Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Local time stepping”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

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

At least 73 records · Page 4
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Extended Lagrangian Born–Oppenheimer molecular dynamics for orbital-free density-functional theory and polarizable charge equilibration models

We report extended Lagrangian Born–Oppenheimer molecular dynamics (XL-BOMD) is formulated for orbital-free Hohenberg–Kohn density-functional theory and for charge equilibration and polarizable force-field models that can be derived from the same orbital-free framework. The purpose is to introduce the most recent features of orbital-based XL-BOMD to molecular dynamics simulations based on charge equilibration and polarizable force-field models. These features include a metric tensor generalization of the extended harmonic potential, preconditioners, and the ability to use only a single Coulomb summation to determine the fully equilibrated charges and the interatomic forces in each time step for the shadow Born–Oppenheimer potential energy surface. The orbital-free formulation has a charge-dependent, short-range energy term that is separate from long-range Coulomb interactions. This enables local parameterizations of the short-range energy term, while the long-range electrostatic interactions can be treated separately. The theory is illustrated for molecular dynamics simulations of an atomistic system described by a charge equilibration model with periodic boundary conditions. The system of linear equations that determines the equilibrated charges and the forces is diagonal, and only a single Ewald summation is needed in each time step. The simulations exhibit the same features in accuracy, convergence, and stability as are expected from orbital-based XL-BOMD.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Time-series forecasting using manifold learning, radial basis function interpolation, and geometric harmonics

We address a three-tier numerical framework based on nonlinear manifold learning for the forecasting of high-dimensional time series, relaxing the “curse of dimensionality” related to the training phase of surrogate/machine learning models. At the first step, we embed the high-dimensional time series into a reduced low-dimensional space using nonlinear manifold learning (local linear embedding and parsimonious diffusion maps). Then, we construct reduced-order surrogate models on the manifold (here, for our illustrations, we used multivariate autoregressive and Gaussian process regression models) to forecast the embedded dynamics. Finally, we solve the pre-image problem, thus lifting the embedded time series back to the original high-dimensional space using radial basis function interpolation and geometric harmonics. The proposed numerical data-driven scheme can also be applied as a reduced-order model procedure for the numerical solution/propagation of the (transient) dynamics of partial differential equations (PDEs). In conclusion, we assess the performance of the proposed scheme via three different families of problems: (a) the forecasting of synthetic time series generated by three simplistic linear and weakly nonlinear stochastic models resembling electroencephalography signals, (b) the prediction/propagation of the solution profiles of a linear parabolic PDE and the Brusselator model (a set of two nonlinear parabolic PDEs), and (c) the forecasting of a real-world data set containing daily time series of ten key foreign exchange rates spanning the time period 3 September 2001–29 October 2020.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2026

TANTE: Time-adaptive operator learning via neural Taylor expansion

Operator learning for time-dependent partial differential equations (PDEs) has seen rapid progress in recent years, enabling efficient approximation of complex spatiotemporal dynamics. However, most existing methods rely on fixed time step sizes during rollout, which limits their ability to adapt to varying temporal complexity and often leads to error accumulation. In this work, we propose the Time-Adaptive Transformer with Neural Taylor Expansion (TANTE), a novel operator-learning framework that produces continuous-time predictions with adaptive step sizes. TANTE predicts future states by performing a Taylor expansion at the current state, where neural networks learn both the higher-order temporal derivatives and the local radius of convergence. This allows the model to dynamically adjust its rollout based on the local behavior of the solution, thereby reducing cumulative error and improving computational efficiency. We demonstrate the effectiveness of TANTE across a wide range of PDE benchmarks, achieving superior accuracy and adaptability compared to fixed-step baselines, delivering accuracy gains of 60-80 % and speed-ups of 30-40 % at inference time.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Quasi-Steady Simulation of Glaze Ice Accretion and Heat Transfer in the Supercooled Large Droplet Regime in Atmospheric Turbulent Air Flow

