SEARCH · Engineering Papers
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.
I am a summer intern. The thesis includes work I’ve done here this summer with EV40. It is on mathematical methods for deblurring images. It doesn't include any radiographic images or data, its all synthetic examples - Dworaczyk, Jordan T.
This project supported cutting-edge nanoscale characterization developments including X-ray Imaging by utilizing the power of Coherent X-ray Diffractive Imaging (CXDI). This research pioneered new directions in studies of nanoscale dynamics using X-ray Photon Correlation Spectroscopy (XPCS) and Ultrafast X-ray Scattering (UXS), extending our understanding of science and technology at the limits of ultrasmall (atomic/nanoscale) and ultrafast (from seconds to femtosecond dynamics). The research funded by this award spans a wide range of condensed matter and materials physics systems, from correlated oxides and nanomaterials to energy-related materials, and quantum materials for neuromorphic computing.
Current state-of-the-art (SOTA) LiDAR-only detectors perform well for 3D object detection tasks, but point cloud data are typically sparse and lacks semantic information. Detailed semantic information obtained from camera images can be added with existing LiDAR-based detectors to create a robust 3D detection pipeline. With two different data types, a major challenge in developing multi-modal sensor fusion networks is to achieve effective data fusion while managing computational resources. With separate 2D and 3D feature extraction backbones, feature fusion can become more challenging as these modes generate different gradients, leading to gradient conflicts and suboptimal convergence during network optimization. To this end, we propose a 3D object detection method, Attention-Enabled Point Fusion (AEPF). AEPF uses images and voxelized point cloud data as inputs and estimates the 3D bounding boxes of object locations as outputs. An attention mechanism is introduced to an existing feature fusion strategy to improve 3D detection accuracy and two variants are proposed. These two variants, AEPF-Small and AEPF-Large, address different needs. AEPF-Small, with a lightweight attention module and fewer parameters, offers fast inference. AEPF-Large, with a more complex attention module and increased parameters, provides higher accuracy than baseline models. Experimental results on the KITTI validation set show that AEPF-Small maintains SOTA 3D detection accuracy while inferencing at higher speeds. AEPF-Large achieves mean average precision scores of 91.13, 79.06, and 76.15 for the car class’s easy, medium, and hard targets, respectively, in the KITTI validation set. Results from ablation experiments are also presented to support the choice of model architecture.
Laser beam welding is a widely used fusion welding process in many industrial applications such as automotive, aerospace, energy, defense, and medical products. Industry has a fundamental need to model the laser welding process to minimize experimental testing and improve confidence in production welds. However, computational models attempting to predict weld formation are limited by an incomplete understanding of the beam-material interactions. As a consequence, these models do not accurately predict the mechanisms associated with laser weld formation. To improve the current state of prediction capabilities, it is vital to better detect/measure the physical aspects of the weld pool during high energy density welding. In this work, a novel real-time laser weld monitoring device using inline coherent imaging (ICI) was used to provide a fundamental understanding of laser weld formation via vaporization. The objective of this work was to investigate and quantify the relationship relating laser weld parameters and the vapor capillary (keyhole) through a state-of-the-art measurement technique. Bead-on-plate laser beam welds were produced with partial penetration on 304L stainless steel, 2205 duplex stainless steel, and Ti-6Al-4V. Keyhole monitoring was performed using a commercially available ICI system to collect keyhole penetration data in real time. These measurements were reconstructed to generate the vapor capillary shape at different welding parameters. Process parameters significantly influenced the keyhole shape and the keyhole root position relative to the process beam. Finally, the keyhole geometry showed distinct differences between the stainless steel alloys and Ti-6Al-4V.
Machine learning (ML)-accelerated discovery requires large amounts of high-fidelity data to reveal predictive structure–property relationships. For many properties of interest in materials discovery, the challenging nature and high cost of data generation has resulted in a data landscape that is both scarcely populated and of dubious quality. Data-driven techniques starting to overcome these limitations include the use of consensus across functionals in density functional theory, the development of new functionals or accelerated electronic structure theories, and the detection of where computationally demanding methods are most necessary. When properties cannot be reliably simulated, large experimental data sets can be used to train ML models. In the absence of manual curation, increasingly sophisticated natural language processing and automated image analysis are making it possible to learn structure–property relationships from the literature. Finally, models trained on these data sets will improve as they incorporate community feedback.
