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.
BUTTER ( An Empirical Deep Learning Experimental Framework) is a system with which empirical deep learning experiments can be conducted and their results, including training and model performance characteristics, can be accumulated for further analysis.
This package contains the code for the paper Combining Deep Learning with Physics Based Features in Explosion-Earthquake Discrimination published at Geophysical Research Letters. There are two Jupyter notebooks, one for building and training the two-branch model, and the other is visualizing the deep learning branch using GradCAM.
SAND2022-14744 O The Deep Learning Denoiser uses a trained deep convolutional neural network (CNN) model to decompose an input waveform into a signal and noise waveform. In the approach, the CNN provides a signal mask and a noise mask for an input signal. The Short-Time Fourier Transform (STFT) of the estimated signal is obtained by multiplying the signal mask with the STFT of the input signal. 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.
The Deep Lynx MATLAB Adapter is a Python application that connects the Deep Lynx data warehouse with any MATLAB simulation.
deep-taxon is a package for training neural networks for classifying genomic sequence data. deep-taxon provides tools for converting genomic sequence data into high performance computing file formats capable of training neural networks at scale.
This project is an implementation of the Deep Active Learning (DeepAL) framework from the paper Deep Active Learning based Experimental Design to Uncover Synergistic Genetic Interactions for Host Targeted Therapeutics
SAND2025-11741O Toadstool is a deep learning framework and support library that provides PyTorch boiler plate training and testing loops. This enables the user to remember parts and customize a callback interface. Toadstool also provides useful callbacks and other methods for deep learning experimentation. The framework also implements publicly available temperature and calibration methods, model initialization methods, learning rate schedulers, and model evaluation methods. 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.
Here, we present a method that uses protein levels to predict times series of metabolite concentrations. Understanding this type of pathway dynamics is important in order to predict the behavior of the pathway and, more pragmatically, to be able to design biological systems (such as strains bioengineered to produce chemical products) reliably. Typically, for this purpose, kinetic models consisting of differential equations based on the Michaelis-Menten dynamics have been used in the past. However, these methods can rarely produce good fits to measured data time series. Possibly, this happens because the kinetic constants are unknown or are different from the ones measured in vivo, or perhaps because Michaelis-Menten dynamics is not a satisfactory description. In order to improve the predictive nature of these kinetic models we have eliminated the Michaelis-Menten description of pathway dynamics and we have substituted it by algorithms that automatically learn these dynamics from previously obtained metabolomics and proteomics data using machine learning approaches. Specifically, kinetic deep learning uses deep learning to map proteomics time series to metabolite concentration time series, instead of learning the first metabolite derivative and integrating in (as in the first version of kinetic learning). This approach is shown to provide good to excellent results with a data set specifically collected for this purpose.
The Amazon rainforest is one of a few regions of the world where continental tropical deep convection occurs. The Amazon’s isolation makes it challenging to observe, but also creates a unique natural laboratory to study anthropogenic impacts on clouds and precipitation in an otherwise pristine environment. Extensive measurements were made upwind and downwind of the large city of Manaus, Brazil during the Observations and Modeling of the Green Ocean Amazon 2014-2015 (GoAmazon2014/5) field campaign. In this study, 15 years of high-resolution satellite data are analyzed to examine the spatial and diurnal variability of convection occurring around the GoAmazon2014/5 sites. Interpretation of anthropogenic differences between the upwind (T0) and downwind (T1-T3) sites is complicated by naturally-occurring spatial variability between the sites. During the rainy season, the inland propagation of the previous day’s sea-breeze front happens to be in phase with the background diurnal cycle near Manaus, but is out of phase elsewhere. Enhanced convergence between the river-breezes and the easterly trade winds generates up to 10% more frequent deep convection at the GoAmazon2014/5 sites east of the river (T0a, T0t/k, and T1) compared to the T3 site which was located near the western bank. In general, the annual and diurnal cycles during 2014 were representative of the 2000-2013 distributions. The only exceptions were in March when the monthly mean rainrate was above the 95th percentile and September when both rain frequency and intensity were suppressed. The natural spatial variability must be accounted for before interpreting anthropogenically-induced differences among the GoAmazon2014/5 sites.
