SEARCH · Engineering Papers
Results for “DEEP SPACE”
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.
New method for detecting fast neutrino flavor conversions in core-collapse supernova models with two-moment neutrino transport
Fast-pairwise neutrino oscillations potentially affect many aspects of core-collapse supernovae (CCSNe): the explosion mechanism, neutrino signals, and nucleosynthesis in the ejecta. This particular mode of collective neutrino oscillations has a deep connection to the angular structure of neutrinos in momentum space; for instance, the appearance of electron-neutrino lepton number (ELN) angular crossings in momentum space is a good indicator of the occurrence of flavor conversions. However, many multidimensional CCSN simulations are carried out with approximate neutrino transport (such as two-moment methods), which limits access to the angular distributions of neutrinos, i.e., inhibits ELNcrossing searches. In this paper, we develop a new method of searching for ELN crossing in these CCSN simulations. The required data is the zeroth and first angular moments of neutrinos and the matter profile, all of which are available in CCSN models with the two-moment method. One of the novelties of our new method is the use of a ray-tracing neutrino transport to determine ELNs in the direction of the stellar center. It is designed to compensate for shortcomings of the crossing searches with only two angular moments. We assess the capability of the method by carrying out a detailed comparison with results of full Boltzmann neutrino transport in 1D and 2D CCSN models. In this work, we find that the ray-tracing neutrino transport improves the accuracy of crossing searches; indeed, the appearance/disappearance of the crossings is accurately detected even in the region of forward-peaked angular distributions. The new method is computationally cheap and has the benefit of efficient parallelization; hence, it will be useful for ELN-crossing searches in any CCSN models that employ two-moment neutrino transport.
Super-resolution image display using diffractive decoders
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.
Towards On-Chip Learning for Low Latency Reasoning with End-to-End Synthesis
The Software Defined Architectures (SODA) Synthesizer is an open-source compiler-based tool able to automatically generate domain-specialized systems targeting Application-Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs) starting from high-level programming. SODA is composed of a frontend, SODA-OPT, which leverages the multilevel intermediate representation (MLIR) framework to interface with productive programming tools (e.g., machine learning frame-works), identify kernels suitable for acceleration, and perform high-level optimizations, and of a state-of-the-art high-level synthesis backend, Bambu from the PandA framework, to generate custom accelerators. One specific application of the SODA Synthesizer is the generation of accelerators to enable ultra-low latency inference and control on autonomous systems for scientific discovery (e.g., electron microscopes, sensors in particle accelerators, etc.). This paper provides an overview of the flow in the context of the generation of accelerators for edge processing to be integrated in transmission electron microscopy (TEM) devices, focusing on use cases from precision material synthesis. We show the tool in action with an example of design space exploration for inference on reconfigurable devices with a conventional deep neural network model (LeNet). Finally, we discuss the research directions and opportunities enabled by SODA in the area of autonomous control for scientific experimental workflows.
Chasing Accreted Structures within Gaia DR2 Using Deep Learning
In previous work, we developed a deep neural network classifier that only relies on phase-space information to obtain a catalog of accreted stars based on the second data release of Gaia (DR2). In this paper, we apply two clustering algorithms to identify velocity substructure within this catalog. We focus on the subset of stars with line-of-sight velocity measurements that fall in the range of Galactocentric radii $r\in [6.5,9.5]\,{\rm{kpc}}$ and vertical distances $| z| \lt 3\,{\rm{kpc}}$. Known structures such as Gaia Enceladus and the Helmi stream are identified. The largest previously unknown structure, Nyx, is a vast stream consisting of at least 200 stars in the region of interest. This study displays the power of the machine-learning approach by not only successfully identifying known features but also discovering new kinematic structures that may shed light on the merger history of the Milky Way.
