Search NASA⌕ Search

SEARCH · Search NASA

Results for “data-driven modeling techniques”

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 415 records · Page 23

GenAI-Based Digital Twins Aided Data Augmentation Increases Accuracy in Real-Time Cokurtosis-Based Anomaly Detection of Wearable Data

Early detection of potential infectious disease outbreaks is crucial for developing effective interventions. In this study, we introduce advanced anomaly detection methods tailored for health datasets collected from wearables, offering insights at both individual and population levels. Leveraging real-world physiological data from wearables, including heart rate and activity, we developed a framework for the early detection of infection in individuals. Despite the availability of data from recent pandemics, substantial gaps remain in data collection, hindering method development. To bridge this gap, we utilized Wasserstein Generative Adversarial Networks (WGANs) to generate realistic synthetic wearable data, augmenting our dataset for training. Subsequently, we use these augmented datasets to implement a cokurtosis-based technique for anomaly detection in multivariate time-series data. Our approach includes a comprehensive assessment of uncertainties in synthetic data compared to the actual data upon which it was modeled, as well as the uncertainty associated with fine-tuning anomaly detection thresholds in physiological measurements. Through our work, we present an enhanced method for early anomaly detection in multivariate datasets, with promising applications in healthcare and beyond. This framework could revolutionize early detection strategies and significantly impact public health response efforts in future pandemics.

Data-Driven Digital Twins↗

ACCELERATED FITTING OF STELLAR SPECTRA

Stellar spectra are often modeled and fitted by interpolating within a rectilinear grid of synthetic spectra to derive the stars’ labels: stellar parameters and elemental abundances. However, the number of synthetic spectra needed for a rectilinear grid grows exponentially with the label space dimensions, precluding the simultaneous and self-consistent fitting of more than a few elemental abundances. Shortcuts such as fitting subsets of labels separately can introduce unknown systematics and do not produce correct error covariances in the derived labels. In this paper we present a new approach—Convex Hull Adaptive Tessellation (chat)—which includes several new ideas for inexpensively generating a sufficient stellar synthetic library, using linear algebra and the concept of an adaptive, data-driven grid. A convex hull approximates the region where the data lie in the label space. A variety of tests with mock data sets demonstrate that chat can reduce the number of required synthetic model calculations by three orders of magnitude in an eight-dimensional label space. The reduction will be even larger for higher dimensional label spaces. In chat the computational effort increases only linearly with the number of labels that are fit simultaneously. Around each of these grid points in the label space an approximate synthetic spectrum can be generated through linear expansion using a set of “gradient spectra” that represent flux derivatives at every wavelength point with respect to all labels. These techniques provide new opportunities to fit the full stellar spectra from large surveys with 15–30 labels simultaneously.

79 ASTRONOMY AND ASTROPHYSICS↗

A Data-driven, Physics-based Transport Model of Solar Energetic Particles Accelerated by Coronal Mass Ejection Shocks Propagating through the Solar Coronal and Heliospheric Magnetic Fields

Abstract In an effort to develop computational tools for predicting radiation hazards from solar energetic particles (SEPs), we have created a data-driven physics-based particle transport model to calculate the injection, acceleration, and propagation of SEPs from coronal mass ejection (CME) shocks traversing through the solar corona and interplanetary magnetic fields. The model runs on an input of corona and heliospheric plasma and magnetic field configuration from a magnetohydrodynamic model driven by solar photospheric magnetic field measurements superposed with observed CME shocks determined from coronagraph images. SEP source particles are injected at the shock using the result of diffusive shock acceleration formulation from a characteristic obliquity-dependent injection from a heated solar wind thermal tail population. With several advanced computation techniques involving stochastic simulation and integration, the model obtains the particle intensity at any location in interplanetary space through the rigorous solution to the time-dependent 5D focus transport equation in the phase space that includes perpendicular diffusion. We apply the model to the 2011 November 3 CME event. The calculation results reproduce multispacecraft SEP observations at Earth and STEREO-B reasonably well without normalization of particle flux. The observations at STEREO-A can be reproduced by rescaling particle energy or modified energy dependence of particle diffusion coefficients. This circumsolar SEP event seen by spacecraft at Earth, STEREO-A, and STEREO-B at widely separated longitudes can be explained by diffusive shock acceleration by a single CME shock with a moderate speed.

