Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learning rate”

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 721 records · Page 40
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Using Active Learning to Rapidly Develop Machine Learned Diffusion Coefficients of CO 2 Conversion Reagents in Metal–Organic Frameworks

Here, we used a combined molecular dynamics/active learning (AL) approach to create machine learning models that can predict the diffusion coefficient of epichlorohydrin and chloropropene carbonate, the reactant and product of a common CO 2 cycloaddition reaction, in metal–organic frameworks (MOFs). Nanoporous MOFs are effective catalysts for the cycloaddition of CO 2 to epoxides. The diffusion rates within nanoporous catalysts can control the rate of reaction as the reactants and products must diffuse to the active sites within the MOF and then out of the nanoporous material for reusability. However, the diffusion process is routinely ignored when searching for new materials in catalytic applications. Here we verified improvement during the AL process by consistently tracking metrics on the same groups of MOFs to ensure consistency. Metal identity was found to have little impact on diffusion rates, while structural features like pore limiting diameter act as a threshold where a minimum value is needed for high diffusion rates. We identified the MOFs with the highest epichlorohydrin and chloropropene carbonate diffusion coefficients which can be used for further studies of reaction energetics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Neural network potential for Zr–Rh system by machine learning

Zr–Rh metallic glass has enabled its many applications in vehicle parts, sports equipment and so on due to its outstanding performance in mechanical property, but the knowledge of the microstructure determining the superb mechanical property remains yet insufficient. Here, we develop a deep neural network potential of Zr–Rh system by using machine learning, which breaks the dilemma between the accuracy and efficiency in molecular dynamics simulations, and greatly improves the simulation scale in both space and time. Here, the results show that the structural features obtained from the neural network method are in good agreement with the cases in ab initio molecular dynamics simulations. Furthermore, we build a large model of 5400 atoms to explore the influences of simulated size and cooling rate on the melt-quenching process of Zr 77 Rh 23 . Our study lays a foundation for exploring the complex structures in amorphous Zr 77 Rh 23 , which is of great significance for the design and practical application.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2023

Predicting oxidation damage of ultra high-temperature carbide ceramics in extreme environments using machine learning

Determining the oxidation resistance of UHTC carbides in extreme environments is challenging theoretically and experimentally due to the high dimensional complexity of influencing variables and intricate testing setups. Herein we demonstrate the use of machine learning (ML) models trained with experimental literature data to predict the oxide thickness of UHTC carbides exposed to air based on composition, mean grain size, relative densification, holding time, and temperature. A multi-dimensional database with 76 occurrences is created containing experimental results of Hf, Zr, and Ta carbides plus additives. In this study, the preprocessed database is then used to train ML models to predict their oxidation behavior. The trained model predicts the oxidation damage in the form of an average oxide thickness in UHTC carbides with a Mean Absolute Error (MAE) of ±65.45 μm for samples in the testing set that developed thicknesses up to 1000 μm. The model successfully predicted oxidation damage for a recession rate lower than 60 μm/min. It is noticed that the ensemble method MAE is increased to ±134.34 μm while forecasting the oxidation of samples with a recession rate higher than the threshold. The unprecedented approach is a novel way to predict the damage through the oxidation of carbide compounds before processing for a smarter design with room for improvement.

