SEARCH · Engineering Papers

Results for “Dimensionality Reduction”

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.

SourceResearch connectionFrom yearThrough year

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

At least 1,711 records · Page 95

All sources NASA NTRS DOE OSTI

DOE OSTI2023

An invertible seven-dimensional Dirichlet cell characterization of lattices

Characterization of crystallographic lattices is an important tool in structure solution, crystallographic database searches and clustering of diffraction images in serial crystallography. Characterization of lattices by Niggli-reduced cells (based on the three shortest non-coplanar lattice vectors) or by Delaunay-reduced cells (based on four non-coplanar vectors summing to zero and all meeting at obtuse or right angles) is commonly performed. The Niggli cell derives from Minkowski reduction. The Delaunay cell derives from Selling reduction. All are related to the Wigner–Seitz (or Dirichlet, or Voronoi) cell of the lattice, which consists of the points at least as close to a chosen lattice point as they are to any other lattice point. The three non-coplanar lattice vectors chosen are here called the Niggli-reduced cell edges. Starting from a Niggli-reduced cell, the Dirichlet cell is characterized by the planes determined by 13 lattice half-edges: the midpoints of the three Niggli cell edges, the six Niggli cell face-diagonals and the four body-diagonals, but seven of the lengths are sufficient: three edge lengths, the three shorter of each pair of face-diagonal lengths, and the shortest body-diagonal length. These seven are sufficient to recover the Niggli-reduced cell.

Chemistry↗

DOE OSTI2015

Properties of three-dimensional structures prepared by Ge dewetting from Si(111) at high temperatures

The formation of three-dimensional (3D) structures during Ge deposition on Si(111) at about 800 °C is studied with scanning tunneling, Kelvin probe and electron microscopies, and scanning tunneling and Raman spectroscopies. The observed surface morphology is formed by dewetting of Ge from Si(111), since it occurs mainly by means of minimization of surface and interfacial energies. The dewetting proceeds through massive Si eroding around growing 3D structures, providing them to be composed of SiGe with about a 30% Ge content, and leads to the significant reduction of the SiGe/Si interface area. It is found that the SiGe top component of 3D structures forms sharp interfaces with the underlying Si. The minimization of interfacial and strain energies occurs on the way that the 3D structures appear to get the dendrite-like shape. The Ge distribution in the 3D SiGe structures is inhomogeneous in the lateral dimension with a higher Ge concentration in their central areas and Ge segregation on their surface.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

DOE OSTI2025

Emulation of radiation transport in 3D stochastic media using 1D planar Monte Carlo stochastic media radiation transport algorithms

A subset of stochastic media radiation transport problems involves those in which radiation is incident on a thin slab of stochastic material. Particle tracking in 3D for such problems is expensive, and 1D planar models lack accuracy because they only allow the material to change in one dimension. Therefore, we propose dimensional emulation, which through a slight modification allows existing 1D planar geometry stochastic media radiation transport models to reproduce results from the equivalent 3D models by allowing the material to change in all three dimensions, reproducing the fidelity of the 3D model for the low computational cost of the 1D planar model. In this work, we apply dimensional emulation to three Monte Carlo stochastic media radiation transport models: Chord Length Sampling (CLS), the Local Realization Preserving method (LRP), and a variant of Conditional Point Sampling (CoPS). For a common Markovian benchmark set, the 3D emulation variants of these algorithms are numerically verified to reproduce the results of the 3D variants within statistics while running 1.3 to 2 times faster in the implementation within Sandia National Laboratories open-source research code PlaybookMC. The 3D emulation variants are also shown to yield a 72%–92% reduction in error for the thin slab problems in comparison to the 1D benchmark. As a result, the 3D emulation variant of CLS and CoPS-1 are shown to reproduce 3D CLS results that were used to approximate results for a 3D spherical inclusion geometry benchmark set.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

