Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Performance prediction”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Cracking the code of multi-layer films to promote circularity in single-use plastic packaging

Multi-layer film packaging (MLF) revolutionized food preservation by combining diverse material layers to optimize barrier properties, mechanical strength, and shelf-life. These materials are essential for transporting perishables across various climates and allow for access to fresh goods in “food deserts”, but they pose significant recycling challenges due to their structural complexity. This perspective examines key structure-property relationships governing barrier performance and highlights innovations in material design. We explore how machine learning can predict performance metrics and propose recyclable alternatives, integrating data-driven approaches with material science insights. By challenging the status quo of MLF design, we advocate for circularity in food packaging, inspiring innovation at the intersection of sustainability, material science, and artificial intelligence.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Evaluating the limitations of Bayesian metabolic control analysis

AbstractBayesian Metabolic Control Analysis (BMCA) has emerged as a promising framework for inferring metabolic control coefficients in data-limited scenarios by integrating Bayesian inference with linlog rate laws. However, its predictive accuracy and limitations remain underexplored. This study systematically evaluates BMCA’s ability to infer elasticity values, flux control coefficients (FCCs), and concentration control coefficients (CCCs) under varying data availability conditions using three synthetic metabolic network models. Our findings highlight the strengths and weaknesses of BMCA, guiding its application in metabolic engineering and emphasizing the need for methodological refinements.Author summaryUnderstanding how enzymes control metabolic pathways is crucial for optimizing biomanufacturing and synthetic biology applications. Bayesian Metabolic Control Analysis (BMCA) is a promising computational method that integrates Bayesian inference with metabolic control analysis to estimate key control parameters, even in cases with limited experimental data. However, the accuracy and limitations of BMCA remain unclear. In this study, we systematically evaluate BMCA using three synthetic metabolic networks to determine how different types of physiological data impact its predictive performance. We find that BMCA requires flux and enzyme concentration data for accurate predictions, while external metabolite concentrations contribute little. Additionally, BMCA fails to predict elasticity values beyond a magnitude of 1.5 and reliably infer allosteric regulation, even when strong regulatory interactions exist. In addition, BMCA does not accurately rank metabolic control points, which may limit its utility in identifying key enzymes in engineered pathways. Our work provides practical insights into when and how BMCA can be applied, guiding future research in metabolic modeling and control analysis.

Shin, Janis (ORCID:0000000216572455)↗
DOE OSTI2020

Electrical Energy Storage Data Submission Guidelines

Energy storage technologies are positioned to play a substantial role in power delivery systems. They are being touted as an effective new resource to maintain reliability and allow for increased penetration of renewable energy. However, due to their relative infancy, there is a lack of knowledge on how these resources truly operate over time. Data analysis can help ascertain the operational and performance characteristics of these emerging technologies. Rigorous testing and data analysis are important for all stakeholders to ensure a safe, reliable system that performs predictably on a macro level. Standardizing testing and analysis approaches to verifying the performance of energy storage devices, equipment, and systems when integrating them into the grid will improve the understanding and benefit of energy storage over time from technical and economic vantage points. Demonstrating the life-cycle value and capabilities of energy storage systems begins with the data the provider supplies for analysis. After review of energy storage data received from several providers, it has become clear that some of these data are inconsistent and incomplete, raising the question of their efficacy for robust analysis. This report reviews and proposes general guidelines such as sampling rates and data points that providers must supply for robust data analysis to take place. Consistent guidelines are the basis of the proper protocol and ensuing standards to (a) reduce the time it takes data to reach those who are providing analysis; (b) allow them to better understand the energy storage installations; and (c) enable them to provide high-quality analysis of the installations. This report is intended to serve as a starting point for what data points should be provided when monitoring. As battery technologies continue to advance and the industry expands, this report will be updated to remain current.

