DoD High Performance Computing Research Articles

Advanced numerical techniques for predicting frequency response functions

Numerical methods have been widely used to predict Frequency Response Functions (FRF) of structures. The FRF can be used to characterize structural dynamics and support sound and vibration control. Generally, frequency sweep is conducted in the FRF calculations to identify damping and resonant frequencies of structures under excitations. However, as numerical models are getting complex and larger, the demands for computational resources (e.g., CPU time, memory, disk space) are greatly increased. Lately, high performance computing (HPC) at the DoD HPC center has been used with advanced numerical techniques to reduce the overall computational time. Those advanced numerical techniques include Krylov subspace and Galerkin Projection (KGP) and Pade Approximation. Significantly CPU time reduction has been demonstrated for relatively smaller FE models. This paper will further discuss adaptive KGP by automating the frequency sweep process via calculated relative residues, then combine with Finite Element Tearing and Interconnecting (FETI) to solve for FRF of a large FEM model in the order of 60M DOF. A flat plate with viscoelastic layer will serve as an example to demonstrate accuracy and efficiency of AKGP with FETI. The benefits of using HPC with advanced numerical techniques to solve for large FEM models are clearly displayed.

Evaluating numerical techniques with HPC systems in predicting vibratory responses of large finite element models

Finite element (FE) analyses have been widely used to support design optimizations in various disciplines to improve performance and possibly save material, cost, and schedule. As computational resources (i.e., CPU, memory, storage, and accelerator) are getting more powerful and cost effective, the complexity and model sizes of numerical models are greatly increasing to capture geometry details and features of the design. Correspondingly, the computational time for complex and large FE models are significantly growing. Traditional brute-force approach is not quickly enough to support design iterations. In this presentation, various numerical techniques are investigated and exercised in the DoD High Performance Computing systems to identify their advantages and limitations in conducting large FE analyses. Those techniques include utilization of GPGPU, direct versus iterative solvers, approximation of frequency sweep, etc. As a benchmark case, frequency response functions of a flat plate with viscoelastic material is exercised at the DoD HPC systems to examine the numerical techniques' impacts.

A Hybrid Learning Approach to Prognostics and Health Management Applied to Military Ground Vehicles Using Time-Series and Maintenance Event Data

Attempts to leverage operational time-series data in Condition Based Maintenance (CBM) approaches to optimize the life cycle management and Reliability, Availability, and Maintainability (RAM) of military vehicles have encountered several obstacles over decades of data collection. These obstacles have beset similar approaches on civilian ground vehicles, as well as on aircraft and other complex systems. Analysis of operational data is critical because it represents a continuous recording of the state of the system. Applying rudimentary data analytics to operational data can provide insights like fuel usage patterns or observed reliability of one vehicle or even a fleet. Monitoring trends and analyzing patterns in this data over time, however, can provide insight into the health of a vehicle, a complex system, or a fleet, predicting mean time to failure or compiling logistic or life cycle needs. Such High-Performance Data Analytics (HPDA) on operational time-series datasets has been historically difficult due to the large amount of data gathered from vehicle sensors, the lack of association between clusters observed in the data and failures or unscheduled maintenance events, and the deficiency of unsupervised learning techniques for time-series data. We present an HPDA environment and a method of discovering patterns in vehicle operational data that determines models for predicting the likelihood of imminent failure, referred to as Parameter-Based Indicators (PBIs). Our method is a data-driven approach that uses both time-series and relational maintenance data. This hybrid approach combines both supervised and unsupervised machine learning and data analytic techniques to correlate labeled, relational maintenance event data with unlabeled operational time-series data utilizing the DoD High Performance Computing (HPC) capabilities at the U.S. Army Engineer Research and Development Center. In leveraging both time-series and relational data, we demonstrate a means of fast, purely data-driven model creation that is more broadly applicable and requires less a priori information than physics informed, data-driven models. By blending these approaches, this system will be able to relate some lifecycle management goals through the workflow to generate specific PBIs that will predict failures or highlight appropriate areas of concern in individual or collective vehicle histories.

Comparison of two and three spatial dimensional solutions of a parabolic approximation of the wave equation at ocean-basin scales in the presence of internal waves: 200–250 Hz.