Abstract To avoid the use of a computationally intensive full unsteady simulation while providing an accurate solution, a quasi-steady simulation has been performed to study glaze and mixed ice accretion in the supercooled large droplet (SLD) regime in turbulent flow of atmospheric air. We have attempted to find the minimum time-step necessary to adequately simulate the icing process on an airfoil surface. Based on node displacement, a mesh morphing scheme has been adopted in the computations to account for the moving boundaries that are caused by the continuous ice buildup. We have modeled ice accretion on an airfoil surface for a time period of 232 s using several time steps. At each time-step we solved the steady-state conservation equations for the air and droplet phases and then used the results as initial conditions for the next time position. The magnitude of the time-step ranged from a least accurate two-shot simulation (where the time-step is 116 s) to a most accurate 2320-shot simulation (where the time-step is 0.1 s). In-between these two extreme time steps, we have performed a three-shot simulation (where the time-step is 77.33 s), a four-shot simulation (where the time-step is 58 s), a six-shot simulation (where the time-step is 38.67 s), a 46-shot simulation (where the time-step is 5 s), and a 232-shot simulation (where the time-step is 1 s). We have done so to find out the degree of accuracy (or inaccuracy) of the multishot simulation approach and to find out the appropriate time-step needed for a successful and valid quasi-steady simulation. A valid quasi-steady simulation needs to use a time-step that is small enough to reproduce the full time-dependent solution within a very small error band. We have found that both the 1 s and the 0.1 s time steps produce virtually identical results. This is the primary litmus test that proves the validity of the quasi-steady-state assumption. The results adopted in this paper are thus all based on the more conservative 0.1 s time-step. In the process of performing the simulations, remeshing was required in order to maintain the grid density in zones of high curvature to be able to capture the full physics in those zones. After successfully modeling glaze ice accretion over the airfoil surface using the 0.1 s quasi-steady simulation approach, the effects of supercooled large droplets (SLDs) impacting the surface have been examined and presented in terms of the variation of the local collection efficiency, the water film thickness, and the heat transfer rate. Examination of the variation of the angle that the ice horn makes with the airfoil chord line demonstrated a 20% improvement in angle prediction when the time-step is reduced from 116 s to 0.1 s. The analysis also reveals a 12% or 8.5% increase in the maximum collection efficiency, βmax, depending on whether the value of the liquid water content (LWC) has been doubled from 0.5 g/m3 to 1 g/m3, or the value of the freestream velocity has been doubled from 75 m/s to 150 m/s, respectively. Because of the need to monitor the local collection efficiency and convective heat fluxes at each shot (regardless of the number of shots employed), the approach adopted here was found to be effective in successively and successfully reproducing the curvature of the glaze ice horn.

Engineering↗
DOE OSTI2023

CFL Optimized Forward–Backward Runge–Kutta Schemes for the Shallow-Water Equations

Abstract We present the formulation and optimization of a Runge–Kutta-type time-stepping scheme for solving the shallow-water equations, aimed at substantially increasing the effective allowable time step over that of comparable methods. This scheme, called FB-RK(3,2), uses weighted forward–backward averaging of thickness data to advance the momentum equation. The weights for this averaging are chosen with an optimization process that employs a von Neumann–type analysis, ensuring that the weights maximize the admittable Courant number. Through a simplified local truncation error analysis and numerical experiments, we show that the method is at least second-order in time for any choice of weights and exhibits low dispersion and dissipation errors for well-resolved waves. Further, we show that an optimized FB-RK(3,2) can take time steps up to 2.8 times as large as a popular three-stage, third-order strong stability-preserving Runge–Kutta method in a quasi-linear test case. In fully nonlinear shallow-water test cases relevant to oceanic and atmospheric flows, FB-RK(3,2) outperforms SSPRK3 in admittable time step by factors roughly between 1.6 and 2.2, making the scheme approximately twice as computationally efficient with little to no effect on solution quality. Significance Statement The purpose of this work is to develop and optimize time-stepping schemes for models relevant to oceanic and atmospheric flows. Specifically, for the shallow-water equations we optimize for schemes that can take time steps as large as possible while retaining solution quality. We find that our optimized schemes can take time steps between 1.6 and 2.2 times larger than schemes that cost the same number of floating point operations, translating directly to a corresponding speedup. Our ultimate goal is to use these schemes in climate-scale simulations.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Patience is a virtue: A data-driven analysis of rooftop solar PV permitting timelines in the United States

© 2020 The Authors Local permitting can ensure the safe installation and operation of rooftop solar photovoltaic (PV) systems. At the same time, burdensome local permitting processes and local variation in requirements may pose challenges to PV deployment. In this article, we explore new data on the durations between key steps in the PV permitting process in the United States. The data suggest that a typical customer can expect to wait around 25–100 days from permit application until an installed system passes inspection. Permit durations vary significantly across jurisdictions, due in part to differences in local permitting policies. However, permit durations vary as significantly within jurisdictions as across them, in part due to significant variation across installers, suggesting that installer strategies and practices play an important role in permitting timelines. Permit durations have declined over time, reflecting progress from permit streamlining policies and jurisdiction learning-by-doing, though durations have stabilized in recent years. The data suggest that typical PV customers still face long and uncertain permitting timelines in the United States.

