Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Generative models”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 271 records · Page 15
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Evaluation of OpenAI Codex for HPC Parallel Programming Models Kernel Generation

We evaluate AI-assisted generative capabilities on fundamental numerical kernels in high-performance computing (HPC), including AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG. We test the generated kernel codes for a variety of language-supported programming models, including (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numpy, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). We use the GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code as of April 2023 to generate a vast amount of implementations given simple + + prompt variants. To quantify and compare the results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. Results suggest that the OpenAI Codex outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general purpose Python can benefit from adding code keywords, while Julia prompts perform acceptably well for its mature programming models (e.g., Threads and CUDA.jl). We expect for these benchmarks to provide a point of reference for each programming model's community. Overall, understanding the convergence of large language models, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

Godoy, William↗
NASA NTRS1991

Model reduction by trimming for a class of semi-Markov reliability models and the corresponding error bound

Semi-Markov processes have proved to be an effective and convenient tool to construct models of systems that achieve reliability by redundancy and reconfiguration. These models are able to depict complex system architectures and to capture the dynamics of fault arrival and system recovery. A disadvantage of this approach is that the models can be extremely large, which poses both a model and a computational problem. Techniques are needed to reduce the model size. Because these systems are used in critical applications where failure can be expensive, there must be an analytically derived bound for the error produced by the model reduction technique. A model reduction technique called trimming is presented that can be applied to a popular class of systems. Automatic model generation programs were written to help the reliability analyst produce models of complex systems. This method, trimming, is easy to implement and the error bound easy to compute. Hence, the method lends itself to inclusion in an automatic model generator.

White, Allan L.↗
DOE OSTI2023

De novo design of protein structure and function with RFdiffusion

Abstract There has been considerable recent progress in designing new proteins using deep-learning methods 1–9 . Despite this progress, a general deep-learning framework for protein design that enables solution of a wide range of design challenges, including de novo binder design and design of higher-order symmetric architectures, has yet to be described. Diffusion models 10,11 have had considerable success in image and language generative modelling but limited success when applied to protein modelling, probably due to the complexity of protein backbone geometry and sequence–structure relationships. Here we show that by fine-tuning the RoseTTAFold structure prediction network on protein structure denoising tasks, we obtain a generative model of protein backbones that achieves outstanding performance on unconditional and topology-constrained protein monomer design, protein binder design, symmetric oligomer design, enzyme active site scaffolding and symmetric motif scaffolding for therapeutic and metal-binding protein design. We demonstrate the power and generality of the method, called RoseTTAFold diffusion (RFdiffusion), by experimentally characterizing the structures and functions of hundreds of designed symmetric assemblies, metal-binding proteins and protein binders. The accuracy of RFdiffusion is confirmed by the cryogenic electron microscopy structure of a designed binder in complex with influenza haemagglutinin that is nearly identical to the design model. In a manner analogous to networks that produce images from user-specified inputs, RFdiffusion enables the design of diverse functional proteins from simple molecular specifications.

Science & Technology - Other Topics↗
DOE OSTI2025

A comparison of probabilistic generative frameworks for molecular simulations

Generative artificial intelligence is now a widely used tool in molecular science. Despite the popularity of probabilistic generative models, numerical experiments benchmarking their performance on molecular data are lacking. Here, in this work, we introduce and explain several classes of generative models, broadly sorted into two categories: flow-based models and diffusion models. We select three representative models: neural spline flows, conditional flow matching, and denoising diffusion probabilistic models, and examine their accuracy, computational cost, and generation speed across datasets with tunable dimensionality, complexity, and modal asymmetry. Our findings are varied, with no one framework being the best for all purposes. In a nutshell, (i) neural spline flows do best at capturing mode asymmetry present in low-dimensional data, (ii) conditional flow matching outperforms other models for high-dimensional data with low complexity, and (iii) denoising diffusion probabilistic models appear the best for low-dimensional data with high complexity. Our datasets include a Gaussian mixture model and the dihedral torsion angle distribution of the Aib9 peptide, generated via a molecular dynamics simulation. We hope our taxonomy of probabilistic generative frameworks and numerical results may guide model selection for a wide range of molecular tasks.

