Engineering Papers⌕ Search

SEARCH · Engineering Papers

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 541 records · Page 30
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

An empirical approach to model selection: weak lensing and intrinsic alignments

ABSTRACT In cosmology, we routinely choose between models to describe our data, and can incur biases due to insufficient models or lose constraining power with overly complex models. In this paper, we propose an empirical approach to model selection that explicitly balances parameter bias against model complexity. Our method uses synthetic data to calibrate the relation between bias and the χ2 difference between models. This allows us to interpret χ2 values obtained from real data (even if catalogues are blinded) and choose a model accordingly. We apply our method to the problem of intrinsic alignments – one of the most significant weak lensing systematics, and a major contributor to the error budget in modern lensing surveys. Specifically, we consider the example of the Dark Energy Survey Year 3 (DES Y3), and compare the commonly used non-linear alignment (NLA) and tidal alignment and tidal torque (TATT) models. The models are calibrated against bias in the Ωm–S8 plane. Once noise is accounted for, we find that it is possible to set a threshold Δχ2 that guarantees an analysis using NLA is unbiased at some specified level Nσ and confidence level. By contrast, we find that theoretically defined thresholds (based on, e.g. p-values for χ2) tend to be overly optimistic, and do not reliably rule out cosmological biases up to ∼1–2σ. Considering the real DES Y3 cosmic shear results, based on the reported difference in χ2 from NLA and TATT analyses, we find a roughly $30{{\ \rm per\ cent}}$ chance that were NLA to be the fiducial model, the results would be biased (in the Ωm–S8 plane) by more than 0.3σ. More broadly, the method we propose here is simple and general, and requires a relatively low level of resources. We foresee applications to future analyses as a model selection tool in many contexts.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Generalizing the Gurson model using symbolic regression and transfer learning to relax inherent assumptions

Abstract To generate material models with fewer limiting assumptions while maintaining closed-form, interpretable solutions, we propose using genetic programming based symbolic regression (GPSR), a machine learning (ML) approach that describes data using free-form symbolic expressions. To maximize interpretability, we start from an analytical, derived material model, the Gurson model for porous ductile metals, and systematically relax inherent assumptions made in its derivation to understand each assumption’s contribution to the GPSR model forms. We incorporate transfer learning methods into the GPSR training process to increase GPSR efficiency and generate models that abide by known mechanics of the system. The results show that regularizing the GPSR fitness function is critical for generating physically valid models and illustrate how GPSR allows a high level of interpretability compared with other ML approaches. The method of systematic assumption relaxation allows the generation of models that address limiting assumptions found in the Gurson model, and the symbolic forms allow conjecture of decreased material strength due to void interaction and non-symmetric void shapes.

36 MATERIALS SCIENCE↗
DOE OSTI2026

MCP-eGridGPT (MCP-Enabled Chatbot with Electrical Power System Analysis and Interactive Visualization Tool) [SWR-25-126]

This software is an advanced chatbot system that integrates the Model Context Protocol (MCP) to provide intelligent electrical power system analysis and automated visualization generation. The system enables users to interact with complex electrical engineering tools through natural language, automatically analyzes power system data for voltage violations and grid health assessment, and generates professional interactive HTML dashboards and reports. Key features include dynamic tool discovery from MCP servers, multi-LLM provider support, intelligent data interpretation using large language models, automated chart generation, and a web-based interface for real-time analysis. The software bridges sophisticated electrical engineering analysis with user-friendly interfaces, making power system diagnostics accessible through conversational AI.