36 MATERIALS SCIENCE↗
DOE OSTI2020

NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

In the last 50 years there has been a tremendous progress in solving numerically the Navier-Stokes equations using finite differences, finite elements, spectral, and even meshless methods. Yet, in many real cases, we still cannot incorporate seamlessly (multi-fidelity) data into existing algorithms, and for industrial-complexity applications the mesh generation is time consuming and still an art. Moreover, solving ill-posed problems (e.g., lacking boundary conditions) or inverse problems is often prohibitively expensive and requires different formulations and new computer codes. Here, we employ physics-informed neural networks (PINNs), encoding the governing equations directly into the deep neural network via automatic differentiation, to overcome some of the aforementioned limitations for simulating incompressible laminar and turbulent flows. We develop the Navier-Stokes flow nets (NSFnets) by considering two different mathematical formulations of the Navier-Stokes equations: the velocity-pressure (VP) formulation and the vorticity-velocity (VV) formulation. Since this is a new approach, we first select some standard benchmark problems to assess the accuracy, convergence rate, computational cost and flexibility of NSFnets; analytical solutions and direct numerical simulation (DNS) databases provide proper initial and boundary conditions for the NSFnet simulations. The spatial and temporal coordinates are the inputs of the NSFnets, while the instantaneous velocity and pressure fields are the outputs for the VP-NSFnet, and the instantaneous velocity and vorticity fields are the outputs for the VV-NSFnet. This is unsupervised learning and, hence, no labeled data are required beyond boundary and initial conditions and the fluid properties. The residuals of the VP or VV governing equations, together with the initial and boundary conditions, are embedded into the loss function of the NSFnets. No data is provided for the pressure to the VP-NSFnet, which is a hidden state and is obtained via the incompressibility constraint without extra computational cost. Unlike the traditional numerical methods, NSFnets inherit the properties of neural networks (NNs), hence the total error is composed of the approximation, the optimization, and the generalization errors. Here, we empirically attempt to quantify these errors by varying the sampling (“residual”) points, the iterative solvers, and the size of the NN architecture. For the laminar flow solutions, we show that both the VP and the VV formulations are comparable in accuracy but their best performance corresponds to different NN architectures. The initial convergence rate is fast but the error eventually saturates to a plateau due to the dominance of the optimization error. For the turbulent channel flow, we show that NSFnets can sustain turbulence at , but due to expensive training we only consider part of the channel domain and enforce velocity boundary conditions on the subdomain boundaries provided by the DNS data base. We also perform a systematic study on the weights used in the loss function for balancing the data and physics components, and investigate a new way of computing the weights dynamically to accelerate training and enhance accuracy. In the last part, we demonstrate how NSFnets should be used in practice, namely for ill-posed problems with incomplete or noisy boundary conditions as well as for inverse problems. We obtain reasonably accurate solutions for such cases as well without the need to change the NSFnets and at the same computational cost as in the forward well-posed problems. As a result, we also present a simple example of transfer learning that will aid in accelerating the training of NSFnets for different parameter settings.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

The drivers and predictability of wildfire re-burns in the western United States (US)

Evidence is mounting that the effectiveness of using prescribed burns as a management tactic may be diminishing due to the higher incidence of wildfire re-burns. The development of predictive models of re-burns is thus essential to better understand their primary drivers so that forest management practices can be updated to account for these events. First, we assess the potential for human activity as a driver of re-burns by evaluating re-burn trends both within and outside of the wildland–urban interface (WUI) of the western US. Next, we investigate the predictability of re-burns through the application of both random forest and the explanatory machine learning non-negative matrix factorization using k-means clustering (NMFk) algorithms to predict re-burn occurrence over California based on a number of climate factors. Our findings indicate that while most states showed increasing trends within the WUI when trends were conducted over longer moving windows (e.g. 20 years), California was the only state where the rate of increase was consistently higher in the WUI, indicating a stronger potential for human activity as a driver in that location. Furthermore, we find model performance was found to be robust over most of California (Testing F1 scores = 0.688), although results were highly variable based on EPA level III Ecoregion (F1 scores = 0.0–0.778). Insights provided from this study will lead to a better understanding of climate and human activity drivers of re-burns and how these vary at broad spatial scales so that improvements in forest management practices can be tuned according to the level of change that is expected for a given region.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Charged particle reconstruction in CLAS12 using Machine Learning

In this work, we present studies of track parameter reconstruction from raw information in CLAS12 detector's Drift Chambers, using Machine Learning (ML). We study the resolution of tracks reconstructed with different types of ML models/algorithms, including Multi-Layer Perceptron (MLP), Extremely Randomized Trees (ERT) and Gradient Boosting Trees (GBT) using simulated data. We find that the resulting ML model is capable of reconstructing track parameters (particle momentum, and polar and azimuthal angles) with accuracy similar to Hit Based (HB) tracking code, but $150$ times faster. Moreover, physics reactions can be identified using the particles reconstructed by the neural network in real-time (with a rate of about $34~kHz$) during experimental data collection. The developed model can be used in numerous applications, such as triggering specific physics reactions in real-time, detector performance monitoring, and real-time detector calibration.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2025

Leveraging large language models to automate the identification of healthcare access barriers for veterans