Artificial intelligence↗
NASA NTRS1987

Satellite camera image navigation

Pixels within a satellite camera (1, 2) image are precisely located in terms of latitude and longitude on a celestial body, such as the earth, being imaged. A computer (60) on the earth generates models (40, 50) of the satellite's orbit and attitude, respectively. The orbit model (40) is generated from measurements of stars and landmarks taken by the camera (1, 2), and by range data. The orbit model (40) is an expression of the satellite's latitude and longitude at the subsatellite point, and of the altitude of the satellite, as a function of time, using as coefficients (K) the six Keplerian elements at epoch. The attitude model (50) is based upon star measurements taken by each camera (1, 2). The attitude model (50) is a set of expressions for the deviations in a set of mutually orthogonal reference optical axes (x, y, z) as a function of time, for each camera (1, 2). Measured data is fit into the models (40, 50) using a walking least squares fit algorithm. A transformation computer (66 ) transforms pixel coordinates as telemetered by the camera (1, 2) into earth latitude and longitude coordinates, using the orbit and attitude models (40, 50).

Kamel, Ahmed A.↗
NASA NTRS1993

The Radar Image Generation (RIG) model

RIG is a modeling system which creates synthetic aperture radar (SAR) and inverse SAR images from 3-D faceted data bases. RIG is based on a physical optics model and includes the effects of multiple reflections. Both conducting and dielectric surfaces can be modeled; each surface is labeled with a material code which is an index into a data base of electromagnetic properties. The inputs to the program include the radar processing parameters, the target orientation, the sensor velocity, and (for inverse SAR) the target angle rates. The current version of RIG can be run on any workstation, however, it is not a real-time model. We are considering several approaches to enable the program to generate realtime radar imagery. In addition to its image generation function, RIG can also generate radar cross-section (RCS) plots as well as range and doppler radar return profiles.

Stenger, Anthony J.↗
DOE OSTI2020

Multiple-objective Reinforcement Learning for Inverse Design and Identification

The aim of inverse chemical design is to develop new molecules with given optimized molecular properties or objectives. Recently, generative deep learning (DL) networks are considered as the state-of-the-art in inverse chemical design and have achieved early success in generating molecular structures with desired properties in the pharmaceutical and material chemistry fields. However, satisfying a large number (> 10 objectives) of molecular objectives is a limitation of current generative models. To improve the model’s ability to handle a large number of molecule design objectives, we developed a Reinforcement Learning (RL) based generative framework to optimize chemical molecule generation. Our use of Curriculum Learning (CL) to fine-tune the pre-trained generative network allowed the model to satisfy up to 21 objectives and increase the generative network’s robustness. The experiments show that the proposed multiple-objective RL-based generative model can correctly identify unknown molecules with an 83% to 100% success rate, compared to the baseline approach of 0%. Additionally, this proposed generative model is not limited to just chemistry research challenges; we anticipate that problems that utilize RL with multiple objectives will benefit from this framework.

reinforcement learning, generative model, inverse ↗
NASA NTRS1994

Quantum statistics of Raman scattering model with Stokes mode generation

The model describing three coupled quantum oscillators with decay of Rayleigh mode into the Stokes and vibration (phonon) modes is examined. Due to the Manley-Rowe relations the problem of exact eigenvalues and eigenstates is reduced to the calculation of new orthogonal polynomials defined both by the difference and differential equations. The quantum statistical properties are examined in the case when initially: the Stokes mode is in the vacuum state; the Rayleigh mode is in the number state; and the vibration mode is in the number of or squeezed states. The collapses and revivals are obtained for different initial conditions as well as the change in time the sub-Poisson distribution by the super-Poisson distribution and vice versa.