79 ASTRONOMY AND ASTROPHYSICS↗

DSCOVR/EPIC-derived global hourly and daily downward shortwave and photosynthetically active radiation data at 0.1° × 0.1° resolution

Downward shortwave radiation (SW) and photosynthetically active radiation (PAR) play crucial roles in Earth system dynamics. Spaceborne remote sensing techniques provide a unique means for mapping accurate spatiotemporally continuous SW–PAR, globally. However, any individual polar-orbiting or geostationary satellite cannot satisfy the desired high temporal resolution (sub-daily) and global coverage simultaneously, while integrating and fusing multisource data from complementary satellites/sensors is challenging because of co-registration, intercalibration, near real-time data delivery and the effects of discrepancies in orbital geometry. The Earth Polychromatic Imaging Camera (EPIC) on board the Deep Space Climate Observatory (DSCOVR), launched in February 2015, offers an unprecedented possibility to bridge the gap between high temporal resolution and global coverage and characterize the diurnal cycles of SW–PAR globally. In this study, we adopted a suite of well-validated data-driven machine-learning models to generate the first global land products of SW–PAR, from June 2015 to June 2019, based on DSCOVR/EPIC data. The derived products have high temporal resolution (hourly) and medium spatial resolution (0.1°×0.1°), and they include estimates of the direct and diffuse components of SW–PAR. We used independently widely distributed ground station data from the Baseline Surface Radiation Network (BSRN), the Surface Radiation Budget Network (SURFRAD), NOAA's Global Monitoring Division and the U.S. Department of Energy's Atmospheric System Research (ASR) program to evaluate the performance of our products, and we further analyzed and compared the spatiotemporal characteristics of the derived products with the benchmarking Clouds and the Earth's Radiant Energy System Synoptic (CERES) data. We found both the hourly and daily products to be consistent with ground-based observations (e.g., hourly and daily total SWs have low biases of -3.96 and -0.71 W m -2 and root-mean-square errors (RMSEs) of 103.50 and 35.40 W m -2 , respectively). The developed products capture the complex spatiotemporal patterns well and accurately track substantial diurnal, monthly, and seasonal variations in SW–PAR when compared to CERES data. They provide a reliable and valuable alternative for solar photovoltaic applications worldwide and can be used to improve our understanding of the diurnal and seasonal variabilities of the terrestrial water, carbon and energy fluxes at various spatial scales.

54 ENVIRONMENTAL SCIENCES↗

Development of a Trip Energy Estimation Model Using Real-World Global Positioning System Driving Data: Preprint

A data-driven technique for estimation of energy requirements for a proposed vehicle trip has been developed. Based on over 700,000 miles of driving data, the technique has been applied to generate a model that estimates trip energy requirements. The model uses a novel binning approach to categorize driving by road type, traffic conditions, and driving profile. The trip-level energy estimations can easily be aggregated to any higher-level transportation system network desired. The model has been tested and validated on the Austin, Texas, data set used to build this model. Ground-truth energy consumption for the data set was obtained from Future Automotive Systems Technology Simulator (FASTSim) vehicle simulation results. The energy estimation model has demonstrated 12.1 percent normalized total absolute error. The energy estimation from the model can be used to inform control strategies in routing tools, such as change in departure time, alternate routing, and alternate destinations, to reduce energy consumption. The model can also be used to determine more accurate energy consumption of regional or national transportation networks if trip origin and destinations are known. Additionally, this method allows the estimation tool to be tuned to a specific driver or vehicle type.