Advances in Spectro‐Microscopy Methods and their Applications in the Characterization of Perovskite Materials
Perovskite materials are promising contenders as the active layer in light-harvesting and light-emitting applications if their long-term stability can be sufficiently increased. Chemical and structural engineering are shown to enhance long-term stability, but the increased complexity of the material system also leads to inhomogeneous functional properties across various length scales. Thus, scanning probe and high-resolution microscopy characterization techniques are needed to reveal the role of local defects and the results promise to act as the foundation for future device improvements. Here, a look at the parameter space: technique-specific sample penetration depth versus probe size highlights a gap in current methods. High spatial resolution combined with a deep penetration depth is not yet achievable. However, multimodal measurement technique may be the key to covering this parameter space. In this perspective, current advanced spectro-microscopy methods which have been applied to perovskite materials are highlighted.
Exploring the 3D nucleon structure with CLAS and CLAS12 at JLAB
Exploring the 3 dimensional structure of the nucleon can help to understand several fundamental questions of nature, such as the origin of the nucleon spin and the charge and density distributions inside the nucleon. In QCD, the 3-dimensional structure of the nucleon is described by Wigner functions. However, experimentally momentum and coordinate space have to be assessed independently. The momentum distribution can be accessed by transverse momentum dependent distribution functions (TMDs) measured in semi-inclusive deep inelastic scattering (SIDIS) or Drell-Yan processes while the distribution in transverse coordinate and longitudinal momentum space is described by generalized parton distributions (GPDs) which can be accessed for example by deeply virtual Compton scattering (DVCS) and hard exclusive meson production (DVMP). Based on the high quality data of CLAS and the recently upgraded CLAS12 detector at Jefferson Laboratory (JLAB), a detailed study of these distribution functions is being performed. With the new CLAS12 data, multidimensional, high precision studies in an extended kinematic range become possible for the first time. The talk will present the results of recent SIDIS, DVCS and DVMP studies with CLAS and CLAS12 and their impact on the understanding of the 3D nucleon structure.
Submersible Multistage Centrifugal Pump for Versatile Test Reactor Cartridge Test Loop
Submersible multistage centrifugal pumps are ideal for pumping in narrow confined spaces and achieving necessary head pressures and flow rates. Once a diameter is determined then manipulation of the number of stages and motor speed are all that are required to meet desired flow conditions. The Versatile Test Reactor (VTR) closed loop cartridge systems will need forced convection cooling independent of the main reactor. A multistage centrifugal pump can meet the necessary flow rates and pumping pressures while minimizing space taken. The pump considered for this work was based off a deep well submersible pump, a variation of a multistage centrifugal pump. We experimented with two pump sizes, 5 cm (2 inch) and 7.5 cm (3 inch) diameters. These diameters were chosen to fit into the inner diameter of standard 5 and 7.5 (2 and 3 inch) Schedule 40 pipe, respectively. This made the design for the test loop both simpler and less expensive as the need for an engineered pump housing was eliminated. Initial test cartridge planning indicated space for only a 5 cm (2 inch) diameter pump, though early testing of this size showed the need for an abnormally high-speed and high-power motor. Fine tuning of the cartridge design allowed a pump size increase to 7.5 cm (3 inches), which was the pump size most extensively tested in this work. The test loop is composed of various sizes of PVC and aluminum piping components in a loop configuration. The pump is driven by a Pittman 250 W (1/3 horsepower) electric motor with maximum speed of 3,450 RPM. Testing consisted of running the pump at a constant motor speed while varying a control valve to restrict flow through the loop, with differential pressure and flow rate recorded. This was done for one and two stage configurations for the 5 cm (2 inch) diameter impeller design and one, two, and three stage configurations for the 7.5 cm (3 inch diameter) impeller design, respectively. Due to pumping power requirements, two and three stage 7.5 cm (3 inch) diameter impeller testing at higher flowrates lowered the motor speed substantially. In regions where motor speed could not be maintained constant, the data were discarded. The test loop was also reconfigured to allow for the pump to be tested for pressure drop in a stalled or inoperable (0 RPM) flow condition. Demonstration of adequate natural convection cooling of the test cartridge fuel type is necessary under accident conditions, and this will depend upon the flow resistance through the impeller assembly when the pump is not