Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Point Optimization”

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 1,441 records · Page 80
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Modeling of the effectiveness of novel edge seal designs for fast, low‐Cap‐Ex manufacturing

Abstract In the manufacturing of photovoltaic (PV) modules, the lamination process can take up to 20 min to complete. In this work, new lamination processes are being developed, and have been prototyped, which hope to be able to cut this time down to as little as 30 s. This could provide significant savings in the cost of lamination equipment, floor space, and energy. PV modules are expected to have a lifespan exceeding 20 to 30 years. For moisture‐sensitive PV technologies, the edge seal between the two layers of glass can be the weakest point of its reliability. There is an inherent challenge when evaluating edge seal materials due to their low permeation rates. As part of Colorado State University's Photovoltaic Research and Development 2, work at the National Renewable Energy Laboratory has developed models to evaluate edge seal configurations in glass‐glass PV modules. Here, this new manufacturing process is evaluated for long‐term moisture durability. Different edge seal design options within glass–glass PV modules are explored. Most of these designs are targeting a superstrate on glass configuration, e.g CdTe, but some designs could be used on conventional crystalline Si cells. Using COMSOL finite element simulation software, we investigated the edge seal and interlayer design configurations containing silicone perimeter edge adhesive, desiccated polyisobutylene‐based edge seal, air, and polyolefin while integrating climate conditions equivalent to a hot and humid environment such as Miami, Florida. We found optimized configurations that will allow the module to prevent moisture ingress over 50 years minimizing the amount of time and material used while utilizing polymers that are easily dispensed.

14 SOLAR ENERGY↗
DOE OSTI2023

Computer-controlled finishing via dynamically constraint position-velocity-time scheduler

In a Computer Numerical Controlled (CNC) finishing process, the target material removal from an optical surface is guided by the convolution between the influence function of a machine tool and its dwell time at certain points over the surface. To reduce dynamics stressing and increase machining efficiency, the dwell time must be converted to varying velocities, which are the actual inputs to the machine tool controller. Conventionally, the conversion assumed constant acceleration and relied on linear motion interpolation, which caused discontinuities in velocities. This unsmooth motion affects the material removal distribution, and, thus, the accuracy of the finished surface shape. Many modern CNC machines support the smoother, cubic-polynomial interpolated Position-Velocity-Time (PVT) motion mode; however, the conventional scheduler may fail to provide suitable velocities for the PVT. Here in this study answers this challenge by proposing a novel PVT-based velocity scheduler that achieves smooth motion while considering CNC dynamic limits. Firstly, the principle of the PVT is explained, and the PVT-based velocity scheduler is formulated. Secondly, a quadratic programming is used to optimize the velocities by imposing the CNC dynamic constraints and the C 1 continuities (zeroth and first derivatives are continuous) simultaneously. Thirdly, the smoothness and accuracy of the scheduled velocities are studied on different kinds of tool paths via simulation. Finally, a sub-0.3 nm level surface finishing experiment using ion beam figuring is demonstrated to verify the feasibility of the proposed method. The PVT-based scheduler and simulator code is open-sourced.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Predicting variations of the least principal stress with depth: Application to unconventional oil and gas reservoirs using a log-based viscoelastic stress relaxation model

Knowledge of layer-to-layer variations of the least principal stress, S hmin , with depth is essential for optimization of multi-stage hydraulic fracturing in unconventional reservoirs. Utilizing a geomechanical model based on viscoelastic stress relaxation in relatively clay rich rocks, we present a new method for predicting continuous S hmin variations with depth. The method utilizes geophysical log data and S hmin measurements from routine diagnostic fracture injection tests (DFITs) at several depths for calibration. We consider a case study in the Wolfcamp formation in the Midland Basin, where both geophysical logs and values of S hmin from DFITs are available. We compute a continuous stress profile as a function of the well logs that fits all of the DFITs well. We utilized several machine learning technologies, such as bootstrap aggregation (or bagging), to improve the generalization of the model and demonstrate that the excellent fit between predicted and observed stress values is not the result of over-fitting the calibration points. The model is then validated by accurately predicting hold-out stress measurements from four wells within the study area and, without recalibration, accurately predicting stress as a function of depth in an offset pad about 6 miles away.