33 ADVANCED PROPULSION SYSTEMS↗

Measurement of Reconstructed Charged Particle Multiplicities of Neutrino Interactions in MicroBooNE

Here, we compare the observed charged particle multiplicity distributions in the MicroBooNE liquid argon time projection chamber from neutrino interactions in a restricted final state phase space to predictions of this distribution from several GENIE models. The measurement uses a data sample consisting of neutrino interactions with a final state muon candidate fully contained within the MicroBooNE detector. These data were collected in 2015-2016 with the Fermilab Booster Neutrino Beam (BNB), which has an average neutrino energy of 800 MeV, using an exposure corresponding to 5e19 protons-on-target. The analysis employs fully automatic event selection and charged particle track reconstruction and uses a data-driven technique to determine the contribution to each multiplicity bin from neutrino interactions and cosmic-induced backgrounds. The restricted phase space employed makes the measurement most sensitive to the higher-energy charged particles expected from primary neutrino-argon collisions and less sensitive to lower energy protons expected to be produced in final state interactions of collision products with the target argon nucleus.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Next Day Building Load Predictions based on Limited Input Features Using an On-Line Laterally Primed Adaptive Resonance Theory Artificial Neural Network.

Optimal integration of thermal energy storage within commercial building applications requires accurate load predictions. Several methods exist that provide an estimate of a buildings future needs. Methods include component-based models and data-driven algorithms. This work implemented a previously untested algorithm for this application that is called a Laterally Primed Adaptive Resonance Theory (LAPART) artificial neural network (ANN). The LAPART algorithm provided accurate results over a two month period where minimal historical data and a small amount of input types were available. These results are significant, because common practice has often overlooked the implementation of an ANN. ANN have often been perceived to be too complex and require large amounts of data to provide accurate results. The LAPART neural network was implemented in an on-line learning manner. On-line learning refers to the continuous updating of training data as time occurs. For this experiment, training began with a singe day and grew to two months of data. This approach provides a platform for immediate implementation that requires minimal time and effort. The results from the LAPART algorithm were compared with statistical regression and a component-based model. The comparison was based on the predictions linear relationship with the measured data, mean squared error, mean bias error, and cost savings achieved by the respective prediction techniques. The results show that the LAPART algorithm provided a reliable and cost effective means to predict the building load for the next day.

25 ENERGY STORAGE↗

A simplified regression building thermal modelling method for detached two-floor house in U.S.

The residential building sector accounts for approximately 37% of total U.S. electricity consumption. Within the residential building sector, heating and cooling is the main target for peak load shifting/reduction since it is the largest contributor to peak demand. In fact, the flexibility of residential HVAC loads can provide continuous variation of demand to provide grid services by varying their demand over a baseline. The performance of HVAC load control to provide grid services relies heavily on the accuracy of indoor air temperature or cooling/heating demand predictions and therefore the quality of building model. Besides forward models, popular building models are data-driven models which can be broken down into two categories: gray-box model, e.g. Resistance-Capacitance (RC) model and black-box model. RC model, also called lumped capacitance or network model, which is constituted with electrical analogue pattern with resistance (R) and capacitance (C). In general, RC models require considerable computation burden and long periods of data to train limited number of model coefficients. Black-box models have gained increasing interest due to their capability in analyzing large-scale data and flexibility in practical applications. But, the data-mining based (machine learning algorithms/techniques based) models tend to have invisible model structures which poses a problem when trying to use the model for optimal control or model predictive control of the HVAC system. Hence, there is a continuing need for efficient online system identification techniques, which can provide explicit parameters for the model. Traditional regression models fit well for this specific purpose. This paper presents an innovative way to predict average indoor temperature in separate floors of typical detached residential house. A rolling horizon linear regression model, which includes online adaptive correction component, is proposed to predict the temperature difference between downstairs and upstairs. A RC model is used to predict the overall mean indoor air temperature. Since the adaptive algorithm needs to be implemented online, a less computation-demanding polynomial fitting algorithm is adopted. This kind of fitting problem can be cast as linear regression problem with multiple variables, parameters of which can be efficiently obtained by well-known gradient descent method.The validation is conducted by comparing the predicted results with the results from data-mining based models as well as measured data from a real typical detached two-floor house. The results show that the developed method has satisfactory performance in predicting the building indoor temperature in 1st and 2nd floors.

