Search NASA⌕ Search

SEARCH · Search NASA

Results for “model interpretability”

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 55 records · Page 3

Model independent interpretation of recent CR lepton data after AMS-02

We model the CR leptonic fluxes above 20 GeV in terms of a superposition of a standard and a charge symmetric extra component, which we generically describe as power-laws in momentum. We investigate under these hypotheses the compatibility between AMS-02, Fermi-LAT and PAMELA datasets on positron fraction, electron+positron spectrum and electron spectrum respectively. We find that it is possible to reconcile AMS and Fermi-LAT data within uncertainties, if energy-dependent effects are present in Fermi-LAT systematics. We also study possible deviations from charge symmetry in the extra component and find no compelling evidence for them.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

The value of soil respiration measurements for interpreting and modeling terrestrial carbon cycling

A recent acceleration of model-data synthesis activities has leveraged many terrestrial carbon (C) datasets, but utilization of soil respiration (RS) data has not kept pace with other types such as eddy covariance (EC) fluxes and soil C stocks. Here we argue that RS data, including non-continuous measurements from survey sampling campaigns, have unrealized value and should be utilized more extensively and creatively in data synthesis and modeling activities. We identify three major challenges in interpreting RS data, and discuss opportunities to address them. The first challenge is that when RS is compared to ecosystem respiration (RECO) measured from EC towers, it is not uncommon to find substantial mismatch, indicating one or both flux methodologies are unreliable. We argue the most likely cause of mismatch is unreliable EC data, and there is an unrecognized opportunity to utilize RS for EC quality control. The second challenge is that RS integrates belowground heterotrophic (RH) and autotrophic (RA) activity, whereas modelers generally prefer partitioned fluxes, and few models include an explicit RS output. Opportunities exist to use the total RS flux for data assimilation and model benchmarking methods rather than less-certain partitioned fluxes. Pushing for more experiments that not only partition RS but also monitor the age of RA and RH, as well as for the development of belowground RA components in models, would allow for more direct comparison between measured and modeled values. The third challenge is that soil respiration is generally measured at a very different resolution than that needed for comparison to EC or ecosystem- to global-scale models. Measuring soil fluxes with finer spatial resolution and more extensive coverage, and downscaling EC fluxes to match the scale of RS, will improve chamber and tower comparisons. Opportunities also exist to estimate RH at regional scales by implementing decomposition functional types, akin to plant functional types. We conclude by discussing the benefits that wide use of RS data will bring to model development, and database developments that will make RS data more robust, useful, and broadly available to the research community.

carbon cycle↗

Machine Learning Augmented Predictive and Generative Model for Rupture Life in Ferritic and Austenitic Steels

The Larson-Miller parameter (LMP) offers an efficient and fast scheme to estimate the creep rupture life of alloy materials for high temperature applications. However, owing to poor generalizability and dependence on the constant C, which is typically not known a-priori, estimations using the Larson-Miller parameter often result in suboptimal performance for a wide range of materials. At best it is useful in comparing alloys of similar composition. In this work, three machine learning (ML) schemes were developed for rupture life prediction for 9-12% Cr ferritic-martensitic steels and austenitic stainless steels, i.e., a hierarchical model to parameterize LMP using the LMP constant C to compute rupture life, a hierarchical model to parameterize both C and LMP to compute rupture life, and a direct prediction of rupture life. Specifically, we show that the third scheme, using a gradient boosting algorithm, can be used to train ML models for very accurate prediction of rupture life in a variety of alloys (Pear-son Correlation Coefficient > 0.9 for 9-12% Cr and > 0.8 for austenitic stainless steels). In addition, the Shapley value was used to quantify feature importance, making the model interpretable by identifying the effect of various features on the model performance. Furthermore, a variational autoencoder-based generative model was built by conditioning on the experimental dataset to sample hypothetical synthetic candidate alloys from the learnt joint distribution not existing in both 9-12 % Cr ferritic-martensitic steel and austenitic stainless steel datasets. Finally, based on the predictive and generative model, a reinforcement learning strategy has been proposed to guide experimentalists into designing better heat resistant alloys.