Abstract not provided.
Full-field data from digital image correlation (DIC) provide rich information for finite-element analysis (FEA) validation. However, there are several inherent inconsistencies between FEA and DIC data that must be rectified before meaningful, quantitative comparisons can be made, including strain formulations, coordinate systems, data locations, strain calculation algorithms, spatial resolutions and data filtering. As such, in this paper, we investigate two full-field validation approaches: (1) the direct interpolation approach, which addresses the first three inconsistencies by interpolating the quantity of interest from one mesh to the other, and (2) the proposed DIC-levelling approach, which addresses all six inconsistencies simultaneously by processing the FEA data through a stereo-DIC simulator to ‘level’ the FEA data to the DIC data in a regularisation sense. Synthetic ‘experimental’ DIC data were generated based on a reference FEA of an exemplar test specimen. The direct interpolation approach was applied, and significant strain errorswere computed, even though therewas no model form error, because the filtering effect of theDIC enginewas neglected. In contrast, the levelling approach provided accurate validation results, with no strain error when no model form error was present. Next, model form error was purposefully introduced via a mismatch of boundary conditions. With the direct interpolation approach, the mismatch in boundary conditions was completely obfuscated, while with the levelling approach, it was clearly observed. Finally, the ‘experimental’ DIC datawere purposefully misaligned slightly fromthe FEA data. Both validation techniques suffered from the misalignment, thus motivating continued efforts to develop a robust alignment process. In conclusion, direct interpolation is insufficient, and the proposed levelling approach is required to ensure that the FEA and the DIC data have the same spatial resolution and data filtering. Only after the FEA data have been ‘levelled’ to the DIC data can meaningful, quantitative error maps be computed.
In this study, while conventional ion-soft landing uses the mass-to-charge (m/z) ratio to achieve molecular selection for deposition, here we demonstrate the use of Structures for Lossless Ion Manipulation (SLIM) for mobility-based ion selection and deposition. The dynamic rerouting capabilities of SLIM were leveraged to enable the rerouting of a selected range of mobilities to a different SLIM path (rather than MS) that terminated at a deposition surface. A selected mobility range from a phosphazene ion mixture was rerouted and deposited with a current pulse (~150 pA) resembling its mobility peak. In addition, from a mixture of tetra-alkyl ammonium (TAA) ions containing chain lengths of C5–C8, selected chains (C6, C7) were collected on a surface, reconstituted into solution-phase, and subsequently analyzed with a SLIM-qToF to obtain an IMS/MS spectrum, confirming the identity of the selected species. Further, this method was used to characterize triply charged tungsten-polyoxometalate anions, PW 12 O 40 3– (WPOM). The arrival time distribution of the IMS/MS showed multiple peaks associated with the triply charged anion (PW 12 O 40 3– ), of which a selected ATD was deposited and imaged using TEM. Additionally, the identity of the deposited WPOM was ascertained using energy-dispersive (EDS) spectroscopy. Further, we present theory and computations that reveal ion landing energies, the ability to modulate the energies, and deposition spot sizes.
We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's ν e -appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. Here, the accuracy achieved on the test sample is ≥ 99%. For full neutrino interaction simulations, the time for processing one image is ≈ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.
As AI workloads increase in scope, generalization capability becomes challenging for small task-specific models and their demand for large amounts of labeled training samples increases. On the contrary, Foundation Models (FMs) are trained with internet-scale unlabeled data via self-supervised learning and have been shown to adapt to various tasks with minimal fine-tuning. Although large FMs have demonstrated significant impact in natural language processing and computer vision, efforts toward FMs for geospatial applications have been restricted to smaller size models, as pretraining larger models requires very large computing resources equipped with state-of-the-art hardware accelerators. Current satellite constellations collect 100+TBs of data a day, resulting in images that are billions of pixels and multimodal in nature. Such geospatial data poses unique challenges opening up new opportunities to develop FMs. We investigate billion scale FMs and HPC training profiles for geospatial applications by pretraining on publicly available data. We studied from end-to-end the performance and impact in the solution by scaling the model size. Our larger 3B parameter size model achieves up to 30% improvement in top1 scene classification accuracy when comparing a 100M parameter model. Moreover, we detail performance experiments on the Frontier supercomputer, America's first exascale system, where we study different model and data parallel approaches using PyTorch's Fully Sharded Data Parallel library. Specifically, we study variants of the Vision Transformer architecture (ViT), conducting performance analysis for ViT models with size up to 15B parameters. By discussing throughput and performance bottlenecks under different parallelism configurations, we offer insights on how to leverage such leadership-class HPC resources when developing large models for geospatial imagery applications.
