SEARCH · Search NASA
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.
Deep learning-based optical flow (DLOF) extracts features in video frames with deep convolutional neural networks to estimate the inter-frame motions of objects. DLOF computes velocity fields more accurately than PIV for densely labeled systems.
Active feedback control in magnetic confinement fusion devices is desirable to mitigate plasma instabilities and enable robust operation. Optical high-speed cameras provide a powerful, non-invasive diagnostic and can be suitable for these applications. Here, in this study, we process high-speed camera data, at rates exceeding 100 kfps, on in situ field-programmable gate array (FPGA) hardware to track magnetohydrodynamic (MHD) mode evolution and generate control signals in real time. Our system utilizes a convolutional neural network (CNN) model, which predicts the n = 1 MHD mode amplitude and phase using camera images with better accuracy than other tested non-deep-learning-based methods. By implementing this model directly within the standard FPGA readout hardware of the high-speed camera diagnostic, our mode tracking system achieves a total trigger-to-output latency of 17.6 μs and a throughput of up to 120 kfps. This study at the High Beta Tokamak-Extended Pulse (HBT-EP) experiment demonstrates an FPGA-based high-speed camera data acquisition and processing system, enabling application in real-time machine-learning-based tokamak diagnostic and control as well as potential applications in other scientific domains.
Understanding the collective behavior of complex spin textures, such as lattices of magnetic skyrmions, is of fundamental importance for exploring and controlling the emergent ordering of these spin textures and inducing phase transitions. It is also critical to understand the skyrmion–skyrmion interactions for applications such as magnetic skyrmion-enabled reservoir or neuromorphic computing. Magnetic skyrmion lattices can be studied using in situ Lorentz transmission electron microscopy (LTEM), but quantitative and statistically robust analysis of the skyrmion lattices from LTEM images can be difficult. In this work, we show that a convolutional neural network, trained on simulated data, can be applied to perform segmentation of spin textures and to extract quantitative data, such as spin texture size and location, from experimental LTEM images, which cannot be obtained manually. This includes quantitative information about skyrmion size, position, and shape, which can, in turn, be used to calculate skyrmion–skyrmion interactions and lattice ordering. We apply this approach to segmenting images of Néel skyrmion lattices so that we can accurately identify skyrmion size and deformation in both dense and sparse lattices. The model is trained using a large set of micromagnetic simulations as well as simulated LTEM images. This entirely open-source training pipeline can be applied to a wide variety of magnetic features and materials, enabling large-scale statistical studies of spin textures using LTEM.
Crystals with complicated geometry are often observed with mixed chemical occupancy among Wyckoff sites, presenting a unique challenge for accurate atomic modeling. Similar systems possessing exact occupancy on all the sites can exhibit superstructural ordering, dramatically inflating the unit cell size. In this work, a crystal graph convolutional neural network (CGCNN) is used to predict optimal atomic decorations on fixed crystalline geometries. This is achieved with a site permutation search (SPS) optimization algorithm based on Monte Carlo moves combined with simulated annealing and basin-hopping techniques. Our approach relies on the evidence that, for a given chemical composition, a CGCNN estimates the correct energetic ordering of different atomic decorations, as predicted by electronic structure calculations. This provides a suitable energy landscape that can be optimized according to site occupation, allowing the prediction of chemical decoration in crystals exhibiting mixed or disordered occupancy, or superstructural ordering. Verification of the procedure is carried out on several known compounds, including the superstructurally ordered clathrate compound Rb8Ga27Sb16 and vacancy-ordered perovskite Cs2SnI6, neither of which was previously seen during the neural network training. In addition, the critical temperature of an order–disorder phase transition in solid solution CuZn is probed with our SPS routines by sampling site configuration trajectories in the canonical ensemble. This strategy provides an accurate method for determining favorable decoration in complex crystals and analyzing site occupation at unprecedented speed and scale.
This study proposes a novel machine learning (ML) methodology for the efficient and cost-effective prediction of high-fidelity three-dimensional velocity fields in the wake of utility-scale turbines. The model consists of an autoencoder convolutional neural network with U-Net skipped connections, fine-tuned using high-fidelity data from large-eddy simulations (LES). The trained model takes the low-fidelity velocity field cost-effectively generated from the analytical engineering wake model as input and produces the high-fidelity velocity fields. The accuracy of the proposed ML model is demonstrated in a utility-scale wind farm for which datasets of wake flow fields were previously generated using LES under various wind speeds, wind directions, and yaw angles. Comparing the ML model results with those of LES, the ML model was shown to reduce the error in the prediction from 20% obtained from the Gauss Curl hybrid (GCH) model to less than 5%. In addition, the ML model captured the non-symmetric wake deflection observed for opposing yaw angles for wake steering cases, demonstrating a greater accuracy than the GCH model. The computational cost of the ML model is on par with that of the analytical wake model while generating numerical outcomes nearly as accurate as those of the high-fidelity LES.