25 ENERGY STORAGE↗
DOE OSTI2021

Electrical Energy Storage Data Submission Guidelines, Version 2

Energy storage technologies are positioned to play a substantial role in power delivery systems. They have the potential to serve as an effective new resource to maintain reliability and allow for increased penetration of renewable energy. However, because of their relative infancy, there is a lack of knowledge about how these resources truly operate over time. A data analysis can help ascertain the operational and performance characteristics of these emerging technologies. Rigorous testing and a data analysis are important for all stakeholders to ensure a safe, reliable system that performs predictably on a macro level. Standardizing testing and analysis approaches to verify the performance of energy storage devices, equipment, and systems when integrating them into the grid will improve the understanding and benefit of energy storage over time from technical and economic vantage points. Demonstrating the life-cycle value and capabilities of energy storage systems begins with the data that the provider supplies for the analysis. After a review of energy storage data received from several providers, some of these data have clearly shown to be inconsistent and incomplete, raising the question of their efficacy for a robust analysis. This report reviews and proposes general guidelines, such as sampling rates and data points, that providers must supply for a robust data analysis to take place. Consistent guidelines are the basis of a proper protocol and ensuing standards to (1) reduce the time that it takes for data to reach those who are providing the analysis; (2) allow them to better understand the energy storage installations; and (3) enable them to provide a high-quality analysis of the installations. The report is intended to serve as a starting point for what data points should be provided when monitoring. Readers are encouraged to use the guidance in the report to develop specifications for new systems, as well as enhance current efforts to ensure optimal storage performance. As battery technologies continue to advance and the industry expands, the report will be updated to remain current.

25 ENERGY STORAGE↗
DOE OSTI2024

Cloud Fusion of Big Data and Multi-Physics Models using Machine Learning for Discovery, Exploration, and Development of Hidden Geothermal Resources

The primary goals of this project are identifying hidden geothermal resources in the USA and designing profitable enhanced geothermal systems (EGS). Many non-obvious processes and parameters could characterize geothermal resources and could control the ultimate energy potential of geothermal fields. Diverse datasets (e.g., geology, geochemistry, geophysics, satellite, airborne geophysics) are available to help characterize geothermal resources, but this data is sparse and multi-scale. This has hindered attempts to leverage the datasets for geothermal exploration and profitable EGS design. Recent advancements in machine learning (ML) give promise to overcome these issues. Modern ML methods and tools can (1) analyze large datasets, (2) assimilate model ensembles that include a multitude of inputs and outputs, (3) process sparse datasets, (4) perform transfer learning between sites with different data quality, (5) extract hidden geothermal signatures from field and simulation data, (6) label geothermal resources and processes, (7) identify high-value data acquisition targets, and (8) guide geothermal exploration and production by selecting optimal exploration, production, and drilling strategies. In this work, we implement ML-based geothermal exploration and an enhanced geothermal systems (EGS) design tool to achieve the above goals. Our exploration tool is GeoThermalCloud (GTC) EGS design tool is GeoDT-ML. GTC (github.com/SmartTensors/GeoThermalCloud.jl) utilizes a LANL unsupervised ML platform called SmartTensors (https://tensors.lanl.gov/) to automate data analyses and interpretations by extracting hidden signatures to identify geothermal prospects. It enables the identification of critical measurements needed to identify geothermal resource signatures. GeoDT-ML (github.com/SmartTensors/GeoThermalCloud.jl/tree/master/) adds coupling to GeoDT (https://github.com/GeoDesignTool/GeoDT.git) for stochastic EGS design optimization and performance prediction. GeoDT-ML leverages recent advances in deep learning and high-performance computing. Contributors to this effort include LANL, PNNL, Google, Stanford, and Julia Computing.

15 GEOTHERMAL ENERGY↗
DOE OSTI2022

A Thermogravimetric Temperature-Programmed Thermal Redox Protocol for Rapid Screening of Metal Oxides for Solar Thermochemical Hydrogen Production

As combinatorial and computational methods accelerate the identification of potentially suitable thermochemically-active oxides for use in solar thermochemical hydrogen production (STCH), the onus shifts to quickly evaluating predicted performance. Traditionally, this has required an experimental setup capable of directly carrying out a two-stage thermochemical water-splitting process. But this can be a difficult endeavor, as most off-the-shelf equipment cannot adequately deal simultaneously with the high temperatures, varying oxygen partial pressures, and high H 2 O partial pressures required; achieving sufficient temporal sensitivity to accurately quantify the kinetics is also a major challenge. However, as proposed here, a less complicated experiment can be used as a first screening for thermochemical water splitting potential. Temperature-Programmed Thermal Redox (TPTR) using thermogravimetry evaluates the thermal reduction behavior of materials. This technique does not require water splitting or CO 2 -splitting analogs but can nonetheless predict water-splitting performance. Three figures of merit are obtained from the TPTR experiment: reduction onset temperature, extent of reduction, and extent of recovery upon reoxidation. These metrics can collectively be used to determine if a material is capable of thermochemical water-splitting, and, to good approximation, predict whether the thermodynamics are favorable for use under more challenging high-conversion conditions. This paper discusses the pros and cons of using TPTR and proposes a protocol for use within the STCH community.

