Publications

Increasingly constrained budgets in the defense community, both DoD and DOE, have created a need to emphasize affordability in the development of future weapons systems and components. Increased use of commercially compatible components will play an important role, but there will always remain a need for specialized production, especially at the system level. We will present on-going work at Sandia National Laboratories (referred to from here as Sandia) aimed at insuring the affordability of low-volume, defence-specific systems.

The TRC code, a mid-course tracking code for ballistic missiles, has previously been implemented on a 1024-processor MIMD (Multiple Instruction -- Multiple Data) massively parallel computer. Measures of Effectiveness (MOE) for this algorithm have been developed for this computing environment. The MOE code is run in parallel with the TRC code. Particularly useful MOEs include the number of missed objects (real objects for which the TRC algorithm did not construct a track); of ghost tracks (tracks not corresponding to a real object); of redundant tracks (multiple tracks corresponding to a single real object); and of unresolved objects (multiple objects corresponding to a single track). All of these are expressed as a function of time, and tend to maximize during the time in which real objects are spawned (multiple reentry vehicles per post-boost vehicle). As well, it is possible to measure the track-truth separation as a function of time. A set of calculations is presented illustrating these MOEs as a function of time for a case with 99 post-boost vehicles, each of which spawns 9 reentry vehicles.

The primary purpose of this paper is to provide a careful evaluation of the diffusion velocity concept with regard to its ability to predict the diffusion of vorticity near a moving wall. A computer code BDIF has been written which simulates the evolution of the vorticity field near a wall of infinite length which is moving in an arbitrary fashion. The simulations generated by this code are found to give excellent results when compared to several exact solutions. We also outline a two-dimensional unsteady viscous boundary layer model which utilizes the diffusion velocity concept and is compatible with vortex methods. A primary goal of this boundary layer model is to minimize the number of vortices generated on the surface at each time step while achieving good resolution of the vorticity field near the wall. Preliminary results have been obtained for simulating a simple two-dimensional laminar boundary layer.

We describe a dynamic load-balancing strategy for parallel finite element methods with adaptive mesh (h-) and order (p-) refinement. The load-balancing algorithm is based on the tiling load-balancing system, where global balance in achieved by performing local balancing within overlapping neighborhoods of processors. Tiling is applied to each mesh level created by the adaptive h-refinement. Weights are used in the migration routines to reflect the nonuniform elemental work loads caused by adaptive p-refinement. The combination of adaptive refinement and tiling significantly reduces total execution time relative to fixed-mesh, fixed-order methods yielding comparable accuracy, as we demonstrate with experiments on an nCUBE/2.

Lead zirconate titanate (PZT) thin films are of technological interest for a variety of electronic and optical applications. Fabrication of PZT films by solution deposition techniques is attractive because of stoichiometric control at the molecular level, ease of processing, and both low capital investment and total cost. Control of phase evolution, microstructure, crystallite size and orientation, and ferroelectric domain assemblage during processing is essential to optimize electrical and/or optical properties of the films. Electron microscopy techniques have been used extensively to correlate microstructural features with film processing.

We have developed a capability to make real time concentration measurements of individual chemicals in a complex mixture using a multispectral laser remote sensing system. Our chemical recognition and analysis software consists of three parts: (1) a rigorous multivariate analysis package for quantitative concentration and uncertainty estimates, (2) a genetic optimizer which customizes and tailors the multivariate algorithm for a particular application, and (3) an intelligent neural net chemical filter which pre-selects from the chemical database to find the appropriate candidate chemicals for quantitative analyses by the multivariate algorithms, as well as providing a quick-look concentration estimate and consistency check. Detailed simulations using both laboratory fluorescence data and computer synthesized spectra indicate that our software can make accurate concentration estimates from complex multicomponent mixtures. even when the mixture is noisy and contaminated with unknowns.

Grid partitioning is the method of choice for decomposing a wide variety of computational problems into naturally parallel pieces. In problems where computational load on the grid or the grid itself changes as the simulation progresses, the ability to repartition dynamically and in parallel is attractive for achieving higher performance. We describe three algorithms suitable for parallel dynamic load-balancing which attempt to partition unstructured grids so that computational load is balanced and communication is minimized. The execution time of algorithms and the quality of the partitions they generate are compared to results from serial partitioners for two large grids. The integration of the algorithms into a parallel particle simulation is also briefly discussed.