58 GEOSCIENCES↗
DOE OSTI2023

Complete Evaluation on Advanced Reactor Machine Learning Subversion Attacks (Final)

Navigating through the world of Artificial Intelligence (AI) in nuclear reactors and their Instrumentation and Control (I&C) systems demands a careful, deliberate journey. AI’s capability to manage massive datasets and streamline control systems has indeed carved out a significant role in various sectors, including nuclear energy. However, while AI, and particularly Large Language Models (LLMs), bring a lot to the table in terms of operational efficiency and anomaly detection, they also expose the sector to a new breed of cybersecurity threats, like Inference Attacks, Adversarial Attacks, and Trojan Attacks. This guide is designed to be a straightforward manual, diving deep into the intertwining worlds of AI and cybersecurity within nuclear reactors, and tailoring insights for three crucial audiences: I&C Vendors/Developers, Nuclear Regulators, and Nuclear Reactor Operators and Cyber Defense Teams. (1) Section 2, directed at I&C Vendors/Developers, will provide a clear and focused look at several cybersecurity attacks, offering practical recommendations and detailed scenarios related to AI cybersecurity. This section isn’t just about identifying problems but also about giving solid, usable solutions. (2) Section 3, meant for Nuclear Regulators, gets straight to the point about regulations, policy suggestions, and guidelines that are needed to lay down a robust, secure, and ethical foundation for the application of AI in nuclear operations. The focus is on making sure that everything adheres to international standards and laws while being practicable and clear-cut. (3) Section 4, aimed at Nuclear Reactor Operators and Cyber Defense Teams, offers an exhaustive exploration and technical reports, with clear recommendations and scenario analyses vital to protect operational environments and guarantee the secure application of AI in nuclear reactor operations. The goal is simple: as we step into an era where AI becomes a fundamental element of our technological and energy infrastructures, this guide is here to act as a clear, direct handbook, ensuring that AI is implemented within the nuclear sector in a manner that is secure, responsible, and practical. It’s about striking a balance – optimizing the undeniable benefits offered by AI while securing and shielding against potential cyber threats as we move through this new and complex landscape.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2020

Adaptive Activation Functions Accelerate Convergence in Deep and Physics-informed Neural Networks

We employ adaptive activation functions for regression in deep and physics-informed neural networks (PINNs) to approximate smooth and discontinuous functions as well as solutions of linear and nonlinear partial differential equations. In particular, we solve the nonlinear Klein-Gordon equation, which has smooth solutions, the nonlinear Burgers equation, which can admit high gradient solutions, and the Helmholtz equation. We introduce a scalable hyper-parameter in the activation function, which can be optimized to achieve best performance of the network as it changes dynamically the topology of the loss function involved in the optimization process. The adaptive activation function has better learning capabilities than the traditional one (fixed activation) as it improves greatly the convergence rate, especially at early training, as well as the solution accuracy. To better understand the learning process, we plot the neural network solution in the frequency domain to examine how the network captures successively different frequency bands present in the solution. We consider both forward problems, where the approximate solutions are obtained, as well as inverse problems, where parameters involved in the governing equation are identified. Our simulation results show that the proposed method is a very simple and effective approach to increase the efficiency, robustness and accuracy of the neural network approximation of nonlinear functions as well as solutions of partial differential equations, especially for forward problems. We theoretically prove that in the proposed method, gradient descent algorithms are not attracted to suboptimal critical points or local minima.

machine leaning, Bad minima, Inverse problems, Phy↗
DOE OSTI2020

Heliostat Aiming Strategies for Arbitrary Flux Profiles for High Temperature Gas-Based Receivers

High temperature gas-based receivers are seen as one potential path toward achieving the next generation of concentrated solar power tower systems. A key challenge associated with high temperature, high pressure heat transfer fluids is the lifetime of receiver containment materials. Achieving the desired lifetime while also designing an efficient receiver requires implementation of unique flux profiles, tailored to place high fluxes at cold fluid injection points. An aimpoint selection methodology was developed in the present study to allow user-defined flux profiles to be prescribed for a receiver. The methodology was implemented in NREL's open-source software SolarPILOT, modifying the existing “image size priority” aimpoint strategy. A case study for a tubular gas-based receiver is presented, examining the effect of various parameters on the ability to replicate the desired flux profile.

