Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Model interpretation”

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 235 records · Page 13

Towards Trustworthy and Interpretable Deep Learning-assisted Ecohydrological Models

The transformational science question we plan to address is: How do we leverage in-situ observations and simulations from process-based ecohydrological model to construct interpretable and trustworthy deep learning (DL) models for improved reliability of prediction of quantities of interest (QoIs) under hydro-climatic extremes?

54 ENVIRONMENTAL SCIENCES↗

Near Infrared Spectroscopy Calibration for Wood Chemistry: Which Chemometric Technique Is Best for Prediction and Interpretation?

This paper addresses the precision in factor loadings during partial least squares (PLS) and principal components regression (PCR) of wood chemistry content from near infrared reflectance (NIR) spectra. The precision of the loadings is considered important because these estimates are often utilized to interpret chemometric models or selection of meaningful wavenumbers. Standard laboratory chemistry methods were employed on a mixed genus/species hardwood sample set. PLS and PCR, before and after 1st derivative pretreatment, was utilized for model building and loadings investigation. As demonstrated by others, PLS was found to provide better predictive diagnostics. However, PCR exhibited a more precise estimate of loading peaks which makes PCR better for interpretation. Application of the 1st derivative appeared to assist in improving both PCR and PLS loading precision, but due to the small sample size, the two chemometric methods could not be compared statistically. This work is important because to date most research works have committed to PLS because it yields better predictive performance. But this research suggests there is a tradeoff between better prediction and model interpretation. Future work is needed to compare PLS and PCR for a suite of spectral pretreatment techniques.

09 BIOMASS FUELS↗

NLIZE: A Perturbation-Driven Visual Interrogation Tool for Analyzing and Interpreting Natural Language Inference Models

With the recent advances in deep learning, neural network models have obtained state-of-the-art performances for many linguistic tasks in natural language processing. However, this rapid progress also brings enormous challenges. The opaque nature of a neural network model leads to hard-to-debug-systems and difficult-to-interpret mechanisms. Here, we introduce a visualization system that, through a tight yet flexible integration between visualization elements and the underlying model, allows a user to interrogate the model by perturbing the input, internal state, and prediction while observing changes in other parts of the pipeline. We use the natural language inference problem to illustrate how a perturbation-driven paradigm can help domain experts assess the potential limitation of a model, probe its inner states, and interpret and form hypotheses about fundamental model mechanisms such as attention.

97 MATHEMATICS AND COMPUTING↗

Interpretation of thermoreflectance measurements with a two-temperature model including non-surface heat deposition

We develop a solution to the two-temperature diffusion equation in axisymmetric cylindrical coordinates to model heat transport in thermoreflectance experiments. Our solution builds upon prior solutions that account for two-channel diffusion in each layer of an N-layered geometry, but adds the ability to deposit heat at any location within each layer. We use this solution to account for non-surface heating in the transducer layer of thermoreflectance experiments that challenge the timescales of electron-phonon coupling. A sensitivity analysis is performed to identify important parameters in the solution and to establish a guideline for when to use the two-temperature model to interpret thermoreflectance data. We then fit broadband frequency domain thermoreflectance (BB-FDTR) measurements of SiO{sub 2} and platinum at a temperature of 300 K with our two-temperature solution to parameterize the gold/chromium transducer layer. We then refit BB-FDTR measurements of silicon and find that accounting for non-equilibrium between electrons and phonons in the gold layer does lessen the previously observed heating frequency dependence reported in Regner et al. [Nat. Commun. 4, 1640 (2013)] but does not completely eliminate it. We perform BB-FDTR experiments on silicon with an aluminum transducer and find limited heating frequency dependence, in agreement with time domain thermoreflectance results. We hypothesize that the discrepancy between thermoreflectance measurements with different transducers results in part from spectrally dependent phonon transmission at the transducer/silicon interface.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

AttentionFire_v1.0: interpretable machine learning fire model for burned-area predictions over tropics