08 HYDROGEN↗
DOE OSTI2022

Non-Boussinesq subgrid-scale model with dynamic tensorial coefficients

A major drawback of Boussinesq-type subgrid-scale stress models used in large-eddy simulations is the inherent assumption of alignment between large-scale strain rates and filtered subgrid-stresses. A priori analyses using direct numerical simulation (DNS) data have shown that this assumption is invalid locally as subgrid-scale stresses are poorly correlated with the large-scale strain rates [J. Bardina, J. Ferziger, and W. Reynolds, Improved subgrid-scale models for large-eddy simulation, in Proceedings of the 13th Fluid and Plasmadynamics Conference, AIAA (1980); C. Meneveau and K. Katz, Scale-invariance and turbulence models for large-eddy simulation, Ann. Rev. Fluid Mech. 32, 1 (2000)]. In the present work, a new, non-Boussinesq subgrid-scale model is presented where the model coefficients are computed dynamically. Some previous non-Boussinesq models have observed issues in providing adequate dissipation of turbulent kinetic energy [e.g., Bardina et al., Proceedings of the 13th Fluid and Plasmadynamics Conference (1980); R. A. Clark, J. Ferziger, and W.C. Reynolds. Evaluation of subgrid-scale models using an accurately simulated turbulent flow, J. Fluid Mech. 91, 1 (1979); S. Stolz and N. A. Adams, An approximate deconvolution procedure for large-eddy simulation, Phys. Fluids 11, 1699 (1999)]; however, the present model is shown to provide sufficient dissipation using dynamic coefficients. Modeled subgrid-scale Reynolds stresses satisfy the consistency requirements of the governing equations for large-eddy simulation (LES), vanish in laminar flow and at solid boundaries, and have the correct asymptotic behavior in the near-wall region of a turbulent boundary layer. The new model, referred to as the dynamic tensor-coefficient Smagorinsky model (DTCSM), has been tested in simulations of canonical flows: decaying and forced homogeneous isotropic turbulence, and wall-modeled turbulent channel flow at high Reynolds numbers. The results show favorable agreement with DNS data. It has been shown that DTCSM offers similar predictive capabilities as the dynamic Smagorinsky model for canonical flows. In order to assess the performance of DTCSM in more complex flows, wall-modeled simulations of high Reynolds number flow over a Gaussian bump (Boeing speed bump) exhibiting smooth-body flow separation are performed. Predictions of surface pressure and skin friction, compared against DNS and experimental data, show improved accuracy from DTCSM in comparison to existing static coefficient (Vreman) and dynamic Smagorinsky model. The computational cost of performing LES with this model is up to 15% higher than the dynamic Smagorinsky model.

42 ENGINEERING↗
DOE OSTI2021

A cross-study analysis of drug response prediction in cancer cell lines

Abstract To enable personalized cancer treatment, machine learning models have been developed to predict drug response as a function of tumor and drug features. However, most algorithm development efforts have relied on cross-validation within a single study to assess model accuracy. While an essential first step, cross-validation within a biological data set typically provides an overly optimistic estimate of the prediction performance on independent test sets. To provide a more rigorous assessment of model generalizability between different studies, we use machine learning to analyze five publicly available cell line-based data sets: National Cancer Institute 60, ancer Therapeutics Response Portal (CTRP), Genomics of Drug Sensitivity in Cancer, Cancer Cell Line Encyclopedia and Genentech Cell Line Screening Initiative (gCSI). Based on observed experimental variability across studies, we explore estimates of prediction upper bounds. We report performance results of a variety of machine learning models, with a multitasking deep neural network achieving the best cross-study generalizability. By multiple measures, models trained on CTRP yield the most accurate predictions on the remaining testing data, and gCSI is the most predictable among the cell line data sets included in this study. With these experiments and further simulations on partial data, two lessons emerge: (1) differences in viability assays can limit model generalizability across studies and (2) drug diversity, more than tumor diversity, is crucial for raising model generalizability in preclinical screening.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Bayesian inference of structured latent spaces from neural population activity with the orthogonal stochastic linear mixing model