41 EE - Solar Energy Technologies Office (EE-4S)↗
DOE OSTI2026

Multibody for Everybody (M4E) - A Linearization Approach to Enable Frequency Domain Analysis, Time Integration and Control Co-Design

1.1 Background/Objectives: Marine energy represents a promising yet underexploited source of power. To increase the harvested power, significant efforts have been made to improve wave energy converter (WEC) modeling capabilities and optimize power take-off (PTO) performance; however, these efforts have often treated WEC dynamics, PTO design, and controller development sequentially. In contrast, control co-design (CCD) is emerging as a promising strategy to address these issues directly, creating a growing need for fast analysis tools suitable for repeated simulation and parametric studies [1]. To support this need, this work presents the Multibody for Everybody (M4E) [2] linearization module, which employs a symbolic toolbox to provide deeper insight of WEC design parameters. The objective is to demonstrate that a minimal-coordinate linearization of articulated WEC dynamics can provide accurate wave response predictions and substantial computational savings relative to nonlinear time-domain simulation, while preserving compatibility with broader wave-energy analysis workflows, enabling CCD. 1.2 Approach/Activities: The proposed approach linearizes the equations of motion, generated by M4E, in minimal coordinates about a selected operating point and combines the resulting system with frequencydomain hydrodynamic terms to incorporate the reduced mass, damping, stiffness, and forcing operators. The linearized model is used for both impedance-based response amplitude operator (RAO) prediction and rapid regular-wave time integration. The methodology is demonstrated on a single-flap device and a FOSWEC configuration, with linearized M4E responses compared against the corresponding nonlinear M4E simulations and WEC-Sim results. Regular-wave time histories, RAO trends, and runtime differences are assessed. The framework is also compatible with broader wave-energy workflows, including coupling to WecOptTool, although that capability is not the focus of this work [3]. 1.3 Results/Lessons: The linearized M4E model reproduces key regularwave response characteristics such as integration and Response Amplitude over multiple frequencies. This module matches nonlinear M4E and WEC-Sim results while substantially reducing integration cost. Thus, the proposed framework can serve as a rapid analysis layer for articulated WEC design, parameter studies, and controls-oriented workflows. The analysis is most appropriate in the near-equilibrium regime, about the linearization point.

16 TIDAL AND WAVE POWER↗
DOE OSTI2021

Distributed deep reinforcement learning for simulation control

Abstract Several applications in the scientific simulation of physical systems can be formulated as control/optimization problems. The computational models for such systems generally contain hyperparameters, which control solution fidelity and computational expense. The tuning of these parameters is non-trivial and the general approach is to manually ‘spot-check’ for good combinations. This is because optimal hyperparameter configuration search becomes intractable when the parameter space is large and when they may vary dynamically. To address this issue, we present a framework based on deep reinforcement learning (RL) to train a deep neural network agent that controls a model solve by varying parameters dynamically. First, we validate our RL framework for the problem of controlling chaos in chaotic systems by dynamically changing the parameters of the system. Subsequently, we illustrate the capabilities of our framework for accelerating the convergence of a steady-state computational fluid dynamics solver by automatically adjusting the relaxation factors of the discretized Navier–Stokes equations during run-time. The results indicate that the run-time control of the relaxation factors by the learned policy leads to a significant reduction in the number of iterations for convergence compared to the random selection of the relaxation factors. Our results point to potential benefits for learning adaptive hyperparameter learning strategies across different geometries and boundary conditions with implications for reduced computational campaign expenses 4 4 Data and codes available at https://github.com/Romit-Maulik/PAR-RL . .

42 ENGINEERING↗
DOE OSTI2020

Addressing Human and Organizational Factors in Nuclear Industry Modernization: An Operationally Focused Approach to Process and Methodology