The electronic components business within the Nuclear Weapons Complex spans organizational and Department of Energy contractor boundaries. An assessment of the current processes indicates a need for fundamentally changing the way electronic components are developed, procured, and manufactured. A model is provided based on a virtual enterprise that recognizes distinctive competencies within the Nuclear Weapons Complex and at the vendors. The model incorporates changes that reduce component delivery cycle time and improve cost effectiveness while delivering components of the appropriate quality.

Cast blasting can be designed to utilize explosive energy effectively and economically for coal mining operations to remove overburden material. The more overburden removed by explosives, the less blasted material there is left to be transported with mechanical equipment, such as draglines and trucks. In order to optimize the percentage of rock that is cast, a higher powder factor than normal is required plus an initiation technique designed to produce a much greater degree of horizontal muck movement. This paper compares two blast models known as DMC (Distinct Motion Code) and SABREX (Scientific Approach to Breaking Rock with Explosives). DMC, applies discrete spherical elements interacted with the flow of explosive gases and the explicit time integration to track particle motion resulting from a blast. The input to this model includes multi-layer rock properties, and both loading geometry and explosives equation-of-state parameters. It enables the user to have a wide range of control over drill pattern and explosive loading design parameters. SABREX assumes that heave process is controlled by the explosive gases which determines the velocity and time of initial movement of blocks within the burden, and then tracks the motion of the blocks until they come to a rest. In order to reduce computing time, the in-flight collisions of blocks are not considered and the motion of the first row is made to limit the motion of subsequent rows. Although modelling a blast is a complex task, the DMC can perform a blast simulation in 0.5 hours on the SUN SPARCstation 10--41 while the new SABREX 3.5 produces results of a cast blast in ten seconds on a 486-PC computer. Predicted percentage of cast and face velocities from both computer codes compare well with the measured results from a full scale cast blast.

In the past few years much effort has been devoted to finding faster and more convenient ways to exchange data between nodes of massively parallel distributed memory machines. One such approach, taken by Thorsten von Eicken et al. is called Active Messages. The idea is to hide message passing latency and continue to compute while data is being sent and delivered. The authors have implemented Active Messages under SUNMOS for the Intel Paragon and performed various experiments to determine their efficiency and utility. In this paper they concentrate on the subset of the Active Message layer that is used by the implementation of the Split-C library. They compare performance to explicit message passing under SUNMOS and explore new ways to support Split-C without Active Messages. They also compare the implementation to the original one on the Thinking Machines CM-5 and try to determine what the effects of low latency and low band-width versus high latency and high bandwidth are on user codes.

SUNMOS is an acronym for Sandia/UNM Operating System. It was originally developed for the nCUBE-2 MIMD supercomputer between January and December of 1991. Between April and August of 1993, SUNMOS was ported to the Intel Paragon. This document provides a quick overview of how to compile and run jobs using the SUNMOS environment on the Paragon. The primary goal of SUNMOS is to provide high performance message passing and process support an example of its capabilities, SUNMOS Release 1.4 occupies approximately 240K of memory on a Paragon node, and is able to send messages at bandwidths of 165 megabytes per second with latencies as low as 42 microseconds using Intel NX calls. By contrast, Release 1.2 of OSF/1 for the Paragon occupies approximately 7 megabytes of memory on a node, has a peak bandwidth of 65 megabytes per second, and latencies as low as 42 microseconds (the communication numbers are reported elsewhere in these proceedings).

We have developed a capability to make real time concentration measurements of individual chemicals in a complex mixture using a multispectral laser remote sensing system. Our chemical recognition and analysis software consists of three parts: (1) a rigorous multivariate analysis package for quantitative concentration and uncertainty estimates, (2) a genetic optimizer which customizes and tailors the multivariate algorithm for a particular application, and (3) an intelligent neural net chemical filter which pre-selects from the chemical database to find the appropriate candidate chemicals for quantitative analyses by the multivariate algorithms, as well as providing a quick-look concentration estimate and consistency check. Detailed simulations using both laboratory fluorescence data and computer synthesized spectra indicate that our software can make accurate concentration estimates from complex multicomponent mixtures, even when the mixture is noisy and contaminated with unknowns.

A parallel unstructured finite element (FE) implementation designed for message passing machines is described. This implementation employs automated problem partitioning algorithms for load balancing unstructured grids, a distributed sparse matrix representation of the global finite element equations and a parallel conjugate gradient (CG) solver. In this paper a number of issues related to the efficient implementation of parallel unstructured mesh applications are presented. These include the differences between structured and unstructured mesh parallel applications, major communication kernels for unstructured CG solvers, automatic mesh partitioning algorithms, and the influence of mesh. partitioning metrics on parallel performance. Initial results are presented for example finite element (FE) heat transfer analysis applications on a 1024 processor nCUBE 2 hypercube. Results indicate over 95% scaled efficiencies are obtained for some large problems despite the required unstructured data communication.

