An official website of the United States government
Summary Energy‐efficient scientific applications require insight into how high performance computing system features impact the applications' power and performance. This insight can result from the development of performance and power models. In this article, we use the modeling and prediction tool MuMMI (Multiple Metrics Modeling Infrastructure) and 10 machine learning methods to model and predict performance and power consumption and compare their prediction error rates. We use an algorithm‐based fault‐tolerant linear algebra code and a multilevel checkpointing fault‐tolerant heat distribution code to conduct our modeling and prediction study on the Cray XC40 Theta and IBM BG/Q Mira at Argonne National Laboratory and the Intel Haswell cluster Shepard at Sandia National Laboratories. Our experimental results show that the prediction error rates in performance and power using MuMMI are less than 10% for most cases. By utilizing the models for runtime, node power, CPU power, and memory power, we identify the most significant performance counters for potential application optimizations, and we predict theoretical outcomes of the optimizations. Based on two collected datasets, we analyze and compare the prediction accuracy in performance and power consumption using MuMMI and 10 machine learning methods.
more »
« less
This content will become publicly available on January 25, 2026
A comparison of techniques to measure the conversion of plastic work to heat in 304L stainless steel under adiabatic conditions
In this work, a cross-laboratory comparison is conducted to critically examine experimental techniques within the High-Strain Rate Mechanics of Materials Laboratory at the University of Utah and the Experimental Solid Mechanics Department at Sandia National Laboratories. The study is aimed at identifying, quantifying, and reducing sources of uncertainty in reported Taylor–Quinney coefficients. Vacuum arc remelted 304L stainless steel specimens extracted from the same ingot are tested in both laboratories under dynamic tension at nominal strain rates of \dot(epsilon) = 450^{-1} and \dot(epsilon) = 900 s^-1 . Independent experiments at both the University of Utah and Sandia National Laboratories report Taylor–Quinney coefficients of \beta_int = 0.85 and \beta_int = 0.9-1.0 for \dot(epsilon) = 450^{-1} and \dot(epsilon) = 900 s^-1, respectively. Sources of variation between labs and practices to mitigate these are also discussed.
more »
« less
- Award ID(s):
- 1847653
- PAR ID:
- 10585259
- Publisher / Repository:
- Elsevier
- Date Published:
- Journal Name:
- International journal of impact engineering
- ISSN:
- 1879-3509
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This paper presents the results of an experimental technique to acquire full-field pressure and strain fields on the windward side of a 10° flap attached to a slender cone-slice model. Tests were conducted in the Hypersonic Wind Tunnel (M = 5, Re= 9 – 14×10^6/m, air) at Sandia National Laboratories. The flap was coated with a fast-response, pressure-sensitive paint sprayed over a photoelastic coating and located near the trailing-edge of an axial slice along the 7° slender cone. This experiment was part of a sponsored project to develop the two-coating luminescent measurement technique and apply to high-speed, fluid-structure interaction environments. Results using a low-speed micropolarizer camera with four polarization orientations show that the technique is sensitive to pressure and strain, measuring an increasing pressure and decreasing strain from leading- to trailing-edge over the surface of the flap. At the low Re condition, the pressure signal captures the separated region near the flap leading edge and compares well with schlieren and oil-film measurements, the latter on a 10° wedge. Aerodynamic heating during the run does affect the pressure signal, likely resulting in an overestimation of pressure. Results using a conventional high-speed camera with a single linear polarizer captures the first bending and torsional modes of vibration when the flap is excited by transient shutdown conditions; however, coupling is difficult to detect in the pressure response due to baseline noise and the slower temporal response of the pressure coating.more » « less
-
