Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “model structural error”

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 OSTI2014

Real-time out-of-plane artifact subtraction tomosynthesis imaging using prior CT for scanning beam digital x-ray system

Purpose: The scanning beam digital x-ray system (SBDX) is an inverse geometry fluoroscopic system with high dose efficiency and the ability to perform continuous real-time tomosynthesis in multiple planes. This system could be used for image guidance during lung nodule biopsy. However, the reconstructed images suffer from strong out-of-plane artifact due to the small tomographic angle of the system. Methods: The authors propose an out-of-plane artifact subtraction tomosynthesis (OPAST) algorithm that utilizes a prior CT volume to augment the run-time image processing. A blur-and-add (BAA) analytical model, derived from the project-to-backproject physical model, permits the generation of tomosynthesis images that are a good approximation to the shift-and-add (SAA) reconstructed image. A computationally practical algorithm is proposed to simulate images and out-of-plane artifacts from patient-specific prior CT volumes using the BAA model. A 3D image registration algorithm to align the simulated and reconstructed images is described. The accuracy of the BAA analytical model and the OPAST algorithm was evaluated using three lung cancer patients’ CT data. The OPAST and image registration algorithms were also tested with added nonrigid respiratory motions. Results: Image similarity measurements, including the correlation coefficient, mean squared error, and structural similarity index, indicated that the BAA model is very accurate in simulating the SAA images from the prior CT for the SBDX system. The shift-variant effect of the BAA model can be ignored when the shifts between SBDX images and CT volumes are within ±10 mm in the x and y directions. The nodule visibility and depth resolution are improved by subtracting simulated artifacts from the reconstructions. The image registration and OPAST are robust in the presence of added respiratory motions. The dominant artifacts in the subtraction images are caused by the mismatches between the real object and the prior CT volume. Conclusions: Their proposed prior CT-augmented OPAST reconstruction algorithm improves lung nodule visibility and depth resolution for the SBDX system.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2020

Influence of sun zenith angle on canopy clumping and the resulting impacts on photosynthesis

Addressing the impact of vegetation architecture on shortwave radiation transfer in land surface models is important for accurate weather forecasting, carbon budget estimates, and climate predictions. This paper investigates to what extent it is possible to retrieve structural parameters of two different parameterization schemes from direct transmittance derived from digital hemispherical photography and 3D radiative transfer modeling for two study sites with different vegetation canopy architectures. Neglecting the representation of 3D canopy structure in radiative transfer schemes leads to significant errors in shortwave radiation partitioning (up to 3.5 times more direct transmittance in the 3D model). Structural parameters, referred to as whole canopy ‘clumping indices’, were obtained in order to evaluate the impact of angular variation in clumping on shortwave radiation transfer. Impacts on photosynthesis were evaluated at site level with the UKESM land surface model, JULES. A comparison between flux tower derived and modeled photosynthesis indicates that considering zenith angular variations of structural parameters in the radiative transfer scheme of the UKESM land surface model significantly improves photosynthesis prediction in light limited ecosystems (from RMSE = 2.91 μmol CO 2 .m -2 .s -1 to RMSE = 1.51 μmol CO 2 .m -2 .s -1 , 48% smaller), typically with enhanced photosynthesis from bottom layers.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2016

WE-AB-209-02: A New Inverse Planning Framework with Principle-Based Modeling of Inter-Structural Dosimetric Tradeoffs

Purpose: Traditional radiotherapy inverse planning relies on the weighting factors to phenomenologically balance the conflicting criteria for different structures. The resulting manual trial-and-error determination of the weights has long been recognized as the most time-consuming part of treatment planning. The purpose of this work is to develop an inverse planning framework that parameterizes the inter-structural dosimetric tradeoff among with physically more meaningful quantities to simplify the search for a clinically sensible plan. Methods: A permissible dosimetric uncertainty is introduced for each of the structures to balance their conflicting dosimetric requirements. The inverse planning is then formulated as a convex feasibility problem, which aims to generate plans with acceptable dosimetric uncertainties. A sequential procedure (SP) is derived to decompose the model into three submodels to constrain the uncertainty in the planning target volume (PTV), the critical structures, and all other structures to spare, sequentially. The proposed technique is applied to plan a liver case and a head-and-neck case and compared with a conventional approach. Results: Our results show that the strategy is able to generate clinically sensible plans with little trial-and-error. In the case of liver IMRT, the fractional volumes to liver and heart above 20Gy are found to be 22% and 10%, respectively, which are 15.1% and 33.3% lower than that of the counterpart conventional plan while maintaining the same PTV coverage. The planning of the head and neck IMRT show the same level of success, with the DVHs for all organs at risk and PTV very competitive to a counterpart plan. Conclusion: A new inverse planning framework has been established. With physically more meaningful modeling of the inter-structural tradeoff, the technique enables us to substantially reduce the need for trial-and-error adjustment of the model parameters and opens new opportunities of incorporating prior knowledge to facilitate the treatment planning process.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2026