Choi, Seong [National Laboratory of the Rockies (N↗
DOE OSTI2021

Electron–Nuclear Dynamics Accompanying Proton-Coupled Electron Transfer

Although photoinduced proton-coupled electron transfer (PCET) plays an essential role in photosynthesis, a full understanding of the mechanism is still lacking due to the complex nonequilibrium dynamics arising from the strongly coupled electronic and nuclear degrees of freedom. Here we report the photoinduced PCET dynamics of a biomimetic model system investigated by means of transient IR and two-dimensional electronic–vibrational (2DEV) spectroscopies, IR spectroelectrochemistry (IRSEC), and calculations utilizing long-range-corrected hybrid density functionals. This collective experimental and theoretical effort provides a nuanced picture of the complicated dynamics and synergistic motions involved in photoinduced PCET. In particular, the evolution of the 2DEV line shape, which is highly sensitive to the mixing of vibronic states, is interpreted by accurate computational modeling of the charge separated state and is shown to represent a gradual change in electron density distribution associated with a dihedral twist that occurs on a 120 fs time scale.

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

Data-Driven Kinetic Reaction Networks for Separation Chemistry

Understanding complex, multistep chemical reactions at the molecular level is a major challenge whose solution would greatly benefit the design and optimization of numerous chemical processes. The separation of rare-earth (4f) and actinide (5f) elements is an example where improving our chemical understanding is important for designing and optimizing new chemistries, even with a limited number of observations. Here, in this work, we leverage data-driven artificial intelligence and machine-learning approaches to develop kinetic reaction networks that describe the liquid–liquid extraction mechanism of uranium using N,N-di-2-ethylhexyl-isobutyramide (DEHiBA). Specifically, we compare and contrast the properties of two classes of models: (1) purely data-driven models that are regularized using chemistry-agnostic, L1 regression and (2) chemistry-informed models that are regularized using relative reaction energies provided by quantum mechanical calculations. We observe that purely data-driven models are unbiased, simple, and accurate in their predictions of experimental measurements when provided with sufficient data but are difficult to fully constrain and interpret. In contrast, chemistry-informed models exhibit significantly improved chemical interpretability and consistency, providing a detailed description of the separation process while achieving high accuracy through ensemble averaging. Overall, the dominant species predicted to be extracted into the organic phase is UO 2 (NO 3 ) 2 (DEHiBA) 2 , agreeing with experimental slope analysis, thermodynamic modeling, EXAFS, and crystal structures. This work demonstrates that leveraging the fundamental structure of the problem can lead to efficient learning schemes that provide both accurate predictions and chemical insights at a low computational cost.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Rare and Experimentally Challenging Supersymmetry Signatures

Supersymmetry has long played a central role in the search for physics beyond the Standard Model at colliders, providing a comprehensive and internally consistent framework for generating well-motivated experimental signatures. For more than 15 years of Large Hadron Collider (LHC) operation, the CMS and ATLAS Collaborations have achieved remarkable sensitivity to a wide range of supersymmetric signatures. Despite this unprecedented reach, no conclusive evidence for supersymmetry has emerged. If supersymmetry is nature's solution to outstanding questions in particle physics, it is necessarily challenging to find. In this article, we review supersymmetric signatures that are particularly rare or otherwise challenging, with a focus on searches at the LHC. We highlight experimental challenges related to detector constraints and analysis difficulties, in addition to model challenges in the interpretation and optimization of searches. Here, we also identify regions of signature space that remain comparatively unconstrained and therefore represent promising targets for future exploration.

Large Hadron Collider↗
DOE OSTI2020

Quantitative Prediction of Perovskite Stability Using Accelerated Testing and Machine Learning