A study was conducted to investigate the temperature dependence of thermomechanical coupling in Inconel 718 (IN718). IN718 was selected as a model material due to deformation being predominantly accommodated by planar slip. Split-Hopkinson (or Kolsky) tension bar experiments were conducted at a nominal strain rate of 750 s^-1 at room temperature and 450 ^o C, representing homologous temperatures (T_H=T/T_melt) of T_H = 0.2 and T_H = 0.5, respectively. During deformation, specimen gauge sections were imaged with a high-speed infrared camera. Using one-dimensional wave analysis, the transient heat conduction equation, and temperature- dependent specific heat capacity values, the temperature rise as a function of plastic strain was used to calculate plastic work, thermal work, and the plastic work to heat conversion efficiency, commonly known as the Taylor–Quinney coefficient (TQC). As expected, a significant reduction in plastic work was observed during testing at elevated temperatures. The temperature rise due to plastic deformation was observed to be lower at room temperature compared to elevated temperature experiments. It is reported here for the first time that the TQC is a temperature-sensitive quantity. At T_H = 0.5, a nearly complete conversion of plastic work to heat was observed (TQC = 1.0). Under ambient conditions of T_H = 0.2, a much lower efficiency TQC = 0.4 was observed.more » « less
-
null (Ed.)Flow of a semidilute neutrally buoyant and non-colloidal suspension is numerically studied in the Taylor–Couette geometry where the inner cylinder is rotating and the outer one is stationary. We consider a suspension with bulk particle volume fraction $${\phi _b} = 0.1$$ , the radius ratio $$(\eta = {r_i}/{r_o} = 0.877)$$ and two particle size ratios $$\mathrm{\epsilon }\,( = \; d\textrm{/}a) = 60,\;200$$ , where d is the gap width ( $$= {r_o} - {r_i}$$ ) between cylinders, a is the suspended particles’ radius and $$r_i$$ and $$r_o$$ are the inner and outer radii of the cylinder, respectively. Numerical simulations are conducted using the suspension balance model (SBM) and rheological constitutive laws. We predict the critical Reynolds number in which counter-rotating vortices arise in the annulus. It turns out that the primary instability appears through a supercritical bifurcation. For the suspension of $$\mathrm{\epsilon } = 200$$ , the circular Couette flow (CCF) transitions via Taylor vortex flow (TVF) to wavy vortex flow (WVF). Additional flow states of non-axisymmetric vortices, namely spiral vortex flow (SVF) and wavy spiral vortex flow (WSVF) are observed between CCF and WVF for the suspension of $$\mathrm{\epsilon } = 60$$ ; thus, the transitions occur following the sequence of CCF → SVF → WSVF → WVF. Furthermore, we estimate the friction and torque coefficients of the suspension. Suspended particles substantially enhance the torque on the inner cylinder, and the axial travelling wave of spiral vortices reduces the friction and torque coefficients. However, the coefficients are practically the same in the WVF regime where particles are almost uniformly distributed in the annulus by the axial oscillating flow.more » « less
-
This research paper presents an experimental, theoretical, and numerical study of the thermomechanical behavior of single-crystal and polycrystal copper under uniaxial stress compression loading at varying rates of deformation. The thermomechanical theory is based on a thermodynamically consistent framework for single-crystal face-centered cubic metals, and assumes that all plastic power is partitioned between stored energy due to dislocation structure evolution (configurational) and thermal (kinetic vibrational) energy. An expression for the Taylor–Quinney factor is proposed, which is a simple function of effective temperature and is allowed by second-law restrictions. This single-crystal model is used for the study of single- and polycrystal copper. New polycrystal thermomechanical experimental results are presented at varying strain rates. The temperature evolution on the surface of the polycrystal samples is measured using mounted thermocouples. Thermomechanical numerical single- and polycrystal simulations were performed for all experimental conditions ranging between 10−3 and 5 × 103 s−1. A Taylor homogenization model is used to represent polycrystal behavior. The numerical simulations of all conditions compare reasonable well with experimental results for both stress and temperature evolution. Given our lack of understanding of the mechanisms responsible for the coupling of dislocation glide and atomic vibration, this implies that the proposed theory is a reasonably accurate approximation of the single-crystal thermomechanics.more » « less