operating. Thus, accurate knowledge of the effective impeller assembly loss coefficient is important for safety evaluations. The test loop was modified to provide water inlet and outlets on either side of the pump impeller stack, and a metered flow of lab water was provided in order to measure the pressure drop across the cartridges as a function of flowrate. Data from the pump head curve testing developed as part of this work and supported by analysis using pump head affinity laws indicates that a three stage 7.5 cm (3 inch) pump impeller design will meet target requirements for coolant flow within the VTR cartridge sodium cartridge at full power conditions [1] with margin; this corresponds to a flowrate of 45 l/min (12 gpm) at a pressure drop of 6.1 m (20 feet) of water head. The results of the pressure loss measurements across the impeller assembly when the pump is stationary (i.e., at 0 RPM) indicate that the pressure loss coefficient is 0.921 for a two impeller stack configuration; this value is calculated based on the flow velocity through the minimum available flow area within a single stage of the impeller which corresponds to 1.4 cm2.
A resolution independent neural operator
The Deep operator network (DeepONet) is a powerful yet simple neural operator architecture that utilizes two deep neural networks to learn mappings between infinite-dimensional function spaces. This architecture is highly flexible, allowing the evaluation of the solution field at any location within the desired domain. However, it imposes a strict constraint on the input space, requiring all input functions to be discretized at the same locations; this limits its practical applications. Here, in this work, we introduce a general framework for operator learning from input–output data with arbitrary number and locations of sensors. This begins by introducing a resolution-independent DeepONet (RI-DeepONet), enabling it to handle input functions that are arbitrarily, but sufficiently finely, discretized. To this end, we propose two dictionary learning algorithms to adaptively learn a set of appropriate continuous basis functions, parameterized as implicit neural representations (INRs), from correlated signals defined on arbitrary point cloud data. These basis functions are then used to project arbitrary input function data as a point cloud onto an embedding space (i.e., a vector space of finite dimensions) with dimensionality equal to the dictionary size, which can be directly used by DeepONet without any architectural changes. In particular, we utilize sinusoidal representation networks (SIRENs) as trainable INR basis functions. The introduced dictionary learning algorithms are then used in a similar way to learn an appropriate dictionary of basis functions for the output function data, which defines a new neural operator architecture referred to as the R esolution I ndependent N eural O perator (RINO). In the RINO, the operator learning task simplifies to learning a mapping from the coefficients of input basis functions to the coefficients of output basis functions. We demonstrate the robustness and applicability of RINO in handling arbitrarily (but sufficiently richly) sampled input and output functions during both training and inference through several numerical examples.
Inverse design of hypoeutectoid pearlite steel microstructures using a deep learning and genetic algorithm optimization framework
Goal-oriented microstructure design in metallic materials is a challenging task due to complex structure-property relationships. Traditional experimental and computational approaches are time-intensive and economically inefficient, limiting their applicability for large-scale design space exploration. Here, in this work, we propose an end-to-end framework that integrates deep learning models with genetic optimization to design microstructures with targeted mechanical properties. Deep learning models enable accurate forward design, while their integration with genetic optimization enables efficient inverse design within a few hours, compared to days or weeks using conventional finite element simulations. The framework combines experimental characterization and finite element modeling to analyze the influence of microstructural features on the mechanical behavior of hypoeutectoid steels. Data from both experiments and simulations are used to train the deep learning models. To demonstrate its effectiveness, we apply the framework to 0.63% carbon steel with proeutectoid ferrite and pearlite phases, commonly used in industrial applications. In this study, 2D microstructures were used for modeling, selected primarily for computational efficiency and to establish proof of concept. The framework successfully optimizes microstructures for targeted yield strength, ultimate strength, and stress concentration factors while significantly reducing computational time. Beyond hypoeutectoid steels, this scalable framework can be extended to other material systems and integrated with additive manufacturing, offering an efficient approach for accelerating microstructure design for specific engineering applications.