DOE OSTI2021

Water and Solute Activities Regulate CO 2 Reduction in Gas-Diffusion Electrodes

Electrolysis of CO 2 at gas-diffusion electrodes (GDEs) has typically been limited by the supply of gas to the electrocatalyst, overshadowing the importance of the supply of water. However, at high current densities that approach 1 A cm –2 , where the electrolyte becomes highly concentrated in the catalyst layer of a GDE, the activity of water and solutes deviate from their bulk dilute solution values, potentially slowing reaction rates and changing reaction equilibrium potentials. In addition, as flow plates for the gas stream are introduced to enable larger electrodes and high single pass conversion of CO 2 to product, variations in the gas composition will become important. By drawing upon literature for the oxygen reduction reaction (ORR), here we explain how to account for these effects in future modeling and experimental work, with particular attention to accurate use of the Nernst equation for electrode potentials and the Arrhenius equation for reaction rates. Specifically, using measurements of KOH solvent and solute activity reported in literature, and assuming the second protonation of CO 2 by water as the rate-determining step, we show the Nernst equation dilute-solution approximation of the CO 2 to CO equilibrium potential to be accurate below 5 M KOH, but it has a 74 mV error when increasing the concentration up to 10 M KOH. Finally, a simple one-dimensional model of a serpentine flow-field on a GDE demonstrated that a reactor with constant pressure of 1 bar and 1 A cm –2 at the inlet had only ~0.3 A cm –2 at the outlet for a conversion in CO 2 partial pressure from 0.90 to 0.48 bar, showing the significant practical implications of this work.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2011

Isotoxic Dose Escalation in the Treatment of Lung Cancer by Means of Heterogeneous Dose Distributions in the Presence of Respiratory Motion

Purpose: To test, in the presence of intrafractional respiration movement, a margin recipe valid for a homogeneous and conformal dose distribution and to test whether the use of smaller margins combined with heterogeneous dose distributions allows an isotoxic dose escalation when respiratory motion is considered. Methods and Materials: Twenty-three Stage II-III non-small-cell lung cancer patients underwent four-dimensional computed tomography scanning. The gross tumor volume and clinical target volume (CTV) were outlined in the mid-ventilation phase. The CTV-to-planning target volume (PTV) margin was calculated by use of a standard margin recipe and the patient-specific respiration pattern. Standard three-dimensional treatment plans were generated and recalculated on the remaining respiration phases. The planning was repeated for a CTV-to-PTV margin decreased by 2.5 and 5 mm relative to the initial margin in all directions. Time-averaged dose-volume histograms (four-dimensional dose-volume histograms) were calculated to evaluate the CTV-to-PTV margin. Finally, the dose was escalated in the plans with decreased PTV such that the mean lung dose (predictor of radiation-induced pneumonitis) was equal to mean lung dose in the plan by use of the initially calculated margin. Results: A reduction of the standard margin by 2.5 mm compared with the recipe resulted in too low of a minimum dose for some patients. A combination of dose escalation and use of heterogeneous dose distribution was able to increase the minimum dose to the target by approximately 10% and 20% for a CTV-to-PTV margin reduction of 2.5 mm and 5.0 mm, respectively. Conclusion: The margin recipe is valid for intrafractional respiration-induced tumor motions. It is possible to increase the dose to the target without increased mean lung dose with an inhomogeneous dose distribution.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

DOE OSTI2016

Asynchronous In Situ Connected-Components Analysis for Complex Fluid flows, In: ISAV '16 Proceedings of the 2nd Workshop on In Situ Infrastructures for Enabling Extreme-scale Analysis and Visualization

The simulation of multiscale physics is an important challenge for scientific computing. For this class of problem, large three-dimensional simulations are performed to advance scientific inquiry. On massively parallel computing systems, the volume of data generated by such approaches can become a productivity bottleneck if the raw data generated from the simulation is analyzed in a post-processing step. To address this, we present a physics-based framework for in situ data reduction that is theoretically grounded in multiscale averaging theory. We show how task parallelism can be exploited to concurrently perform a variety of analysis tasks with data-dependent costs, including the generation of iso-surfaces, morphological analyses, and connected components analysis. All analyses are performed in parallel using distributed memory and use the same domain decomposition as the simulation. A task management framework is constructed to leverage available parallelism within a node for analysis. The capabilities of the framework are to launch asynchronous analysis threads, manage dependencies between different tasks, promote data locality and minimize the impact of data transfers. The framework is applied to analyze GPU-based simulations of two-fluid-phase flow in porous media, generating a set of averaged measures that represents the overall system behavior. We demonstrate how the approach can be applied to perform physically-consistent analysis over fluid sub-regions determined from connected components analysis. Simulations performed on Oak Ridge National Lab's Titan supercomputer are profiled to demonstrate the performance of the associated multi-threaded in situ analysis approach for typical production simulation of two-fluid-phase flow.