Abstract. African and South American (ASA) wildfires account for more than 70 % of global burned areas and have strong connection to local climate for sub-seasonal to seasonal wildfire dynamics. However, representation of the wildfire–climate relationship remains challenging due to spatiotemporally heterogenous responses of wildfires to climate variability and human influences. Here, we developed an interpretable machine learning (ML) fire model (AttentionFire_v1.0) to resolve the complex controls of climate and human activities on burned areas and to better predict burned areas over ASA regions. Our ML fire model substantially improved predictability of burned areas for both spatial and temporal dynamics compared with five commonly used machine learning models. More importantly, the model revealed strong time-lagged control from climate wetness on the burned areas. The model also predicted that, under a high-emission future climate scenario, the recently observed declines in burned area will reverse in South America in the near future due to climate changes. Our study provides a reliable and interpretable fire model and highlights the importance of lagged wildfire–climate relationships in historical and future predictions.

58 GEOSCIENCES↗

Collaborative Research: Enabling multi-scale studies of magnetic reconnection with interpretable data-driven models

The development of accurate reduced descriptions and improved closures for magnetic reconnection is an important and a long‐standing challenge in plasma physics. The four‐fluid approach, and associated closures, that were investigated have the potential to improve the accuracy of plasma fluid models, capturing physical effects which would otherwise require a kinetic description. If successful, this approach could have an important impact for the modeling of laboratory and space plasmas. The major goals of this project were to develop new machine learning (ML) tools based on sparse and symbolic regression techniques, and to extract interpretable and generalizable reduced models (e.g., in the form of partial differential equations - PDEs) from data generated by first principles plasma simulations. Preserving interpretability of such data‐driven models is key to addressing the long‐standing theoretical and numerical challenges. Prior proof‐of‐principle studies have demonstrated the enormous potential of this approach, by recovering the well‐established hierarchy of plasma equations (from Vlasov to MHD) from data produced by particle‐in‐cell (PIC) simulations. Our goal in this project was to extend and apply these new tools to construct better kinetic closures for magnetic reconnection; to derive better models of particle injection and acceleration by this fundamental plasma process; and to use this understanding to accelerate the development of multi‐scale plasma algorithms. While our immediate focus was on the problem of magnetic reconnection, the tools that were will developed are general and applicable to other areas of plasma physics, and more broadly to many‐body phenomena. We anticipate that the development of these multi‐scale models will have a significant impact across different areas of plasma science, from fusion to space and astrophysical plasmas.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Demystifying Cyberattacks: Potential for Securing Energy Systems With Explainable AI : Preprint

Modernization of energy systems has led to in- creased interactions among multiple critical infrastructures and diverse stakeholders making the challenge of operational decision making more complex and at times beyond cognitive capabilities of human operators. The state-of-the-art machine learning and deep learning approaches show promise of supporting users with complex decision-making challenges, such as those occurring in our rapidly transforming cyber-physical energy systems. However, successful adoption of data-driven decision support technology for critical infrastructure will be dependent on the ability of these technologies to be trustworthy and contextually interpretable. In this paper, we investigate the feasibility of implementing XAI for interpretable detection of cyberattacks in the energy system. Leveraging a proof-of-concept simulation use case of detection of a data falsification attack on a photovoltaic system using XGBoost algorithm, we demonstrate how Local Interpretable Model-Agnostic Explanations (LIME), a flavor XAI approach, can help provide contextual and actionable interpretation of cyberattack detection.

artificial intelligence↗

Single photon simultaneous K-shell ionization and K-shell excitation. I. Theoretical model applied to the interpretation of experimental results on H{sub 2}O

We present in detail a theoretical model that provides absolute cross sections for simultaneous core-ionization core-excitation (K{sup −2}V ) and compare its predictions with experimental results obtained on the water molecule after photoionization by synchrotron radiation. Two resonances of different symmetries are assigned in the main K{sup −2}V peak and comparable contributions from monopolar (direct shake-up) and dipolar (conjugate shake-up) core-valence excitations are identified. The main peak is observed with a much greater width than the total experimental resolution. This broadening is the signature of nuclear dynamics.