Deep learning-based predictive models for laser direct drive at the Omega Laser Facility
The rich and complex physics of inertial confinement fusion provides a unique and challenging space for high-fidelity first-principles modeling. Consequently, simulation codes that are used to design experiments are computationally expensive and lack the predictive capability required for extensive parameter exploration in search of a high-performing design for laser direct drive. In this article, we present two deep-learning-based predictive models intended to address these difficulties. The first model (TL DNN) acts as a fast emulator of simulations as well as experiments at the Omega Laser Facility. This model is trained on a simulation database and subsequently calibrated on experimental data using transfer learning. To facilitate the development of this model, an autoencoder is developed to reduce the dimensionality of the input space by compressing the laser pulse input. The model predicts key experimental scalar observables of Omega experiments with high accuracy and minimal computational cost. This deep neural net enables rapid exploration of a high-dimensional input parameter space for an optimal implosion design. The second model (DNN SM+) aims to extend the statistical modeling work of Lees et al. [Phys. Rev. Lett. 127, 105001 (2021)], by increasing the complexity of the model space and allowing for coupling between degradation terms. Since the model capacity of DNN SM+ is higher than the model of Lees et al., DNN SM+ can potentially provide an improvement in predictive capability, and we use this model to provide insight into complicated degradation dependencies.
DeepAdversaries: examining the robustness of deep learning models for galaxy morphology classification
With increased adoption of supervised deep learning methods for work with cosmological survey data, the assessment of data perturbation effects (that can naturally occur in the data processing and analysis pipelines) and the development of methods that increase model robustness are increasingly important. In the context of morphological classification of galaxies, we study the effects of perturbations in imaging data. In particular, we examine the consequences of using neural networks when training on baseline data and testing on perturbed data. We consider perturbations associated with two primary sources: (a) increased observational noise as represented by higher levels of Poisson noise and (b) data processing noise incurred by steps such as image compression or telescope errors as represented by one-pixel adversarial attacks. We also test the efficacy of domain adaptation techniques in mitigating the perturbation-driven errors. We use classification accuracy, latent space visualizations, and latent space distance to assess model robustness in the face of these perturbations. For deep learning models without domain adaptation, we find that processing pixel-level errors easily flip the classification into an incorrect class and that higher observational noise makes the model trained on low-noise data unable to classify galaxy morphologies. On the other hand, we show that training with domain adaptation improves model robustness and mitigates the effects of these perturbations, improving the classification accuracy up to 23% on data with higher observational noise. Domain adaptation also increases up to a factor of ${\approx}2.3$ the latent space distance between the baseline and the incorrectly classified one-pixel perturbed image, making the model more robust to inadvertent perturbations. Successful development and implementation of methods that increase model robustness in astronomical survey pipelines will help pave the way for many more uses of deep learning for astronomy.
We Know Where We Don't Know: 3D Bayesian CNNs for Credible Geometric Uncertainty
Deep learning has been successfully applied to the segmentation of 3D Computed Tomography (CT) scans. Establishing the credibility of these segmentations requires uncertainty quantification (UQ) to identify untrustworthy predictions. Recent UQ architectures include Monte Carlo dropout networks (MCDNs), which approximate deep Gaussian processes, and Bayesian neural networks (BNNs), which learn the distribution of the weight space. BNNs are advantageous over MCDNs for UQ but are thought to be computationally infeasible in high dimension, and neither architecture has produced interpretable geometric uncertainty maps. We propose a novel 3D Bayesian convolutional neural network (BCNN), the first deep learning method which generates statistically credible geometric uncertainty maps and scales for application to 3D data. We present experimental results on CT scans of graphite electrodes and laser-welded metals and show that our BCNN outperforms an MCDN in recent uncertainty metrics. The geometric uncertainty maps generated by our BCNN capture distributions of sigmoid values that are interpretable as confidence intervals, critical for applications that rely on deep learning for high-consequence decisions.
