracted via the linguistic inquiry and word count (LIWC) dictionary, and semantic embeddings. In the TIB, static models using logistic regression and random forests were not able to predict task accuracy, but predicted team efficacy and cohesion during both the decision making and relational tasks to a moderate level (60-70%). Majority voting on the individual turns to predict day long team efficacy further increased accuracies (70-80%). Finally, long short-term memory (LSTM) models showed the best performance across all variables (80-91%), including task performance. In the MMSEV-EVA, static models achieved an accuracy of 60% with majority voting, which increased to 80% through the incorporation of mission day as a variable, accounting for the learning effect. A key finding across both tasks was the "team-dependent" nature of these interactions; models achieved much higher accuracy when trained on prior days of the same team's data rather than attempting to generalize across entirely different teams, with even 1-2 days of prior data per team achieving 5-15% improvement over team-independent models. In addition, the incorporation of pre-task data from the same team also improves model performance, e.g., incorporating data from the decision-making task of the TIB, which preceded the relational task, improved the prediction of team efficacy and cohesion during the latter. We compared model performance when trained on machine-generated data compared to data that had been further corrected by human annotators. Overall, models trained on human-corrected data exhibited a modest improvement in performance, particularly when acoustic features were used. We found no significant correlation between word error rate (WER) and model accuracy (r(55) = -0.08, p = 0.51), but model’s accuracy was significantly higher for medium/high quality transcription (0.74 (SD = 0.48)) compared to the low-quality group (0.64 (SD = 0.36)) (t(63)=2.82, p = 0.006). Based on these, several design recommendation emerge, that could inform Standards at NASA. Models predicting team functioning should incorporate at least one to two days of historical interaction data, include brief pre-task discussions, and explicitly model temporal learning effects, especially for longer operational tasks. Minimum quality standards for automated speech-processing pipelines are needed, given the performance gains observed with manually corrected acoustic data. Finally, systems should leverage both acoustic features and language embeddings in complementary ways, with modality choices and fusion strategies tailored to mission context, task demands, and data quality requirements.

Shrivatsa Mishra↗

DOE OSTI2025

Plasma phosphorylated tau217 strongly associates with memory deficits in the Alzheimer’s disease spectrum

Abstract Plasma phosphorylated tau (p-tau) biomarkers open unprecedented opportunities for identifying carriers of Alzheimer’s disease pathophysiology in early disease stages using minimally invasive techniques. Plasma p-tau biomarkers are believed to reflect tau phosphorylation and secretion. However, it remains unclear to what extent the magnitude of plasma p-tau abnormalities reflects neuronal network disturbance in the form of cognitive impairment. To address this question, we included 103 cognitively unimpaired elderly and 40 cognitively impaired, amyloid-β-positive individuals from the TRIAD cohort, in addition to 336 cognitively unimpaired and 216 cognitively impaired, amyloid-β-positive older adults from the BioFINDER-2 cohort. Participants had tau PET scans, amyloid PET scans or amyloid CSF, p-tau217, p-tau181 and p-tau231 blood measures, structural T1-MRI and cognitive assessments. In this cross-sectional study, we used regression models and correlation analyses to assess the relationship between plasma biomarkers and cognitive scores. Furthermore, we applied receiver operating characteristic curves to assess cognitive impairment across plasma biomarkers. Finally, we categorized participants into amyloid (A), p-tau (T1) and tau PET (T2) positive (+) or negative (−) profiles and ran non-parametric comparisons to assess differences across cognitive domains. We found that plasma p-tau217 was more associated with cognitive performance than p-tau181 and p-tau231 and that this relationship was particularly strong for memory scores (TRIAD: βp-tau217 = −0.53, βp-tau181 = −0.35 and βp-tau231 = −0.24; BioFINDER-2: βp-tau217 = −0.52, βp-tau181 = −0.24 and βp-tau231 = −0.29). Associations in amyloid-β-positive participants resembled these results, but other cognitive scores also showed strong associations in cognitively impaired individuals. Moreover, plasma p-tau217 outperformed plasma p-tau181 and plasma p-tau231 in identifying memory impairment (area under the curve values for TRIAD: p-tau217 = 0.86, p-tau181 = 0.77 and p-tau231 = 0.75; and for BioFINDER-2: p-tau217 = 0.86, p-tau181 = 0.76 and p-tau231 = 0.81) and in identifying executive function impairment only in the BioFINDER-2 cohort (p-tau217 = 0.82, p-tau181 = 0.76 and p-tau231 = 0.76). Lastly, we showed that subtle memory deficits were present in A+T1+T2− participants for plasma p-tau217 (P = 0.007) and plasma p-tau181 (P = 0.01) in the TRIAD cohort and for all biomarkers across cognitive domains in A+T1+T2− and A+T1+T2− individuals (P < 0.001 in all) in the BioFINDER-2 cohort. The A+T1+T2− individuals showed cognitive deficits in both cohorts (P < 0.001 in all). Together, our results suggest that plasma p-tau217 stands out as a biomarker capable of identifying memory deficits attributable to Alzheimer’s disease and that memory impairment certainly occurs in amyloid-β- and plasma p-tau-positive individuals who have no significant amounts of tau in the neocortex.