The brain produces diverse functions, from perceiving sounds to producing arm reaches, through the collective activity of populations of many neurons. Determining if and how the features of these exogenous variables (e.g., sound frequency, reach angle) are reflected in population neural activity is important for understanding how the brain operates. Often, high-dimensional neural population activity is confined to low-dimensional latent spaces. However, many current methods fail to extract latent spaces that are clearly structured by exogenous variables. This has contributed to a debate about whether or not brains should be thought of as dynamical systems or representational systems. Here, we developed a new latent process Bayesian regression framework, the orthogonal stochastic linear mixing model (OSLMM) which introduces an orthogonality constraint amongst time-varying mixture coefficients, and provide Markov chain Monte Carlo inference procedures. We demonstrate superior performance of OSLMM on latent trajectory recovery in synthetic experiments and show superior computational efficiency and prediction performance on several real-world benchmark data sets. We primarily focus on demonstrating the utility of OSLMM in two neural data sets: μ ECoG recordings from rat auditory cortex during presentation of pure tones and multi-single unit recordings form monkey motor cortex during complex arm reaching. We show that OSLMM achieves superior or comparable predictive accuracy of neural data and decoding of external variables (e.g., reach velocity). Most importantly, in both experimental contexts, we demonstrate that OSLMM latent trajectories directly reflect features of the sounds and reaches, demonstrating that neural dynamics are structured by neural representations. Together, these results demonstrate that OSLMM will be useful for the analysis of diverse, large-scale biological time-series datasets.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Airborne hyperspectral imaging of cover crops through radiative transfer process-guided machine learning

Cover cropping between cash crop growing seasons is a multifunctional conservation practice. Timely and accurate monitoring of cover crop traits, notably aboveground biomass and nutrient content, is beneficial to agricultural stakeholders to improve management and understand outcomes. Currently, there is a scarcity of spatially and temporally resolved information for assessing cover crop growth. Remote sensing has a high potential to fill this need, but conventional empirical regression operated with coarse-resolution multispectral data has large uncertainties. Therefore, this study utilized airborne hyperspectral imaging techniques and developed new process-guided machine learning approaches (PGML) for cover crop monitoring. Specifically, we deployed an airborne hyperspectral system covering visible to shortwave-infrared wavelengths (400–2400 nm) to acquire high spatial (0.5 m) and spectral (3–5 nm) resolution reflectance over 23 cover crop fields across Central Illinois in March and April of 2021. Airborne hyperspectral surface reflectance with high spectral and spatial resolution can be well matched with field data to quantify cover crop traits. Furthermore, the PGML models were pre-trained by synthetic data from soil-vegetation radiative transfer modeling (one million records), and then fine-tuned with field data of cover crop biomass and nutrient content. Results show that airborne hyperspectral data with PGML can achieve high accuracy to predict cover crop aboveground biomass (R 2 = 0.72, relative RMSE = 15.16%) and nitrogen content (R 2 = 0.69, relative RMSE = 16.59%) through leave-one-field-out cross-validation. Unlike the pure data-driven approach (e.g., partial least-squares regression), PGML incorporated radiative transfer knowledge and obtained higher predictive performance with fewer field data. Meanwhile, with field data for model fine-tuning, PGML predicted biomass more accurately than the inversion of radiative transfer models. Here we also found that the red edge has a high contribution in quantifying aboveground biomass and nitrogen content, followed by green and shortwave spectra. This study demonstrated the first attempt of utilizing hyperspectral remote sensing to accurately quantify cover crop traits. We highlight the strength of PGML in exploiting sensing data to quantify ecosystem variables to advance agroecosystem monitoring for sustainable agricultural management.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2020

Remaining Useful Strength (RUS) Prediction of SiCf-SiCm Composite Materials Using Deep Learning and Acoustic Emission