Computer Science↗

DOE OSTI2020

Reduced Models for Chemical Kinetics derived from Parallel Ensemble Simulations of Stirred Reactors

Detailed chemical kinetic models are essential to improve the predictive capability of computational fluid dynamics (CFD) simulations and to impact the development of next generation combustion technologies. However, such detailed kinetic models are beyond the reach of CFD codes due to the computational cost of transporting additional chemical degrees of freedom. Reduced order models for combustion kinetics exploit the presence of lower dimensional manifolds in the state space and require far fewer number of scalars to be transported as part of CFD simulations. In this paper, we describe the software infrastructure developed to allow large ensembles of stirred reactor configurations with detailed chemical kinetics. The canonical reactor simulations are incorporated in a workflow with sparse grid and data reduction methodologies aimed at development of reduced order models. We present results from analyzing the results of the stirred reactor calculations for syngas mixture using principal component analysis (PCA) and neural network methodologies. The results show that nonlinear autoencoder model can surpass PCA in representing the original dataset with far fewer degrees of freedom.

Zhang, Pei↗

NASA NTRS1981

Comparative study of flare control laws

The development of a digital, three dimensional, automatic control law designed to achieve an optimal transition of a B-737 aircraft between glide slope conditions and the desired final touchdown condition is presented. The digital control law is a time invariant, state estimate feedback law, and the design is capable of using the microwave landing system. Major emphasis is placed on the reduction of aircraft noise in communities surroundings airports, the reduction of fuel consumption, the reduction of the effects of adverse weather conditions on aircraft operations, and the efficient use of airspace in congested terminal areas. Attention is also given to the development of the capability to perform automatic flares from steep glide slopes to precise touchdown locations.

Nadkarni, A. A.↗

NASA NTRS1983

Vibration characteristics of a deployable controllable-geometry truss boom

An analytical study was made to evaluate changes in the fundamental frequency of a two dimensional cantilevered truss boom at various stages of deployment. The truss could be axially deployed or retracted and undergo a variety of controlled geometry changes by shortening or lengthening the telescoping diagonal members in each bay. Both untapered and tapered versions of the truss boom were modeled and analyzed by using the finite element method. Large reductions in fundamental frequency occurred for both the untapered and tapered trusses when they were uniformly retracted or maneuvered laterally from their fully deployed position. These frequency reductions can be minimized, however, if truss geometries are selected which maintain cantilever root stiffness during truss maneuvers.

Dorsey, J. T.↗

NASA NTRS1986

Computer simulations of electromagnetic cool ion beam instabilities

Electromagnetic ion beam instabilities driven by cool ion beams at propagation parallel or antiparallel to a uniform magnetic field are studied using computer simulations. The elements of linear theory applicable to electromagnetic ion beam instabilities and the simulations derived from a one-dimensional hybrid computer code are described. The quasi-linear regime of the right-hand resonant ion beam instability, and the gyrophase bunching of the nonlinear regime of the right-hand resonant and nonresonant instabilities are examined. It is detected that in the quasi-linear regime the instability saturation is due to a reduction in the beam core relative drift speed and an increase in the perpendicular-to-parallel beam temperature; in the nonlinear regime the instabilities saturate when half the initial beam drift kinetic energy density is converted to fluctuating magnetic field energy density.

Gary, S. P.↗

NASA NTRS1989

An analytic study of nonsteady two-phase laminar boundary layer around an airfoil

Recently, NASA, FAA, and other organizations have focused their attention upon the possible effects of rain on airfoil performance. Rhode carried out early experiments and concluded that the rain impacting the aircraft increased the drag. Bergrum made numerical calculation for the rain effects on airfoils. Luers and Haines did an analytic investigation and found that heavy rain induces severe aerodynamic penalties including both a momentum penalty due to the impact of the rain and a drag and lift penalty due to rain roughening of the airfoil and fuselage. More recently, Hansman and Barsotti performed experiments and declared that performance degradation of an airfoil in heavy rain is due to the effective roughening of the surface by the water layer. Hansman and Craig did further experimental research at low Reynolds number. E. Dunham made a critical review for the potential influence of rain on airfoil performance. Dunham et al. carried out experiments for the transport type airfoil and concluded that there is a reduction of maximum lift capability with increase in drag. There is a scarcity of published literature in analytic research of two-phase boundary layer around an airfoil. Analytic research is being improved. The following assumptions are made: the fluid flow is non-steady, viscous, and incompressible; the airfoil is represented by a two-dimensional flat plate; and there is only a laminar boundary layer throughout the flow region. The boundary layer approximation is solved and discussed.