O'Shaughnessy, E↗
DOE OSTI2020

Locally Engineering and Interrogating the Photoelectrochemical Behavior of Defects in Transition Metal Dichalcogenides

Transition metal dichalcogenides (TMDs) are attractive materials for a variety of applications in solar energy conversion and electrocatalysis, due to their favorable optical and electrical properties and their unique two-dimensional structures which facilitate the fabrication of wide-area, ultrathin layers. Unfortunately, the basal planes which make up the majority of these materials are relatively inert, and thus a great deal of effort has been directed to engineering favorable, catalytically active defects into these materials. In this paper, we demonstrate how probe-based electrochemical techniques can be employed as multifunctional tools for locally modifying TMD materials and probing the electrochemical behavior of the resulting defects. Scanning Electrochemical Cell Microscopy (SECCM) was employed to locally anodize exfoliated p-type WSe 2 nanosheets, creating hole-like defects within individual basal planes in a highly controllable fashion. Photoelectrochemical SECCM imaging was then employed to characterize the chemical behavior of these engineered defects, revealing significantly enhanced activity toward the Hydrogen Evolution Reaction (HER). Atomic force microscopy studies are presented which suggest these enhancements result from an increased density of monolayer-high step features within the anodized defects. Analysis of the SECCM data in the context of finite element simulations revealed that these enhancements increased with increasing anodization time, with local kinetic rates over 2 orders of magnitude higher than unaltered basal planes.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Exponentially Reduced Circuit Depths Using Trotter Error Mitigation

Product formulas are a popular class of digital quantum simulation algorithms due to their conceptual simplicity, low overhead, and performance, which often exceeds theoretical expectations. Recently, Richardson extrapolation and polynomial interpolation have been proposed to mitigate the Trotter error incurred by the use of these formulas. This work provides a rigorous, general analysis of these techniques for computing time-evolved observables, simplifying the interpolation algorithm in the process, and shows that extrapolation generically improves the performance of product formulas for this task. We demonstrate that, to achieve error 𝜖 in a simulation of time 𝑇 using a 𝑝 ⁢th-order product formula with extrapolation, circuit depths of 𝑂⁡(𝑇 1+1/𝑝 ⁢polylog (1/𝜖)) are sufficient—an exponential improvement in the precision over product formulas alone. Furthermore, we prove that these algorithms achieve commutator scaling, and improve the 𝑇 complexity for the interpolation algorithm. By relaxing the requirement of performing exact Chebyshev interpolation, our simplified algorithm eliminates the need for fractional implementations of Trotter steps, reducing computational overhead. Finally, we show these techniques can be combined with the classical shadows method to estimate many time-evolved local observables. Taken together, our findings provide the strongest evidence yet for the utility of Trotter error-mitigation techniques in algorithmic applications.

quantum algorithms & computation↗
DOE OSTI2026

Sap Velocity Data for Urban Trees in Chicago, Illinois (2024-2025)