The work in this program covered four primary areas: solid modeling, path planning, modular fixturing, and stability analysis. This report contains highlights of results from the program, references to published reports, and, in an appendix, a currently unpublished report which has been accepted for journal publication, but has yet to appear.

A 3-D finite element analysis was performed to evaluate the stability of the SPR upper and lower oil storage levels at Weeks Island. The mechanical analysis predicted stresses and strains from which pillar stability was inferred using a fracture criterion developed from previous testing of Weeks Island salt. This analysis simulated the sequential mining of the two levels and subsequent oil fill of the mine. The predicted subsidence rates compare well to those measured over the past few years. Predicted failure mechanisms agree with observations made at the time the mine was being modified for oil storage. The modeling technique employed here treats an infinite array of pillars and is a reasonable representation of the behavior at the center of the mine. This analysis predicts that the lower level pillars, at the center of the mine, have fractured and their stability at this time is questionable. Localized pillar fracturing is predicted and implies that the mine is entering a phase of continual time dependent deterioration. Continued and expanded monitoring of the facility and development of methods to assess and predict its behavior are more important now than ever.

We consider bounding the cardinality of an arbitrary triangulation with smallest angle {alpha}. We show that if the local feature size (i.e. distance between disjoint vertices or edges) of the triangulation is within a constant factor of the local feature size of the input, then N < O(1/{alpha})M, where N is the cardinality of the triangulation and M is the cardinality of any other triangulation with smallest angle at least {alpha}. Previous results had an O(1/{alpha}{sup 1/{alpha}}) dependence. Our O(1/{alpha}) dependence is tight for input with a large length to height ratio, in which triangles may be oriented along the long dimension.

A method which is capable of an efficient calculation of the three-dimensional flow field produced by a large system of vortons (discretized regions of vorticity) is presented in this report. The system of vortons can, in turn, be used to model body surfaces, container boundaries, free-surfaces, plumes, jets, and wakes in unsteady three-dimensional flow fields. This method takes advantage of multipole and local series expansions which enables one to make calculations for interactions between groups of vortons which are in well-separated spatial domains rather than having to consider interactions between every pair of vortons. In this work, series expansions for the vector potential of the vorton system are obtained. From such expansions, the three components of velocity can be obtained explicitly. A Fortran computer code FAST3D has been written to calculate the vector potential and the velocity components at selected points in the flow field. In this code, the evaluation points do not have to coincide with the location of the vortons themselves. Test cases have been run to benchmark the truncation errors and CPU time savings associated with the method. Non-dimensional truncation errors for the magnitudes of the vector potential and velocity fields are on the order of 10{sup {minus}4}and 10{sup {minus}3} respectively. Single precision accuracy produces errors in these quantities of up to 10{sup {minus}5}. For less than 1,000 to 2,000 vortons in the field, there is virtually no CPU time savings with the fast solver. For 100,000 vortons in the flow, the fast solver obtains solutions in 1 % to 10% of the time required for the direct solution technique depending upon the configuration.

MELCOR is a fully integrated, engineering-level computer code being developed at Sandia National Laboratories for the USNRC, that models the entire spectrum of severe accident phenomena in a unified framework for both BWRs and PWRs. As a part of an ongoing assessment program, MELCOR has been used to model the MP-1 and MP-2 experiments, which provided data for late-phase melt progression in PWR geometries. Core temperature predicted by MELCOR were within 250--500 K of measured data in both MP-1 and MP-2. Relocation in the debris bed and metallic crust regions of MP-2 was predicted accurately compared to PIE data. Temperature gradients in lower portions of the test bundle were not predicted well in both MP-1 and MP-2, due to the lack of modeling of the heat transfer path to the cooling jacket in those portions of the test bundles. Fifteen sensitivity studies were run on various core (COR), control volume hydrodynamics (CVH) and heat structures (HS) package parameters. No unexpected sensitivities were found, and in particular there were no sensitivities to reduced time step, finer nodalization or to computer platform. Calculations performed by the DEBRIS and TAC2D codes for MP-1 and MP-2 showed better agreement with measured data than those performed by MELCOR. This was expected, through, due to the fully 2-dimensional modeling used in the other codes.