Mamun, Md Osman G.↗

Accelerated Materials Design for Molten Salt Technologies Using Innovative High-Throughput Methods

The focus of the project is on building an innovative accelerated materials design platform for molten salt technologies using novel high-throughput methods coupled to data analytics. The main objectives is to predict a FeCrMnNi alloy compositional space with better corrosion resistance than stainless steel 316 and identify new molten salt corrosion mechanisms. The project demonstrates the feasibility to use high-throughput methods coupled to data analytics to accelerate alloy design for extreme environments applications. Using a trained and tested machine learning (ML) model, 2000 FeCrMnNi alloy corrosion rate in molten chloride salts were predicted and a compositional field with corrosion rate lower than 316 stainless steel was identified. The ML model interpretability unveiled multiple features of importance in the model prediction. Some features were expected to be of relative significance, such as work function, surface energy and alloy electronegativity, and the ML model interpretability analysis confirmed those. On the other hand, the most important feature is the diffusion coefficient of Ni in the bulk alloy which indicates that a surface diffusion mechanism plays an important role in the overall corrosion mechanism in molten salts.

36 MATERIALS SCIENCE↗

Interpretable Machine Learning for Characterizing Electric Vehicle Charging Behavior: Insights from Real-World Data

As electric vehicle (EV) adoption rises globally, concerns about the impact on aging electrical grids grow, particularly regarding the charging behavior of EV drivers. This study analyzes real-world driving and charging data from Ford battery electric vehicles (BEVs) collected between 2018 and 2019 to develop interpretable models that characterize charging behavior and quantify influencing factors. Prior research has relied on assumptions regarding driver behavior, often overlooking actual charging patterns. By employing generalized linear mixed models (GLMMs), this work offers insights into how various elements, such as next trip distance and state of charge (SOC), influence charging decisions. The dataset comprises over three million park-trip pairs from 1,997 vehicles, revealing that features related to driving behavior significantly dictate charging behavior, while infrastructure and regional factors have lesser impacts. The findings suggest that existing simulation models may oversimplify EV charging behavior assumptions. This work utilizes real-world EV driving and charging data to train interpretable models that describe charging behavior and quantify the factors most associated with how drivers use charging infrastructure. This research underscores the need for interpretable, data-driven methodologies to inform future EV infrastructure planning and grid management.

29 - ENERGY PLANNING, POLICY AND ECONOMY↗

Interpretive MHD modeling of dispersive shell pellet injection for rapid shutdown in tokamaks

Dispersive shell pellet (DSP) injection is modeled with the extended-MHD code NIMROD for interpretive insight into the results of recent DIII-D DSP experiments and to explore the dynamics of an inside-out thermal quench for disruption mitigation in tokamaks. Simulations of the pre-thermal quench (TQ) phase indicate that the upper bound for the quantity of ablated carbon shell material that will not perturb the flux surfaces is in the ballpark of, but somewhat below the experimental quantity. Even below this quantity, sufficient electrons are added to the plasma by the shell material to produce significant dilution cooling before the TQ is triggered. Simulations carried through the end of the TQ have very large amplitude MHD fluctuations (δB/B>10 -2 ) at the time of the plasma current spike associated with current profile redistribution. Finally, after the plasma current spike, which is of comparable amplitude to that measured in DIII-D experiments, none of the runaway electron test-particles whose orbits are tracked throughout the simulation remain confined.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A SEMI-ANALYTICAL LINE TRANSFER MODEL TO INTERPRET THE SPECTRA OF GALAXY OUTFLOWS