3D Deep Learning Joint Inversion of Active Seismic Full Waveform and Passive Seismic Traveltime Data for Reservoir Imaging and Uncertainty Quantification

Here, we present deep learning (DL) networks for three-dimensional (3D) joint inversion of active seismic full waveform and passive seismic traveltime data to image reservoirs and their properties and quantify imaging uncertainties. Active seismic full-waveform data can provide high-resolution monitoring images but are collected only intermittently because of their high acquisition cost. In contrast, passive seismic data can be gathered at relatively low cost between regular active surveys, although their imaging quality can be compromised by factors such as low signal-to-noise ratios and limited ray coverage of the target. Although these datasets are routinely acquired together at CO 2 storage sites, their combined inversion within a 3D DL framework has not been previously demonstrated. To our knowledge, this is the first study to address this gap, combining the strength of both data types. For efficient data storage and DL training with large 3D seismic datasets, we use a 3D data matrix in which a random number of passive seismic traveltime data are stored as parabolic envelopes using one-hot encoding and a 3D full-waveform data matrix in which multiple shot gathers are summed. Two network architectures are evaluated: a single-encoder U-Net for single-data type inversion and a dual-encoder U-Net for joint inversion of active and passive seismic data. We also evaluate the single-encoder U-Net for joint inversion by concatenating full-waveform data and traveltime data. We propose a systematic approach for selecting an optimal dropout rate that balances regularization during training and Monte Carlo dropout-based uncertainty quantification during prediction by examining the correlation coefficient between standard deviation and prediction error, along with the training misfit, across a range of dropout rates. 3D DL inversion experiments include five different network configurations, with evaluations under ideal, noisy and dropout-enabled conditions. Both model and data uncertainties are assessed, as well as their combined effects. Across all conditions, the networks consistently predict accurate CO 2 saturation models with low prediction errors, such as a structural similarity index of 0.993 and CO 2 difference of 1.1%. Uncertainty estimates show strong spatial correlation with prediction errors, confirming the effectiveness of the proposed dropout selection approach. The results demonstrate that our DL approach, utilizing compact data representations and appropriate uncertainty quantification, yields accurate subsurface images under various inversion conditions and provides valuable insights into the reliability of predictions.

Um, Evan Schankee [Lawrence Berkeley National Labo↗
DOE OSTI2014

The Sloan Digital Sky Survey COADD: 275 deg{sup 2} of deep Sloan Digital Sky Survey imaging on stripe 82

We present details of the construction and characterization of the coaddition of the Sloan Digital Sky Survey (SDSS) Stripe 82 ugriz imaging data. This survey consists of 275 deg{sup 2} of repeated scanning by the SDSS camera over –50° ≤ α ≤ 60° and –1.°25 ≤ δ ≤ +1.°25 centered on the Celestial Equator. Each piece of sky has ∼20 runs contributing and thus reaches ∼2 mag fainter than the SDSS single pass data, i.e., to r ∼ 23.5 for galaxies. We discuss the image processing of the coaddition, the modeling of the point-spread function (PSF), the calibration, and the production of standard SDSS catalogs. The data have an r-band median seeing of 1.''1 and are calibrated to ≤1%. Star color-color, number counts, and PSF size versus modeled size plots show that the modeling of the PSF is good enough for precision five-band photometry. Structure in the PSF model versus magnitude plot indicates minor PSF modeling errors, leading to misclassification of stars as galaxies, as verified using VVDS spectroscopy. There are a variety of uses for this wide-angle deep imaging data, including galactic structure, photometric redshift computation, cluster finding and cross wavelength measurements, weak lensing cluster mass calibrations, and cosmic shear measurements.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Parallel transport dynamics for mixed quantum states with applications to time-dependent density functional theory