PV technologies based on hybrid perovskites offer the potential for reducing solar cell costs, but they are particularly vulnerable to degradation by environmental factors such as moisture, oxygen, and illumination. Commercialization will require not only stable materials and device architectures but also accelerated testing protocols and models that can predict degradation from the accelerated testing data. Here, we report results from in situ photoluminescence (PL) and photoconductivity (PC) measurements during perovskite degradation with simultaneous optical transmittance (Tr) measurements or reflected dark field (DF) imaging. From PL, PC, Tr, and DF, we determine (respectively) the steady-state quasi-Fermi level splitting, the mean effective carrier diffusion length, the extent of conversion of perovskite to higher bandgap degradation products, and the extent of scattering from domains with different orientation or composition, all as a function of time during degradation. Simultaneous measurement of PL-PC-Tr or PL-PC-DF on perovskite absorbers in an environmental chamber over a wide range of humidity, oxygen, temperature, and illumination levels yields a rich data set. We use machine learning to develop a model that accurately predicts (within 10%) the time for the diffusion length to decrease to 85% of its initial value. The model takes the environmental conditions and the first few measurements from the PL-PC-Tr or PL-PC-DF experiment as input. Thus, the model provides a framework to interpret the results of accelerated testing of absorber materials. One of the dominant features in the model of degradation for CH 3 NH 3 PbI 3 is the initial rate at which transmittance increase. For devices with opaque contacts, the pixel-averaged rate of change of the intensity in dark-field images can be used in place of transmittance. Results are also presented on full PV devices with in-situ current-voltage (JV) measurements. Here, the data from simultaneous PL-DF-JV under environmental stresses reveal intimate connection between degradation and shunts and provide a framework to extend accelerated testing to devices. Further, the presentation reveals underlying universal behavior in degradation pathways over a broad range of environmental stresses and perovskite compositions.

14 SOLAR ENERGY↗
DOE OSTI2021

Logistic classification for tool life modeling in machining

This paper describes the application of logistic classification for tool life modeling and prediction in an industrial setting using shop floor data. Tool life is treated as a classification problem since tool wear can only be measured at the time of tool replacement in a production environment. Laboratory tool wear experiments are used to simulate shop floor wear data by two states: not worn (class 0); and worn (class 1). To incorporate non-linearity in logistic classification, a log-transformation of input features is performed. The logistic classification approach, results, and interpretability of the logistic model are presented.

Karandikar, Jaydeep↗
DOE OSTI2025

eDNAjoint: An R package for interpreting paired or semi‐paired environmental DNA and traditional survey data in a Bayesian framework

Abstract Environmental DNA (eDNA) sampling is increasingly used in surveys of species distribution as a potentially sensitive and efficient monitoring method. Yet access to modelling tools designed specifically for interpreting this new data type lags behind its ubiquity. While occupancy modelling software has dominated the analytical landscape for eDNA data analysis of single species, this type of model may not always be the most appropriate. The rate of eDNA detection often corresponds to species density, rather than just occupancy, and researchers often have access to observations from non‐genetic sampling methods at the same sites. To provide users access to a modelling framework designed to maximize the use of all available data, we developed an R package, eDNAjoint . The package provides an easy‐to‐use interface for fitting a ‘joint’ model that integrates data from paired or semi‐paired eDNA and traditional surveys in a Bayesian framework. The model can be used to estimate parameters like the probability of a false positive eDNA detection and mean catch rate at a site, and the package allows access to multiple model variations and Bayesian prior customization. Additional functionality can be used for model selection, summarising posteriors and comparing the relative sensitivities of the two survey methods. We demonstrate the use of eDNAjoint by fitting a variation of the model with site‐level covariates that scale the sensitivity of eDNA sampling relative to traditional sampling. The example workflow uses binary eDNA and seine count data for the endangered tidewater goby ( Eucyclogobius newberryi ) from a study by Schmelzle and Kinziger (2016). This use case includes a prior sensitivity analysis and an evaluation of the relationship between detection rates and environmental variables. eDNAjoint has the potential to greatly increase the range of users who will be able to rigorously analyse eDNA and traditional survey data in a Bayesian framework, understand if and how eDNA can improve monitoring practices, and gain confidence in the interpretability of eDNA data.