Hsu, Yu-Kao↗

DOE OSTI2017

An In-depth Review On the Scientific and Policy Issues Associated with Additive Manufacturing

Historically, manufacturing processes have been predominantly subtractive, i.e. three-dimensional objects were created by successively cutting material away from a solid block, by scraping, machining, turning or dissolving. Additive manufacturing (AM) or three-dimensional (3D) printing, in contrast, is controlled material addition, implemented by successively depositing layers of material until a predesigned shape is formed. AM represents an innovative manufacturing technology, and is set to transform production processes from design to manufacture, and to eventual distribution to end users, ultimately leading to an increase in energy efficiency and a reduction in gas emissions for future generation of industries. The unique capability of this technology to produce intricate geometries with customizable material properties has made it a widely interesting and welcome development among scientists, industry and the general public. Its wide acceptance has continued to make 3D printing technology more openly accessible, and low-cost desktop printing, with the capability to reproduce 3D objects from medical prostheses to weapons, is rapidly increasing in availability to the public. However, most research and media attention has been focused on the ingenuity of this ground-breaking technology and its wide range of possibilities. Very little consideration is being given to the adverse effects of the seemingly unstoppable advancement of AM technology and unrestricted access to 3D printing techniques. Also, proponents of AM technology rarely take into account the overall life-cycle cost and risks of failure of the manufactured part. Despite the promising prospects of this novel development, there are still concerns about how printed products will perform over time, the consistency of their quality, and the types and safety of materials used with this technology, especially with very large scale additive-manufactured products. This paper brings to light some scientific and policy risks and challenges concerning the material science and engineering aspects of these issues. Matters discussed include life-cycle cost analysis, end-product safety and quality assurance/control, regulation gaps, digital piracy, and resulting loopholes in safety and national security. The paper also presents potential options for curbing these risks, and otherwise adapting to the eventualities which, with certain inevitable factors in play, may lie beyond control.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

DOE OSTI2015

The impact of dust evolution and photoevaporation on disk dispersal

Protoplanetary disks are dispersed by viscous evolution and photoevaporation in a few million years; in the interim small, sub-micron-sized dust grains must grow and form planets. The time-varying abundance of small grains in an evolving disk directly affects gas heating by far-ultraviolet (FUV) photons, while dust evolution affects photoevaporation by changing the disk opacity and resulting penetration of FUV photons in the disk. Photoevaporative flows, in turn, selectively carry small dust grains, leaving the larger particles—which decouple, from the gas—behind in the disk. We study these effects by investigating the evolution of a disk subject to viscosity, photoevaporation by EUV, FUV, and X-rays, dust evolution, and radial drift using a one-dimensional (1D) multi-fluid approach (gas + different dust grain sizes) to solve for the evolving surface density distributions. The 1D evolution is augmented by 1+1D models constructed at each epoch to obtain the instantaneous disk structure and determine photoevaporation rates. The implementation of a dust coagulation/fragmentation model results in a marginal decrease in disk lifetimes when compared to models with no dust evolution; the disk lifetime is thus found to be relatively insensitive to the evolving dust opacity. We find that photoevaporation can cause significant reductions in the gas/dust mass ratio in the planet-forming regions of the disk as it evolves, and may result in a corresponding increase in heavy element abundances relative to hydrogen. We discuss implications for theories of planetesimal formation and giant planet formation, including the formation of gas-poor giants. After gas disk dispersal, ∼3×10{sup −4} M {sub ⊙} of mass in solids typically remain, comparable to the solids inventory of our solar system.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2012

Heavy ion beam probe advances from the first installation of the diagnostic on an RFP (invited)