Cui, Borui↗

GMLS-Nets: Scientific Machine Learning Methods for Unstructured Data

Data fields sampled on irregularly spaced points arise in many applications in the sciences and engineering. For regular grids, Convolutional Neural Networks (CNNs) have been successfully used to gaining benefits from weight sharing and invariances. We generalize CNNs by introducing methods for data on unstructured point clouds based on Generalized Moving Least Squares (GMLS). GMLS is a nonparametric technique for estimating linear bounded functionals from scattered data, and has recently been used in the literature for solving partial differential equations. By parameterizing the GMLS estimator, we obtain learning methods for operators with unstructured stencils. In GMLS-Nets the necessary calculations are local, readily parallelizable, and the estimator is supported by a rigorous approximation theory. We show how the framework may be used for unstructured physical data sets to perform functional regression to identify associated differential operators and to regress quantities of interest. The results suggest the architectures to be an attractive foundation for data-driven model development in scientific machine learning applications

97 MATHEMATICS AND COMPUTING↗

GMLS-Nets: A Framework for Learning from Unstructured Data

Data fields sampled on irregularly spaced points arise in many applications in the sciences and engineering. For regular grids, Convolutional Neural Networks (CNNs) have been successfully used to gaining benefits from weight sharing and invariances. We generalize CNNs by introducing methods for data on unstructured point clouds based on Generalized Moving Least Squares (GMLS). GMLS is a non-parametric technique for estimating linear bounded functionals from scattered data, and has recently been used in the literature for solving partial differential equations. By parameterizing the GMLS estimator, we obtain learning methods for operators with unstructured stencils. In GMLS-Nets the necessary calculations are local, readily parallelizable, and the estimator is supported by a rigorous approximation theory. We show how the framework may be used for unstructured physical data sets to perform functional regression to identify associated differential operators and to regress quantities of interest. The results suggest the architectures to be an attractive foundation for data-driven model development in scientific machine learning applications

97 MATHEMATICS AND COMPUTING↗

Efficient Reinforcement Learning for Real-Time Hardware-Based Energy System Experiments: Preprint

In the context of urgent climate challenges and the pressing need for rapid technology development, Reinforcement Learning (RL) stands as a compelling data-driven method for controlling real-world physical systems. However, RL implementation often entails time-consuming and computationally intensive data collection and training processes, rendering them inefficient for real-time applications that lack non-real-time models. To address these limitations, real-time emulation techniques have emerged as valuable tools for the lab-scale rapid prototyping of intricate energy systems. While emulated systems offer a bridge between simulation and reality, they too face constraints, hindering comprehensive characterization, testing, and development. In this research, we construct a surrogate model using limited data from simulated systems, enabling an efficient and effective training process for a Double Deep Q-Network (DDQN) agent for future deployment. Our approach is illustrated through a hydropower application, demonstrating the practical impact of our approach on climate-related technology development.

deep Q-learning↗

Measurement of the muon anomalous precession frequency at the Muon g − 2 Experiment at Fermilab

The anomalous magnetic moment of the muon, $a_\mu = \frac{g-2}{2}$ is the fractional deviation of the muon $g$-factor from the value of 2. It emerges as the cumulative effect of the virtual particles participating in the muon interaction with a magnetic field via quantum loop corrections. Its value encodes all the possible interactions between the virtual particles and, for this reason, represents an important test of the Standard Model (SM). In particular, any deviation from the SM theoretical evaluation could be due to new physics contributions. The new Muon $g-2$ (E989) Experiment at Fermilab is currently operating to repeat and improve the previous E821 experiment at Brookhaven National Laboratory (BNL), aiming to reduce the experimental error by a factor of 4 to the final accuracy of 140 parts per billion (ppb). On April 7th, 2021, the E989 collaboration published the first result based on the first year of data taking (Run-1), measuring $a_\mu = 0.001~165~920~40(54)$ with a precision of 460 ppb. The measured value is consistent with the BNL measurement and strengthens the long-standing tension with the data-driven SM prediction to a combined discrepancy of 4.2$\sigma$. On the theory side, however, new efforts involving lattice-QCD techniques are starting to question the current consensus on the theoretical prediction, demanding new improvements on both the experimental and theoretical sides. The E989 collaboration is now finalizing the analysis of Run-2 and Run-3 data and a new publication is expected in the first half of 2023 with a combined statistical uncertainty of 200 ppb. The anomalous magnetic moment $a_\mu$ is measured as the ratio between the muon spin anomalous precession frequency, $\omega_a$, and the average magnetic field experienced by the muons as they circulate in the storage ring. This thesis presents a precession frequency analysis of the Run-1 data and an evaluation of the related systematic uncertainties. A new positron reconstruction developed for the analysis of the subsequent data-taking periods, aiming to reduce some of the major systematic uncertainties of the $\omega_a$ measurement, is presented. The author's involvement in the production of the Run-2/5 data and in the precise calibration of the detectors is discussed. Finally, the complete Run-1 $a_\mu$ result is presented.