Cloud-scale models apply two drastically different methods to represent condensation of water vapor to form and grow cloud droplets. Maintenance of water saturation inside liquid clouds is assumed in the computationally efficient saturation adjustment approach used in most bulk microphysics schemes. When super- or subsaturations are allowed, condensation/evaporation can be calculated using the predicted saturation ratio and (either predicted or prescribed) mean droplet radius and concentration. Here, we investigate differences between simulations of deep unorganized convection applying a saturation adjustment condensation scheme (SADJ) and a scheme with supersaturation prediction (SPRE). A double-moment microphysics scheme with CCN activation parameterized as a function of the local vertical velocity is applied to compare cloud fields simulated applying SPRE and SADJ. Clean CCN conditions are assumed to demonstrate upper limits of the SPRE and SADJ difference. Microphysical piggybacking is used to extract the impacts with confidence. Results show a significant impact on deep convection dynamics, with SADJ featuring more cloud buoyancy and thus stronger updrafts. This leads to around a 3% increase of the surface rain accumulation in SADJ. Upper-tropospheric anvil cloud fractions are much larger in SPRE than in SADJ because of the higher ice concentrations and thus longer residence times of anvil particles in SPRE, as demonstrated by sensitivity tests. Higher ice concentrations in SPRE come from significantly larger ice supersaturations in strong convective updrafts that feature water supersaturations of several percent.
Abstract Contiguous time–height cloud objects at the Department of Energy Atmospheric Radiation Measurement Southern Great Plains (SGP) site are matched with surface condensation nuclei (CN) concentrations and retrieved thermodynamic and kinematic vertical profiles for warm-cloud-base, cold-cloud-top systems in convectively unstable environments. Statistical analyses show that previously published conclusions that increasing CN concentrations cause a decrease in minimum cloud-top temperature (CTT) at the SGP site through the aerosol convective invigoration effect are unfounded. The CN–CTT relationship is statistically insignificant, while correlations between convective available potential energy (CAPE), level of neutral buoyancy (LNB), and CN concentration account for most of the change in the CN–CTT positive correlation. Removal of clouds with minimum CTTs > −36°C from the analysis eliminates the CN–CTT correlation. Composited dirty conditions at the SGP have ~1°C-warmer low levels and ~1°C-cooler upper levels than clean conditions. This correlation between aerosol concentrations and thermodynamic profiles may be caused by an increase in regional rainfall preceding deep convective conditions as CN concentration decreases. Increased rainfall can be expected to increase wet deposition of aerosols, cool low-level temperatures, and warm upper-level temperatures. The masking of a potential aerosol effect by such small thermodynamic changes implies that the strategy of analyzing subsets of aerosol data by binned meteorological factor values is not a valid method for discerning an aerosol effect in some situations. These findings highlight the need for more careful, detailed, and strategic observations to confidently isolate and quantify an aerosol deep convective invigoration effect.
Abstract We developed an entraining parcel approach that partitions parcel buoyancy into contributions from different processes (e.g., adiabatic cooling, condensation, freezing, and entrainment). Applying this method to research-quality radiosonde profiles provided by the Atmospheric Radiation Measurement (ARM) program at six sites, we evaluated how atmospheric thermodynamic conditions and entrainment influence various physical processes that determine the vertical buoyancy structure across different climate regimes as represented by these sites. The differences of morning buoyancy profiles between the deep convection (DC)/transition cases and shallow convection (SC)/nontransition cases were used to assess preconditions important for shallow-to-deep convection transition. Our results show that for continental sites such as the U.S. Southern Great Plains (SGP) and west-central Africa, surface conditions alone are enough to account for the buoyancy difference between DC and SC cases, although entrainment further enhances the buoyancy difference at SGP. For oceanic sites in the tropical west Pacific, humidity dilution in the lower to middle free troposphere (~1–6 km) and temperature mixing in the middle to upper troposphere (>4 km) have the most important influences on the buoyancy difference between DC and SC cases. For the humid central Amazon region, entrainment in both the boundary layer and the lower free troposphere (~0–4 km) have significant contributions to the buoyancy difference; the upper-tropospheric influence seems unimportant. In addition, the integral of the condensation term, which represents the parcel’s ability to transform available water vapor into heat through condensation, provides a better discrimination between DC and SC cases than the integral of buoyancy or the convective available potential energy (CAPE).