This dataset contains uncorrected sap velocity measurements using the heat ratio method (HRM) collected using ICT International SFM1x sensors at five urban sites in Chicago, Illinois, as part of the DOE CROCUS project. The data includes continuous monitoring of sap velocity from various tree species, including Maples (Acer spp.): Sugar Maple (Acer saccharum), Silver Maple (Acer saccharinum), and Red Maple (Acer rubrum); Oaks (Quercus spp.): Swamp White Oak (Quercus bicolor); American Elm (Ulmus americana); Honey Locust (Gleditsia triacanthos); Cottonwood (Populus deltoides); and Tree of Heaven (Ailanthus altissima) across Chicago State University (CSU), Northeastern Illinois University (NEIU), Northwestern University (NU), University of Illinois Chicago (UIC), and West Woodlawn "Blacks in Green" (BIG). These include both street trees and those in urban park locations. Measurements were collected at 15-20 minute intervals, depending on the sensor, and transmitted via Long Range Wide Area Network (LoRaWAN) protocols. The wireless data was collected by Sage Network (https://sagecontinuum.org/) nodes. The dataset includes sensor ID, Global Positioning System (GPS) coordinates, tree species (common and scientific names), tree identification number, diameter at breast height (DBH in cm), uncorrected sap velocity measurements (cm/hr) from both inner and outer probes, and Sage Node identifiers so the data can be mapped to related variables such as air quality and wind speed that were collected on the Sage nodes. All timestamps are in local Chicago time (CDT/CST). Quality control flags are provided using a 3-bit binary system indicating physical range violations (< -10 or > 60 cm/hr), step spikes (absolute difference > 36 cm/hr), and stuck sensor conditions (> 10 consecutive identical values). These are raw data, not corrected for wood anatomy or species-specific characteristics. Data is provided in comma separated (CSV) format. This dataset is part of a larger collection of CROCUS environmental monitoring data, including linked datasets from Air Quality Transmitter (AQT) sensors, Weather Transmitter (WXT) sensors, and Multi-Function Research LoRaWAN (MFR) Nodes. DOIs for the supporting data are provided as part of this data package.

Chicago↗
DOE OSTI2020

An analysis of the spatio-temporal resolution of the immersed boundary method with direct forcing

The immersed boundary method (IBM) with direct forcing is very popular in the simulation of rigid particulate flows. In the IBM, an interaction force is introduced at the interface between fluid and particle in order to approximate the no-slip boundary condition. The interaction force is calculated through dividing the velocity difference (or error) between fluid and particle at the interface by the time step. Here, a dynamic equation for the velocity difference is derived. Additionally, analyses on the dynamic equation provide a few new findings: (i) The interaction force is the solution of a least-squares error problem, with the direct implication that the Lagrangian marker distribution has no effect on the large scale flow structure once the distribution of Lagrangian markers become saturated along the interface (i.e., each marker remains properly correlated with all its neighbors); (ii) The Lagrangian volume-weight is a relaxation factor to control how fast the velocity error decays to the ideal value of zero; (iii) The optimal choice of the Lagrangian volume-weight is the largest value permissible by a stability condition. A comprehensive convergence analysis with regard to the spatial and temporal resolution is presented for the velocity error and also for the shear-stress and surface pressure. In three simple canonical problems, it is analytically and numerically shown that the IBM results converge to the theoretical solutions obtained with precise imposition of no-slip and no-penetration boundary conditions. It is observed that it is not necessary to match the Lagrangian marker volume-weight to that of the local Eulerian cell volume and in fact this matching leads to lower than optimal computational efficiency. However, it is found that extremely high Eulerian grid resolution and small time step have to be used to obtain high precision simulation results. Especially, the time step should be inversely proportional to the particle Reynolds number for low Reynolds number flows. For high frequency oscillation problems, the grid size needs to be reduced by a factor of the square root of the frequency, and the time step to be reduced by a factor of the frequency. The theoretical findings here can be used to alleviate the technical difficulties in simulating non-spherical particles by not requiring the Lagrangian marker distribution to match the Eulerian grids and also in the implementation of IBM on non-uniform Eulerian grids. The present work also provides simple practical guidance on the choice of temporal and spatial resolution so as to control the simulation error a priori.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2023

Scalable manufacturing of quantum light emitters in silicon under rapid thermal annealing

Quantum light sources play a fundamental role in quantum technologies ranging from quantum networking to quantum sensing and computation. The development of these technologies requires scalable platforms, and the recent discovery of quantum light sources in silicon represents an exciting and promising prospect for scalability. The usual process for creating color centers in silicon involves carbon implantation into silicon, followed by rapid thermal annealing. However, the dependence of critical optical properties, such as the inhomogeneous broadening, the density, and the signal-to-background ratio, on centers implantation steps is poorly understood. We investigate the role of rapid thermal annealing on the dynamic of the formation of single color centers in silicon. We find that the density and the inhomogeneous broadening greatly depend on the annealing time. We attribute the observations to nanoscale thermal processes occurring around single centers and leading to local strain fluctuations. Our experimental observation is supported by theoretical modeling based on first principles calculations. The results indicate that annealing is currently the main step limiting the scalable manufacturing of color centers in silicon.