74 ATOMIC AND MOLECULAR PHYSICS↗

Comparison of permeability predictions on cemented sandstones with physics-based and machine learning approaches

Permeability prediction has been an important problem since the time of Darcy. Most approaches to solve this problem have used either idealized physical models or empirical relations. In recent years, machine learning (ML) has led to more accurate and robust, but less interpretable empirical models. Using 211 core samples collected from 12 wells in the Garn Sandstone from the North Sea, this study compared idealized physical models based on the Carman-Kozeny equation to interpretable ML models. We found that ML models trained on estimates of physical properties are more accurate than physical models. The results show evidence of a threshold of about 10% volume fraction, above which pore-filling cement strongly affects permeability.

03 NATURAL GAS↗

PuPO4(cr, hyd.) Solubility Product and Pu3+ Complexes With Phosphate and Ethylenediaminetetraacetic Acid

To determine the solubility product of PuPO4(cr, hyd.) and the complexation constants of Pu(III) with phosphate and EDTA, the solubility of PuPO4(cr, hyd.) was investigated as a function of: 1) time and pH varying from 1.0 to 12.0 and at a fixed 0.00032 M phosphate concentration; 2) NaH2PO4 concentrations varying from 0.0001 M to 1.0 M and at a fixed pH value of 2.5; 3) time and pH varying from 1.3 to 13.0 at fixed concentrations of 0.00032 M phosphate and 0.0004 M or 0.002 M Na2H2EDTA; and 4) Na2H2EDTA concentrations varying from 0.00005 M to 0.0256 M at a fixed 0.00032 M phosphate concentration and at pH values of approximately 3.5, 10.6, and 12.6. A combination of solvent extraction and spectrophotometric techniques confirmed that the use of hydroquinone and Na2S2O4 helped maintain Pu as Pu(III). The solubility data were interpreted using Pitzer and SIT models, and both provided similar values for the solubility product of PuPO4(cr, hyd.) and for the formation constant of PuEDTA-. The log10 of the solubility product of PuPO4(cr, hyd.) (PuPO4(cr, hyd.) = Pu3+ + PO4 ) was determined to be –(24.42 ± 0.38). Pitzer modeling showed that phosphate interactions with Pu3+ were extremely weak and did not require any phosphate complexes (e.g., PuPO4(aq), PuH2PO42+, Pu(H2PO4)2+, Pu(H2PO4)3(aq), and Pu(H2PO4)4-), as proposed in existing literature, to explain the experimental data. SIT modeling, however, required the inclusion of PuH2PO42+ to explain the data in high NaH2PO4 concentrations; this illustrates the differences one can expect when using these two chemical models to interpret the data. As the Pu(III)-EDTA species, only PuEDTA- was needed to interpret the experimental data in a large range in pH values (1.3–12.9) and EDTA concentrations (0.00005–0.256 M). Calculations based on density functional theory support the existence of PuEDTA- (with prospective stoichiometry as Pu(OH2)3EDTA-) as the chemically and structurally stable species. The log10 of the complexation constant for the formation of PuEDTA- [Pu3+ + EDTA4- = PuEDTA-] determined in this study is -20.15 ± 0.59. The data also showed that PuHEDTA(aq), Pu(EDTA)25-, Pu(EDTA)(HEDTA)4-, Pu(EDTA)(H2EDTA)3-, and Pu(EDTA)(H3EDTA)2-, reported in the literature, have no region of dominance in the experimental range of variables investigated in this study.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Search for narrow resonances in dilepton mass spectra in proton-proton collisions at $\sqrt{s}$ = 13 TeV and combination with 8 TeV data

A search for narrow resonances in dielectron and dimuon invariant mass spectra has been performed using data obtained from proton-proton collisions at $ \sqrt{s} = $ 13 TeV collected with the CMS detector. The integrated luminosity for the dielectron sample is 2.7 fb$^{-1}$ and for the dimuon sample 2.9 fb$^{-1}$. The sensitivity of the search is increased by combining these data with a previously analysed set of data obtained at $ \sqrt{s} = $ 8 TeV and corresponding to a luminosity of 20 fb$^{-1}$. No evidence for non-standard-model physics is found, either in the 13 TeV data set alone, or in the combined data set. Upper limits on the product of production cross section and branching fraction have also been calculated in a model-independent manner to enable interpretation in models predicting a narrow dielectron or dimuon resonance structure. Limits are set on the masses of hypothetical particles that could appear in new-physics scenarios. For the $\mathrm{Z}'_{\text{SSM}}$ particle, which arises in the sequential standard model, and for the superstring inspired $\mathrm{Z}'_{\psi}$ particle, 95% confidence level lower mass limits for the combined data sets and combined channels are found to be 3.37 and 2.82 TeV, respectively. The corresponding limits for Kaluza-Klein gravitons arising in the Randall-Sundrum model of extra dimensions with coupling parameters 0.01 and 0.10 are 1.46 and 3.11 TeV, respectively. Lastly, these results significantly extend previous limits.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Geophysical Impacts and Spectroscopic Identification of a Hydrous Iron Sulfate on Icy Worlds