In this paper, we explore the operational map of a lean axial-staged combustor of premixed and partially premixed reacting jet-in-crossflow flames at high -pressure (5 atm). This study attempts to expand the data to relatively high pressure and could significantly aid scaling to real gas turbine engine conditions at 20–30 atm. High-speed camera, particle image velocimetry (PIV), CH* chemiluminescence, temperature, and pressure measurements were taken and processed to allow accurate reconstruction of six operating points relative to computational fluid dynamics (CFD) simulations under minimal adjustments. Variation of lean main stage (φ = 0.575 and 0.73) and rich jet (φ = 1.1, 4, and 8) equivalence ratio has been investigated for a four mm axial jet. The fully premixed flames were found to be controlled by the crossflow temperature before ignition and the crossflow oxygen content during combustion. Analysis of flame shape and position for the partially premixed operating points describes a lee stabilized as well as a more unsteady windward flame branch. Adjustment of added jet fuel and crossflow temperature along with its corresponding oxygen level is required to attain a compact flame body. Here, the risk of delaying combustion progress is significantly increased at a richer jet φ = 8 and an overshooting, spatially divided flame was attained with a main stage φ = 0.73. Control toward a compact flame body is critical to allow combustion at reasonable reaction rate.
High-resolution image projection over a large field of view (FOV) is hindered by the restricted space-bandwidth product (SBP) of wavefront modulators. We report a deep learning–enabled diffractive display based on a jointly trained pair of an electronic encoder and a diffractive decoder to synthesize/project super-resolved images using low-resolution wavefront modulators. The digital encoder rapidly preprocesses the high-resolution images so that their spatial information is encoded into low-resolution patterns, projected via a low SBP wavefront modulator. The diffractive decoder processes these low-resolution patterns using transmissive layers structured using deep learning to all-optically synthesize/project super-resolved images at its output FOV. This diffractive image display can achieve a super-resolution factor of ~4, increasing the SBP by ~16-fold. We experimentally validate its success using 3D-printed diffractive decoders that operate at the terahertz spectrum. This diffractive image decoder can be scaled to operate at visible wavelengths and used to design large SBP displays that are compact, low power, and computationally efficient.
Deep Neural Networks (DNNs) have become increasingly capable of performing tasks ranging from image recognition to content generation. The training and inference of DNNs heavily rely on GPUs, as GPUs' massively parallel architecture delivers extremely high computing capability. With the growing complexity of DNNs and the size of training datasets, training DNNs with a large number of GPUs is becoming a prevalent strategy. Researchers have been exploring how to design software and hardware systems for GPU farms to achieve the best utilization, efficiency, and DNN accuracy during training or inference. However, when designing and deploying such systems, designers usually rely on testing on physical hardware platforms equipped with many GPUs, incurring high costs that are almost prohibitive for system designers to test different configurations and designs, even for highly resourceful companies. While an alternative solution is to test on GPU simulators, they are often too slow for these l
The software-tutorial was developed within the 2024 Remote Sensing and Fluxes Upscaling for Real-world Impact workshop as part of the hands-on session. The workshop was supported by AmeriFlux, National Ecological Observatory Network (NEON) and CarbonDew. The software provides basic tools to perform the following tasks: - gather remote sensing images using Google Earth Engine API; - gather flux data; - perform basic functions, such as plotting time-series, perform QA of the data, compute vegetation indices; - perform correlation analysis between flux data and remote sensing data; - perform flux predictions based on remote sensing data integrated in different modalities.
SAND2025-11465O VoroClust, also known as Voronoi Clustering, is a fast, density-based unsupervised clustering algorithm applicable to high-resolution and high-dimensional data. It operates as quickly as distance-based clustering methods while effectively capturing complex regional geometries, matching the performance of current density-based methods. VoroClust employs a data-centered sphere cover to reduce computational demands while preserving data topology. It propagates clusters outward from local density peaks. Although supervised machine learning is powerful for applications like image classification and segmentation, it requires comprehensive, consistent datasets, which many applications lack. Unsupervised clustering algorithms analyze the structure of each dataset rather than relying on similarities with other examples, making them well-suited for practical applications with insufficient or inappropriate data for supervised learning. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.