42 ENGINEERING↗
DOE OSTI2023

Convergence analysis of single rate and multirate fixed stress split iterative coupling schemes in heterogeneous poroelastic media

Recently, the accurate modeling of flow–structure interactions has gained more attention and importance for both petroleum and environmental engineering applications. Of particular interest is the coupling between subsurface flow and reservoir geomechanics. Different single rate and multirate iterative and explicit coupling schemes have been proposed and analyzed in the past. In addition, Banach fixed point contraction results were obtained for iterative coupling schemes, and conditionally stable results were obtained for explicit coupling schemes. In this work, we will consider the mathematical analysis of the single rate and multirate fixed stress split iterative coupling schemes for spatially heterogeneous poroelastic media. We will re–establish the contractivity for both schemes in the localized case, and we will show that heterogeneities come at the expense of imposing more restricted conditions on the number of fine flow time steps that can be taken within one coarse mechanics time step in the multirate case. Our mathematical analysis is supplemented by numerical simulations validating our derived upper bounds. Finally, to the best of our knowledge, this is the first rigorous mathematical analysis of the multirate fixed–stress split iterative coupling scheme in heterogeneous poroelastic media.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

A coupled multipoint stress–multipoint flux mixed finite element method for the Biot system of poroelasticity

In this work, we present a mixed finite element method for a five-field formulation of the Biot system of poroelasticity that reduces to a cell-centered pressure–displacement system on simplicial and quadrilateral grids. A mixed stress–displacement–rotation formulation for elasticity with weak stress symmetry is coupled with a mixed velocity–pressure Darcy formulation. The spatial discretization is based on combining the multipoint stress mixed finite element (MSMFE) method for elasticity and the multipoint flux mixed finite element (MFMFE) method for Darcy flow. It uses the lowest order Brezzi–Douglas–Marini mixed finite element spaces for the poroelastic stress and Darcy velocity, piecewise constant displacement and pressure, and continuous piecewise linear or bilinear rotation. A vertex quadrature rule is applied to the velocity, stress, and stress–rotation bilinear forms, which block-diagonalizes the corresponding matrices and allows for local velocity, stress, and rotation elimination. This leads to a cell-centered positive-definite system for pressure and displacement at each time step. We perform error analysis for the semidiscrete and fully discrete formulations, establishing first order convergence for all variables in their natural norms. The numerical tests confirm the theoretical convergence rates and illustrate the locking-free property of the method.

42 ENGINEERING↗
DOE OSTI2021

Dispersion curve regression – assisted wideband local wavenumber analysis for characterizing three-dimensional (3D) profile of hidden corrosion damage

Corrosion as common damage in civil, petrochemical, nuclear, and aerospace structures affects the integrity and safety of the structures and may lead to catastrophic failures. This paper presents dispersion curve regression-assisted local wavenumber analysis method, which can analyze the time–space wavefield containing wideband information of wave-damage interaction and further extract the structural information carried by such wavefield for characterizing hidden corrosion damage in an isotropic plate. To acquire the time–space wavefield, a noncontact scanning laser Doppler vibrometer is used. In our analysis method, the acquired time–space wavefield is firstly processed to generate a local wavenumber function. The derived wavenumber function is further analyzed with a dispersion curve regression step, which searches in a set of theoretical frequency-wavenumber dispersion curves for different plate thicknesses and identifies the optimal plate thickness whose theoretical curve best matches the frequency-wavenumber relation contained in the local wavenumber function. By this means, a 3D profile including both in plane and thickness dimensions of the structure can be constructed for corrosion visualization and quantification. The experimental study demonstrates that our method can quantify the profile of hidden uniform corrosion in metal plates in terms of its in-plane shape and size as well as its out-of-plane depth in the subwavelength scale. Moreover, compared to previous Lamb wave-based corrosion inspection methods, our technique allows for noncontact 3D characterization of hidden corrosion from a far distance to the structure. Finally, we believe this work will inspire new 3D damage quantification methods that are based on wavefield analysis as well as enable potential applications for the quantification of hidden corrosion in civil, petrochemical, nuclear, and aerospace structures.