43 PARTICLE ACCELERATORS↗

GIS-Based Modeling of Contaminated Soil Volumes at Multiple Sites in the Formerly Utilized Sites Remedial Action Program - 20149

The remediation of hazardous, toxic, and radioactive waste (HTRW) sites produces cost-related risks associated with the estimation of contaminated soil or debris volumes. Historical risk-management techniques include cost contingencies to cover volume uncertainties that affect project budgeting and decision-making. The Buffalo District teamed with project partners to lessen volume uncertainty and reduce project risks at multiple HTRW sites managed under the Formerly Utilized Sites Remedial Action Program (FUSRAP). Historical remedial investigations under FUSRAP commonly identified the presence of radiological material in site media, the associated human health risk, and then areas of remediation. To manage remedial execution and reduce risk, pre-design or remediation-phase sampling essentially 'chased' contamination, which was not conducive to efficient predictive budgeting derived from Feasibility Study (FS) cost analyses. The Buffalo District first optimized their approach to better understand volume uncertainty by utilizing the Argonne National Laboratory's Bayesian Approaches for Adaptive Spatial Sampling (BAASS) software [1]. BAASS processed soft data (e.g., gamma walk-over data) and spatial sampling data to estimate the lateral extent of contaminated soil irrespective of depth (i.e., gross contamination extent) and define areas of contaminant uncertainty. The software performed a binary transformation of contaminant concentrations at all sampling points based upon remedial action goals or a sum of ratios approach (i.e., clean, impacted, or range of impacts in soil). The model produced two-dimensional (horizontal) contaminant probability contours and statistical uncertainty in the sampling coverage and resulting contaminant extents. This method was translated vertically by partitioning the sampling data into depth brackets that produced a stacked representation of contaminant extents and uncertainty in the subsurface (i.e., similar to construction lifts). The results commonly led to a better understanding of project uncertainty and the need for sampling strategies that produce high-confidence soil volumes, which control costs. The BAASS-based delineations were eventually replaced by Empirical Bayesian Kriging (EBK) methods available in ArcGIS Spatial or 3D Analysts [2]. The EBK method calculates contaminant probability zones derived from user-controlled semivariograms of the spatial datasets. The resulting probability zones (e.g., 50% or 80% of contaminant probability) represent the two-dimensional surface delineation of the overall horizontal remedial area, similarly to BAASS. However, unlike BAASS, the vertical sampling data within these probability zones became vertical control points to contour a subterranean surface that connects subsurface points to the land-surface delineations of contamination. The resulting representation of horizontal and vertical impacts within an enclosed envelop (volume) of soil included uncertainty distributions that are used to plan uncertainty-reduction sampling. These data-driven and math-based models of three-dimensional sampling results produced well-bounded remedial volumes for project planning and better uncertainty predictions during project budgeting. The EBK method was applied to several FUSRAP sites managed by the Buffalo District and compared to less rigorously modeled sites previously remediated by the District. The comparison of modeled to actual remediated volumes provide a basis for validating the volume-estimation method. This comparison is important to ensure modeled volumes match physical boundaries of site remediation. FUSRAP sites with denser investigative sampling and lesser volume uncertainty proved useful in remedial planning and contracting. The Buffalo District noted that historical sites with sparser sampling arrays had greater disparity between estimated volumes and final remedial volumes. The benefit achieved over the cost of detailed soil sampling appears positive for FUSRAP projects, especially where impacts vary widely and appear unbounded by investigation-phase sampling. The subsequent Empirical Bayesian Kriging of contamination coupled with vertical contouring for soil estimations reduces uncertainty in soil volumes or indicates where sampling is required to reduce uncertainty, which together optimize remedial planning and budgeting. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