Numerical simulations of the impact of ultrafine cloud condensation nuclei (CCN) on deep convection are analyzed herein to investigate the idea proposed by Fan et al. that addition of ultrafine CCN to an otherwise pristine environment leads to convective invigoration. The piggybacking methodology is applied, allowing rigorous separation of the impact of aerosols from different flow realizations that typically occur when even a small element of the model physics or modeling setup is changed. The setup follows the case of daytime convective development over land based on observations during the Large-Scale Biosphere–Atmosphere (LBA) experiment in Amazonia. Overall, the simulated impacts of ultrafine CCN are similar to the previous study by the authors on the impact of pollution on deep convection. There is no convective invigoration above the freezing level, but there is a small invigoration (increase in vertical velocities) below due to the supersaturation and buoyancy differences in conditions with additional ultrafine CCN compared to unperturbed pristine conditions. As in the previous study, the most significant impact is on the upper-tropospheric convective anvils that feature higher cloud fractions in conditions with ultrafine CCN. The increase comes from purely microphysical considerations as the increased cloud droplet concentrations from ultrafine CCN lead to increased ice crystal concentrations and, consequently, smaller sizes and fall velocities, and longer residence times. Mesoscale organization due to low-level shear has a small effect on the simulated ultrafine CCN impacts. Finally, an alternative explanation of increased lightning above oceanic shipping lines seen in satellite observations and argued to result from convective invigoration is provided.
Abstract It is often assumed in parcel theory calculations, numerical models, and cumulus parameterizations that moist static energy (MSE) is adiabatically conserved. However, the adiabatic conservation of MSE is only approximate because of the assumption of hydrostatic balance. Two alternative variables are evaluated here: MSE − IB and MSE + KE, wherein IB is the path integral of buoyancy ( B ) and KE is kinetic energy. Both of these variables relax the hydrostatic assumption and are more precisely conserved than MSE. This article quantifies the errors that result from assuming that the aforementioned variables are conserved in large-eddy simulations (LES) of both disorganized and organized deep convection. Results show that both MSE − IB and MSE + KE better predict quantities along trajectories than MSE alone. MSE − IB is better conserved in isolated deep convection, whereas MSE − IB and MSE + KE perform comparably in squall-line simulations. These results are explained by differences between the pressure perturbation behavior of squall lines and isolated convection. Errors in updraft B diagnoses are universally minimized when MSE − IB is assumed to be adiabatically conserved, but only when moisture dependencies of heat capacity and temperature dependency of latent heating are accounted for. When less accurate latent heat and heat capacity formulae were used, MSE − IB yielded poorer B predictions than MSE due to compensating errors. Our results suggest that various applications would benefit from using either MSE − IB or MSE + KE instead of MSE with properly formulated heat capacities and latent heats.
The Remote Sensing of Electrification, Lightning, and Mesoscale/Microscale Processes with Adaptive Ground Observations (RELAMPAGO) and Cloud, Aerosol, and Complex Terrain Interactions (CACTI) projects deployed a high-spatiotemporal-resolution radiosonde network to examine environments supporting deep convection in the complex terrain of central Argentina. This study aims to characterize atmospheric profiles most representative of the near-cloud environment (in time and space) to identify the mesoscale ingredients affecting storm initiation and growth. Spatiotemporal autocorrelation analysis of the soundings reveals that there is considerable environmental heterogeneity, with boundary layer thermodynamic and kinematic fields becoming statistically uncorrelated on scales of 1–2 h and 30 km. Using this as guidance, we examine a variety of environmental parameters derived from soundings collected within close proximity (30 km in space and 30 min in time) of 44 events over 9 days where the atmosphere either: 1) supported the initiation of sustained precipitating convection, 2) yielded weak and short-lived precipitating convection, or 3) produced no precipitating convection in disagreement with numerical forecasts from convection-allowing models (i.e., Null events). There are large statistical differences between the Null event environments and those supporting any convective precipitation. Null event profiles contained larger convective available potential energy, but had low free-tropospheric relative humidity, higher freezing levels, and evidence of limited horizontal convergence near the terrain at low levels that likely suppressed deep convective growth. Here, we also present evidence from the radiosonde and satellite measurements that flow–terrain interactions may yield gravity wave activity that affects CI outcome.