We present a semi-analytical line transfer model, (SALT), to study the absorption and re-emission line profiles from expanding galactic envelopes. The envelopes are described as a superposition of shells with density and velocity varying with the distance from the center. We adopt the Sobolev approximation to describe the interaction between the photons escaping from each shell and the remainder of the envelope. We include the effect of multiple scatterings within each shell, properly accounting for the atomic structure of the scattering ions. We also account for the effect of a finite circular aperture on actual observations. For equal geometries and density distributions, our models reproduce the main features of the profiles generated with more complicated transfer codes. Also, our SALT line profiles nicely reproduce the typical asymmetric resonant absorption line profiles observed in starforming/starburst galaxies whereas these absorption profiles cannot be reproduced with thin shells moving at a fixed outflow velocity. We show that scattered resonant emission fills in the resonant absorption profiles, with a strength that is different for each transition. Observationally, the effect of resonant filling depends on both the outflow geometry and the size of the outflow relative to the spectroscopic aperture. Neglecting these effects will lead to incorrect values of gas covering fraction and column density. When a fluorescent channel is available, the resonant profiles alone cannot be used to infer the presence of scattered re-emission. Conversely, the presence of emission lines of fluorescent transitions reveals that emission filling cannot be neglected.

79 ASTRONOMY AND ASTROPHYSICS↗

Simplified Models for LHC New Physics Searches

This document proposes a collection of simplified models relevant to the design of new-physics searches at the LHC and the characterization of their results. Both ATLAS and CMS have already presented some results in terms of simplified models, and we encourage them to continue and expand this effort, which supplements both signature-based results and benchmark model interpretations. A simplified model is defined by an effective Lagrangian describing the interactions of a small number of new particles. Simplified models can equally well be described by a small number of masses and cross-sections. These parameters are directly related to collider physics observables, making simplified models a particularly effective framework for evaluating searches and a useful starting point for characterizing positive signals of new physics. This document serves as an official summary of the results from the 'Topologies for Early LHC Searches' workshop, held at SLAC in September of 2010, the purpose of which was to develop a set of representative models that can be used to cover all relevant phase space in experimental searches. Particular emphasis is placed on searches relevant for the first {approx} 50-500 pb{sup -1} of data and those motivated by supersymmetric models. This note largely summarizes material posted at http://lhcnewphysics.org/, which includes simplified model definitions, Monte Carlo material, and supporting contacts within the theory community. We also comment on future developments that may be useful as more data is gathered and analyzed by the experiments.

43 PARTICLE ACCELERATORS↗

Strongly intermittent edge plasma transport: Issues with modeling and interpretation of experimental data

An impact of blobby transport on plasma dynamics along the magnetic field lines, cross-field plasma transport including ExB drift, and the rates of atomic processes is discussed. It is shown that nonlinearities of plasma transport processes and strong intermittency associated with blobs result in the problems with both interpretation of experimental data based on averaged plasma parameters and with edge plasma modeling with two-dimensional plasma transport codes.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Recurrent Neural Network Attention Mechanisms for Interpretable System Log Anomaly Detection

Deep learning has recently demonstrated state-of-the art performance on key tasks related to the maintenance of computer systems, such as intrusion detection, denial of service attack detection, hardware and software system failures, and malware detection. In these contexts, model interpretability is vital for administrator and analyst to trust and act on the automated analysis of machine learning models. Deep learning methods have been criticized as black box oracles which allow limited insight into decision factors. In this work we seek to "bridge the gap'' between the impressive performance of deep learning models and the need for interpretable model introspection. To this end we present recurrent neural network (RNN) language models augmented with attention for anomaly detection in system logs. Our methods are generally applicable to any computer system and logging source. By incorporating attention variants into our RNN language models we create opportunities for model introspection and analysis without sacrificing state-of-the art performance. We demonstrate model performance and illustrate model interpretability on an intrusion detection task using the Los Alamos National Laboratory (LANL) cyber security dataset, reporting upward of 0.99 area under the receiver operator characteristic curve despite being trained only on a single day's worth of data.