Tanatar, Bilal↗
DOE OSTI2025

Predicting critical heat flux with uncertainty quantification and domain generalization using conditional variational autoencoders and deep neural networks

Deep generative models (DGMs) can generate synthetic data samples that closely resemble the original dataset, addressing data scarcity. In this work, we developed a conditional variational autoencoder (CVAE) to augment critical heat flux (CHF) data used for the 2006 Groeneveld lookup table. To compare with traditional methods, a fine-tuned deep neural network (DNN) regression model was evaluated on the same dataset. Both models achieved small mean absolute relative errors, with the CVAE showing more favorable results. Uncertainty quantification (UQ) was performed using repeated CVAE sampling and DNN ensembling. The DNN ensemble improved performance over the baseline, while the CVAE maintained consistent results with less variability and higher confidence. Both models achieved small errors inside and outside the training domain, with slightly larger errors outside. Altogether, the CVAE performed better than the DNN in predicting CHF and exhibited better uncertainty behavior.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2026

Artificial intelligence-driven approaches for materials design and discovery

Materials design is an important component of modern science and technology, yet traditional approaches rely heavily on trial and error and can be inefficient. Computational techniques, enhanced by modern artificial intelligence, have reshaped the landscape of designing new materials. Among these approaches, inverse design has shown great promise in designing materials that meet specific property requirements. Here, in this Review, we present key computational advances in materials design over the past few decades. We follow the evolution of relevant materials design techniques, from high-throughput forward machine learning methods and evolutionary algorithms, to advanced artificial intelligence strategies such as reinforcement learning and deep generative models. We highlight the paradigm shift from conventional screening approaches to inverse generation driven by deep generative models. Finally, we discuss current challenges and future perspectives of materials inverse design. This Review may serve as a brief guide to the approaches, progress and outlook of designing future functional materials with technological relevance.

computational methods↗
DOE OSTI2026

Generative Thermodynamic Computing

Here, we introduce a generative modeling framework for thermodynamic computing, in which structured data are synthesized from noise by the natural time evolution of a physical system governed by Langevin dynamics. While conventional diffusion models use neural networks to perform denoising, here the information needed to generate structure from noise is encoded by the dynamics of a thermodynamic system. Training proceeds by maximizing the probability with which the computer generates the reverse of a noising trajectory, which ensures that the computer generates data with minimal heat emission. We demonstrate this framework within a digital simulation of a thermodynamic computer. If realized in analog hardware, such a system would function as a generative model that produces structured samples without the need for artificially injected noise or active control of denoising.

Whitelam, Stephen [Lawrence Berkeley National Labo↗
NASA NTRS1979

Alternatives for jet engine control

The research is classified in two categories: (1) the use of modern multivariable frequency domain methods for control of engine models in the neighborhood of a set-point, and (2) the use of nonlinear modelling and optimization techniques for control of engine models over a more extensive part of the flight envelope. Progress in the first category included the extension of CARDIAD (Complex Acceptability Region for Diagonal Dominance) methods developed with the help of the grant to the case of engine models with four inputs and four outputs. A suitable bounding procedure for the dominance function was determined. Progress in the second category had its principal focus on automatic nonlinear model generation. Simulations of models produced satisfactory results where compared with the NASA DYNGEN digital engine deck.

Sain, M. K.↗
NASA NTRS1995

Is walking a random walk? Evidence for long-range correlations in stride interval of human gait

Complex fluctuation of unknown origin appear in the normal gait pattern. These fluctuations might be described as being (1) uncorrelated white noise, (2) short-range correlations, or (3) long-range correlations with power-law scaling. To test these possibilities, the stride interval of 10 healthy young men was measured as they walked for 9 min at their usual rate. From these time series we calculated scaling indexes by using a modified random walk analysis and power spectral analysis. Both indexes indicated the presence of long-range self-similar correlations extending over hundreds of steps; the stride interval at any time depended on the stride interval at remote previous times, and this dependence decayed in a scale-free (fractallike) power-law fashion. These scaling indexes were significantly different from those obtained after random shuffling of the original time series, indicating the importance of the sequential ordering of the stride interval. We demonstrate that conventional models of gait generation fail to reproduce the observed scaling behavior and introduce a new type of central pattern generator model that sucessfully accounts for the experimentally observed long-range correlations.

