Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “gradient methods”

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 199 records · Page 11
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Improving Deep Neural Networks’ Training for Image Classification With Nonlinear Conjugate Gradient-Style Adaptive Momentum

Momentum is crucial in stochastic gradient-based optimization algorithms for accelerating or improving training deep neural networks (DNNs). In deep learning practice, the momentum is usually weighted by a well-calibrated constant. However, tuning the hyperparameter for momentum can be a significant computational burden. In this article, we propose a novel adaptive momentum for improving DNNs training; this adaptive momentum, with no momentum-related hyperparame- ter required, is motivated by the nonlinear conjugate gradient (NCG) method. Stochastic gradient descent (SGD) with this new adaptive momentum eliminates the need for the momentum hyperparameter calibration, allows using a significantly larger learning rate, accelerates DNN training, and improves the final accuracy and robustness of the trained DNNs. For example, SGD with this adaptive momentum reduces classification errors for training ResNet110 for CIFAR10 and CIFAR100 from 5.25% to 4.64% and 23.75% to 20.03%, respectively. Furthermore, SGD, with the new adaptive momentum, also benefits adversarial training and, hence, improves the adversarial robustness of the trained DNNs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

ARC-DPA 1-MeV Neutron Fluence Equivalence Model for GaAs

Here, the legacy data used in the production of the American Society for Testing and Materials (ASTM) E722 standard for gallium arsenide (GaAs) were fit with the athermal recombination corrected-displacements per atom (ARC-DPA) model for the development of a novel 1-MeV neutron equivalent fluence metric. Improving on previous work, the coefficients for the ARC-DPA model were optimized using a conjugate gradient method, and the uncertainties in the observed damage in different benchmark neutron fields were sampled to quantify their contribution to the damage metric. The optimized parameters of −0.474 ± 0.03 for b arcdpa and 0.016 ± 0.002 for c arcdpa provided good agreement and indicated that the solution is not sensitive to the corresponding fluence uncertainties. Compared to the current 1-MeV neutron fluence equivalence standard for displacement damage in GaAs, our results indicate that for a thermal reactor neutron environment, displacement damage has been underestimated by around 25% and that the saturation value for the damage efficiency had been higher than previously modeled.

1-MeV equivalent fluence↗
DOE OSTI2024

From Sim to Real: A Pipeline for Training and Deploying Traffic Smoothing Cruise Controllers

Designing and validating controllers for connected and automated vehicles to enhance traffic flow presents significant challenges, from the complexity of replicating real-world stop-and-go traffic dynamics in simulation, to the intricacies involved in transitioning from simulation to actual deployment. In this work, we present a full pipeline from data collection to controller deployment. Specifically, we collect 772 km of driving data from the I-24 in Tennessee, and use it to build a one-lane simulator, placing simulated vehicles behind real-world trajectories. Using policy-gradient methods with an asymmetric critic, we improve fuel efficiency by over 10% when simulating congested scenarios. Our comprehensive approach includes reinforcement learning for controller training, software verification, hardware validation and setup, and navigating various sim-to-real challenges. Furthermore, we analyze the controller's behavior and wave-smoothing properties, and deploy it on four Toyota Rav4’s in a real-world validation experiment on the I-24. Lastly, we release the driving dataset, the simulator and the trained controller, to enable future benchmarking and controller design.

42 ENGINEERING↗
DOE OSTI2025

Fast and Scalable FFT-Based GPU-Accelerated Algorithms for Block-Triangular Toeplitz Matrices with Application to Linear Inverse Problems Governed by Autonomous Dynamical Systems

In this work, we present an efficient and scalable algorithm for performing matrix-vector multiplications (matvecs) for block Toeplitz matrices. Such matrices, which are shift-invariant with respect to their blocks, arise in the context of solving inverse problems governed by autonomous systems, and time-invariant systems in particular. In this article, we consider inverse problems that infer unknown parameters from observational data of a linear time-invariant dynamical system given in the form of partial differential equations (PDEs). Matrix-free Newton-conjugate-gradient methods are often the gold standard for solving these inverse problems, but they require numerous actions of the Hessian on a vector. Matrix-free adjoint-based Hessian matvecs require solution of a pair of linearized forward/adjoint PDE solves per Hessian action, which may be prohibitive for large-scale inverse problems. Time invariance of the forward PDE problem leads to a block Toeplitz structure of the discretized parameter-to-observable (p2o) map defining the mapping from inputs (parameters) to outputs (observables) of the PDEs. This block Toeplitz structure enables us to exploit two key properties: (1) compact storage of the p2o map and its adjoint, and (2) efficient fast Fourier transform–based Hessian matvecs. The proposed algorithm is mapped onto large multi-GPU clusters and achieves more than 80% of peak bandwidth on NVIDIA A100 GPUs. Excellent weak scaling is shown for up to 48 A100 GPUs. For the targeted problems, the implementation executes Hessian matvecs within fractions of a second, which is orders of magnitude faster than can be achieved by conventional matrix-free Hessian matvecs via forward/adjoint PDE solves.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Grid-Interactive Multi-Zone Building Control Using Reinforcement Learning with Global-Local Policy Search