The Abisko project aims to develop an energy-efficient spiking neural network (SNN) computing architecture and software system capable of autonomous learning and operation. The SNN architecture explores novel neuromorphic devices that are based on resistive-switching materials, such as memristors and electrochemical RAM. Equally important, Abisko uses a deep codesign approach to pursue this goal by engaging experts from across the entire range of disciplines: materials, devices and circuits, architectures and integration, software, and algorithms. Here, the key objectives of our Abisko project are threefold. First, we are designing an energy-optimized high-performance neuromorphic accelerator based on SNNs. This architecture is being designed as a chiplet that can be deployed in contemporary computer architectures and we are investigating novel neuromorphic materials to improve its design. Second, we are concurrently developing a productive software stack for the neuromorphic accelerator that will also be portable to other architectures, such as field-programmable gate arrays and GPUs. Third, we are creating a new deep codesign methodology and framework for developing clear interfaces, requirements, and metrics between each level of abstraction to enable the system design to be explored and implemented interchangeably with execution, measurement, a model, or simulation. As a motivating application for this codesign effort, we target the use of SNNs for an analog event detector for a high-energy physics sensor.
Background Histopathology images of tumor biopsies present unique challenges for applying machine learning to the diagnosis and treatment of cancer. The pathology slides are high resolution, often exceeding 1GB, have non-uniform dimensions, and often contain multiple tissue slices of varying sizes surrounded by large empty regions. The locations of abnormal or cancerous cells, which may constitute a small portion of any given tissue sample, are not annotated. Cancer image datasets are also extremely imbalanced, with most slides being associated with relatively common cancers. Since deep representations trained on natural photographs are unlikely to be optimal for classifying pathology slide images, which have different spectral ranges and spatial structure, we here describe an approach for learning features and inferring representations of cancer pathology slides based on sparse coding. Results We show that conventional transfer learning using a state-of-the-art deep learning architecture pre-trained on ImageNet (RESNET) and fine tuned for a binary tumor/no-tumor classification task achieved between 85% and 86% accuracy. However, when all layers up to the last convolutional layer in RESNET are replaced with a single feature map inferred via a sparse coding using a dictionary optimized for sparse reconstruction of unlabeled pathology slides, classification performance improves to over 93%, corresponding to a 54% error reduction. Conclusions We conclude that a feature dictionary optimized for biomedical imagery may in general support better classification performance than does conventional transfer learning using a dictionary pre-trained on natural images.
Background: Microbial colonization of subsurface shales following hydraulic fracturing offers the opportunity to study coupled biotic and abiotic factors that impact microbial persistence in engineered deep subsurface ecosystems. Shale formations underly much of the continental USA and display geographically distinct gradients in temperature and salinity. Complementing studies performed in eastern USA shales that contain brine-like fluids, here we coupled metagenomic and metabolomic approaches to develop the first genome-level insights into ecosystem colonization and microbial community interactions in a lower-salinity, but high-temperature western USA shale formation. Results: We collected materials used during the hydraulic fracturing process (i.e., chemicals, drill muds) paired with temporal sampling of water produced from three different hydraulically fractured wells in the STACK (Sooner Trend Anadarko Basin, Canadian and Kingfisher) shale play in OK, USA. Relative to other shale formations, our metagenomic and metabolomic analyses revealed an expanded taxonomic and metabolic diversity of microorganisms that colonize and persist in fractured shales. Importantly, temporal sampling across all three hydraulic fracturing wells traced the degradation of complex polymers from the hydraulic fracturing process to the production and consumption of organic acids that support sulfate- and thiosulfate-reducing bacteria. Furthermore, we identified 5587 viral genomes and linked many of these to the dominant, colonizing microorganisms, demonstrating the key role that viral predation plays in community dynamics within this closed, engineered system. Lastly, top-side audit sampling of different source materials enabled genome-resolved source tracking, revealing the likely sources of many key colonizing and persisting taxa in these ecosystems. Conclusions: These findings highlight the importance of resource utilization and resistance to viral predation as key traits that enable specific microbial taxa to persist across fractured shale ecosystems. We also demonstrate the importance of materials used in the hydraulic fracturing process as both a source of persisting shale microorganisms and organic substrates that likely aid in sustaining the microbial community. Moreover, we showed that different physicochemical conditions (i.e., salinity, temperature) can influence the composition and functional potential of persisting microbial communities in shale ecosystems. Together, these results expand our knowledge of microbial life in deep subsurface shales and have important ramifications for management and treatment of microbial biomass in hydraulically fractured wells.