A Study of Physics-Informed Deep Learning for System Fluid Dynamics Closures

The sub-grid-scale (SGS) physics models, or so-called closure relations (CR), are essential in thermal-fluid modeling and simulation codes, ranging from large-eddy simulation to system-level thermo-fluid dynamic problems. While details vary, all thermo-fluid codes solve coupled mass-momentum-energy conservation partial differential equations (PDE). Traditionally, the analytical form of closure models has to be known, and then we calibrate the model using data from a set of applicable experiments. Characteristically, it takes long effort to gain insights and mechanistic understanding needed for developing closure models. Such a model is a compact form of data. The long time required for new model development limits the pace of model application while dealing with new geometries, new coolants or system conditions, particularly in newly designed systems. Alternative to mechanistic and semi-analytical models, it is possible to discover the pattern of closure relations by applying the modern machine learning methods to the body of relevant experimental data. Deep learning (DL) is a relatively new branch in the machine learning family, and a major breakthrough has been achieved since Hinton introduces a fast algorithm to efficiently train a feed-forward neural network (NN) with multilayer perceptrons. Generally, any NN with more than two layers is considered as deep learning, and it contains a large number of hyper-parameters to achieve pattern recognition or regression for complex datasets. Hornik shows that multilayer NN is compatible with universal approximation theorem. There is no constraint for a multilayer NN to capture the properties of any measurable information. This capability makes DL attractive for closure models in a system thermal-hydraulic code. As system geometry and/or operating conditions change, new physical experiments (increasingly sophisticated) are conducted to support evaluation and revision of closure models. Such a so-called data-driven modeling relaxes the structural form of system codes to directly assimilate data instead of hard-wiring the model to pre-calibrated correlations. The objective of this research is to enable cost-effective development of a DL-based closure that reflects the data. The previous work on applying machine learning techniques to fluid dynamics focused on statistical learning with basic NN from direct numerical simulations (DNS) dataset. In the present work, we show that DL-based closure models can cause the fluid calculation to diverge if multilayer NN is used. To address the issue, we investigate conditions by which the DL-based closure models work compatibly, stably, and effectively with PDE-constrained forward prediction problems. Optimally meeting the compatibility conditions constitutes a physics-informed deep learning (PIDL) strategy. We refer to this process as PDE-constrained machine learning or PDE-ML. To develop the technical basis for PDE-ML using DL-based fluid dynamics closure, numerical experiments are performed for manufactured problems of increasing complexity. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DUNE-PRISM – A New Method to Measure Neutrino Oscillations

The Deep Underground Neutrino Experiment (DUNE) is a next-generation long baseline neutrino oscillation experiment designed to make precision measurements in a 1.2 2.4 MW neutrino beam, which is directed 1285 km from the Fermi National Accelerator Laboratory (Fermilab) to the Sandford Underground Research Facility (SURF) in South Dakota. Neutrinos are measured at two detector facilities: a near detector located at Fermilab close where the beam is produced and a far detector at SURF. The neutrino beam can be configured to be composed primarily of either muon or anti-muon neutrinos. DUNE measures the disappearance of muon and anti-muon neutrinos and appearance electron and anti-electron neutrinos in the neutrino beam. Measuring these neutrino flavour transitions provides DUNE with sensitivity to the neutrino mass ordering, $\delta_{CP}$, $\theta_{13}$, $\theta_{23}$ and the magnitude of $\Delta m^2_{23}$. The DUNE Precision Reaction Independent Spectrum Measurement (DUNE-PRISM) concept presents a novel way to perform a neutrino oscillation analysis, which has the potential to significantly reduce the impact of large systematic uncertainties in the neutrino interaction model. The PRISM method linearly combines measurements of off-axis neutrino interactions at the DUNE near detector to produce data-driven predictions of the oscillated neutrino event rate spectrum at the far detector. By building an oscillated far detector prediction directly from data, any unknown or poorly modelled neutrino interaction effects will be naturally incorporated into the measurement of the parameters of the neutrino oscillation model. This thesis presents the first complete neutrino oscillation analysis for DUNE using the PRISM method. Details of the methodology are fully explained and the prospects for further improvements to the techniques described are highlighted. The expected impact and relative importance of the neutrino flux, cross section and detector systematic uncertainties are described in detail. Finally, this thesis demonstrates that the PRISM method is capable of performing a measurement of the oscillation parameters that is robust against neutrino interaction modelling errors.