Utility owners and operators of commercial nuclear power plants in the United States (U.S.) are and will be modernizing their nuclear power plants by performing a digital transformation involving design of an integrated set of systems that together enable a technology centric operating plant. The Plant Modernization Pathway of the U.S. Department of Energy Light Water Reactor Sustainability Program has a strategic action plan that lays the groundwork for a digital transformation of the nuclear industry. The model for this transformation is an advanced concept of operations, with an end point vision, “To achieve the maximum aggregate benefit enabled by this digital transformation.” To achieve this, the digital infrastructure for a nuclear plant must be designed as an integrated set of systems that together enable a technology centric operating model. The digital transformation process obviously needs to involve technology considerations and systems engineering, but it also needs to include human and organizational expertise. Thus, human and organizational factors, including sociotechnical systems methods and techniques (e.g., Cognitive Systems Engineering, Systems Theoretic Accident Modeling and Processes, human systems integration, and Macroergonomics) need to be considered for digital transformation projects in order to effectively integrate human and organizational expertise efforts into the new work system that results from nuclear power plant digital modernization. That is, the work system is the basic unit of sociotechnical systems analysis and contains three components: personnel, technical, and organization and management. These components should be jointly optimized with respect to the interdependence of systems performance criteria of effectiveness, efficiency and safety. Joint optimization can be achieved through the application of three human and organization functions: knowledge representation, knowledge elicitation, and cross-functional integration. This report provides a strategic framework for effective integration of human and organizational expertise within nuclear power plant digital modernization efforts.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2026

Tantalum alloy–based resonators for quantum information systems

Utilizing tantalum (Ta) in superconducting circuits has led to significant improvements, such as high qubit lifetime (T 1 ) and quality factors in both qubits and resonators, suggesting that material optimization plays an important role in the development of superconducting circuits. Thus we here explore superconducting gap engineering in Ta-based devices as a powerful strategy for expanding the range of suitable host materials. By alloying 20 atomic percent (at.%) hafnium into Ta thin films, we achieve a superconducting transition temperature (T c ) of 6.09 K as observed in direct current (DC) transport measurements, reflecting an increase in the superconducting gap. We systematically vary deposition conditions to control film orientation and transport properties of Ta-Hf alloy thin films. We then confirm the enhancement in T c via microwave measurements at millikelvin temperatures. We verify the ~40 increase in T c relative to bare Ta devices, while the loss contributions from two-level systems and quasi-particles remain unchanged in the low temperature regime. These findings emphasize the promise of material engineering in superconducting circuits and point to many potential material candidates for further exploration.

Ta-Hf alloy thin films↗
DOE OSTI2021

Plateau Phenomenon in Gradient Descent Training of RELU Networks: Explanation, Quantification, and Avoidance

The ability of neural networks to provide ‘best in class’ approximation across a wide range of applications is well-documented. Nevertheless, the powerful expressivity of neural networks comes to naught if one is unable to effectively train (choose) the parameters defining the network. In general, neural networks are trained by gradient descent type optimization methods,a stochastic variant thereof. In practice, such methods result in the loss function decreases rapidly at the beginning of training but then, after a relatively small number of steps, significantly slow down. The loss may even appear to stagnate over the period of a large number of epochs, only to then suddenly start to decrease fast again for no apparent reason. This so-called plateau phenomenon manifests itself in many learning tasks. The present work aims to identify and quantify the root causes of plateau phenomenon.analysis is carried out in the setting of univariate ReLU networks. No assumptions are made on the number of neurons relative to the number of training data, and our results hold for both the lazy and adaptive regimes. Here, the main findings are: plateaux correspond to periods during which activation patterns remain constant, where activation pattern refers to the number of data points that activate a given neuron; quantification of convergence of the gradient flow dynamics; and, characterization stationary points in terms solutions of local least squares regression lines over subsets of the training data. Based on these conclusions, we propose a new iterative training method, the Active Neuron Least Squares (ANLS), characterised by the explicit adjustment of the activation pattern at each step, which is designed to enable a quick exit from a plateau. Illustrative numerical examples are included throughout.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Impact of toroidal magnetic field direction on integrated ELM-stable operation and divertor power exhaust via boron powder injection in EAST