Direct simulation of the von Neumann dynamics for a general (pure or mixed) quantum state can often be expensive. One prominent example is the real-time time-dependent density functional theory (rt-TDDFT), a widely used framework for the first principle description of many-electron dynamics in chemical and materials systems. Practical rt-TDDFT calculations often avoid the direct simulation of the von Neumann equation, and solve instead a set of Schrödinger equations, of which the dynamics is equivalent to that of the von Neumann equation. However, the time step size employed by the Schrödinger dynamics is often much smaller. Here, in order to improve the time step size and the overall efficiency of the simulation, we generalize a recent work of the parallel transport (PT) dynamics for simulating pure states [An, Lin, Multiscale Model. Simul. 18, 612, 2020] to general quantum states. The PT dynamics provides the optimal gauge choice, and can employ a time step size comparable to that of the von Neumann dynamics. Going beyond the linear and near adiabatic regime in previous studies, we find that the error of the PT dynamics can be bounded by certain commutators between Hamiltonians, density matrices, and their derived quantities. Such a commutator structure is not present in the Schrödinger dynamics. We demonstrate that the parallel transport-implicit midpoint (PT-IM) method is a suitable method for simulating the PT dynamics, especially when the spectral radius of the Hamiltonian is large. The commutator structure of the error bound, and numerical results for model rt-TDDFT calculations in both linear and nonlinear regimes, confirm the advantage of the PT dynamics.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

Towards robust autonomous impedance spectroscopy analysis: A calibrated hierarchical Bayesian approach for electrochemical impedance spectroscopy (EIS) inversion

Distribution-based analyses, such as the distribution of relaxation times (DRT) and the distribution of diffusion times (DDT), present model-free alternatives to equivalent circuit modeling for analysis of electrochemical impedance spectroscopy (EIS) data. However, reconstructing such distributions from noisy impedance data is an ill-posed problem that must be solved with specialized inversion algorithms, requiring careful control and tuning. Furthermore, most inversion algorithms developed to date can only solve problems of limited complexity. Herein, we present a new hierarchical Bayesian method for EIS inversion, leveraging efficient algorithms for optimization and Hamiltonian Monte Carlo (HMC) sampling to solve models of arbitrary complexity. We overcome the challenge of ad-hoc parameter tuning by encoding intrinsic characteristics of the DRT and DDT into flexible prior distributions and “pre-calibrating” the model to simulated data. This approach is versatile, highly robust to noise, and provides quantitative estimates of both the error structure of the data and the uncertainty in the recovered distributions. The model is validated with simulated data to demonstrate accurate recovery of the DRT and the DDT. The method also shows promise for simultaneous recovery of multiple distributions, raising the intriguing possibility of semi-autonomous EIS analysis and ad-hoc model construction. Finally, the practical utility of the method is illustrated with experimental data. Throughout, we draw comparisons to several recently published EIS inversion methodologies.

36 MATERIALS SCIENCE↗
DOE OSTI2016

Quantifying the impact of material-model error on macroscale quantities-of-interest using multiscale a posteriori error-estimation techniques

An a posteriori error-estimation framework is introduced to quantify and reduce modeling errors resulting from approximating complex mesoscale material behavior with a simpler macroscale model. Such errors may be prevalent when modeling welds and additively manufactured structures, where spatial variations and material textures may be present in the microstructure. We consider a case where a <100> fiber texture develops in the longitudinal scanning direction of a weld. Transversely isotropic elastic properties are obtained through homogenization of a microstructural model with this texture and are considered the reference weld properties within the error-estimation framework. Conversely, isotropic elastic properties are considered approximate weld properties since they contain no representation of texture. Errors introduced by using isotropic material properties to represent a weld are assessed through a quantified error bound in the elastic regime. Lastly, an adaptive error reduction scheme is used to determine the optimal spatial variation of the isotropic weld properties to reduce the error bound.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Applications of the Dulmage–Mendelsohn decomposition for debugging nonlinear optimization problems

Nonlinear modeling and optimization is a valuable tool for aiding decisions by engineering practitioners, but programming an optimization problem based on a complex electrical, mechanical, or chemical process is a time-consuming and error-prone activity. Therefore, there is a need for model analysis and debugging tools that can detect and diagnose modeling errors. One such tool is the Dulmage–Mendelsohn decomposition, which identifies structurally under- and over-determined subsets in systems of equations and variables by partitioning the bipartite graph of the system. This work provides the necessary background to understand the Dulmage–Mendelsohn decomposition and its application to the analysis of nonlinear optimization problems, demonstrates its use in diagnosing a variety of modeling errors, and introduces software implementations for analyzing nonlinear optimization problems in the Pyomo and JuMP algebraic modeling languages.