ML-based Dimension Reduction Strategies
Deep learning (DL)--based surrogate models have achieved success in various applications in carbon capture and storage (CCS). However, the model training on high-dimensional spaces is computationally expensive and impractical for large-scale and complex geological models, because the models usually contain hundreds of thousands to millions of grid cells, each with a set of parameters. Furthermore, the high cost of generating training data with sufficient variation is another limitation of model training on high-dimensional spaces, which may result in overfitting and reduce the model efficiency and prediction performance. We proposed the workflow incorporating dimension reduction methods and deep learning models, which aim to extract the latent variables of input parameters and output state variables, and then build the mapping function at the latent spaces. The proposed workflow can significantly reduce the computational complexity in solving both forward and inverse problems compared to models trained on high-dimensional spaces. Dimensionality reduction models showed great potential in workflows for fast reservoir simulation, history matching, prior model generation, visualization, and more, ultimately enhancing DL model performance in related SMART Work Packages.
Wire-cell 3D pattern recognition techniques for neutrino event reconstruction in large LArTPCs: algorithm description and quantitative evaluation with MicroBooNE simulation
Wire-Cell is a 3D event reconstruction package for liquid argon time projection chambers. Through geometry, time, and drifted charge from multiple readout wire planes, 3D space points with associated charge are reconstructed prior to the pattern recognition stage. Pattern recognition techniques, including track trajectory and dQ/dx (ionization charge per unit length) fitting, 3D neutrino vertex fitting, track and shower separation, particle-level clustering, and particle identification are then applied on these 3D space points as well as the original 2D projection measurements. A deep neural network is developed to enhance the reconstruction of the neutrino interaction vertex. Compared to traditional algorithms, the deep neural network boosts the vertex efficiency by a relative 30% for charged-current ν e interactions. Therefore, this pattern recognition achieves 80–90% reconstruction efficiencies for primary leptons, after a 65.8% (72.9%) vertex efficiency for charged-current ν e (ν μ ) interactions. Based on the resulting reconstructed particles and their kinematics, we also achieve 15-20% energy reconstruction resolutions for charged-current neutrino interactions.
Beryllium Isotopes in Rainfall for Stratospheric-Tropospheric Exchange Dynamics Experiment (BIRDIE) (Field Campaign Report)
The Beryllium Isotopes in Rainfall for Stratospheric-Tropospheric Exchange Dynamics Experiment or BIRDIE was a precipitation collection campaign designed to measure cosmogenic isotopes of beryllium in rainfall and other forms of precipitation across the U.S. Department of Energy Atmospheric Radiation Measurement (ARM) user facility’s Southern Great Plains (SGP) atmospheric observatory. Precipitation samples were collected over three weeks (May 21-June 10, 2022) at 11 SGP sites across a 120-km 2 domain in northern and central Oklahoma. The field campaign targeted summertime convection rain events. We aimed to capture any spatial and/or temporal variability in the concentrations of beryllium-10 (10Be) and beryllium-7 (7Be) across the ARM SGP site during these rainfall events. The ratio of 10Be to 7Be (10/7Be) has been shown to be a powerful tracer for transport processes in the stratosphere-troposphere exchange (STE) system (e.g., Koch and Rind 1998). Any observed variability is analyzed to see if it could be linked to STE processes, including overshooting of deep convection clouds that penetrate the lower stratosphere. Six wet weather system events were collected within our three-week span. Five out of six of these rainfall events are classified as moderately to strongly convective. Those events are studied in more detail. The five systems also follow each other over a short period of 11 days. In our rainfall samples we observed spatial and event variability in the concentrations of beryllium-10 and -7 that resulted in variability of 10/7Be ratios. Each weather event appeared to exhibit