Heavy ion beam probes have been installed on a variety of toroidal devices, but the first and only application on a reversed field pinch is the diagnostic on the Madison Symmetric Torus. Simultaneous measurements of spatially localized equilibrium potential and fluctuations of density and potential, previously inaccessible in the core of the reversed field pinch (RFP), are now attainable. These measurements reflect the unique strength of the heavy ion beam probe (HIBP) diagnostic. They will help determine the characteristics and evolution of electrostatic fluctuations and their role in transport, and determine the relation of the interior electric field and flows. Many aspects of the RFP present original challenges to HIBP operation and inference of plasma quantities. The magnetic field contributes to a number of the issues: the comparable magnitudes of the toroidal and poloidal fields and edge reversal result in highly three-dimensional beam trajectories; partial generation of the magnetic field by plasma current cause it and hence the beam trajectories to vary with time; and temporal topology and amplitude changes are common. Associated complications include strong ultraviolet radiation and elevated particle losses that can alter functionality of the electrostatic systems and generate noise on the detectors. These complexities have necessitated the development of new operation and data analysis techniques: the implementation of primary and secondary beamlines, adoption of alternative beam steering methods, development of higher precision electrostatic system models, refinement of trajectory calculations and sample volume modeling, establishment of stray particle and noise reduction methods, and formulation of alternative data analysis techniques. These innovative methods and the knowledge gained with this system are likely to translate to future HIBP operation on large scale stellarators and tokamaks.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

DOE OSTI2012

Facile synthesis of {alpha}-MnO{sub 2} one-dimensional (1D) nanostructure and energy storage ability studies

The dense manganese oxide nanorods with an extremely narrow distribution are synthesized at a low temperature using first cathodic electrodeposition subsequently heat treatment. Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) images show that the nanorods have bar shapes, and their average diameter is less than 50 nm. The Fourier transform infrared (FT-IR) study, the selected area electron diffraction (SAED) pattern in TEM images and the X-ray diffraction (XRD) result show that the nanorods are {alpha}-MnO{sub 2} single crystal. The results of N{sub 2} adsorption-desorption analysis indicate that the BET surface area of the {alpha}-MnO{sub 2} nanorods is 93 m{sup 2} g{sup -1}. By recording the potential-time curve during the electrodeposition process, it is revealed that water reduction reaction has a major role in the electrogeneration of base at the cathode surface under the applied electrochemical conditions. Finally, based on the H{sub 2} bubbling on the cathode surface, the mechanism of the formation and the growth of {alpha}-MnO{sub 2} nanorods are proposed and discussed. For the electrochemical supercapacitor application, electrochemically prepared {alpha}-MnO{sub 2} is found to be stable for a large number of cycles with high specific capacitance, 338 F g{sup -1} at a scan rate of 10 mV s{sup -1}. Finally, the charge-discharge mechanism is discussed. - Graphical abstract: Highlights: Black-Right-Pointing-Pointer New nanostructures of MnO{sub 2} is synthesized by simple method of cathodicelectrodeposition. Black-Right-Pointing-Pointer The product has unique one-dimensional morphology with average diameter size of 50 nm. Black-Right-Pointing-Pointer The experiment conditions (temperature, current density) has not been reported. Black-Right-Pointing-Pointer The one-nanostructures obtained without using of hard template or surfactant.

36 MATERIALS SCIENCE↗

DOE OSTI2015

Geometric Verification of Dynamic Wave Arc Delivery With the Vero System Using Orthogonal X-ray Fluoroscopic Imaging