Keller, Abigail G. [Department of Environment Scie↗
DOE OSTI2024

Investigating the effects of precise mass measurements of Ru and Pd isotopes on machine learning mass modeling

Atomic masses are a foundational quantity in our understanding of nuclear structure, astrophysics, and fundamental symmetries. The longstanding goal of creating a predictive global model for the binding energy of a nucleus remains a significant challenge, however, and prompts the need for precise measurements of atomic masses to serve as anchor points for model developments. We present precise mass measurements of neutron-rich Ru and Pd isotopes performed at the Californium Rare Isotope Breeder Upgrade facility at Argonne National Laboratory using the Canadian Penning Trap mass spectrometer. The masses of 108 Ru, 110 Ru, and 116 Pd were measured to a relative mass precision $\delta$⁢$m/m$ ≈ 10 -8 via the phase-imaging ion-cyclotron-resonance technique, and represent an improvement of approximately an order of magnitude over previous measurements. Further, these mass data were used in conjunction with the physically interpretable machine learning (PIML) model, which uses a mixture density neural network to model mass excesses via a mixture of Gaussian distributions. The effects of our new mass data on a Bayesian-updating of a PIML model are presented.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2020

Multi-jet gas cooling of in-beam foils or specimens: CFD predictions of the convective heat-transfer coefficient

An experiment was designed to investigate the possible enhancement of the convective heat-transfer coefficient by utilizing multiple, parallel jets in the cooling of a small heated surface, such as typically induced by an accelerated ion beam on a thin foil or specimen. The hot spot was provided using a small electrically heated plate. It was found that heat-transfer calculations by means of simple empirical methods based on dimensional analysis are not useful in this case and that advanced computational fluid dynamics (CFD) modelling is essential to interpret the results. It is shown that enhanced convective cooling can indeed be obtained with a multi-jet configuration as compared to a single-jet configuration but only under very selective conditions. An improperly designed multi-jet configuration can also provide significantly reduced cooling relative to the single-jet case and the estimation of the behavior of any particular jet geometry is eminently non-intuitive. CFD provides acceptable quantitative results and seem to be the only tool available to gain an understanding of these complex flows where simple models and “rules of thumb” cannot be relied upon.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2021

Machine-Learned Manifold-Based Models for Large Eddy Simulation of Turbulent Combustion

Reduced-order manifold combustion models are commonly used to lower the cost of reacting Large Eddy Simulations (LES) and may be derived either from physical principles as in flamelet models or from data-driven methods like Principal Component Analysis (PCA). In either case, neural networks are increasingly used as part of these models to provide a nonlinear mapping between a small set of pre-defined variables that parameterize the manifold and outputs of interest, such as reaction rates. In this work, we propose a new manifold-based modeling approach that combines the definition of the manifold-parameterizing variables (linear combinations of species), the nonlinear mapping to the outputs, and closure of filtered quantities for LES into the structure of a single neural network. This allows the process used to train the neural network to simultaneously optimize both the functional form of the model and the identities of the inputs to the model. The new approach can flexibly incorporate thermochemical data from any combustion system; if trained on data from 1D flames it can be interpreted as an optimized flamelet model, but it can also be used to learn models from data from more complex configurations. This work presents a priori evaluations of the new approach in both contexts. Evaluation using data from 1D premixed flames demonstrates the physical interpretability of the manifold variables generated by the new approach. Evaluation using data from direct numerical simulations of turbulent flames shows improved predictions relative to either flamelet or PCA-based models in a more complex configuration.

47 OTHER INSTRUMENTATION↗
DOE OSTI2024

Towards interpretable Cryo-EM: disentangling latent spaces of molecular conformations