Explore the source record for details and available documents.
A deep-learning convolutional neural network (CNN) is used to infer, from x-ray images along multiple lines of sight, the low-mode shape of the hot-spot emission of deuterium–tritium (DT) laser-direct-drive cryogenic implosions on OMEGA. The motivation of this approach is to develop a physics-informed 3-D reconstruction technique that can be performed within minutes to facilitate the use of the results to inform changes to the initial target and laser conditions for the subsequent implosion. The CNN is trained on a 3D radiation-hydrodynamic simulation database to relate 2D x-ray images to 3D emissivity at stagnation. The CNN accounts for the lack of an absolute spatial reference and the different bands of photon energies in the x-ray images. While previous works studied the effect of mode-1 asymmetries on implosion performance using nuclear diagnostics, this work focuses on the effect of mode 2 inferred from x-ray diagnostics on implosion performance. A current analysis of 19 DT cryogenic implosions indicates there is an upper limit of ~20% reduction in the neutron yield caused by an ℓ = 2 amplitude for ℓ 2 /ℓ 0 ≤ 0.32. Here, these conclusions are supported by 2D simulations.
The emergence of high-repetition-rate x-ray free-electron lasers (XFELs), such as SLAC’s LCLS-II, serves as our canonical example for autonomous controls that necessitate high-throughput diagnostics paired with streaming computational pipelines capable of single-shot analysis with extremely low latency. We present the deterministic characterization with an integrated parallelizable hybrid resolver architecture, a hybrid machine learning framework designed for fast, accurate analysis of XFEL diagnostics using angular streaking-based sinogram images. This architecture integrates convolutional neural networks and bidirectional long short-term memory models to denoise input, identify x-ray sub-spike features, and extract sub-spike relative delays with sub-30 attosecond temporal resolution. Deployed on low-latency hardware, it achieves over 10 kHz throughput with 168.3 μs inference latency, indicating scalability to 14 kHz with field-programmable gate array integration. By transforming regression tasks into classification problems and leveraging optimized error encoding, we achieve high precision with low-latency performance that is critical for real-time streaming event selection and experimental control feedback signals. This represents a key development in real-time control pipelines for next-generation autonomous science, generally, and high repetition-rate x-ray experiments in particular.
Modern cosmological surveys probe the Universe deep into the nonlinear regime, where massive neutrinos suppress cosmic structure. Traditional cosmological analyses, which use the 2-point correlation function to extract information, are no longer optimal in the nonlinear regime, and there is thus much interest in extracting beyond-2-point information to improve constraints on neutrino mass. Quantifying and interpreting the beyond-2-point information is thus a pressing task. We study the field-level information in weak lensing convergence maps using convolution neural networks. We find that the network performance increases as higher source redshifts and smaller scales are considered — investigating up to a source redshift of 2.5 and ℓ max ≃ 10 4 — verifying that massive neutrinos leave a distinct effect on weak lensing. However, the performance of the network significantly drops after scaling out the 2-point information from the maps, implying that most of the field-level information can be found in the 2-point correlation function alone. We quantify these findings in terms of the likelihood ratio and also use Integrated Gradient saliency maps to interpret which parts of the map the network is learning the most from. We find that, in the absence of noise, the network extracts a similar amount of information from the most overdense and underdense regions. However, upon adding noise, the information in underdense regions is distorted as noise disproportionately washes out void-like structures.
The Dark Energy Spectroscopic Instrument (DESI) survey uses an automatic spectral classification pipeline to classify spectra. QuasarNET is a convolutional neural network used as part of this pipeline originally trained using data from the Baryon Oscillation Spectroscopic Survey (BOSS). In this paper we implement an active learning algorithm to optimally select spectra to use for training a new version of the QuasarNET weights file using only DESI data, with the goal of improving classification accuracy. This active learning algorithm includes a novel outlier rejection step using a Self-Organizing Map to ensure we label spectra representative of the larger quasar sample observed in DESI. We perform two iterations of the active learning pipeline, assembling a final dataset of 5600 labeled spectra, a small subset of the approximately 1.3 million quasar targets in DESI's Data Release 1. When splitting the spectra into training and validation subsets we achieve similar performance to the previously trained weights file in completeness and purity calculated on the validation dataset but do so with less than one tenth of the amount of training data. The new weights also more consistently classify objects in the same way when used on unlabeled data compared to the old weights file. In the process of improving QuasarNET's classification accuracy we discovered a systemic error in QuasarNET's redshift estimation and used our findings to improve our understanding of QuasarNET's redshifts.