Over geologic time-scales, large volumes of exogenic sulfur ions from Io's plasma torus have been supplied to the surface of Europa and Ganymede, which, combined with recent interpretations of orbiter images, dynamical modeling, and surface-subsurface exchange, suggests further sulfur transport into the interior of the icy worlds. These observations motivate mixed-phase spectral modeling for interpreting orbiter spectroscopy data and determination of hydration states of candidate surface materials including hydrous sulfates. In this work, we present a combined experimental and theoretical study of the low temperature and high pressure vibrational spectral signature of the iron-sulfate monohydrate endmember, szomolnokite (FeSO 4 ·H 2 O). By employing synchrotron Fourier-transform infrared spectroscopy (FTIR) in the diamond anvil cell up to 23 GPa and down to 20 K, we explore the extreme range of pressure-temperature domains relevant to icy environments throughout our solar system and beyond. Combined with our density-functional theory quantum-mechanics molecular dynamics results, we demonstrate that experimentally observed infrared features in the O-H stretching region commonly associated with nH 2 O (n > 1) hydration states can be attributed to a pure monohydrate without the need for pressure-induced exsolved ice, other coexisting hydrous iron sulfates, or strong overtone and combination modes. We further discuss the possibility of lateral variations in density and shear properties on icy worlds associated with temperature variations and the high-pressure phases of kieserite group monohydrated sulfates.

Geosciences↗

An attempt to calibrate and validate a simple ductile failure model against axial-torsion experiments on Al 6061-T651

This report details a work in progress. We have attempted to calibrate and validate a Von Mises plasticity model with the Johnson-Cook failure criterion ( Johnson & Cook , 1985 ) against a set of experiments on various specimens of Al 6061-T651. As will be shown, the effort was not successful, despite considerable attention to detail. When the model was com- pared against axial-torsion experiments on tubes, it over predicted failure by 3 x in tension, and never predicted failure in torsion, even when the tube was twisted by 4 x further than the experiment. While this result is unfortunate, it is not surprising. Ductile failure is not well understood. In future work, we will explore whether more sophisticated material mod- els of plasticity and failure will improve the predictions. Selecting the appropriate advanced material model and interpreting the results of said model are not trivial exercises, so it is worthwhile to fully investigate the behavior of a simple plasticity model before moving on to an anisotropic yield surface or a similarly complicated model.

42 ENGINEERING↗

The persistent shadow of the supermassive black hole of M87. II. Model comparisons and theoretical interpretations

The Event Horizon Telescope (EHT) observation of M87∗ in 2018 has revealed a ring with a diameter that is consistent with the 2017 observation. The brightest part of the ring is shifted to the southwest from the southeast. In this paper, we provide theoretical interpretations for the multi-epoch EHT observations for M87∗ by comparing a new general relativistic magnetohydrodynamics model image library with the EHT observations for M87∗ in both 2017 and 2018. The model images include aligned and tilted accretion with parameterized thermal and nonthermal synchrotron emission properties. The 2018 observation again shows that the spin vector of the M87∗ supermassive black hole is pointed away from Earth. A shift of the brightest part of the ring during the multi-epoch observations can naturally be explained by the turbulent nature of black hole accretion, which is supported by the fact that the more turbulent retrograde models can explain the multi-epoch observations better than the prograde models. The EHT data are inconsistent with the tilted models in our model image library. Assuming that the black hole spin axis and its large-scale jet direction are roughly aligned, we expect the brightest part of the ring to be most commonly observed 90 deg clockwise from the forward jet. This prediction can be statistically tested through future observations.