Neurosciences & Neurology↗

DOE OSTI2025

Field Validation of Thermoelectric Generation System at Holcim Cement Plant in Alpena, Michigan

Executive Summary Project Background The Industrial Technology Validation (ITV) program aims to identify and demonstrate the performance of new, emerging, and underutilized energy-saving technologies in the industrial sector to help inform decisions to help accelerate their commercialization and deployment, as well as to help make industries more competitive. This ITV demonstration evaluated a thermoelectric generation (TEG) technology at a cement plant, aiming to reduce energy demand in the cement industry. A median cement plant consumes 5.73 million British thermal units per ton of clinker production (resulting in 0.838 metric tons of carbon dioxide [CO₂] emissions per ton of clinker) (Boyd and Zhang 2011, EPA 2021), equivalent to approximately 6.9 trillion British thermal units (TBtu) per year in energy consumption at a cement plant producing 3,300 tons of clinker per day.¹ Collaborating with Holcim, Advanced Thermovoltaic Systems (ATS) developed and deployed a pilot-scale thermoelectric power system to efficiently capture and convert waste heat to electricity. The system leverages the Seebeck effect to convert temperature differences on two sides of semiconductor cartridges into electrical power (ScienceDirect, n.d.). This generation is realized with minimal moving parts compared to existing waste-heat-to-generation solutions and allows capture from heat sources with temperatures as low as 150°C. This project aimed to validate a scalable solution applicable for capturing medium-temperature waste heat, including ambient losses from other high-temperature processes, and high-temperature sources less suitable for other waste-heat-to-power solutions. By recovering this otherwise wasted heat, this project intends to validate improvements to overall process efficiency through reduction in purchased electricity, thereby reducing operational costs while enhancing resiliency and competitiveness. Description and Scope This study evaluated the performance of a TEG system from ATS as a solution to convert waste heat into useful power at a Holcim cement plant in Alpena, Michigan. This plant is a fully integrated cement plant that has been operating since 1907. The facility operates continuously (24/7/365) with approximately 250 employees and five long dry kilns, yielding a total production capacity of 7,852 tons of cement per day (EPA 2023). Currently, the Alpena plant uses waste heat boilers to convert waste heat from the exhaust of each kiln into steam, which drives steam turbine generators. The ATS TEG is being evaluated for its potential to supplement the steam turbines by capturing the remaining lower grade heat. This technology is also being considered for other Holcim plants where steam turbines are not a viable option. ATS installed a pilot-scale TEG unit with an array of 582 individual thermoelectric semiconductor cartridges, of which 573 were operational. The cartridges are sandwiched between 48 hot plates and 49 cold plates. Each cartridge is designed to generate 20 watts (W) of gross power at a hot-side temperature of 240°C and cold-side temperature of 20°C. As such, the total gross generation capacity of the installed system is 11.5 kilowatts (kW) at design conditions. The system configuration for the evaluation was designed to prioritize convenience of installation and minimize disruption to production at the site, while ensuring that the heat required can be obtained for evaluating the TEG system at various operational conditions. To accomplish this, a portion of the steam supplied to Alpena’s steam turbine generation system was diverted to be used as the heat source for the TEG system, while water was supplied to the cold side of the system from nearby Lake Huron. This configuration was designed for the evaluation of the pilot-scale system to assess the performance at different conditions. A commercial-scale system will likely vary from the pilot system depending on typical configurations, including both scale and application. Future commercial applications of the ATS system would involve integrating the system into the exhaust from kiln preheaters, clinker coolers, or radiant heat capture from kiln shells for the heat source. For the cold source, a range of cooling solutions can be considered, including a mechanical cooling system, depending on the location and the application. To increase the generation capacity for commercial applications, the technology provider is working toward developing a commercial-scale TEG system, which would combine multiple TEG units (each similar in design to the pilot system) together. The scope of this evaluation includes the pilot-scale TEG system and all impacted equipment including pumps, controllers, and power handling equipment. Study Objectives The evaluation's goal was to assess the potential of the ATS TEG system to generate useful electrical power by capturing waste heat from cement production kilns. The objectives of this study are to evaluate and verify the following claims made by ATS regarding the pilot-scale system installed at the Holcim Alpena plant. The following design parameters and claims are also outlined in