Background Radiopharmaceutical therapy (RPT) uses radionuclides that decay via one of three therapeutically relevant decay modes (alpha, beta, and internal conversion (IC) / Auger electron (AE) emission) to deliver short range, highly damaging radiation inside of diseased cells, maintaining localized dose distribution and sparing healthy cells. Antimony-119 ( 119 Sb, t 1/2 = 38.19 h, EC = 100%) is one such IC/AE emitting radionuclide, previously limited to in silico computational investigation due to barriers in production, chemical separation, and chelation. A theranostic (therapeutic/diagnostic) pair can be formed with 119 Sb’s radioisotopic imaging analogue 117 Sb (t 1/2 = 2.80 h, E γ = 158.6 keV, I γ = 85.9%, β + = 262.4 keV, I β+ = 1.81%). Results Within, we report techniques for sustainable and cost-effective production of pre-clinical quality and quantity, radionuclidically pure 119 Sb and 117 Sb, novel low energy photon measurement techniques for 119 Sb activity determination, and physical yields for various tin target isotopic enrichments and thicknesses using (p, n) and (d, n) nuclear reactions. Additionally, we present a two-column separation providing a radioantimony yield of 73.1% ± 6.9% (N = 3) and tin separation factor of (6.8 ± 5.5) x 10 5 (N = 3). Apparent molar activity measurements for deuteron produced 117 Sb using the chelator TREN-CAM were measured at 42.4 ± 25 MBq 117 Sb/µmol (1.14 ± 0.68 mCi/µmol), and we recovered enriched 119 Sn target material at a recycling efficiency of 80.2% ± 5.5% (N = 6) with losses of 11.6 mg ± 0.8 mg (N = 6) per production. Conclusion We report significant steps in overcoming barriers in 119 Sb production, chemical isolation and purification, enriched target material recycling, and chelation, helping promote accessibility and application of this promising therapeutic radionuclide. We describe a method for 119 Sb activity measurement using its low energy gamma (23.87 keV), negating the need for attenuation correction. Finally, we report the largest yet-measured 119 Sb production yields using proton and deuteron irradiation of natural and enriched targets and radioisotopic purity > 99.8% at end of purification.
Abstract Barium titanate (BTO) is a ferroelectric perovskite material used in energy storage applications because of its high dielectric constant. A previous study showed that the dielectric constant for BTO nanoparticles drastically increases to over 15,000 at a particle size of 70 nm. This result is highly contested, but its implications to energy storage motivated our investigation into the dielectric constants of BTO nanoparticles that are incorporated into a polymer matrix. We developed a novel method of using image processing techniques on transmission electron microscope images of BTO-polymer nanocomposites. Data on the positions, shapes, sizes, and orientations of BTO nanoparticles were used to build more realistic computational models that simulate the dielectric behavior of the nanocomposites. Here, we investigate the relationship between regions of enhanced electric field and the composite dielectric constant.
The ceramic microstructure strongly influences its properties. During manufacturing, the online monitoring of microstructure is critical to ensure the desired material properties. So far, the microstructure on the relevant scale is usually characterized offline using scanning electron microscopy (SEM), which is time and cost-consuming. In this work, we demonstrate a cost-effective, machine learning (ML)-based approach to simulate the SEM micrographs in real-time from the laser spot brightness. We experimentally observed a strong correlation between the laser spot brightness and the corresponding microstructure at the exact locations. The brightness values obtained from thermal emission images and the corresponding SEM micrographs were used in the training datasets. The ML algorithm was a style-based conditional generative adversarial network (CGAN). After training, the ML model could generate high-fidelity microstructure images within 0.1 seconds based on in-situ captured brightness at the laser sintering spot. We used the average grain sizes as the metric to evaluate the accuracy of the ML-predicted micrographs. Here, the ML-predicted microstructures were in good agreement, with less than 5% in difference from the real SEM images. In conclusion, we demonstrate the cost-effective, online microstructure estimation during laser sintering with a simple setup (a camera, a regular computer, and the ML model).