a unique 10/7Be pattern across the domain that was not directly linked to storm total precipitation amount. To better understand the causes for this spatial variability, cloud characteristics in the domain were looked at to assess cloud type, cloud height, and depth information. We also ran Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model back trajectories to identify possible air mass and moisture origins across our events. Detailed analysis is ongoing. The purpose of this technical report is to provide background on using cosmogenic isotopes as atmospheric tracers, describe the experimental design, and present the beryllium concentration and ratio data. Lastly, we note that BIRDIE coincided with two other deep convection field campaigns in the area: The National Aeronautics and Space Administration (NASA)’s Dynamics and Chemistry of the Summer Stratosphere (DCOTSS) experiment (https://dcotss.org/) centered in Salina, Kansas, and the Targeted Observation by Radars and (uncrewed aerial systems) UAS of Supercells (TORUS) experiment (https://catalog.eol.ucar.edu/torus_2022) across the U.S. Central Plains. Additionally, a Lawrence Livermore National Laboratory ( LLNL) team at DCOTSS attempted to measure 10Be/7Be ratios on aerosols collected in situ in the lower and mid-stratosphere over Kansas using high-altitude balloons. Precipitation samples were also collected on the ground in Salina, Kansas.
Deep-field analytical calibration
The next generation of imaging surveys, including the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST), Euclid, and the Nancy Grace Roman Space Telescope, will provide unprecedented constraints on cosmology using weak gravitational lensing. To fully exploit this statistical power, shear measurement methods must achieve sub- per cent accuracy while mitigating systematic biases from noise, the point-spread function (PSF), blending, and shear-dependent detection. The analytical calibration framework (AnaCal) has demonstrated such accuracy but requires adding noise to images, reducing effective depth. We introduce Deep-Field Analytical Calibration (DEEP-FIELD AnaCal), an extension of AnaCal that uses deep-field images to compute shear responses while preserving the statistical power of wide-field data. We validate DEEP-FIELD AnaCal on isolated and blended galaxy image simulations with LSST-like conditions, finding it meets the stringent requirement of multiplicative bias $|m| < 3\times 10^{-3}$ at 99.7 per cent confidence. Compared to standard AnaCal applied to wide-field images, DEEP-FIELD AnaCal increases the effective galaxy number density from 17 to 30 arcmin$^{-2}$ for simulated 10-yr LSST data. With deep fields $10\times$ longer than the wide field, we find pixel noise variance in shear estimation is reduced by 30 per cent and overall uncertainty by $\sim 25~{{\ \rm per\ cent}}$. Finally, using the LSST Deep Drilling Fields strategy, we assess sample variance and find an equivalent calibration uncertainty of $\lesssim 0.3~{{\ \rm per\ cent}}$. These results demonstrate that DEEP-FIELD AnaCal offers a promising path to achieve the required shear calibration for upcoming weak lensing surveys.
Improved Architectures and Training Algorithms for Deep Operator Networks
Operator learning techniques have recently emerged as a powerful tool for learning maps between infinite-dimensional Banach spaces. Trained under appropriate constraints, they can also be effective in learning the solution operator of partial differential equations (PDEs) in an entirely self-supervised manner. In this work we analyze the training dynamics of deep operator networks (DeepONets) through the lens of Neural Tangent Kernel theory, and reveal a bias that favors the approximation of functions with larger magnitudes. To correct this bias we propose to adaptively re-weight the importance of each training example, and demonstrate how this procedure can effectively balance the magnitude of back-propagated gradients during training via gradient descent. We also propose a novel network architecture that is more resilient to vanishing gradient pathologies. Taken together, our developments provide new insights into the training of DeepONets and consistently improve their predictive accuracy by a factor of 10-50x, demonstrated in the challenging setting of learning PDE solution operators in the absence of paired input-output observations.