Table ES- 1 and Table ES- 2: • Gross Power: The thermoelectric system converts heat into power to create gross power, the total measured power generated by the system. The 573 active cartridge pilot-scale system is expected to generate 11.5 kW of gross power at the designed hot-side temperature of 240°C and cold-side temperature of 20°C. Power production is dependent on the temperature difference between the heat source (ultimately from the waste heat) and cold temperature supply source. • Net Power: The net power is the total usable power provided to the site by the TEG system after deducting parasitic power loads from the gross generated power. Supplementary equipment is required to operate the TEG system including pumps, controllers, and, in certain anticipated applications, mechanical cooling, which introduce parasitic loads to system operation. After deducting the parasitic loads from the gross power generation, ATS anticipates achieving a net power generation of 7.5 kW from the pilot-scale system. • Thermal Efficiency: The thermal efficiency is the percent of the total heat transferred to the TEG system that is converted to gross power. Historically, TEGs have a thermal efficiency of 2%–5% (DOE 2008). Prior industrial-scale TEG systems, such as the E1 TEG offered by Alphabet Energy, operated at an efficiency of 2.5% (Lamonica, 2014). ATS anticipates achieving an average efficiency of 4.8% or higher in converting heat energy to usable electricity. • Cartridge Performance: The TEG system comprises 573 active individual semiconductor cartridges, each of which generates a portion of the total power. Cartridge optimization and selection is an important design consideration for potential future TEG system design performance. Therefore, understanding the distribution of gross power and efficiency within the pilot system is vital to understanding what is achievable. At a design hot-side temperature of 240°C and cold-side temperature of 20°C, ATS anticipates a cartridge performance of 20 W of gross power per cartridge at an efficiency of 4.8% per cartridge. In addition to evaluating the claimed performance of the TEG pilot-scale unit, the study estimated the potential annual impacts of a scaled-up commercial system used to capture kiln waste heat over annual operations. The evaluation estimated the gross and net annual electric generation achievable by capturing heat from the two proposed tap-in points: the kiln exhaust and the clinker cooler exhaust; see Section 2.1 for details. Two use cases were examined: • Holcim Alpena: The Holcim Alpena site consists of long dry kilns with superheater boilers, which differs from the rest of Holcim’s cement plant portfolio and results in lower waste heat temperatures. The study estimates gross and net annual generation using the superheater boiler exhaust and clinker cooler exhaust, based on 2023 operational data. • Typical Installation: Common cement plants have preheater kilns with higher exhaust temperatures than Holcim Alpena across a range of production rates. The study estimates gross and net annual generation using the preheater exhaust and clinker cooler exhaust, with a sensitivity analysis to account for the typical range of preheater exhaust temperatures, clinker cooler exhaust temperatures, and clinker production rates. Methodology The evaluation methodology followed a measurement and verification (M&V) strategy based on the International Performance Measurement and Verification Protocol Option B through comprehensive measurements and analyses of the affected systems. Evaluation data was collected from March 9 to March 11, 2024, the test period of the pilot TEG system. During the test period, in coordination with the ITV team, the ATS team adjusted system operations to capture the range of variability expected for each of the variables pertinent to performance of the system. The methodology consisted of two parts: evaluating the performance of the pilot unit's TEG system and estimating the annual TEG impact in terms of gross and net power based on a given waste heat profile. First, the evaluation of the thermoelectric generation performance of the pilot unit relative to the claims was performed by analyzing the collected test data. Gross power of the pilot TEG system was directly measured. Net power was determined by deducting the measured parasitic power from the gross power. The gross power generation was compared to heat transferred to the system by the working fluid (which was heated by steam generated from the kiln waste heat) to calculate the thermal efficiency achieved by the system. Performance of individual semiconductor cartridges within the pilot array was also assessed in terms of measured gross cartridge power and calculated cartridge thermal efficiency. The second part of the evaluation estimated the annual TEG impacts in terms of gross power and net power (calculated from the difference between gross power and parasitic power). This analysis comprised development of mathematical regression models for gross power and parasitic power, with assessment of each model’s goodness-of-fit characteristics to ensure satisfaction of statistical requirements. The models predicted the gross power generation, the parasitic load based on the temperature