Anomaly detection, Attention, Recurrent Neural Net↗

Design of Digital Twin Sensing Strategies Via Predictive Modeling and Interpretable Machine Learning

This work develops a methodology for sensor placement and dynamic sensor scheduling decisions for digital twins. The digital twin data assimilation is posed as a classification problem, and predictive models are used to train optimal classification trees that represent the map from observed data to estimated digital twin states. In addition to providing a rapid digital twin updating capability, the resulting classification trees yield an interpretable mathematical representation that can be queried to inform sensor placement and sensor scheduling decisions. The proposed approach is demonstrated for a structural digital twin of a 12 ft wingspan unmanned aerial vehicle. Offline, training data are generated by simulating scenarios using predictive reduced-order models of the vehicle in a range of structural states. Furthermore, these training data can be further augmented using experimental or other historical data. In operation, the trained classifier is applied to observational data from the physical vehicle, enabling rapid adaptation of the digital twin in response to changes in structural health. Within this context, we study the performance of the optimal tree classifiers and demonstrate how they enable explainable structural assessments from sparse sensor measurements and also inform optimal sensor placement.

47 OTHER INSTRUMENTATION↗

PBPK and population modelling to interpret urine cadmium concentrations of the French population

As cadmium accumulates mainly in kidney, urinary concentrations are considered as relevant data to assess the risk related to cadmium. The French Nutrition and Health Survey (ENNS) recorded the concentration of cadmium in the urine of the French population. However, as with all biomonitoring data, it needs to be linked to external exposure for it to be interpreted in term of sources of exposure and for risk management purposes. The objective of this work is thus to interpret the cadmium biomonitoring data of the French population in terms of dietary and cigarette smoke exposures. Dietary and smoking habits recorded in the ENNS study were combined with contamination levels in food and cigarettes to assess individual exposures. A PBPK model was used in a Bayesian population model to link this external exposure with the measured urinary concentrations. In this model, the level of the past exposure was corrected thanks to a scaling function which account for a trend in the French dietary exposure. It resulted in a modelling which was able to explain the current urinary concentrations measured in the French population through current and past exposure levels. Risk related to cadmium exposure in the general French population was then assessed from external and internal critical values corresponding to kidney effects. The model was also applied to predict the possible urinary concentrations of the French population in 2030 assuming there will be no more changes in the exposures levels. This scenario leads to significantly lower concentrations and consequently lower related risk. - Highlights: • Interpretation of urine cadmium concentrations in France • PBPK and Bayesian population modelling of cadmium exposure • Assessment of the historic time-trend of the cadmium exposure in France • Risk assessment from current and future external and internal exposure.

60 APPLIED LIFE SCIENCES↗

Stability analysis of WEST L-mode discharges with improved confinement from boron powder injection

WEST L-mode plasmas with dominant electron heating and no core torque source have observed improvements in confinement during boron (B) powder injection. These results are reminiscent of previous powder injection experiments on other devices and gaseous impurity seeding experiments on WEST. During powder injection, the stored energy increased up to 25% due to enhanced ion and electron heat and particle confinement. The improvements in confinement were not indicative of an L-H transition. To identify the dominant mechanisms and the causality chain behind these improvements in confinement, we employ interpretative modeling using METIS, predictive integrated modeling using a high-fidelity plasma simulator (HFPS), and stand-alone gyrokinetic simulations using quasi-linear gyrokinetic code. Interpretative modeling with METIS allowed for the estimation of missing data while maintaining good overall consistency with experiment. These results provided the initial conditions for fully predictive flux driven simulations using the HFPS. From these simulations, quasi-linear gyrokinetic analysis was performed at ρ = 0.5 and ρ = 0.65. Collisionality was found to be a strong candidate for the turbulence suppression mechanism at ρ = 0.5, while a combination of collisionality and the T e /T i ratio was found to be the likely mechanism at ρ = 0.65. The results additionally suggested that increased Z eff (through main ion dilution) could play a role in the improved confinement, but this could not be confirmed due to a lack of experimental measurements. The modeling framework established here can now be used to evaluate and exploit a variety of future powder injection experiments.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Evaluation of Deep Learning Model Architectures for Point-of-Care Ultrasound Diagnostics