The CUBIT mesh generation environment is a two- and three-dimensional finite element mesh generation tool which is being developed to pursue the goal of robust and unattended mesh generation--effectively automating the generation of quadrilateral and hexahedral elements. It is a solid-modeler based preprocessor that meshes volume and surface solid models for finite element analysis. A combination of techniques including paving, mapping, sweeping, and various other algorithms being developed are available for discretizing the geometry into a finite element mesh. CUBIT also features boundary layer meshing specifically designed for fluid flow problems. Boundary conditions can be applied to the mesh through the geometry and appropriate files for analysis generated. CUBIT is specifically designed to reduce the time required to create all-quadrilateral and all-hexahedral meshes. This manual is designed to serve as a reference and guide to creating finite element models in the CUBIT environment.

Funded as a laboratory-directed research and development (LDRD) project, the work reported here focuses on the development of a computational methodology to determine the dynamic response of heterogeneous solids on the basis of their composition and microstructural morphology. Using the solid dynamics wavecode CTH, material response is simulated on a scale sufficiently fine to explicitly represent the material`s microstructure. Conducting {open_quotes}numerical experiments{close_quotes} on this scale, the authors explore the influence that the microstructure exerts on the material`s overall response. These results are used in the development of constitutive models that take into account the effects of microstructure without explicit representation of its features. Applying this methodology to a glass-reinforced plastic (GRP) composite, the authors examined the influence of various aspects of the composite`s microstructure on its response in a loading regime typical of impact and penetration. As a prerequisite to the microscale modeling effort, they conducted extensive materials testing on the constituents, S-2 glass and epoxy resin (UF-3283), obtaining the first Hugoniot and spall data for these materials. The results of this work are used in the development of constitutive models for GRP materials in transient-dynamics computer wavecodes.

The compute power of the individual nodes of massively parallel systems increases steadily, while network latencies and bandwidth have not improved as quickly. Many researches believe that it is necessary to use explicit message passing in order to get the best possible performance out of these systems. High level parallel languages are shunned out of fear they might compromise performance. In this paper we have a look at one such language called Split-C. It fits into a middle ground between efforts such as High Performance Fortran (HPF) and explicit message passing. HPF tries to hide the underlying architecture from the programmer and let the compiler and the run time system make decision about parallelization, location of data, and the mechanisms used to transfer the data from one node to another. On the other hand, explicit message passing leaves all the decision to the programmer. Split-C allows access to a global address space, but leaves the programmer in control of the location of data, and offers a clear cost model for data access. Split-C is based on Active Messages. We have implemented both under the SUNMOS operating system on the Intel Paragon. We will discuss performance issues of Split-C and make direct comparisons to the Thinking Machines CM-5 implementation. We will also scrutinize Active Messages, discuss their properties and drawbacks, and show that other mechanisms can be used to support Split-C.

We present a theory for transforming the system-theory-based realization models into the corresponding physical coordinate-based structural models. The theory has been implemented into computational procedure and applied to several example problems. Our results show that the present transformation theory yields an objective model basis possessing a unique set of structural parameters from an infinite set of equivalent system realization models. For proportionally damped systems, the transformation directly and systematicaly yields the normal modes and modal damping. Moreover, when nonproportional damping is present, the relative magnitude and phase of the damped mode shapes are separately characterized, and a corrective transformation is then employed to capture the undamped normal modes and nondiagonal modal damping matrix.

The blast-induced movement and final location of geologic layers that may cause environmental problems can be predicted using discrete element methods. This prediction capability can be used by mine operators to locate the material in the muck pile during excavation which would allow encapsulation to prevent groundwater infiltration.

This report presents the results of an inspection around fastener holes in simulated lap splice specimens using a Nortec-30 Eddyscan inspection system. The inspector performing the tests had no prior knowledge of the extent or location of cracks in the specimens examined. The results of the inspection are presented in terms of various probability of detection curve models and are compared to various other eddy current inspections performed on the same set of test specimens. Results indicate that the system is capable, with high confidence, of detecting 60 to 70 mil cracks from under countersink fasteners.

The computer program, DMC (Distinct Motion Code), which was developed for simulating the rock motion associated with blasting, has been used to study the influence of row delay timing on rock motion. The numerical simulations correspond with field observations in that very short delays (< 50ms) and very long delays (> 300ms) produce a lower percent-cast than a medium delay (100 to 200 ms). The DMC predicted relationship between row delay timing and percent-cast is more complex than expected with a dip in the curve where the optimum timing might be expected. More study is required to gain a full understanding of this phenomenon.