Molecules are essential building blocks of life and their different conformations (i.e., shapes) crucially determine the functional role that they play in living organisms. Cryogenic Electron Microscopy (cryo-EM) allows for acquisition of large image datasets of individual molecules. Recent advances in computational cryo-EM have made it possible to learn latent variable models of conformation landscapes. However, interpreting these latent spaces remains a challenge as their individual dimensions are often arbitrary. The key message of our work is that this interpretation challenge can be viewed as an Independent Component Analysis (ICA) problem where we seek models that have the property of identifiability. That means, they have an essentially unique solution, representing a conformational latent space that separates the different degrees of freedom a molecule is equipped with in nature. Thus, we aim to advance the computational field of cryo-EM beyond visualizations as we connect it with the theoretical framework of (nonlinear) ICA and discuss the need for identifiable models, improved metrics, and benchmarks. Moving forward, we propose future directions for enhancing the disentanglement of latent spaces in cryo-EM, refining evaluation metrics and exploring techniques that leverage physics-based decoders of biomolecular systems. Moreover, we discuss how future technological developments in time-resolved single particle imaging may enable the application of nonlinear ICA models that can discover the true conformation changes of molecules in nature. The pursuit of interpretable conformational latent spaces will empower researchers to unravel complex biological processes and facilitate targeted interventions. This has significant implications for drug discovery and structural biology more broadly. More generally, latent variable models are deployed widely across many scientific disciplines. Thus, the argument we present in this work has much broader applications in AI for science if we want to move from impressive nonlinear neural network models to mathematically grounded methods that can help us learn something new about nature.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2025

Search for electroweak production of vector-like leptons in $\tau$-lepton and b-jet final states in pp collisions at $\sqrt{s}$ = 13 TeV with the ATLAS detector

A search for pair-production of vector-like leptons is presented, considering their decays into a third-generation Standard Model (SM) quark and a vector leptoquark ( U 1 ) as predicted by an ultraviolet-complete extension of the SM, referred to as the ‘4321’ model. Given the assumed decay of U 1 into third-generation SM fermions, the final state can contain multiple τ-leptons and b-quarks. This search is based on a dataset of pp collisions at $\sqrt{s}=13$ TeV recorded with the ATLAS detector during Run 2 of the Large Hadron Collider, corresponding to an integrated luminosity of up to $140~\textrm{fb}^{-1}$. No significant excess above the SM background prediction is observed, and 95% confidence level limits on the cross-section times branching ratio are derived as a function of the vector-like lepton mass. A lower observed (expected) limit of 910 GeV (970 GeV) is set on the vector-like lepton mass. Additionally, the results are interpreted for a supersymmetric model with an R-parity violating coupling to the third-generation quarks and leptons. Lower observed (expected) limits are obtained on the higgsino mass at 880 GeV (940 GeV) and on the wino mass at 1170 GeV (1170 GeV).

Aad, G. [CPPM, Aix-Marseille Université, CNRS/IN2P↗
DOE OSTI2022

PyOECP: A flexible open-source software library for estimating and modeling the complex permittivity based on the open-ended coaxial probe (OECP) technique

Here, we present PyOECP, a Python-based flexible open-source software for estimating and modeling the complex permittivity obtained from the open-ended coaxial probe (OECP) technique. The transformation of the measured reflection coefficient to complex permittivity is performed based on three different methods. The software library contains the dielectric spectra of common reference liquids, which can be used to transform the reflection coefficient into the dielectric spectra. Several Python routines that are commonly employed (e.g., SciPy and NumPy) in the field of science and engineering are required only so that the users can alter the software structure depending on their needs. The modeling algorithm exploits the Markov Chain Monte Carlo method for the data regression. The discrete relaxation models can be built by a proper combination of well-known relaxation models. In addition to these models, electrode polarization, a typical measurement artifact for interpreting dielectric spectra, can be incorporated into the modeling algorithm. A continuous relaxation model, which solves the Fredholm integral equation of the first kind (a mathematically ill-posed problem), is also included. This open-source software enables users to freely adjust the physical parameters to obtain physical insight into their materials under test and will be consistently updated for more accurate measurement and interpretation of dielectric spectra in an automated manner. This work describes the theoretical and mathematical background of the software, lays out the workflow, and validates the software functionality based on both synthetic and empirical data included in the software.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Megahertz Rate Optical Diagnostics of Explosively Generated Soot