42 ENGINEERING↗
DOE OSTI2018

An integrated phenology modelling framework in r

Abstract Phenology is a first‐order control on productivity and mediates the biophysical environment by altering albedo, surface roughness length and evapotranspiration. Accurate and transparent modelling of vegetation phenology is therefore key in understanding feedbacks between the biosphere and the climate system. Here, we present the phenor r package and modelling framework. The framework leverages measurements of vegetation phenology from four common phenology observation datasets, the PhenoCam network, the USA National Phenology Network ( USA ‐ NPN ), the Pan European Phenology Project ( PEP 725), MODIS phenology ( MCD 12Q2) combined with (global) retrospective and projected climate data. We show an example analysis, using the phenor modelling framework, which quickly and easily compares 20 included spring phenology models for three plant functional types. An analysis of model skill using the root mean squared ( RMSE ) error shows little or no difference regardless of model structure, corroborating previous studies. We argue that addressing this issue will require novel model development combined with easy data assimilation as facilitated by our framework. In conclusion, we hope the phenor phenology modelling framework in the r language and environment for statistical computing will facilitate reproducibility and community driven phenology model development, in order to increase their overall predictive power, and leverage an ever growing number of phenology data products.

Hufkens, Koen↗
DOE OSTI2024

Structure Sensitive Reaction Kinetics of Chiral Molecules on Intrinsically Chiral Surfaces

Enantiospecific heterogeneous catalysis utilizes chiral surfaces to resolve enantiomers via structure sensitive surface chemistry. The catalyst design challenge is the identification of chiral surface structures that maximize enantiospecificity. Herein, we develop data driven models for the enantiospecificity of tartaric acid reactions on chiral Cu(hkl) R&S surfaces. Measurements of enantiospecific rate constants were obtained by using curved Cu(hkl) R&S surfaces that enable kinetic measurements on hundreds of chiral surface orientations. One model uses feature vectors derived from generalized coordination numbers to capture the local structure around Cu atoms exposed by the Cu(hkl) R&S surfaces. The second model introduces the use of chiral cubic harmonic functions to capture the symmetry constraints of the face-centered cubic Cu structure. The model using 58 generalized coordination numbers has a fitting error similar to that of the model using only 5 cubic harmonic functions. The two models predict maxima in the enantiospecificity on surfaces with very similar surface orientations. The models developed in this work are applicable for any enantiospecific reaction happening on any chiral material with a cubic lattice structure, opening the way to understanding the surface structure sensitivity of the enantiospecific reaction kinetics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2013

On Optimal Input Design and Model Selection for Communication Channels

In this paper, the optimal model (structure) selection and input design which minimize the worst case identification error for communication systems are provided. The problem is formulated using metric complexity theory in a Hilbert space setting. It is pointed out that model selection and input design can be handled independently. Kolmogorov n-width is used to characterize the representation error introduced by model selection, while Gel fand and Time n-widths are used to represent the inherent error introduced by input design. After the model is selected, an optimal input which minimizes the worst case identification error is shown to exist. In particular, it is proven that the optimal model for reducing the representation error is a Finite Impulse Response (FIR) model, and the optimal input is an impulse at the start of the observation interval. FIR models are widely popular in communication systems, such as, in Orthogonal Frequency Division Multiplexing (OFDM) systems.

fast identification↗
DOE OSTI2015

Automated contouring error detection based on supervised geometric attribute distribution models for radiation therapy: A general strategy