A discrete element computer program, DMC (Distinct Motion Code) was developed to simulate blast-induced rock motion. To simplify the complex task of entering material and explosive design parameters as well as bench configuration, a full-featured graphical interface has been developed. DMC is currently executed on both Sun SPARCstation 2 and Sun SPARCstation 10 platforms and routinely used to model bench and crater blasting problems. This paper will document the design and development of the full-featured interface to DMC. The development of the interface will be tracked through the various stages, highlighting the adjustments made to allow the necessary parameters to be entered in terms and units that field blasters understand. The paper also discusses a novel way of entering non-integer numbers and the techniques necessary to display blasting parameters in an understandable visual manner. A video presentation will demonstrate the graphics interface and explains its use.

MELCOR is a fully integrated, engineering-level computer code being developed at Sandia National Laboratories for the USNRC, that models the entire spectrum of severe accident phenomena in a unified framework for both BWRs and PWRs. As a part of an ongoing assessment, program, MELCOR has been used to model the ACRR in-pile DF-4 Damaged Fuel experiment. DF-4 provided data for early phase melt progression in BWR fuel assemblies, particularly for phenomena associated with eutectic interactions in the BWR control blade and zircaloy oxidation in the canister and cladding. MELCOR provided good agreement with experimental data in the key areas of eutectic material behavior and canister and cladding oxidation. Several shortcomings associated with the MELCOR modeling of BWR geometries were found and corrected. Twenty-five sensitivity studies were performed on COR, HS and CVH parameters. These studies showed that the new MELCOR eutectics model played an important role in predicting control blade behavior. These studies revealed slight time step dependence and no machine dependencies. Comparisons made with the results from four best-estimate codes showed that MELCOR did as well as these codes in matching DF-4 experimental data.

MIMD massively parallel computers promise unique power and flexibility for engineering and scientific simulations. In this paper we review the development of a number of software methods and algorithms for scientific and engineering problems which are helping to realize that promise. We discuss new domain decomposition, load balancing, data layout and communications methods applicable to simulations in a broad range of technical field including signal processing, multi-dimensional structural and fluid mechanics, materials science, and chemical and biological systems.

Given a planar straight-line graph, we find a covering triangulation whose maximum angle is as small as possible. A covering triangulation is a triangulation whose vertex set contains the input vertex set and whose edge set contains the input edge set. Such a triangulation differs from the usual Steiner triangulation in that we may not add a Steiner vertex on any input edge. Covering triangulations provide a convenient method for triangulating multiple regions sharing a common boundary, as each region can be triangulated independently. As it is possible that no finite covering triangulation is optimal in terms of its maximum angle, we propose an approximation algorithm. Our algorithm produces a covering triangulation whose maximum angle {gamma} is probably close to {gamma}{sub opt}, a lower bound on the maximum angle in any covering triangulation of the input graph. Note that we must have {gamma} {le} 3{gamma}{sub opt}, since we always have {gamma}{sub opt} {ge} {pi}/3 and no triangulation can contain an angle of size greater than {pi}. We prove something significantly stronger. We show that {pi} {minus} {gamma} {ge} ({pi} {minus} {gamma}{sub opt})/6, i.e., our {gamma} is not much closer to {pi} than is {gamma}{sub opt}. This result represents the first nontrivial bound on a covering triangulation`s maximum angle. We require a subroutine for the following problem: Given a polygon with holes, find a Steiner triangulation whose maximum angle is bounded away from {pi}. No angle larger than 8{pi}/9 is sufficient for the bound on {gamma} claimed above. The number of Steiner vertices added by our algorithm and its running time are highly dependent on the corresponding bounds for the subroutine. Given an n-vertex planar straight-line graph, we require O(n + S(n)) Steiner vertices and O(n log n + T(n)) time, where S(n) is the number of Steiner vertices added by the subroutine and T(n) is its running time for an O(n)-vertex polygon with holes.

The multiplication of a vector by a matrix is the kernel computation of many algorithms in scientific computation. A fast parallel algorithm for this calculation is therefore necessary if one is to make full use of the new generation of parallel supercomputers. This paper presents a high performance, parallel matrix-vector multiplication algorithm that is particularly well suited to hypercube multiprocessors. For an n x n matrix on p processors, the communication cost of this algorithm is O(n/{radical}p + log(p)), independent of the matrix sparsity pattern. The performance of the algorithm is demonstrated by employing it as the kernel in the well-known NAS conjugate gradient benchmark, where a run time of 6.09 seconds was observed. This is the best published performance on this benchmark achieved to date using a massively parallel supercomputer.