Prognosis techniques for prediction of remaining useful life (RUL) are of crucial importance to the management of complex systems for they can lead to appropriate maintenance interventions and improvements in reliability. While various data-driven methods have been introduced to predict the remaining useful life (RUL) of machinery systems or batteries, no research has been reported on the remaining useful strength (RUS) prediction of silicon carbide fiber reinforced silicon carbide matrix (SiCf-SiCm) materials with pivotal role in its potential usage as a structural material in nuclear reactors and turbine engines. Knowledge of its degradation process is of the utmost importance to the manufacturers. For this purpose, two approaches based on the machine-learning techniques of random-forest (RF) and convolutional neural network (CNN) are proposed to predict the RUS of SiCf-SiCm using only acoustic emission (AE) signals generated during the material’s stress applying process. Experimental results show that the CNN models achieved better predictive performance than the RF models but the latter with expert-engineered features achieves better prediction for AE signals in the early stage of degradation. Additionally, our results demonstrate that both models can correctly predict the SiCf-SiCm RUS as evaluated by our robust testing method from which the best average root mean square error (RMSE) and Pearson correlation coefficient of 3.55 ksi units and 0.85 were obtained.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Operation of helium sub-atmospheric multistage cryogenic centrifugal compressor trains: Part 2 – Transient modeling and pump-down path selection

Low-pressure conditions required for operation of helium cryogenic systems below the normal boiling point (i.e. 4.2 K) are established through a transient process, commonly referred to as ‘pump-down’. This process is defined as the transition from pressures above atmospheric conditions to the saturation pressure which corresponds to a specified operational temperature. The FRIB 2 K system consists of five cryogenic centrifugal compressors which are operated in series. Historically, the pump-down process path has been established through empirical methods and system operator experience. Investigation into the pump-down process at FRIB aimed to develop a pump-down methodology which relies on theoretical model predictions rather than empirically developed process paths. Ensuring stable operation during the pump-down process involved application of a centrifugal compressor performance prediction model, which is described in Part 1 of this paper. Compressor performance maps can be directly used to evaluate the stability of a selected pump-down path and anticipate the overall reliability of the selected path. In conjunction with the compressor performance maps, a system pressure model was developed to estimate the transient pressure response during the pump-down process. Lastly, an explicit equation was developed to establish a mass flow rate profile for the pump-down process. Implementation of the presented methodology (including the developed models) allows for the system operator to determine a continuous pump-down path which maintains compressor stability while conforming to overall system capabilities. Altogether, the methodology presented has resulted in simplification of transient pump-down operations and increased the reliability, stability and efficiency of the pump-down process.

Compressor train control↗
DOE OSTI2025

Solar Decarbonization of Paraffin Dehydrogenation Through Particle Heat Carriers (Final Technical Report)

This project focuses on solutions to decarbonize high-temperature catalytic processes using solar thermal heat. The primary project goal is to show the validity of a moving packed bed reactor for propane dehydrogenation using catalyst particles as the heat carrier for the reaction, which can be heated by concentrated solar energy in a particle receiver. This concept, if further developed, may provide a cost-effective pathway for converting lower value gases to important chemical precursors for industrial materials using only renewable energy. The project was divided into six tasks. In Task 1, DFT calculations were performed to understand the role of Pt and Sn in the catalytic dehydrogenation reaction. In Task 2 chemical kinetics measurements were made for several catalyst formulations at high temperatures. In Task 3, the solar absorptance of catalyst particles was compared to the absorptance of commonly used materials in particle receivers. In Task 4, numerical models were developed which could predict performance of the complete system and predict specific temperatures in the system. In Task 5, a prototype system was designed, fabricated, and tested to show the validity of the concept. Task 6 concerned project management activities. Experiments with the prototype showed repeatable thermal performance at temperatures targeted for the reaction. A limited set of tests were done with active catalyst and propane dehydrogenation, showing conversion of propane to propylene with a range of conversions and selectivities. The results are promising, and the prototype designed was reliable during testing, and the team expects that further development of the prototype would yield improved results. A numerical model framework based on coupled fluid and particle mechanics was developed with high computational efficiency using GPU calculations. The model may prove highly useful for evaluating other high-temperature particle systems. However, it was determined that simpler porous media models were good fits for the needs of the current moving packed bed concept. Data showing strong solar absorption of the particles validates the plan of using existing solar particle receivers with only a change in the particle type. Catalyst investigation showed that Pt 1 Sn 3 is the most viable candidate for developing PtSn catalysts for high temperature propane dehydrogenation, considering the balance of activity, selectivity, and deactivation. This project completed an initial study of various factors needed to incorporate a moving bed catalytic reactor for propane dehydrogenation into a concentrated solar thermal particle system. Future developments may allow this technology to be scaled up and help to use solar thermal energy to decarbonize not only the propane dehydrogenation reaction, but other gas-solid catalytic reactions at similar temperatures.