We report the first in-depth comparison of the impact of toroidal magnetic field direction on solid boron injection used for Edge-Localized Mode (ELM) control, power exhaust, and core high-Z impurity control in the Experimental Advanced Superconducting Tokamak. With favorable ion ∇B drift towards the upper X-point in an upper-single-null configuration, boron injection effectively suppresses ELMs, produces a detachment of the inner divertor target, and leads to improved energy confinement. ELM suppression in this configuration is accompanied by the excitation of an Edge Harmonic Mode. In contrast, with unfavorable ion ∇B drift away from the upper X-point, boron injection also suppresses ELMs but leads to a more symmetric detachment state of both the inner and outer divertor targets, while plasma energy confinement is slightly degraded despite similar boron injection levels; a different low-frequency coherent mode without multiple harmonics is observed. Measurements from toroidally separated views show that the divertor response to boron injection is essentially toroidally symmetric, supporting the use of two-dimensional SOLPS-ITER modeling with a toroidally uniform impurity source. These experimental observations are qualitatively consistent with SOLPS-ITER simulations, which highlight the critical role of E × B drift effects in setting the Bt-dependent in–out asymmetry of detachment and in asymmetrically transporting particles and injected impurities within the scrape-off layer and private-flux region. These findings underscore the importance of drift physics and real-time wall conditioning in controlling low-Z impurity transport and optimizing edge solutions for integrated, ELM-stable, high-performance tokamak operation.

E × B drift↗
DOE OSTI2023

Wassersplines for Neural Vector Field-Controlled Animation

Much of computer-generated animation is created by manipulating meshes with rigs. While this approach works well for animating articulated objects like animals, it has limited flexibility for animating less structured free-form objects. Here we introduce Wassersplines, a novel trajectory inference method for animating unstructured densities based on recent advances in continuous normalizing flows and optimal transport. The key idea is to train a neurally-parameterized velocity field that represents the motion between keyframes. Trajectories are then computed by advecting keyframes through the velocity field. We solve an additional Wasserstein barycenter interpolation problem to guarantee strict adherence to keyframes. Our tool can stylize trajectories through a variety of PDE-based regularizers to create different visual effects. We demonstrate our tool on various keyframe interpolation problems to produce temporally-coherent animations without meshing or rigging.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

ElectroCat: Platinum-Group-Metal-Free Engineered Framework Nano-Structure Catalyst (Final Technical Report)

Traditional synthesis methodologies that produce durable, highly active PGM-free electrocatalysts typically consist of a top-down approach utilizing a nitrogen-coordinating precursor, a metal precursor, and a support material. Nitrogen-coordinated transition metal complexes typically are absorbed onto a support material, which is then activated via high-temperature pyrolysis. Selection of the support and the nitrogen coordinating precursor has long been the focal point of efforts to maximize the number of accessible, catalytically active sites. In this project, GWE in collaboration with SRNL and NU will develop durable, highly active, low-cost, PGM-free electrocatalysts for polymer electrolyte membrane fuel cells designed and synthesized using a unique, bottom-up, rationally designed approach. Electrocatalysts are produced via polymerization reaction followed by a high-temperature pyrolysis step. Additionally, this project focuses on catalyst site and ORR mechanistic modeling to better understand the structure-property relationship.

08 HYDROGEN↗
DOE OSTI2023

Calculated covariance matrices for fission product yields using BeoH

Fission product yields (FPY) are important for a variety of applications (reactor neutronics, spent fuel, dosimetry, radiochemistry, etc.) and are currently included in many of the evaluated libraries around the world. The FPYs in the current US evaluation, ENDF/B-VIII.0, are mainly based on the 1994 evaluation of England and Rider and have only had slight updates—such as the inclusion of a 2 MeV point for 239 Pu—since their development. Additionally, only mean values and uncertainties are included in the evaluation, not full correlations. Los Alamos National Laboratory, in collaboration with several other institutes, has been working on an updated evaluation for the FPYs of 239 Pu(n,f), 235 U(n,f), 238 U(n,f), and 252 Cf(sf) using the deterministic, Hauser-Feshbach, fission fragment decay code, BeoH. BeoH calculates the FPYs consistently with many other prompt and delayed fission observables, explicitly taking into account multi-chance fission and ensuring consistency between observables. In addition to providing updated means and uncertainties for the FPYs on a pointwise energy grid from thermal to 20 MeV, we calculate correlations between all FPYs at each incident energy and across incident energies. Here, we discuss the development of these covariance matrices, differences in the correlations between FPYs based on the parameters that are included in the model optimization, and correlations across incident energies for neutron-induced fission.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Dean vortex-enhanced blood plasma separation in self-driven spiral microchannel flow with cross-flow microfilters