Purpose: The purpose of this study was to define an independent verification method based on on-board orthogonal fluoroscopy to determine the geometric accuracy of synchronized gantry–ring (G/R) rotations during dynamic wave arc (DWA) delivery available on the Vero system. Methods and Materials: A verification method for DWA was developed to calculate O-ring-gantry (G/R) positional information from ball-bearing positions retrieved from fluoroscopic images of a cubic phantom acquired during DWA delivery. Different noncoplanar trajectories were generated in order to investigate the influence of path complexity on delivery accuracy. The G/R positions detected from the fluoroscopy images (DetPositions) were benchmarked against the G/R angulations retrieved from the control points (CP) of the DWA RT plan and the DWA log files recorded by the treatment console during DWA delivery (LogActed). The G/R rotational accuracy was quantified as the mean absolute deviation ± standard deviation. The maximum G/R absolute deviation was calculated as the maximum 3-dimensional distance between the CP and the closest DetPositions. Results: In the CP versus DetPositions comparison, an overall mean G/R deviation of 0.13°/0.16° ± 0.16°/0.16° was obtained, with a maximum G/R deviation of 0.6°/0.2°. For the LogActed versus DetPositions evaluation, the overall mean deviation was 0.08°/0.15° ± 0.10°/0.10° with a maximum G/R of 0.3°/0.4°. The largest decoupled deviations registered for gantry and ring were 0.6° and 0.4° respectively. No directional dependence was observed between clockwise and counterclockwise rotations. Doubling the dose resulted in a double number of detected points around each CP, and an angular deviation reduction in all cases. Conclusions: An independent geometric quality assurance approach was developed for DWA delivery verification and was successfully applied on diverse trajectories. Results showed that the Vero system is capable of following complex G/R trajectories with maximum deviations during DWA below 0.6°.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

DOE OSTI2011

Structure of an Odorant-Vinding Protein form the Mosquito Aedes aegypti Suggests a Binding Pocket Covered by a pH-Sensitive

The yellow fever mosquito, Aedes aegypti, is the primary vector for the viruses that cause yellow fever, mostly in tropical regions of Africa and in parts of South America, and human dengue, which infects 100 million people yearly in the tropics and subtropics. A better understanding of the structural biology of olfactory proteins may pave the way for the development of environmentally-friendly mosquito attractants and repellents, which may ultimately contribute to reduction of mosquito biting and disease transmission. Previously, we isolated and cloned a major, female-enriched odorant-binding protein (OBP) from the yellow fever mosquito, AaegOBP1, which was later inadvertently renamed AaegOBP39. We prepared recombinant samples of AaegOBP1 by using an expression system that allows proper formation of disulfide bridges and generates functional OBPs, which are indistinguishable from native OBPs. We crystallized AaegOBP1 and determined its three-dimensional structure at 1.85 {angstrom} resolution by molecular replacement based on the structure of the malaria mosquito OBP, AgamOBP1, the only mosquito OBP structure known to date. The structure of AaegOBP1 (= AaegOBP39) shares the common fold of insect OBPs with six {alpha}-helices knitted by three disulfide bonds. A long molecule of polyethylene glycol (PEG) was built into the electron-density maps identified in a long tunnel formed by a crystallographic dimer of AaegOBP1. Circular dichroism analysis indicated that delipidated AaegOBP1 undergoes a pH-dependent conformational change, which may lead to release of odorant at low pH (as in the environment in the vicinity of odorant receptors). A C-terminal loop covers the binding cavity and this 'lid' may be opened by disruption of an array of acid-labile hydrogen bonds thus explaining reduced or no binding affinity at low pH.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2018

Machine learning for graph-based representations of three-dimensional discrete fracture networks

Structural and topological information play a key role in modeling flow and transport through fractured rock in the subsurface. Discrete fracture network (DFN) computational suites such as dfnWorks (Hyman et al. Comput. Geosci. 84, 10–19 2015) are designed to simulate flow and transport in such porous media. Flow and transport calculations reveal that a small backbone of fractures exists, where most flow and transport occurs. Restricting the flowing fracture network to this backbone provides a significant reduction in the network’s effective size. However, the particle-tracking simulations needed to determine this reduction are computationally intensive. Such methods may be impractical for large systems or for robust uncertainty quantification of fracture networks, where thousands of forward simulations are needed to bound system behavior. In this paper, we develop an alternative network reduction approach to characterizing transport in DFNs, by combining graph theoretical and machine learning methods. We consider a graph representation where nodes signify fractures and edges denote their intersections. Using random forest and support vector machines, we rapidly identify a subnetwork that captures the flow patterns of the full DFN, based primarily on node centrality features in the graph. Our supervised learning techniques train on particle-tracking backbone paths found by dfnWorks, but run in negligible time compared to those simulations. In conclusion, we find that our predictions can reduce the network to approximately 20% of its original size, while still generating breakthrough curves consistent with those of the original network.

97 MATHEMATICS AND COMPUTING↗