Numerical solutions are given for a parabolic approximation to the acoustic wave equation at 200 and 250 Hz in two and three spatial dimensions to determine if azimuthal coupling in the horizontal coordinate significantly affects horizontal correlation in the presence of internal gravity waves in the sea. Coupling is a small effect at distances of 4000 km or less. This implies that accurate solutions are possible using computations from uncoupled vertical slices. Shapes of horizontal correlation are not far from shapes given by several theories. Estimates of horizontal correlation at 4000 km and 200 and 250 Hz are about 0.4 and 0.3 km, respectively. [Work supported by the Office of Naval Research Contract Nos. N00014-06-C-0031 and N00014-10-C-0480 and by a grant of computer time from the DOD High Performance Computing Modernization Program at the Naval Oceanographic Office.]

Time-domain equations for sound propagation in or reflection from rigid porous media

In acoustics, many absorbing materials can be considered as porous media with rigid frames. Therefore, studies of sound propagation in or reflection from rigid porous media are important in many applications. These studies are usually done in the frequency domain where equations describing sound propagation in or reflection from porous media are well established. However, there are also problems where such equations are needed in the time domain. Examples include studies of material properties by acoustic impulses and finite-difference time-domain (FDTD) simulation of sound propagation in the presence of absorbing surfaces. In this paper, using Wilson’s relaxation model for acoustical properties of porous media [D. K. Wilson, Appl. Acoust. 50, 171–188 (1997)], we derive a closed set of time-domain, integro-differential equations for the sound pressure and particle velocity in rigid porous media. In the limiting cases of low and high frequencies, these equations coincide with those known in the literature. Furthermore, in these limiting cases, using the relation model, we derive time-domain boundary conditions for sound reflection from a porous medium. This result can be used in FDTD simulation of sound propagation outdoor. [Work supported by the DoD High Performance Computing Modernization Office and ERDC-CRREL.]

A Scalable Version of the Navy Operational Global Atmospheric Prediction System Spectral Forecast Model

The Navy Operational Global Atmospheric Prediction System (NOGAPS) includes a state-of-the-art spectral forecast model similar to models run at several major operational numerical weather prediction (NWP) centers around the world. The model, developed by the Naval Research Laboratory (NRL) in Monterey, California, has run operational at the Fleet Numerical Meteorological and Oceanographic Center (FNMOC) since 1982, and most recently is being run on a Cray C90 in a multi-tasked configuration. Typically the multi-tasked code runs on 10 to 15 processors with overall parallel efficiency of about 90%. resolution is T159L30, but other operational and research applications run at significantly lower resolutions. A scalable NOGAPS forecast model has been developed by NRL in anticipation of a FNMOC C90 replacement in about 2001, as well as for current NOGAPS research requirements to run on DOD High-Performance Computing (HPC) scalable systems. The model is designed to run with message passing (MPI). Model design criteria include bit reproducibility for different processor numbers and reasonably efficient performance on fully shared memory, distributed memory, and distributed shared memory systems for a wide range of model resolutions. Results for a wide range of processor numbers, model resolutions, and different vendor architectures are presented. Single node performance has been disappointing on RISC based systems, at least compared to vector processor performance. This is a common complaint, and will require careful re-examination of traditional numerical weather prediction (NWP) model software design and data organization to fully exploit future scalable architectures.

Limitations on adiabatic normal modes in range-varying environments

Curvilinear coordinates are used to extend adiabatic normal modes to weak range-dependent environments including those with sloping bottoms. Reciprocal vertical wave numbers of a reference mode provide the coordinate scale factors. This approach is valid for negligible horizontal derivatives of intermodal vertical-wave number ratios. Large gradients in the bottom impedance make these ratios large, implying significant mode coupling in oceanic regions spanning both soft and hard terrains. Additionally, a large number of eigenvalues are needed to compute adiabatic normal-mode solutions for high-resolution acoustic fields. The direct solution of eigenvalue problems on a fine grid is often prohibitive. Alternatively, the eigenvalues may be interpolated to the fine grid. This approach usually requires extrapolation toward edges across which the horizontal wave number becomes imaginary at mode cutoff. One choice for an interpolation function is the integrated reciprocal vertical wave number. This function, proportional to water depth, is smooth and real throughout the cutoff region, while causing the horizontal wave number to vanish along the cutoff line. [Work supported by the Office of Naval Research and partially by DOD High Performance Computing facilities at NRL.]