In this paper, we develop a grid-interactive multi-zone building controller based on a deep reinforcement learning (RL) approach. The controller is designed to facilitate building operation during normal conditions and demand response events, while ensuring occupants comfort and energy efficiency. We leverage a continuous action space RL formulation, and devise a two-stage global-local RL training framework. In the first stage, a global fast policy search is performed using a gradient-free RL algorithm. In the second stage, a local fine-tuning is conducted using a policy gradient method. In contrast to the state-of-the-art model predictive control (MPC) approach, the proposed RL controller does not require complex computation during real-time operation and can adapt to nonlinear building models. We illustrate the controller performance numerically using a five-zone commercial building.

30 DIRECT ENERGY CONVERSION↗
NASA NTRS1972

Conjugate gradient optimization programs for shuttle reentry

Two computer programs for shuttle reentry trajectory optimization are listed and described. Both programs use the conjugate gradient method as the optimization procedure. The Phase 1 Program is developed in cartesian coordinates for a rotating spherical earth, and crossrange, downrange, maximum deceleration, total heating, and terminal speed, altitude, and flight path angle are included in the performance index. The programs make extensive use of subroutines so that they may be easily adapted to other atmospheric trajectory optimization problems.

Powers, W. F.↗
NASA NTRS1972

A discrete variable approximation to minimum weight panel designs subject to a supersonic flutter-speed constraint.

A numerical technique is presented that determines the minimum-weight thickness distribution for a one-dimensional, simply supported panel subjected to a parallel supersonic flow on one side. An aeroelastic eigenvalue, which characterizes flutter speed, is held constant. The governing differential equations are approximated by sets of difference equations adjoined to the weight function via a penalty function. A conjugate gradient method then solves the resulting sequence of unconstrained minimization problems. Numerical results are obtained without imposing a minimum thickness constraint; the effect of constant inplane tensile and compressive stresses is also investigated.

Pierson, B. L.↗
NASA NTRS1972

Iterative explicit guidance for low thrust spacecraft.

A retargeting procedure is developed for use as a nonlinear low thrust guidance scheme. The selection of a control program composed of a sequence of inertially fixed thrust-acceleration vectors permits all trajectory computations to be made with closed form expressions, and allows the controls to be represented by constant parameters, thrust-acceleration vectors and thrusting times. By requiring each trajectory to be time optimal, the guidance problem is transformed into a parameter optimization problem which is solved by the conjugate gradient method. The scheme is applied to a low thrust capture mission, and the results of computer simulations are presented.

Jacobson, R. A.↗
NASA NTRS1973

FORTRAN program for calculating velocities and streamlines on the hub-shroud mid-channel flow surface of an axial- or mixed-flow turbomachine. 1: User's manual

A FORTRAN 4 computer program has been developed that obtains a subsonic or shock-free transonic flow solution on the hub-shroud mid-channel flow surface of a turbomachine. The blade row may be fixed or rotating, and may be twisted and leaned. Flow may be axial or mixed, up to 45 deg from axial. Upstream and downstream flow variables may vary from hub to shroud, and provision is made to correct for loss of stagnation pressure. The results include velocities, streamlines, and flow angles on the flow surface; and approximate blade surface velocities. Subsonic solutions are obtained by a finite-difference stream-function solution. Transonic solutions are obtained by a velocity-gradient method, using information from a finite-difference stream-function solution at a reduced mass flow.

Katsanis, T.↗
NASA NTRS1973

Parabolic cylinder antennas

Some of the features of single and dual parabolic-cylinder reflector antenna systems are discussed in terms of wave conversions, field divergence, and wavefronts in Fresnel and Fraunhofer regions. Beam-squinting, by mechanical displacement and electrical phase gradient methods, is introduced together with the combination of these methods. In the case of dual parabolic cylinders there is also a discussion of surface-truncation, parametric representation of surface intersections, main-aperture blockage by the subsystem, and beam squinting. A few diffraction patterns are presented to illustrate the type of data available via a formulation equivalent to the complex-vector Kirchhoff-Kottler formulation. Main and cross-polarization components of the solution are available as they are inherently part of the field solution in the Cartesian coordinate system. A means of displaying main and cross-polarization components for arbitrary beam-scanning in space is outlined. Extensions to monopulse tracking with variable-beamwidth (zooming) capability are introduced.