Hausdorff, Jeffrey M.↗
NASA NTRS1995

On-line estimation of error covariance parameters for atmospheric data assimilation

A simple scheme is presented for on-line estimation of covariance parameters in statistical data assimilation systems. The scheme is based on a maximum-likelihood approach in which estimates are produced on the basis of a single batch of simultaneous observations. Simple-sample covariance estimation is reasonable as long as the number of available observations exceeds the number of tunable parameters by two or three orders of magnitude. Not much is known at present about model error associated with actual forecast systems. Our scheme can be used to estimate some important statistical model error parameters such as regionally averaged variances or characteristic correlation length scales. The advantage of the single-sample approach is that it does not rely on any assumptions about the temporal behavior of the covariance parameters: time-dependent parameter estimates can be continuously adjusted on the basis of current observations. This is of practical importance since it is likely to be the case that both model error and observation error strongly depend on the actual state of the atmosphere. The single-sample estimation scheme can be incorporated into any four-dimensional statistical data assimilation system that involves explicit calculation of forecast error covariances, including optimal interpolation (OI) and the simplified Kalman filter (SKF). The computational cost of the scheme is high but not prohibitive; on-line estimation of one or two covariance parameters in each analysis box of an operational bozed-OI system is currently feasible. A number of numerical experiments performed with an adaptive SKF and an adaptive version of OI, using a linear two-dimensional shallow-water model and artificially generated model error are described. The performance of the nonadaptive versions of these methods turns out to depend rather strongly on correct specification of model error parameters. These parameters are estimated under a variety of conditions, including uniformly distributed model error and time-dependent model error statistics.

Dee, Dick P.↗
NASA NTRS2011

Range 7 Scanner Integration with PaR Robot Scanning System

An interface bracket and coordinate transformation matrices were designed to allow the Range 7 scanner to be mounted on the PaR Robot detector arm for scanning the heat shield or other object placed in the test cell. A process was designed for using Rapid Form XOR to stitch data from multiple scans together to provide an accurate 3D model of the object scanned. An accurate model was required for the design and verification of an existing heat shield. The large physical size and complex shape of the heat shield does not allow for direct measurement of certain features in relation to other features. Any imaging devices capable of imaging the entire heat shield in its entirety suffers a reduced resolution and cannot image sections that are blocked from view. Prior methods involved tools such as commercial measurement arms, taking images with cameras, then performing manual measurements. These prior methods were tedious and could not provide a 3D model of the object being scanned, and were typically limited to a few tens of measurement points at prominent locations. Integration of the scanner with the robot allows for large complex objects to be scanned at high resolution, and for 3D Computer Aided Design (CAD) models to be generated for verification of items to the original design, and to generate models of previously undocumented items. The main components are the mounting bracket for the scanner to the robot and the coordinate transformation matrices used for stitching the scanner data into a 3D model. The steps involve mounting the interface bracket to the robot's detector arm, mounting the scanner to the bracket, and then scanning sections of the object and recording the location of the tool tip (in this case the center of the scanner's focal point). A novel feature is the ability to stitch images together by coordinates instead of requiring each scan data set to have overlapping identifiable features. This setup allows models of complex objects to be developed even if the object is large and featureless, or has sections that don't have visibility to other parts of the object for use as a reference. In addition, millions of points can be used for creation of an accurate model [i.e. within 0.03 in. (=0.8 mm) over a span of 250 in. (=635 mm)].