Modern wide-field time-domain surveys facilitate the study of transient, variable and moving phenomena by conducting image differencing and relaying alerts to their communities. Machine learning tools have been used on data from these surveys and their precursors for more than a decade, and convolutional neural networks (CNNs), which make predictions directly from input images, saw particularly broad adoption through the 2010s. Since then, continually rapid advances in computer vision have transformed the standard practices around using such models. It is now commonplace to use standardized architectures pre-trained on large corpora of everyday images (e.g., ImageNet). In contrast, time-domain astronomy studies still typically design custom CNN architectures and train them from scratch. Here, we explore the effects of adopting various pre-training regimens and standardized model architectures on the performance of alert classification. We find that the resulting models match or outperform a custom, specialized CNN like what is typically used for filtering alerts. Moreover, our results show that pre-training on galaxy images from Galaxy Zoo tends to yield better performance than pre-training on ImageNet or training from scratch. We observe that the design of standardized architectures are much better optimized than the custom CNN baseline, requiring significantly less time and memory for inference despite having more trainable parameters. On the eve of the Legacy Survey of Space and Time and other image-differencing surveys, these findings advocate for a paradigm shift in the creation of vision models for alerts, demonstrating that greater performance and efficiency, in time and in data, can be achieved by adopting the latest practices from the computer vision field.
Liquid argon time projection chambers are often used in neutrino physics and dark-matter searches because of their high spatial resolution. The images generated by these detectors are extremely sparse, as the energy values detected by most of the detector are equal to 0, meaning that despite their high resolution, most of the detector is unused in a particular interaction. Instead of representing all of the empty detections, the interaction is usually stored as a sparse matrix, a list of detection locations paired with their energy values. Traditional machine learning methods that have been applied to particle reconstruction such as convolutional neural networks (CNNs), however, cannot operate over data stored in this way and therefore must have the matrix fully instantiated as a dense matrix. Operating on dense matrices requires a lot of memory and computation time, in contrast to directly operating on the sparse matrix. We propose a machine learning model using a point set neural network that operates over a sparse matrix, greatly improving both processing speed and accuracy over methods that instantiate the dense matrix, as well as over other methods that operate over sparse matrices. Compared to competing state-of-the-art methods, our method improves classification performance by 14%, segmentation performance by more than 22%, while taking 80% less time and using 66% less memory. Compared to state-of-the-art CNN methods, our method improves classification performance by more than 86%, segmentation performance by more than 71%, while reducing runtime by 91% and reducing memory usage by 61%.
Dual-phase xenon time projection chambers (TPCs) are widely used in searches for rare dark matter and neutrino interactions, in part because of their excellent position reconstruction capability in 3D. Despite their millimeter-scale resolution along the charge drift axis, xenon TPCs face challenges in resolving single-site (SS) and multi-site (MS) interactions in the transverse plane. In this paper, we build a generic TPC model with an idealized signal readout, and use Fisher Information (FI) to predict its theoretical capability of differentiating SS and MS events using the electroluminescence signal. We also demonstrate via simulation that, when only statistical photon noise is present, the theoretical limits can be approached with conventional reconstruction algorithms like maximum likelihood estimation, and with a convolutional neural network classifier. The implications of this study on future TPC experiments will be discussed.
We present a new class of AI models for the detection of quasi-circular, spinning, non-precessing binary black hole mergers whose waveforms include the higher order gravitational wave modes ($\ell$, |m|) = {(2,2), (2,1), (3,3), (3,2), (4,4)}, and mode mixing effects in the $\ell$ = 3, |m| = 2 harmonics. These AI models combine hybrid dilated convolution neural networks to accurately model both short- and long-range temporal sequential information of gravitational waves; and graph neural networks to capture spatial correlations among gravitational wave observatories to consistently describe and identify the presence of a signal in a three detector network encompassing the Advanced LIGO and Virgo detectors. We first trained these spatiotemporal-graph AI models using synthetic noise, using 1.2 million modeled waveforms to densely sample this signal manifold, within 1.7 h using 256 NVIDIA A100 GPUs in the Polaris supercomputer at the Argonne Leadership Computing Facility. This distributed training approach exhibited optimal classification performance, and strong scaling up to 512 NVIDIA A100 GPUs. With these AI ensembles we processed data from a three detector network, and found that an ensemble of 4 AI models achieves state-of-the-art performance for signal detection, and reports two misclassifications for every decade of searched data. We distributed AI inference over 128 GPUs in the Polaris supercomputer and 128 nodes in the Theta supercomputer, and completed the processing of a decade of gravitational wave data from a three detector network within 3.5 h. Finally, we fine-tuned these AI ensembles to process the entire month of February 2020, which is part of the O3b LIGO/Virgo observation run, and found 6 gravitational waves, concurrently identified in Advanced LIGO and Advanced Virgo data, and zero false positives. This analysis was completed in one hour using one NVIDIA A100 GPU.