The numerical stability and truncation error of a family of differencing schemes for viscoplastic constitutive relations in wavecodes is investigated. A von Neumann stability analysis is performed for a one-dimensional model problem. This analysis identifies two differencing methods that have no restriction on the time step size beyond the usual Courant-Friedrichs-Lewy condition. One of these methods is first-order accurate, and the other is second-order accurate. Implementation of one of these methods in the three-dimensional wavecode CTH is discussed. © 1993 Academic Press. All rights reserved.

A method of solving the two-phase fluid flow equations using a genetic algorithm on a NCUBE multiprocessor computer is presented. The topics discussed are the two-phase flow equations, the genetic representation of the unknowns, the fitness function, the genetic operators, and the implementation of the algorithm on the NCUBE computer. The efficiency of the implementation is investigated using a pipe blowdown problem. Effects of varying the genetic parameters and the number of processors are presented.

A method is described for parallelizing molecular dynamics (MD) simulations by block-decomposing the matrix of bonded and non-bonded force computations. It is particularly useful for organic simulations (polymers, proteins) because unlike spatial-decomposition methods, it requires no geometric information about the simulation domain. Because its communication cost scales as N/{radical}P. rather than N as in the all-to-all broadcast or ring-exchange techniques commonly used in this type of MD simulation, larger numbers of processors can be used effectively, yielding greater parallel speed-ups.

Performance and functionality increases in network environments have in the need for readily accessible mass storage. UNIX{reg_sign}-based networks and mass storage systems are providing the required connectivity and interoperability, however, how UNIX-based mass storage systems are being used is not well documented. This paper describes a study of the usage of the UNIX-based Network Storage Service at Sandia National Laboratories.

A discrete element computer program, DMC (Distinct-Motion Code), has been used for several years to simulate blasting-induced rock motion. Recent enhancements of DMC`s capabilities have included addition of an algorithm that couples together rock motion and gas flow. This allows the user to specify a particular explosive which also specifies equation-of-state and other parameters necessary to model explosive gas flow from the blastwell. Rock loading by the flowing gas is calculated automatically. The mechanism for calculating the rock loading is the subject of this paper. The rock motion effects the gas flow calculation by changing the porosity. DMC is currently being used on a SLTN SPARCstation 2 computer workstation.

Multiple sources of variation will often affect the stability of a manufacturing process. Items from different batches may vary because of variation both within a batch and among different batches. Potential sources of variation include within run, run-to-run and week-to-week differences in a manufacturing process. If multiple sources of variation are present, traditional control chart methods may not be appropriate. In this report we develop control charts for monitoring these sources of variation as well as the process average. An example of how to use the control charts is given, using Field 89 data from functional testing of the MC3854 neutron tube.

Efficient use of a distributed memory parallel computer requires that the computational load be balanced across processors in a way that minimizes interprocessor communication. We present a new domain mapping algorithm that extends recent work in which ideas from spectral graph theory have been applied to this problem. Our generalization of spectral graph bisection involves a novel use of multiple eigenvectors to allow for division of a computation into four or eight parts at each stage of a recursive decomposition. The resulting method is suitable for scientific computations like irregular finite elements or differences performed on hypercube or mesh architecture machines. Experimental results confirm that the new method provides better decompositions arrived at more economically and robustly than with previous spectral methods. We have also improved upon the known spectral lower bound for graph bisection.

In 1991, Sandia National Laboratories acquired a Network Storage Service (NSS) as a result of a fully competitive procurement. The Network Storage Service, which provides access to over a terabyte of data storage in a two-tiered hierarchy, had minimal software security features. Before the NSS could be placed into production, it had to be accredited by the Department of Energy, Sandia`s accrediting authority. Sandia was faced with implementing security features to allow the NSS to be operated in its secure computing network, which is a single security clearance, multiple data security level environment. This paper describes the software security design alternatives that were considered and what was ultimately implemented.

Archimedes is a prototype mechanical assembly system which generates and executes robot assembly programs from a CAD model input. The system seeks to increase flexibility in robotic mechanical assembly applications by automating the programming task. Input is a solid model of the finished assembly, augmented by additional design information such as weld specifications. Parts relationships and geometric constraints are deduced from the solid model. A rule-based planner generates a ``generic`` assembly plan that satisfies the geometric constraints, as well as other constraints embodied in the rules. A plan compiler then converts the generic plan into code specific to an application environment. Other outputs include fixture designs, workcell layout information, object-recognition (vision) routines, grasp plans, and executable code for controlling the robot and workcell accessories. Lessons from operating and demonstrating the system are presented, with a particular emphasis on the implications for future systems. 12 refs.