14 SOLAR ENERGY↗
DOE OSTI2026

Empirical Comparison of Machine Learning Approaches for Black-Box Modeling of Power Conversion System Dynamics

Inverter-based resources are key components in modern power systems, but accurately modeling their complex behavior can be challenging. Standard, generic converter models often oversimplify inverter dynamics, leading to significant errors in predicting performance. In this work, we compare several data-driven machine learning (ML) approaches for inverter modeling, performing experiments on power conversion systems, systematically varying input conditions, and recording the resulting voltages and currents. The ML models were then trained on this measured data to capture the inverter's dynamic response and to predict the inverter's output current. A performance comparison between the four ML models under study is conducted, laying the foundation for future work on hardware implementation for real-time inference.

30 DIRECT ENERGY CONVERSION↗
DOE OSTI2024

Model orthogonalization and Bayesian forecast mixing via principal component analysis

One can improve predictability in the unknown domain by combining forecasts of imperfect complex computational models using a Bayesian statistical machine learning framework. In many cases, however, the models used in the mixing process are similar. In addition to contaminating the model space, the existence of such similar, or even redundant, models during the multimodeling process can result in misinterpretation of results and deterioration of predictive performance. In this paper we describe a method based on the principal component analysis that eliminates model redundancy. We show that by adding model orthogonalization to the proposed Bayesian model combination framework, one can arrive at better prediction accuracy and reach excellent uncertainty quantification performance.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2022

Composition design of high-entropy alloys with deep sets learning

High entropy alloys (HEAs) are an important material class in the development of next-generation structural materials, but the astronomically large composition space cannot be efficiently explored by experiments or first-principles calculations. Machine learning (ML) methods might address this challenge, but ML of HEAs has been hindered by the scarcity of HEA property data. In this work, the EMTO-CPA method was used to generate a large HEA dataset (spanning a composition space of 14 elements) containing 7086 cubic HEA structures with structural properties, 1911 of which have the complete elastic tensor calculated. The elastic property dataset was used to train a ML model with the Deep Sets architecture. The Deep Sets model has better predictive performance and generalizability compared to other ML models. Association rule mining was applied to the model predictions to describe the compositional dependence of HEA elastic properties and to demonstrate the potential for data-driven alloy design.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Data-driven electrolyte design for lithium metal anodes

Improving Coulombic efficiency (CE) is key to the adoption of high energy density lithium metal batteries. Liquid electrolyte engineering has emerged as a promising strategy for improving the CE of lithium metal batteries, but its complexity renders the performance prediction and design of electrolytes challenging. Here, we develop machine learning (ML) models that assist and accelerate the design of high-performance electrolytes. Using the elemental composition of electrolytes as the features of our models, we apply linear regression, random forest, and bagging models to identify the critical features for predicting CE. Our models reveal that a reduction in the solvent oxygen content is critical for superior CE. We use the ML models to design electrolyte formulations with fluorine-free solvents that achieve a high CE of 99.70%. This work highlights the promise of data-driven approaches that can accelerate the design of high-performance electrolytes for lithium metal batteries.

25 ENERGY STORAGE↗
DOE OSTI2025

Transformers and Long Short-Term Memory Transfer Learning for GenIV Reactor Temperature Time Series Forecasting

Automated monitoring of the coolant temperature can enable autonomous operation of generation IV reactors (GenIV), thus reducing their operating and maintenance costs. Automation can be accomplished with machine learning (ML) models trained on historical sensor data. However, the performance of ML usually depends on the availability of large amount of training data, which is difficult to obtain for GenIV, as this technology is still under development. We propose the use of transfer learning (TL), which involves utilizing knowledge across different domains, to compensate for this lack of training data. TL can be used to create pre-trained ML models with data from small-scale research facilities, which can then be fine-tuned to monitor GenIV reactors. In this work, we develop pre-trained Transformer and long short-term memory (LSTM) networks by training them on temperature measurements from thermal hydraulic flow loops operating with water and Galinstan fluids at room temperature at Argonne National Laboratory. The pre-trained models are then fine-tuned and re-trained with minimal additional data to perform predictions of the time series of high temperature measurements obtained from the Engineering Test Unit (ETU) at Kairos Power. The performance of the LSTM and Transformer networks is investigated by varying the size of the lookback window and forecast horizon. The results of this study show that LSTM networks have lower prediction errors than Transformers, but LSTM errors increase more rapidly with increasing lookback window size and forecast horizon compared to the Transformer errors.

LSTM↗