Abstract Accelerating cavities are an integral part of the continuous electron beam accelerator facility (CEBAF) at Jefferson Laboratory. When any of the over 400 cavities in CEBAF experiences a fault, it disrupts beam delivery to experimental user halls. In this study, we propose the use of a deep learning model to predict slowly developing cavity faults. By utilizing pre-fault signals, we train a long short-term memory-convolutional neural network binary classifier to distinguish between radio-frequency (RF) signals during normal operation and RF signals indicative of impending faults. We optimize the model by adjusting the fault confidence threshold and implementing a multiple consecutive window criterion to identify fault events, ensuring a low false positive rate. Results obtained from analysis of a real dataset collected from the accelerating cavities simulating a deployed scenario demonstrate the model’s ability to identify normal signals with 99.99% accuracy and correctly predict 80% of slowly developing faults. Notably, these achievements were achieved in the context of a highly imbalanced dataset, and fault predictions were made several hundred milliseconds before the onset of the fault. Anticipating faults enables preemptive measures to improve operational efficiency by preventing or mitigating their occurrence.
High-energy Two dimensional (2D) synchrotron x-ray diffractometry provides important insights into the atomistic structure and phase evolution of materials, yet traditional analysis methods remain complex, knowledge-intensive, and computationally demanding. Deep-learning models offer a powerful alternative for automating their analysis. Institutions that hold these datasets may be unwilling to share their data due to privacy and security policies, as well as the challenges associated with large-scale data transfer. As a result, models trained on local datasets often perform well only on their own data but exhibit bias and poor generalization across different instruments or facilities. To overcome these limitations, we explore federated learning (FL) for 2D synchrotron diffractograms, enabling collaborative model training without exchanging raw data. In this study, 2D synchrotron diffractograms of Ti–6Al–4V alloy collected from two independent facilities are used to train convolutional neural networks for predicting the β-phase volume fraction. Experimental results show that federated global models significantly outperform locally trained models in terms of generalization and achieve accuracy comparable to centralized trained models. These findings demonstrate the potential of FL to enable secure, cross-institutional collaboration and enhance the scalability of deep-learning-based materials characterization.
Accurately quantifying the energy use of artificial intelligence (AI) training is critical for infrastructure planning, carbon accounting, and sustainable data center operation, but few studies have directly measured the power consumption of production workloads on contemporary hardware. By combining empirical measurements from Brookhaven National Laboratory during AI training on 8-graphics-processing-unit H100 systems with open-source benchmarking data, we develop statistical models relating computational intensity to node-level power consumption. We measure the gap between manufacturer-rated thermal design power (TDP) and actual power demand during AI training. Our analysis reveals that even computationally intensive workloads operate at only 76% of the 10.2 kW TDP rating. Our architecture-specific model, calibrated to floating-point operations, predicts energy consumption with 11.4% mean absolute percentage error, significantly outperforming TDP-based approaches (27%–37% error). We identified distinct power signatures between transformer and convolutional neural network architectures, with transformers showing characteristic fluctuations that may impact grid stability. These results provide a measurement-grounded basis for improving AI training energy estimates, enabling more reliable infrastructure sizing, cost projections, and environmental impact assessments.
The cycle-skipping problem that plagues full waveform inversion (FWI) can be at least partially mitigated if low frequencies (which encode the kinematics of wave propagation in seismic data) are recorded. However, seismic sources and receivers are band-limited, so seismic data does not generally include signals down to 0 Hz. To improve our ability to solve the seismic inverse problem, one can synthesize this missing low-frequency (LF) content from the recorded high-frequency (HF) data using machine learning (ML) models. Deep learning models such as convolutional neural networks (CNNs) demonstrate impressive ability to perform low frequency extrapolation. However, such models require powerful hardware (GPU machines) and careful training. We assess the extrapolation capabilities of three different ML models that do not require GPU machines, namely, random forest, Gaussian process regression and gradient boosting, on both synthetic and real data. Experimental results on two synthetic data sets (generated from a low velocity lens embedded in a homogeneous medium, and the Marmousi model) demonstrate that FWI applied to the extrapolated data consistently improves inversion accuracy relative to FWI applied to the original data sets that do not contain low frequencies. Application of low-frequency extrapolation to real data from the Northwest Shelf of Australia demonstrates that tree-based ML models such as gradient boosting can outperform CNNs in terms of both accuracy and computational cost on non-GPU architectures.