Point-of-care (POC) diagnostic devices have been developing rapidly in recent years, but they are mainly using saliva instead of blood as a test sample. A highly efficient self-separation during the self-driven flow without power systems is desired for expanding the point-of-care diagnostic devices. Microfiltration stands out as a promising technique for blood plasma separation but faces limitations due to blood cell clogging, resulting in reduced separation speed and efficiency. These limitations are mainly caused by the high viscosity and hematocrit in the blood flow. A small increment in the hematocrit of the blood significantly increases the pressure needed for the blood plasma separation in the micro-filters and decreases the separation speed and efficiency. Addressing this challenge, this study explores the feasibility of diluting whole blood within a microfluidic device without external power systems. This study implemented a spiral microchannel utilizing the inertial focusing and Dean vortex effects to focus the red blood cells and extract the blood with lower hematocrit. The inertial migration of the particles during the capillary flow was first investigated experimentally; a maximum of 88% of the particles migrated to the bottom and top equilibrium positions in the optimized 350 × 60 μm (cross-sectional area, 5.8 aspect ratio) microchannel. With the optimized dimension of the microchannel, the whole blood samples within the physiological hematocrit range were tested in the experiments, and more than 10% of the hematocrit reduction was compared between the outer branch outlet and inner branch outlet in the 350 × 60 μm microchannel.

Biochemistry & Molecular Biology↗
DOE OSTI2026

Regression Analysis with the Directed Infusion of Data

Integrating artificial intelligence and machine learning tools into industry necessitates large-scale collaborative efforts that ensure the robust and accurate execution of downstream analytics such as time series prediction, uncertainty quantification, grid optimization, and condition monitoring. However, concerns related to data privacy pervade the nuclear industry due to the proprietary nature of its data and the possibility of data leakage. Legacy techniques such as encryption often require the explicit transmission of data to trustworthy parties, thereby inviting data leakage concerns. The ideal collaboration scenario avoids the explicit dissemination of data/code while maintaining experimental fidelity, which is currently accomplished using various techniques such as trusted execution environments, homomorphic encryption, differential privacy, and multimatrix masking. These techniques, however, often necessitate a trade-off between trust, efficiency, and utility. This article extends a previously proposed technique called the directed infusion of data (DIOD) that ensures data privacy, allows for scalable obfuscation, and combats the risk of data leakage without compromising utility. The experiments discussed in this article examine a regression-type scenario using DIOD with the goal of preserving the inferential link between two variables. Using the point-kinetics equations, regression experiments compare the performance of a model trained using the original data to that of a model trained using the obfuscated data, which produced identical results. Our claim is further strengthened by an information theoretic proof and experiment, which showed that the inferential content between variables remains the same after obfuscation, thereby avoiding the required communication of the proprietary data.

47 - OTHER INSTRUMENTATION↗
DOE OSTI2025

Signal-preserving CMB component separation with machine learning

Analysis of microwave sky signals, such as the cosmic microwave background, often requires component separation using multifrequency methods, whereby different signals are isolated according to their different frequency behaviors. Many so-called blind methods, such as the internal linear combination (ILC), make minimal assumptions about the spatial distribution of the signal or contaminants, and only assume knowledge of the frequency dependence of the signal. The ILC produces a minimum-variance linear combination of the measured frequency maps. In the case of Gaussian, statistically isotropic fields, this is the optimal linear combination, as the variance is the only statistic of interest. However, in many cases the signal we wish to isolate, or the foregrounds we wish to remove, are non-Gaussian and/or statistically anisotropic (in particular for the case of Galactic foregrounds). In such cases, it is possible that machine learning (ML) techniques can be used to exploit the non-Gaussian features of the foregrounds and thereby improve component separation. However, many ML techniques require the use of complex, difficult-to-interpret operations on the data. We propose a hybrid method whereby we train an ML model using only combinations of the data that , and combine the resulting ML-predicted foreground estimate with the ILC solution to reduce the error from the ILC. We demonstrate our methods on simulations of extragalactic temperature and Galactic polarization foregrounds and show that our ML model can exploit non-Gaussian features, such as point sources and spatially varying spectral indices, to produce lower-variance maps than ILC—e.g., reducing the variance of the B-mode residual by factors of up to 5—while preserving the signal of interest in an unbiased manner. Moreover, we often find improved performance even when applying our ML technique to foreground models on which it was not trained. Published by the American Physical Society 2025

McCarthy, Fiona (ORCID:0000000253893565)↗