79 ASTRONOMY AND ASTROPHYSICS↗

Data for FUN-PROSE: A Deep Learning Approach to Predict Condition-Specific Gene Expression in Fungi

mRNA levels of all genes in a genome is a critical piece of information defining the overall state of the cell in a given environmental condition. Being able to reconstruct such condition-specific expression in fungal genomes is particularly important to metabolically engineer these organisms to produce desired chemicals in industrially scalable conditions. Most previous deep learning approaches focused on predicting the average expression levels of a gene based on its promoter sequence, ignoring its variation across different conditions. Here we present FUN-PROSE—a deep learning model trained to predict differential expression of individual genes across various conditions using their promoter sequences and expression levels of all transcription factors. We train and test our model on three fungal species and get the correlation between predicted and observed condition-specific gene expression as high as 0.85. We then interpret our model to extract promoter sequence motifs responsible for variable expression of individual genes. We also carried out input feature importance analysis to connect individual transcription factors to their gene targets. A sizeable fraction of both sequence motifs and TF-gene interactions learned by our model agree with previously known biological information, while the rest corresponds to either novel biological facts or indirect correlations.

Genomics↗

Nuclear Power's Future Role in a Decarbonized U.S. Electricity System

This study explores the potential future role of nuclear energy in a decarbonized U.S. electricity system through a multi-model comparison approach. We employ four state-of-the-art CEMs with native and harmonized input assumptions, layered with different policy and technology trajectories. Comparing outputs across models, technology assumptions, and policy scenarios informs model understanding, interpretation, and development decisions. Under current policies, models differ in their projections for nuclear retirements, but nuclear power plants consistently run with high capacity factors and new builds only occur in scenarios with very low nuclear costs. String power sector carbon policies drive models to align in keeping existing nuclear capacity and employing nuclear plant flexibility, but they may not be enough to bring new nuclear capacity online in the absence of significant cost declines. Therefore, significant economic deployment of new nuclear capacity requires both a stringent electric sector CO2 policy and very low cost assumptions for new nuclear. While these scenarios should not be interpreted as predictions, they are informative for understanding differing model assessments of the relative competitiveness of nuclear energy under a range of policy and technology conditions.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Variogram Identification Aided by a Structural Framework for Improved Geometric Modeling of Faulted Reservoirs: Jeffara Basin, Southeastern Tunisia

This article describes a proposed work-sequence to generate accurate reservoir-architecture models, describing the geometry of bounding surfaces (i.e., fault locations and extents), of a structurally complex geologic setting in the Jeffara Basin (South East Tunisia) by means of geostatistical modeling. This uses the variogram as the main tool to measure the spatial variability of the studied geologic medium before making any estimation or simulation. However, it is not always easy to fit complex experimental variograms to theoretical models. Thus, our primary purpose was to establish a relationship between the geology and the components of the variograms to fit a mathematically consistent and geologically interpretable variogram model for improved predictions of surface geometries. We used a three-step approach based on available well data and seismic information. First, we determined the structural framework: a seismo-tectonic data analysis was carried out, and we showed that the study area is cut mainly by NW-SE-trending normal faults, which were classified according to geometric criteria (strike, throw magnitude, dip, and dip direction). We showed that these normal faults are at the origin of a large-scale trend structure (surfaces tilted toward the north-east). At a smaller scale, the normal faults create a distinct compartmentalization of the reservoirs. Then, a model of the reservoir system architecture was built by geostatistical methods. An efficient methodology was developed, to estimate the bounding faulted surfaces of the reservoir units. Emphasis was placed on (i) elaborating a methodology for variogram interpretation and modeling, whereby the importance of each variogram component is assessed in terms of probably geologic factor controlling the behavior of each structure; (ii) integrating the relevant fault characteristics, which were deduced from the previous fault classification analysis, as constraints in the kriging estimation of bounding surfaces to best reflect the geologic structure of the study area. Finally, the estimated bounding surfaces together with seismic data and variogram interpretations were used to obtain further insights into the tectonic evolution of the study area that has induced the current reservoirs configuration.

58 GEOSCIENCES↗