Detonation of a solid explosive produces a turbulent and luminous post‐detonation fireball containing condensed carbon soot. Diagnostics of soot dynamics are needed for model validation and to interpret emission signals. Diffuse back‐illumination extinction imaging (DBI‐EI) and laser‐induced incandescence (LII) are two common optical diagnostics for flame soot. This work extends both to measure time‐resolved soot dynamics from a 12 mm HMX hemisphere. DBI‐EI measured line‐of‐sight optical extinction exceeds 99% in some regions. Results are tomographically inverted to obtain a signal proportional to the soot volume fraction. Pulse‐burst LII measures are presented at 1 MHz. For the first time, LII results are combined with DBI‐EI extinction measurements to quantify and correct for signal trapping. Following this, spatially and temporally resolved DBI‐EI and LII measures are shown to be in reasonable agreement. Finally, experimental results are compared against recently developed simulations. Quantitative differences in the soot mixing dynamics are resolved. These findings motivate future model improvements and demonstrate ongoing needs for diagnostic advancements for heavily sooting environments.

diffuse back-illumination extinction imaging↗
DOE OSTI2025

Modeling a High-Temperature Electrochemically Driven Water-Gas-Shift Process Using a Mixed-Conducting Membrane without External Electrical Power

This paper develops a model to predict and interpret the performance of an elevated-temperature, electrochemical, membrane-assisted, water-gas-shift process. The process uses separated feed streams of H 2 O and CO to produce separated streams of H 2 and CO 2 , without an external electrical power source. The dense ceramic membrane is mixed ionic-electronic-conducting (MIEC) gadolinium-doped ceria (GDC) and the porous composite electrodes are Ni-YSZ. At elevated temperature, GDC conducts both oxygen ions and small polarons. The present process uses chemical potential to drive the process. Electrochemical oxidation of CO proceeds within the composite anode and H 2 O reduction proceeds within the composite cathode. At high temperature (e.g., T > 700 °C), GDC has significant electronic leakage in the form of a reduced-cerium small polaron, which supports the charge-transfer reactions. In a typical electrolyzer or fuel cell, this leakage is significantly problematic. However, the present process depends on the leakage current to complete the electrochemical circuit. Model development and validation is based on measured material properties and reactor performance. Potential applications include using CO-rich blast-furnace off gases in steel processing, producing separated streams of H 2 and CO 2 .

Zhu, Huayang↗
DOE OSTI2020

Deep neural network improves the estimation of polygenic risk scores for breast cancer

Polygenic risk scores (PRS) estimate the genetic risk of an individual for a complex disease based on many genetic variants across the whole genome. Here, we compared a series of computational models for estimation of breast cancer PRS. A deep neural network (DNN) was found to outperform alternative machine learning techniques and established statistical algorithms, including BLUP, BayesA, and LDpred. In the test cohort with 50% prevalence, the Area Under the receiver operating characteristic Curve (AUC) were 67.4% for DNN, 64.2% for BLUP, 64.5% for BayesA, and 62.4% for LDpred. BLUP, BayesA, and LPpred all generated PRS that followed a normal distribution in the case population. However, the PRS generated by DNN in the case population followed a bimodal distribution composed of two normal distributions with distinctly different means. This suggests that DNN was able to separate the case population into a high-genetic-risk case subpopulation with an average PRS significantly higher than the control population and a normal-genetic-risk case subpopulation with an average PRS similar to the control population. This allowed DNN to achieve 18.8% recall at 90% precision in the test cohort with 50% prevalence, which can be extrapolated to 65.4% recall at 20% precision in a general population with 12% prevalence. Interpretation of the DNN model identified salient variants that were assigned insignificant p values by association studies, but were important for DNN prediction. These variants may be associated with the phenotype through nonlinear relationships.

59 BASIC BIOLOGICAL SCIENCES↗