Schuler, Jason↗
NASA NTRS2005

Measuring Low-Order Aberrations in a Segmented Telescope

The in-focus PSF optimizer (IPO) is an algorithm for use in monitoring and controlling the alignment of the segments of a segmented-mirror astronomical telescope. IPO is so named because it computes wave-front aberrations of the telescope from digitized pointspread functions (PSFs) measured in infocus images. Inasmuch as distant astronomical objects that behave optically as point sources can typically be seen in almost any astronomical image, the main benefit afforded by IPO may be to enable maintenance of mirror-segment alignments without detracting from valuable scientific-observation time. IPO evolved from prescription-retrieval type algorithms. Prescription retrieval uses in-focus and out-of-focus PSFs to infer the state of an imaging optical system. The state, in this context, refers to the positions, orientations, and low-order figure errors of the optical elements in the system. Both prescription- retrieval and IPO use an iterative, nonlinear, least-squares optimizer to compute the optimal state parameters such that a digital computer-generated model image matches the digitized image acquired from the real system. The difference between IPO and prescription- retrieval algorithms is that IPO is specifically designed to utilize infocus images only. Although the restriction to in-focus images limits IPO to calculating only the lowest-order wave front aberrations, it also causes the resulting computation to take much less time because fewer degrees of freedom are included in the optimization process. In the prescription retrieval software developed at JPL, the model images are generated using the ray-trace/physical optics program, MACOS. IPO, on the other hand, uses a linear sensitivity matrix to compute the exit-pupil wave front from the system parameters; the wave front is then converted into a complex pupil field, which is then propagated to the image plane via a fast Fourier transform. This approach is computationally faster and requires less computer memory than is needed for prescription retrieval.

Ohara, Catherine↗
NASA NTRS2015

Relative Impacts of Mitigation, Temperature, and Precipitation on 21st-Century Megadrought Risk in the American Southwest

Megadroughts are comparable in severity to the worst droughts of the 20th century but are of much longer duration. A megadrought in the American Southwest would impose unprecedented stress on the limited water resources of the area, making it critical to evaluate future risks not only under different climate change mitigation scenarios but also for different aspects of regional hydroclimate. We find that changes in the mean hydroclimate state, rather than its variability, determine megadrought risk in the American Southwest. Estimates of megadrought probabilities based on precipitation alone tend to underestimate risk. Furthermore, business-as-usual emissions of greenhouse gases will drive regional warming and drying, regardless of large precipitation uncertainties. We find that regional temperature increases alone push megadrought risk above 70, 90, or 99% by the end of the century, even if precipitation increases moderately, does not change, or decreases, respectively. Although each possibility is supported by some climate model simulations, the latter is the most common outcome for the American Southwest in Coupled Model Intercomparison 5 generation models. An aggressive reduction in global greenhouse gas emissions cuts megadrought risks nearly in half.

risk↗
NASA NTRS2021

Optimized MODIS Channel Selection to Improve the NASA CERES FluxByCloudType Product Fluxes

The NASA Clouds and the Earth’s Radiant Energy System (CERES) project has provided the climate community more than a 20-year record of top of the atmosphere (TOA) reflected shortwave (SW) and emitted longwave (LW) fluxes. The fluxes used to monitor the Earth’s energy balance as well as for climate model validation and cloud feedback studies. One of the largest uncertainties in climate models is the response of clouds in feedback studies. To reduce this uncertainty requires more stringently validation of model generated fluxes by cloud-type. Rather than relying on radiative transfer model generated fluxes from observed cloud-type retrievals, the CERES FluxByCloudType (FBCT) product, relies on MODIS empirical narrowband to broadband relationships to convert the cloudy portions of the CERES footprints into fluxes. The overcast footprint and sub-footprint cloud layer fluxes are then stratified by 7 pressure layers and 6 optical depth bins and temporally averaged into daily and monthly regional cloud-type fluxes. The CERES project analyzed a total of 19 MODIS channels for use in the next generation FBCT product narrowband to broadband conversion. This was accomplished by comparing the narrowband to broadband RMS errors from all possible 5-channel combinations. The SW and LW optimized channels were selected by also considering analogous geostationary, MODIS and VIIRS imager channels as well as channel combinations that provided the lowest RMS errors across all surface types. The optimized channel combinations will be tested in the FBCT algorithm to investigate any cloud fraction, effective pressure, optical depth, PW, solar and view angle, surface type and regional narrowband to broadband dependencies.

CERES↗