difference between the hot working fluid and the cold-side fluid (cold water from Lake Huron) entering the system, the volumetric flow rate of the cold-side fluid at the inlet, and the volumetric flow rate of the hot working fluid at the inlet. The annual impact analysis considered a theoretical commercial-scale system sized to capture the available waste heat at a cement plant, consisting of linked pilot-scale units that receive heat from a theoretical gas-to-working-fluid heat exchanger. To estimate annual impacts at the Alpena plant, the gross power and parasitic power regression models were applied to the arrays in the theoretical commercial-scale system. The heat supplied to the unit was calculated based on the kiln run time, annual production, kiln exhaust waste heat, and clinker cooler waste heat derived from 2023 Holcim Alpena kiln operational data. Net power impacts were calculated by deducting the resulting parasitic power from the estimated gross power. Inputs for the model were generated from a combination of hourly data, assumed design considerations for TEG system scale-up from the pilot-scale unit, and assumptions regarding TEG system operations. This analysis was then used as the basis for estimating annual impacts of typical TEG installation at cement plants, by applying sensitivity analyses to key kiln operational characteristics including kiln preheater exhaust temperatures, cooler clinker exhaust temperatures, and plant daily production rates across a range of expected values. Project Results/Findings Table ES- 2 and Table ES- 2 provide a summary of the operating conditions and evaluation results compared to the stated claims from the technology provider. Key takeaways include: • Gross Power: The peak gross power achieved during the testing period was 10.0 kW, compared to the 11.5 kW expected for 573 active cartridges. The claimed gross power was associated with a target hot side of 240°C; however, the system only received a maximum hot-side mean plate temperature of 212°C during the testing period. • Net Power: The pilot-scale unit exceeded the claims for net power, achieving a peak of 7.7 kW net compared to a claim of 7.5 kW. One factor contributing to the higher achieved net power is the relatively high water pressure available through Lake Huron. The pilot TEG system did not require cold-side pumps during the test, whereas most installations would. This reduced the parasitic loads on the system, ultimately contributing to higher net power relative to the gross power. • Thermal Efficiency: The pilot-scale unit outperformed the claimed efficiency, achieving a peak system efficiency of 5.0% thermal efficiency compared to the stated 4.8%. • Cartridge Performance: To compare cartridge performance against claims, the study focused on the third day of testing, which aimed for conditions closest to the design specifications, with a hot side of 240°C and cold-side exit temperature of 6.4°–30°C. On this day, the mean gross power observed in the cartridges within the TEG array was 18.1 W/cartridge, and the peak performance was 34.7 W/cartridge. The estimated mean cartridge efficiency was 5.2%, and the estimated efficiency at peak gross cartridge power was 10%. The regression models developed for gross power generation and parasitic loads were used to estimate the generation impact for given heat input to the TEG from the working fluid (captured from the waste heat) and from the cold loop (Lake Huron) on an hourly basis for a year of operation. Based on this analysis, installation of a commercial-scale TEG system at the Holcim cement plant in Alpena, Michigan, with a waste heat exchanger of 0.85 effectiveness, would generate up to 391 kW of net power, translating to between 920,000 and 1,800,000 kilowatt-hours (kWh) in net electricity per year. Based on typical grid emissions for Alpena, this would avoid estimated net emissions by 752 metric tons of CO₂ annually.² The sensitivity analysis estimated that typical TEG system installations at cement plants could generate an average of 56–1,040 kW of net power, or between 488,000 and 9,110,000 kWh of net energy. This generation potential is most significantly affected by plant production rates and also influenced by preheater and clinker cooler exhaust temperatures. Applying the national average emission rate, typical commercial-scale installations at Holcim plants are projected to avoid between 182 and 3,401 metric tons of CO₂ annually per site. Table ES- 3 shows a summary of the estimated annual impacts.³ While parasitic loads are significant and vary by application, this analysis assumed the use of heating loop pumps and access to Lake Huron as a cold sink. This setup assumed no need for cooling loop pumps due to the available water pressure at the test site. Applications that require cooling towers or additional equipment are likely to experience higher parasitic loads. Therefore, the study’s estimates are most applicable to scenarios with similar parasitic load configurations—namely, access to a high-pressure cold sink. Applicability to other locations may be limited, as differing conditions could necessitate additional pumps and cooling systems, potentially impacting performance significantly.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