42 ENGINEERING↗
DOE OSTI2024

Thermal relaxation of strain and twist in ferroelectric hexagonal boron nitride moiré interfaces

New properties can arise at van der Waals (vdW) interfaces hosting a moiré pattern generated by interlayer twist and strain. However, achieving precise control of interlayer twist/strain remains an ongoing challenge in vdW heterostructure assembly, and even subtle variation in these structural parameters can create significant changes in the moiré period and emergent properties. Characterizing the rate of interlayer twist/strain relaxation during thermal annealing is critical to establish a thermal budget for vdW heterostructure construction and may provide a route to improve the homogeneity of the interface or to control its final state. Here, we characterize the spatial and temporal dependence of interfacial twist and strain relaxation in marginally-twisted hBN/hBN interfaces heated under conditions relevant to vdW heterostructure assembly and typical sample annealing. We find that the ferroelectric hBN/hBN moiré at very small twist angles (θ ≤ 0.1°) relaxes minimally during annealing in air at typical assembly temperatures of 170°C. However, at 400°C, twist angle relaxes significantly, accompanied by a decrease in spatial uniformity. Uniaxial heterostrain initially increases and then decreases over time, becoming increasingly non-uniform in direction. Structural irregularities such as step edges, contamination bubbles, or contact with the underlying substrate result in local inhomogeneity in the rate of relaxation.

2D materials↗
DOE OSTI2020

Quantum algorithms for disordered physics

We show how a quantum computer may efficiently simulate a disordered Hamiltonian, by incorporating a pseudo-random number generator directly into the time evolution circuit. This technique is applied to quantum simulation of few-body disordered systems in the large volume limit; in particular, Anderson localization. The method requires a number of (error corrected) qubits proportional to the logarithm of the volume of the system, and each time evolution step requires a number of gates polylogarithmic in the volume. We simulate the method to observe the metal-insulator transition on a three-dimensional lattice. Additionally, we demonstrate the algorithm on a one-dimensional lattice, using physical quantum processors.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2025

On High-Order/Low-Order and Micro-Macro Methods for Implicit Time-Stepping of the BGK Model

In this paper, a high-order/low-order (HOLO) method is combined with a micro-macro (MM) decomposition to accelerate iterative solvers in fully implicit time-stepping of the Bhatnagar–Gross–Krook (BGK) equation for gas dynamics. The MM formulation represents a kinetic distribution as the sum of a local Maxwellian and a perturbation. In highly collisional regimes, the perturbation away from initial and boundary layers is small and can be compressed to reduce the overall storage cost of the distribution. The convergence behavior of the MM methods, the usual HOLO method, and the standard source iteration method is analyzed on a linear BGK model. Both the HOLO and MM methods are implemented using a discontinuous Galerkin (DG) discretization in phase space, which naturally preserves the consistency between high- and low-order models required by the HOLO approach. Furthermore, the accuracy and performance of these methods are compared on the Sod shock tube problem and a sudden wall heating boundary layer problem. Overall, the results demonstrate the robustness of the MM and HOLO approaches and illustrate the compression benefits enabled by the MM formulation when the kinetic distribution is near equilibrium.

BGK model↗
DOE OSTI2021

In Situ Machine Learning for Intelligent Data Capture on Exascale Platforms. Final Report

In many dynamic systems, interesting events occur locally in time and space. Examples of such systems include ignition events in combustion simulations, material fractures in mechanics simulations, and extreme weather events in climate simulations. Due to memory constraints and data I/O costs, current simulation workflows save data at regularly spaced time-steps, at a fixed rate determined before the start of the simulation. Often this mode of operation results in missed events of interest, necessitating a simulation restart from before an event occurred with more frequent data saves. This data saving workflow is grossly inefficient and is already a bottleneck in the computing process. We propose to develop machine learning algorithms that can detect when interesting dynamical events are occurring, triggering data saves. These machine learning algorithms will perform in situ anomaly detection to flag regions with different dynamical properties than those previously recorded. The adaptive data saves would be local in time and space to match the event of interest, thereby enabling a much more efficient workflow that will reduce data I/O costs and data storage memory requirements. The algorithms will be tested on two applications: auto-ignition simulations and climate simulations. A critical component of this project will be developing machine learning algorithms that can be deployed efficiently in situ on HPC platforms with out-of-the-box functionality. The development of in situ machine learning methods to detect anomalous events would enable a more efficient and effective workflow, in which all the relevant data are saved in a single simulation run, without re-starts or scientist intervention.

42 ENGINEERING↗