Purpose: One of the most critical steps in radiation therapy treatment is accurate tumor and critical organ-at-risk (OAR) contouring. Both manual and automated contouring processes are prone to errors and to a large degree of inter- and intraobserver variability. These are often due to the limitations of imaging techniques in visualizing human anatomy as well as to inherent anatomical variability among individuals. Physicians/physicists have to reverify all the radiation therapy contours of every patient before using them for treatment planning, which is tedious, laborious, and still not an error-free process. In this study, the authors developed a general strategy based on novel geometric attribute distribution (GAD) models to automatically detect radiation therapy OAR contouring errors and facilitate the current clinical workflow. Methods: Considering the radiation therapy structures’ geometric attributes (centroid, volume, and shape), the spatial relationship of neighboring structures, as well as anatomical similarity of individual contours among patients, the authors established GAD models to characterize the interstructural centroid and volume variations, and the intrastructural shape variations of each individual structure. The GAD models are scalable and deformable, and constrained by their respective principal attribute variations calculated from training sets with verified OAR contours. A new iterative weighted GAD model-fitting algorithm was developed for contouring error detection. Receiver operating characteristic (ROC) analysis was employed in a unique way to optimize the model parameters to satisfy clinical requirements. A total of forty-four head-and-neck patient cases, each of which includes nine critical OAR contours, were utilized to demonstrate the proposed strategy. Twenty-nine out of these forty-four patient cases were utilized to train the inter- and intrastructural GAD models. These training data and the remaining fifteen testing data sets were separately employed to test the effectiveness of the proposed contouring error detection strategy. Results: An evaluation tool was implemented to illustrate how the proposed strategy automatically detects the radiation therapy contouring errors for a given patient and provides 3D graphical visualization of error detection results as well. The contouring error detection results were achieved with an average sensitivity of 0.954/0.906 and an average specificity of 0.901/0.909 on the centroid/volume related contouring errors of all the tested samples. As for the detection results on structural shape related contouring errors, an average sensitivity of 0.816 and an average specificity of 0.94 on all the tested samples were obtained. The promising results indicated the feasibility of the proposed strategy for the detection of contouring errors with low false detection rate. Conclusions: The proposed strategy can reliably identify contouring errors based upon inter- and intrastructural constraints derived from clinically approved contours. It holds great potential for improving the radiation therapy workflow. ROC and box plot analyses allow for analytically tuning of the system parameters to satisfy clinical requirements. Future work will focus on the improvement of strategy reliability by utilizing more training sets and additional geometric attribute constraints.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2016

The use of noncrystallographic symmetry averaging to solve structures from data affected by perfect hemihedral twinning

Molecular replacement and noncrystallographic symmetry averaging were used to detwin a data set affected by perfect hemihedral twinning. The noncrystallographic symmetry averaging of the electron-density map corrected errors in the detwinning introduced by the differences between the molecular-replacement model and the crystallized structure. Hemihedral twinning is a crystal-growth anomaly in which a specimen is composed of two crystal domains that coincide with each other in three dimensions. However, the orientations of the crystal lattices in the two domains differ in a specific way. In diffraction data collected from hemihedrally twinned crystals, each observed intensity contains contributions from both of the domains. With perfect hemihedral twinning, the two domains have the same volumes and the observed intensities do not contain sufficient information to detwin the data. Here, the use of molecular replacement and of noncrystallographic symmetry (NCS) averaging to detwin a 2.1 Å resolution data set for Aichi virus 1 affected by perfect hemihedral twinning is described. The NCS averaging enabled the correction of errors in the detwinning introduced by the differences between the molecular-replacement model and the crystallized structure. The procedure permitted the structure to be determined from a molecular-replacement model that had 16% sequence identity and a 1.6 Å r.m.s.d. for C{sup α} atoms in comparison to the crystallized structure. The same approach could be used to solve other data sets affected by perfect hemihedral twinning from crystals with NCS.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2020

Calibration, measurement, and characterization of soil moisture dynamics in a central Amazonian tropical forest

Soil moisture plays a key role in hydrological, biogeochemical, and energy budgets of terrestrial ecosystems. Accurate soil moisture measurements in remote ecosystems such as the Amazon are difficult and limited because of logistical constraints. Time domain reflectometry (TDR) sensors are widely used to monitor soil moisture and require calibration to convert the TDR's dielectric permittivity measurement (K a ) to volumetric water content (θ v ). In this study, our objectives were to develop a field-based calibration of TDR sensors in an old-growth upland forest in the central Amazon, to evaluate the performance of the calibration, and then to apply the calibration to determine the dynamics of soil moisture content within a 14.2 -m-deep vertical soil profile. Depth-specific TDR calibration using local soils in a controlled laboratory setting yielded a novel K a –θ v third-degree polynomial calibration. The sensors were later installed to their specific calibration depth in a 14.2-m pit. The widely used K a –θ v relationship (Topp model) underestimated the site-specific θ v by 22–42%, indicating significant error in the model when applied to these well-structured, clay-rich tropical forest soils. The calibrated wet- and dry-season θ v data showed a variety of depth and temporal variations highlighting the importance of soil textural differentiation, root uptake depths, as well as event to seasonal precipitation effects. Data such as these are greatly needed for improving our understanding of ecohydrological processes within tropical forests and for improving models of these systems in the face of changing environmental conditions.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2018