This paper presents a method to solve partial differential equations governing two-phase fluid flow by using a genetic algorithm on the NCUBE/2 multiprocessor computer. Genetic algorithms represent a significant departure from traditional approaches of solving fluid flow problems. The inherent parallelism of genetic algorithms offers the prospect of obtaining solutions faster than ever possible. The paper discusses the two-phase flow equations, the genetic representation of the unknowns, the fitness function, the genetic operators, and the implementation of the genetic algorithm on the NCUBE/2 computer. The paper investigates the implementation efficiency using a pipe blowdown test and presents the effects of varying both the genetic parameters and the number of processors. The results show that genetic algorithms provide a major advancement in methods for solving two-phase flow problems. A desired goal of solving these equations for a specific simulation problem in real time or faster requires computers with an order of magnitude more processors or faster than the NCUBE/2`s 1024.

A new algorithm for the treatment of sliding interfaces between solids with or without friction in an Eulerian wavecode is described. The algorithm has been implemented in the two-dimensional version of the CTH code. The code was used to simulate penetration and perforation of aluminum plates by rigid, conical-nosed tungsten projectiles. Comparison with experimental data is provided.

Many complex physical processes are modeled by coupled systems of partial differential equations (PDEs). Often, the numerical approximation of these PDEs requires the solution of large sparse nonsymmetric systems of equations. In this paper we compare the parallel performance of a number of preconditioned Krylov subspace methods on a large-scale MIMD machine. These methods are among the most robust and efficient iterative algorithms for the solution of large sparse linear systems. They are easy to implement on various architectures and work well on a wide variety of important problems. In this comparison we focus on the parallel issues associated with both local preconditioners (those that combine information from the entire domain). The various preconditioners are applied to a variety of PDE problems within the GMRES, CCGS, BiCGSTAB, and QMRCGS methods. Conclusions are drawn on the effectiveness of the different schemes based on results obtained from a 1024 processor a nCUBE 2 hypercube.

A series of experiments has been performed on the Sandia Hypervelocity Launcher to determine the performance limits of conventional Whipple shields against representative 0.8 g aluminum orbital debris plate-like fragments with velocities of 7 and 10 km/s. Supporting diagnostics include flash X-rays, high speed photography and transient digitizers for timing correlation. Two Whipple shield designs were tested with either a 0.030 cm or a 0.127 cm thick front sheet and a 0.407 cm thick backsheet separated by 30.5 cm. These two designs bracket the ballistic penetration limit curve for protection against these debris simulants for 7 km/s impacts.

Acoustic telemetry has been a dream of the drilling industry for the past 50 years. It offers the promise of data rates which are one-hundred times greater than existing technology. Such a system would open the door to true logging-while-drilling technology and bring enormous profits to its developers. The basic idea is to produce an encoded sound wave at the bottom of the well, let it propagate up the steel drillpipe, and extract the data from the signal at the surface. Unfortunately, substantial difficulties arise. The first difficult problem is to produce the sound wave. Since the most promising transmission wavelengths are about 20 feet, normal transducer efficiencies are quire low. Compounding this problem is the structural complexity of the bottomhole assembly and drillstring. For example, the acoustic impedance of the drillstring changes every 30 feet and produces an unusual scattering pattern in the acoustic transmission. This scattering pattern causes distortion of the signal and is often confused with signal attenuation. These problems are not intractable. Recent work has demonstrated that broad frequency bands exist which are capable of transmitting data at rates up to 100 bits per second. Our work has also identified the mechanism which is responsible for the observed anomalies in the patterns of signal attenuation. Furthermore in the past few years a body of experience has been developed in designing more efficient transducers for application to metal waveguides. The direction of future work is clear. New transducer designs which are more efficient and compatible with existing downhole power supplies need to be built and tested; existing field test data need to be analyzed for transmission bandwidth and attenuation; and the new and less expensive methods of collecting data on transmission path quality need to be incorporated into this effort. 11 refs.