Hasnip, Ciaran↗

Invertible Temper Modeling using Normalizing Flows and the Effects of Structure Preserving Loss

Advanced manufacturing research and development is typically small-scale, owing to costly experiments associated with these novel processes. Deep learning techniques could help accelerate this development cycle but frequently struggle in small-data regimes like the advanced manufacturing space. While prior work has applied deep learning to modeling visually plausible advanced manufacturing microstructures, little work has been done on data-driven modeling of how microstructures are affected by heat treatment, or assessing the degree to which synthetic microstructures are able to support existing workflows. We propose to address this gap by using invertible neural networks (normalizing flows) to model the effects of heat treatment, e.g., tempering. The model is developed using scanning electron microscope imagery from samples produced using shear-assisted processing and extrusion (ShAPE) manufacturing. This approach not only produces visually and topologically plausible samples, but also captures information related to a sample’s material properties or experimental process parameters. We also demonstrate that topological data analysis, used in prior work to characterize microstructures, can also be used to stabilize model training, preserve structure, and improve downstream results. We assess directions for future work and identify our approach as an important step towards end-to-end deep learning system for accelerating advanced manufacturing research and development.

Howland, Sylvia↗

Temporal sequence transformer to advance long-term streamflow prediction

Accurate streamflow prediction is crucial for understanding climate change impacts on water resources and for effective management of extreme hydrological events. While Long Short-Term Memory (LSTM) networks have been the dominant data-driven approach for streamflow forecasting, recent advancements in transformer architectures for time series tasks have shown promise in outperforming traditional LSTM models. This study introduces a transformer-based model that integrates historical streamflow data with climatic variables to enhance streamflow prediction accuracy. We evaluated our transformer model against a benchmark LSTM across five diverse basins in the United States. Results demonstrate that the transformer architecture consistently outperforms the LSTM model across all evaluation metrics, highlighting its potential as a more effective tool for hydrological forecasting. This research contributes to the ongoing development of advanced AI techniques for improved water resource management and climate change adaptation strategies.

Singh, Ruhaan [Farragut High School]↗

Demonstration and performance of an online data selection algorithm for liquid argon time projection chambers using MicroBooNE

The MicroBooNE detector is a liquid argon time projection chamber (LArTPC) that produces three-dimensional images of particle interactions using ionization charge collected by anode wire plane arrays and scintillation light collected by a light detection system. In addition to testing long-standing experimental neutrino anomalies and performing measurements of neutrino interactions with argon nuclei using the Fermilab Booster Neutrino Beam, MicroBooNE aims to develop methodologies for rare beyond the Standard Model and off-beam physics searches. Looking ahead to the upcoming Deep Underground Neutrino Experiment (DUNE), with MicroBooNE serving as a valuable testbed, achieving high sensitivity and livetime for off-beam physics while satisfying data processing and storage constraints will require data-driven, intelligent, and online or real-time data selection techniques. These techniques are essential for reducing data rates and preserving rare signals with high accuracy. In this paper, we describe a fast data selection algorithm suitable for online execution to identify electrons from stopping cosmic ray muons in the MicroBooNE detector utilizing ionization charge information, and present its performance. This represents the first demonstration of online data selection in a LArTPC using real data and charge information exclusively and provides an important proof-of-principle for applying such techniques to other LArTPC experiments such as the Short-Baseline Near Detector and DUNE.

Abratenko, P. [Tufts U. (main)]↗