Point-of-care ultrasound imaging is a critical tool for patient triage during trauma for diagnosing injuries and prioritizing limited medical evacuation resources. Specifically, an eFAST exam evaluates if there are free fluids in the chest or abdomen but this is only possible if ultrasound scans can be accurately interpreted, a challenge in the pre-hospital setting. In this effort, we evaluated the use of artificial intelligent eFAST image interpretation models. Widely used deep learning model architectures were evaluated as well as Bayesian models optimized for six different diagnostic models: pneumothorax (i) B- or (ii) M-mode, hemothorax (iii) B- or (iv) M-mode, (v) pelvic or bladder abdominal hemorrhage and (vi) right upper quadrant abdominal hemorrhage. Models were trained using images captured in 27 swine. Using a leave-one-subject-out training approach, the MobileNetV2 and DarkNet53 models surpassed 85% accuracy for each M-mode scan site. The different B-mode models performed worse with accuracies between 68% and 74% except for the pelvic hemorrhage model, which only reached 62% accuracy for all model architectures. These results highlight which eFAST scan sites can be easily automated with image interpretation models, while other scan sites, such as the bladder hemorrhage model, will require more robust model development or data augmentation to improve performance. With these additional improvements, the skill threshold for ultrasound-based triage can be reduced, thus expanding its utility in the pre-hospital setting.

47 OTHER INSTRUMENTATION↗

Laboratory Experiments and Modeling for Interpreting Field Studies of Secondary Organic Aerosol Formation Using an Oxidation Flow Reactor

This grant was originally funded for deployment of a suite of aerosol instrumentation by our group in collaboration with other research groups and DOE/ARM to the Ganges Valley in India (GVAX) to study aerosols sources and processing. Much of the first year of this grant was focused on preparations for GVAX. That campaign was cancelled due to political reasons and with the consultation with our program manager, the research of this grant was refocused to study the applications of oxidation flow reactors (OFRs) for investigating secondary organic aerosol (SOA) formation and organic aerosol (OA) processing in the field and laboratory through a series of laboratory and modeling studies. We developed a gas-phase photochemical model of an OFR which was used to 1) explore the sensitivities of key output variables (e.g., OH exposure, O 3 , HO 2 /OH) to controlling factors (e.g., water vapor, external reactivity, UV irradiation), 2) develop simplified OH exposure estimation equations, 3) investigate under what conditions non-OH chemistry may be important, and 4) help guide design of future experiments to avoid conditions with undesired chemistry for a wide range of conditions applicable to the ambient, laboratory, and source studies. Uncertainties in the model were quantified and modeled OH exposure was compared to tracer decay measurements of OH exposure in the lab and field. Laboratory studies using OFRs were conducted to explore aerosol yields and composition from anthropogenic and biogenic VOC as well as crude oil evaporates. Various aspects of the modeling and laboratory results and tools were applied to interpretation of ambient and source measurements using OFR. Additionally, novel measurement methods were used to study gas/particle partitioning. The research conducted was highly successful and details of the key results are summarized in this report through narrative text, figures, and a complete list of publications acknowledging this grant.

54 ENVIRONMENTAL SCIENCES↗

Model-independent versus model-dependent interpretation of the SDSS-III BOSS power spectrum: Bridging the divide

The traditional clustering analyses of galaxy redshift surveys compress the clustering data into a set of late-time physical variables in a model-independent way. This approach has recently been extended by an additional shape variable encoding early-time physics information. Here we apply this new technique, ShapeFit, to SDSS-III BOSS data and show that it matches the constraining power of alternative, model-dependent approaches, which directly constrain the model’s parameters adopting a cosmological model ab initio. ShapeFit is ~30 times faster, model independent, naturally splits early- and late-time variables, and enables a better control of observational systematics.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Charmonium in lattice QCD and the non-relativistic quark model

We compare the results of a numerical lattice QCD calculation of the charmonium spectrum with the structure of a general non-relativistic potential model. To achieve this we form the non-relativistic reduction of derivative-based fermion bilinear interpolating fields used in lattice QCD calculations and compute their overlap with $c\bar{c}$ meson states at rest constructed in the non-relativistic quark model, providing a bound-state model interpretation for the lattice data. Essential gluonic components in the bound-states, usually called hybrids, are identified by considering interpolating fields that involve the gluonic field-strength tensor and which have zero overlap onto simple $c\bar{c}$ model states.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