Uncertainty quantification for incident helium flux in plasma-exposed tungsten

In this work, the surface response of a tungsten plasma-facing component was simulated by a cluster-dynamics code, Xolotl, with a focus on quantifying the impact of uncertainty in one of the input parameters to Xolotl, namely, the incident helium flux. The simulated conditions involve a tungsten surface exposed to 100 eV helium ion implantations with a flux of either 4 × 10 22 or 4 × 10 25 He m -2 s -1 . Furthermore, two sources were used to describe the implanted helium depth distribution in tungsten, either molecular dynamics (MD) or a binary collision approximation code, the stopping and range of ions in matter (SRIM). The aim of this work is to evaluate and examine uncertain predictions on the helium retention based on these two different modeling methodologies that either neglect electronic energy loss or the crystalline structure of the solid, respectively. An embedded model-form error approach was pursued here in order to arrive at predictions that account for variability due to the two different data sources, and the impact of this model-form uncertainty in incident helium flux on Xolotl output was presented for the two implantation fluxes.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2025

MS25: Materials Science-Focused Benchmark Data Set for Machine Learning Interatomic Potentials

Here, we present MS25, a benchmark data set for evaluating machine learning interatomic potentials (MLIPs) across diverse materials-relevant systems including MgO surfaces, liquid water, zeolites, a catalytic Pt surface reaction, high-entropy alloys (HEAs), and disordered Zr-oxides. Five MLIP architectures (MACE, NequIP, Allegro, MTP, and Torch-ANI) are trained and tested, focusing not only on traditional metrics (energies, forces, and stresses) but also explicitly validating derived physical observables such as lattice constants, volumes, and reaction barriers. We find that most models reach comparable accuracy on standard error metrics across the simple systems, although equivariant MLIPs offer 1.5–2× improvements over nonequivariant MLIPs in energy and force error for structurally complex or compositionally disordered environments such as HEAs and Zr–O systems. Our analysis highlights that low errors in energy and force predictions do not guarantee reliable observables, emphasizing the necessity of explicit validation. We demonstrate limitations in cross-framework transferability, as models trained on one zeolite framework (CHA) fail to reliably generalize to predictions of structurally distinct frameworks (e.g., MFI). Size-extensive tests show some dependence on system size for MgO, resulting from forced periodicity. The HEA and Zr–O data sets are identified as challenging tests for future benchmarks and MLIP model architecture developments as they show significant differentiation in error between MLIP architectures and are still relatively difficult at 1000 training images. Moving forward, we recommend that benchmarking efforts shift their focus from marginal accuracy improvements in energy and force errors toward identifying and understanding model failure modes, rigorously assessing transferability, and evaluating how their errors affect observable predictions. For researchers looking to choose an MLIP architecture, we suggest selecting equivariant MLIP architectures if the complexity of the system is a challenge. For simple materials problems, auxiliary features such as integration with molecular dynamics engines, trade-offs between computational data set generation cost vs MLIP inference speed, and framework integration may play a more important decision factor than small differences in error metrics that are unlikely to matter for production-level research.

chemical structure↗
DOE OSTI2021

AtomSets as a hierarchical transfer learning framework for small and large materials datasets

Abstract Predicting properties from a material’s composition or structure is of great interest for materials design. Deep learning has recently garnered considerable interest in materials predictive tasks with low model errors when dealing with large materials data. However, deep learning models suffer in the small data regime that is common in materials science. Here we develop the AtomSets framework, which utilizes universal compositional and structural descriptors extracted from pre-trained graph network deep learning models with standard multi-layer perceptrons to achieve consistently high model accuracy for both small compositional data (<400) and large structural data (>130,000). The AtomSets models show lower errors than the graph network models at small data limits and other non-deep-learning models at large data limits. They also transfer better in a simulated materials discovery process where the targeted materials have property values out of the training data limits. The models require minimal domain knowledge inputs and are free from feature engineering. The presented AtomSets model framework can potentially accelerate machine learning-assisted materials design and discovery with less data restriction.

Chen, Chi (ORCID:0000000180087043)↗