Schmidt, R. F.↗
NASA NTRS1974

FORTRAN program for calculating velocities and streamlines on the hub-shroud mid-channel flow surface of an axial-or mixed-flow turbomachine. 2: Programmer's manual

A FORTRAN-IV computer program, MERIDL, has been developed that obtains a subsonic or shock-free transonic flow solution on the hub-shroud mid-channel flow surface of a turbomachine. The blade row may be fixed or rotating and may be twisted and leaned. Flow may be axial or mixed, up to 45 deg from axial. Upstream and downstream flow variables can vary from hub to shroud, and provision is made to correct for loss of stagnation pressure. The results include velocities, streamlines, and flow angles on the flow surface and approximate blade surface velocities. Subsonic solutions are obtained by a finite-difference stream-function solution. Transonic solutions are obtained by a velocity-gradient method, using information from a finite-difference stream-function solution at a reduced mass flow.

Katsanis, T.↗
NASA NTRS1977

Revised FORTRAN program for calculating velocities and streamlines on the hub-shroud midchannel stream surface of an axial-, radial-, or mixed-flow turbomachine or annular duct. 1: User's manual

A FORTRAN 4 computer program was developed that obtains a detailed subsonic or shock-free transonic flow solution on the hub-shroud midchannel stream surface of a turbomachine. The blade row may be fixed or rotating, and the blades may be twisted and leaned. Flow may be axial, mixed, or radial. Upstream and downstream flow variables may vary from hub to shroud, and provision is made to correct for loss of stagnation pressure. The results include velocities, streamlines, and flow angles on the stream surface as well as approximate blade surface velocities. Subsonic solutions are obtained by a finite-difference, stream-function solution. Transonic solutions are obtained by a velocity-gradient method that uses information from a finite-difference, stream-function solution at a reduced mass flow.

Katsanis, T.↗
NASA NTRS1978

Flow velocities and streamlines

Computer program calculates subsonic or transonic flow on hubshroud, midchannel, stream surface of single-blade row of turbomachine. Program uses finite-different and quasi-orthogonal (velocity-gradient) methods. Program is reported in two volumes: Part I is User's Manual, Part II is Programmer's Manual.

Katsanis, T.↗
NASA NTRS1977

NASCAP user's manual

The NASCAP (NASA Charging Analyzer Program) code simulates the charging process for a complex object in either tenuous plasma or ground test environment. Detailed specifications needed to run the code are presented. The object definition section, OBJDEF, allows the test object to be easily defined in the cubic mesh. The test object is composed of conducting sections which may be wholly or partially covered with thin dielectric coatings. The potential section, POTENT, obtains the electrostatic potential in the space surrounding the object. It uses the conjugate gradient method to solve the finite element formulation of Poisson's equation. The CHARGE section of NASCAP treats charge redistribution among the surface cells of the object as well as charging through radiation bombardment. NASCAP has facilities for extensive graphical output, including several types of object display plots, potential contour plots, space charge density contour plots, current density plots, and particle trajectory plots.

Mandell, M. J.↗
NASA NTRS1977

General purpose computer program for interacting supersonic configurations. User's manual

The input data required to execute the computer program ISCON are described. The program generates a numerical procedure for the determination of unsteady aerodynamic forces on arbitrarily interacting wings and tails in supersonic flow. A velocity potential gradient method is used. Constant Mach number is assumed throughout the flow field. Lifting surfaces are represented by trapezoidal elements which can be generated automatically by the program. The wake field is represented by rectangular strip elements. The formulation is reviewed as well as input overview and input format. Instruction on how to use ISCON, a sample problem, and the restart feature are discussed. Program size limitations, computer program flow, and error messages are also included along with a description of the SS31 program used to compute the coefficients of surface spline.

Crill, W.↗
NASA NTRS1981

Driving bubbles out of glass

Surface tension gradient in melt forces gas bubbles to surface, increasing glass strength and transparency. Conventional chemical and buoyant fining are extremely slow in viscous glasses, but tension gradient method moves 250 um bubbles as rapidly as 30 um/s. Heat required for high temperature part of melt is furnished by stationary electrical or natural-gas heater; induction and laser heating are also possible. Method has many applications in industry processes.

Mattox, D. M.↗