DOE OSTI2026

HydraGNN_Predictive_GFM_2026 - Ensemble of predictive graph foundation models for atomistic materials modeling

This release contains data and parameters of HydraGNN-based graph foundation models trained as a result of the work published in the pre-print "Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data" by M. Lupo Pasini et al. (https://arxiv.org/abs/2604.15380). We jointly train on 16 open first-principles datasets (544+ million structures covering 85+ elements) using a multi-task architecture with per-dataset heads and a scalable ADIOS2/DDStore data pipeline. On Frontier, we execute six large-scale DeepHyper hyperparameter optimization campaigns in FP64 and promote the top-performing message-passing models to sustained 2,048-node training, yielding a PaiNN-based lead model. The version of HydraGNN used to generate the outputs provided in this release is HydraGNN v5.0 (https://github.com/ORNL/HydraGNN/releases/tag/v5.0) The list of datasets used for the training of the graph foundation model is the following: 1) Alexandria [1] 2) ANI1x [2] 3) MPTrj [3] 4) Open Catalyst 2020 (OC20) [4] 5) Open Catalyst 2022 (OC22) [5] 6) Open Catalyst 2025 (OC25) [6] 7) Open Direct ir Capture 2023 (ODAC23) [7] 8) Open Materials 2024 (OMat24) [8] 9) Open Molecules 2025 (OMol25) [9] 10) OMol25-neutral (subset of OMol25 that contains only molecules with zero total charge) 11) OMol25-non-neutral (subset of OMol25 that contains only molecules with non-zero total charge) 12) Open Polymers 2026 (OPoly2026) [10] 13) Nabla2DFT [11] 14) QCML [12] 15) QM7X [reference 13] 16) transition1x [14] Dataset references: [1] J. Schmidt et al., “A dataset of 175k stable and metastable materials calculated with the PBEsol and SCAN functionals,” Scientific Data, vol. 9, p. 64, 2022. [2] J. S. Smith et al., “The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules,” Scientific Data, vol. 7, p. 134, 2020. [Online]. Available: https: //www.nature.com/articles/s41597-020-0473-z [3] A. Jain et al., “Commentary: The Materials Project: A materials genome approach to accelerating materials innovation,” APL Materials, vol. 1, no. 1, p. 011002, 07 2013. [Online]. Available: https://doi.org/10.1063/1.4812323 [4] L. Chanussot et al., “Open catalyst 2020 (oc20) dataset and community challenges,” ACS Catalysis, vol. 11, no. 10, pp. 6059–6072, 2021. [Online]. Available: https://doi.org/10.1021/acscatal.0c04525 [5] K. Tran et al., “Open catalyst 2022 (oc22) dataset and challenges for oxidation electrocatalysts,” ACS Catalysis, vol. 13, no. 5, pp. 3066–3084, 2023. [Online]. Available: https://doi.org/10.1021/acscatal.2c05426 [6] S. J. Sahoo et al., “The open catalyst 2025 (oc25) dataset and models for solid-liquid interfaces,” arXiv preprint arXiv:2509.17862, 2025. [Online]. Available: https://arxiv.org/abs/2509.17862 [7] A. Sriram et al., “The open DAC 2023 dataset and challenges for sorbent discovery in direct air capture,” ACS Central Science, vol. 10, no. 5, pp. 923–941, 2024. [8] L. Barroso-Luque et al., “Open materials 2024 (omat24) inorganic materials dataset and models,” 2024. [Online]. Available: https://arxiv.org/abs/2410.12771 [9] D. S. Levine et al., “The open molecules 2025 (OMol25) dataset, evaluations, and models,” 2025. [Online]. Available: https://arxiv.org/abs/2505.08762 [10] D. S. Levine et al., The open polymers 2026 (OPoly26) dataset and evaluations,” arXiv preprint arXiv:2512.23117, 2025. [Online]. Available: https://arxiv.org/abs/2512.23117 [11] K. Khrabrov et al., “Nabla2dft: A universal quantum chemistry dataset of drug-like molecules and a benchmark for neural network potentials,” in NeurIPS 2024 Datasets and Benchmarks Track, 2024. [Online]. Available: https://openreview.net/forum?id=ElUrNM9U8c [12] S. Ganscha et al., “The QCML dataset, quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations,” Scientific Data, vol. 12, p. 406, 2025. [13] J. Hoja et al., “QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules,” Scientific Data, vol. 8, p. 43, 2021. [Online]. Available: https://www.nature.com/articles/s41597-021-00812-2 [14] M. Schreiner et al., “Transition1x - a dataset for building generalizable reactive machine learning potentials,” Scientific Data, vol. 9, p. 779, 2022. The folder "datasets_ADIOS2_format" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "datasets_ADIOS2_format" directory contains 2 sub-directories, one for the version "v1" of the datasets and one for the version "v2" of the datasets. The version "v1" of the datasets provides values of the total energy as they are extracted from the original data as it was released by the respective institutions. The version "v2" of the datasets provides values of the energy that have been realigned. The realignment was performed by training a linear regression model that predicts the total energy as a function of the chemical composition of the atomistic structure, and then subtract such prediction from the original value of the total energy. Both folders "v1" and "v2" contain 16 sub-directories, each corresponding to an ADIOS2-formatted dataset The folder "DeepHyper-results" contains the configurational files and model's parameters for all the 186 HPO trials that were successfully completed by the scalable hyperparameter optimization (HPO) runs on Frontier. The content of the folder "DeepHyper-results" I structured as follows: 1) task-list.txt: list of mpnn name, jobid, and deephyper task id 2) gfm_${MPNN}_${JOBID}_0.${TASKID}: run directory with checkpoint files 3) gfm_${MPNN}: deephyper summary directory (*.csv) for each specific MPNN type 4) deephyper-experiment-${JOBID}: output and error logs for each job The file "deephyper-sorted.csv" contains the details of each HydraGNN model built and tested by HPO, obtained by merging the (*.csv) filed from each HPO run executed. Out of all the HPO trials, we selected 10 to continue the training of the respective HydraGNN models. Due to limited computational budget available in the LRN070 allocation we could not complete the training till convergence for all these 10 selected models. The folder "models" contains multiple sub-folders, one per each HydraGNN model trained. Each model sub-folder contains the parameters of each HydraGNN model, with multiple checkpoint-restarts. The list of sub-folders are as follows: 1) multidataset_hpo-BEST1-fp64 2) multidataset_hpo-BEST2-fp64 3) multidataset_hpo-BEST3-fp64 4) multidataset_hpo-BEST4-fp64 5) multidataset_hpo-BEST5-fp64 6) multidataset_hpo-BEST6-fp64 7) multidataset_hpo-BEST7-fp64 8) multidataset_hpo-BEST8-fp64 9) multidataset_hpo-BEST9-fp64 10) multidataset_hpo-BEST10-fp64 Within each one of these folders, additional auxiliary log files are provided with descriptions about how the training proceeded. The lead PaiNN-model is contained inside "multidataset_hpo-BEST6-fp64". The file "mlp_branch_weights" contains the parameters of the multi-layer perceptron (MLP) used to reconcile the predictions of the 16 output decoding heads of the HydragNN architectures. The MLP takes in input the chemical composition of the atomistic structure and predicts averaging weights to linearly mix the predictions of each output decoding head toward consolidating them into a single one. The folder "1.1billion-structure-inference" contains 1.1 billion atomistic structures randomly generated. Each structures is associated with energy and forces predicted with the lead-PaiNN model combined with the MLP model for reconciliation of the multi-branch predictions generated by the 16 output decoding heads. The folder "1.1billion-structure-inference" contains 9,300 (*.tar.gz) subdirectories, one per Frontier compute node used to execute the inference at exascale. Once uncompressed, each (*.tar.gz) subdirectory contains an ADIOS2 (*.bp) file container, where each atomistic structure is stored as a PyTorch-Geometric Data object. The file "export_dataset_environment_variables.sh" contains the environment variables that need to be set before running the HydraGNN code to reproduce the results provided in this dataset release. The code that can be used to load the ADIOS2 files, load HydraGNN models, and run inference is available at: https://github.com/ORNL/HydraGNN/releases/tag/v5.0

36 MATERIALS SCIENCE↗