Geologic materials are often modeled with discrete spheres because the material is not continuous and discrete spherical models simplify the mathematics. Spherical element models have been created using assemblages of spheres with a specified particle size distribution or by assuming the particles are all the same size and making the assemblage a close-packed array of spheres. Both of these approaches yield a considerable amount of material dilatation upon movement. This has proven to be unsatisfactory for sedimentary rock formations that contain bedding planes where shear movement can occur with minimal dilatation of the interface. A new concept referred to as packing angle has been developed to allow the modeler to build arrays of spheres that are the same size but have the rows of spheres offset from each other. ne row offset is a function of the packing angle and allows the modeler to control the dilatation as rows of spheres experience relative horizontal motion.

Sandia National Laboratories and ICI Explosives USA have worked together since 1987 to develop computer modeling techniques for Rock Blasting. A result of this effort is the computer program DMC (Distinct Motion Code) which was developed for two-dimensional simulation of rock motion following a blast (Taylor and Preece, 1989 1992). This program has been used to study blasting-induced rock motion resulting from oil shale mining and has been coupled with a gas flow computation capability for better treatment of the explosive behavior. This past year it has been customized for simulations of bench blasting in coat mines and rock quarries (Preece and Knudsen, 1992b). The explicit descretized nature of DMC gives it an advantage over previous blast modeling programs because subtle differences, such as row delay timing, have an influence on the results. This paper will present a DMC study of the influence on percent cast of row delay timing in a typical coal mine bench blast.

Relativistic high current electron beams can be transported long distances across the geomagnetic field using the IFR (Ion focused Regime) technique. IFR is a method of providing strong electrostatic focusing and guiding of the beam. The guiding is sufficiently strong to allow the beam to transport any angle with respect to geomagnetic field. In the IFR method, first an ionizing laser (or any ionizing method) is used to create a preionized cylindrical channel.

Blade fatigue life is an important element in determining the economic viability of the Vertical-Axis Wind Turbine (VAWT). A principal source of blade fatigue is thought to be the stochastic (i.e., random) aerodynamic loads created by atmospheric turbulence. This report describes the theoretical background of the VAWT Stochastic Aerodynamic Loads (VAWT-SAL) computer code, whose purpose is to numerically simulate these random loads, given the rotor geometry, operating conditions, and assumed turbulence properties. A Double-Multiple-Stream Tube (DMST) analysis is employed to model the rotor's aerodynamic response. The analysis includes the effects of Reynolds number variations, different airfoil sections and chord lengths along the blade span, and an empirical model for dynamic stall effects. The mean ambient wind is assumed to have a shear profile which is described by either a power law or a logarithmic variation with height above ground. Superimposed on this is a full 3-D field of turbulence: i.e., in addition to random fluctuations in time, the turbulence is allowed to vary randomly in planes perpendicular to the mean wind. The influence of flow retardation on the convection of turbulence through the turbine is also modeled. Calculations are presented for the VAWT 34-m Test Bed currently in operation at Bushland, Texas. Predicted time histories of the loads, as well as their Fourier spectra, are presented and discussed. Particular emphasis is placed on the differences between so-called steady-state'' (mean wind only) predictions, and those produced with turbulence present. Somewhat surprisingly, turbulence is found to be capable of either increasing or decreasing the average output power, depending on the turbine's tip-speed ratio. A heuristic explanation for such behavior is postulated, and a simple formula is derived for predicting the magnitude of this effect without the need for a full stochastic simulation. 41 refs., 32 figs., 1 tab.

The hypervelocity impact of a particle on a surface generates a jet of shocked material which is thrown from the impact site. A simple analytic model has been developed to obtain expressions for the evolution of this jet of ejecta. The analysis is based on applying the conservation equations of mass and momentum to the problem of a normal impact of a sphere against a semi-infinite flat target. Expressions are developed for the evolution of the jet velocity, jet release point and the locus of points which describe the ejecta envelope. These analytical ejecta profiles are compared with high speed photographs of impact jet formation.

Our charged particle simulation models a relativistic electron beam for which the field solution is local and thus requires no grid. We have implemented the simulation on a CRAY and on two parallel machines, a nCUBE 2 and Connection Machine. We present implementation details and contrast the approaches necessary for the three architectures. On the parallel machines a dynamic load-balancing problem arises because the beam grows uniformly in one dimension from a few hundred to hundreds of thousands of particles as the simulation progresses. We discuss a folded Gray-code mapping of the processors to the length scale of the simulation that expands (or shrinks) as the beam changes length so as to minimize inter-processor communication. This improves the efficiency of the nCUBE version of the simulation which runs at 10x the speed of the vectorized CRAY version.