Objective: To develop and evaluate an automated system for identifying healthcare barriers focusing on transportation issues in veterans’ clinical notes using large language models (LLMs) and to assess the impact of different prompting strategies on classification performance and explanation consistency. Methods: We developed a hybrid system combining pattern matching for templated notes with LLM analysis for free-text notes. Using 2000 manually annotated clinical notes, we compared four prompting strategies (dual-role short, dual-role long, analysis-first, analysis-only) across Mistral-7B and Llama-3.1 models. We evaluated classification performance using standard metrics and assessed explanation consistency through embedding similarity analysis. Results: The analysis-first strategy achieved superior performance, with Mistral-7B reaching an F1 score of 0.914, outperforming traditional machine learning approaches (GBM: 0.786, BERT: 0.811). LLMs demonstrated higher explanation consistency within models (mean cosine similarity 0.887–0.908) compared to cross-model similarities (0.767–0.872). Pattern matching successfully handled 6.7% of templated notes deterministically. Mistral-7B showed greater internal consistency but higher abstention rates compared to Llama-3.1. Conclusion: Requiring LLMs to analyze evidence before classification improves both accuracy and explanation consistency for identifying transportation barriers in clinical notes. This approach enables automated barrier detection at scale while providing clinically relevant explanations, supporting both population-level healthcare planning and individual patient care decisions.

Healthcare access barriers↗
DOE OSTI2024

Real-time confinement regime detection in fusion plasmas with convolutional neural networks and high-bandwidth edge fluctuation measurements

Abstract A real-time detection of the plasma confinement regime can enable new advanced plasma control capabilities for both the access to and sustainment of enhanced confinement regimes in fusion devices. For example, a real-time indication of the confinement regime can facilitate transition to the high-performing wide-pedestal (WP) quiescent H-mode, or avoid unwanted transitions to lower confinement regimes that may induce plasma termination. To demonstrate real-time confinement regime detection, we use the 2D beam emission spectroscopy (BES) diagnostic system to capture localized density fluctuations of long wavelength turbulent modes in the edge region at a 1 MHz sampling rate. BES data from 330 discharges in either L-mode, H-mode, quiescent H (QH)-mode, or WP QH-mode were collected from the DIII-D tokamak and curated to develop a high-quality database to train a deep-learning classification model for real-time confinement detection. We utilize the 6×8 spatial configuration with a time window of 1024 µ s and recast the input to obtain spectral-like features via fast Fourier transform preprocessing. We employ a shallow 3D convolutional neural network for the multivariate time-series classification task and utilize a softmax in the final dense layer to retrieve a probability distribution over the different confinement regimes. Our model classifies the global confinement state on 44 unseen test discharges with an average F 1 score of 0.94, using only ∼1 ms snippets of BES data at a time. This activity demonstrates the feasibility for real-time data analysis of fluctuation diagnostics in future devices such as ITER, where the need for reliable and advanced plasma control is urgent.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2020

Automated, Unmanned Detection and Quantification of Fugitive Methane Emissions

Compressor stations used to move natural gas are one of the largest sources of fugitive methane emissions in the midstream sector, accounting for approximately 50% of all fugitive emissions (Zimmerle et al., 2015). This problem is most widespread at reciprocating compressors (Subramanian et al., 2015) where faulty seals are a key contributor to methane emissions (Johnson et al., 2015). As such, there is a significant need for a robust technology that could provide an early indication of an unexpected emission. Equally important, the technology needs to be able to account for biogenic versus anthropogenic sources of methane. One means of indirectly making this determination, is to leverage optical technologies that can autonomously pinpoint the source of such leaks. This presentation discusses recent work funded by the U.S. Department of Energy (DOE) National Energy Technology Laboratory (NETL), focused on the development of an innovative remote sensing technology that can reliably and autonomously detect fugitive methane emissions in near real-time, using computer vision and deep learning. The technology called the Smart Methane Leak Detection (SLED/M) system was initially developed to monitor facilities such as compressor stations in a stationary, pan-tilt-zoom configuration. The system has recently been adapted to monitor facilities from an unmanned aerial system (UAS). The speed and maneuverability of UAS platforms are attractive to leak detection and repair program operators, but introduce several challenges. Many existing methane detection algorithms rely on mostly static backgrounds becoming unusable with motion. In addition, top-down views of fugitive methane emissions present differently in Optical Gas Imagers (OGI) compared to looking across the plume. Our work has focused on overcoming these challenges, enhancing the operators ability to detect methane emissions, and pinpoint their sources. Another recent adaptation to SLED/M is the ability to quantify methane emissions using passive sensors (OGI, thermal camera), environmental conditions, plume modeling, and deep learning. SLED/M advances the state-of-the-art for methane emission detection and quantification by focusing on three key critical criteria for effective methane emission mitigation: (1) autonomy (no need for a human to be in the loop), (2) high reliability (low false alarm rates), and (3) real-time performance. Results from this work will be presented.

03 NATURAL GAS↗
DOE OSTI2022

Extracting the Subhalo Mass Function from Strong Lens Images with Image Segmentation

Abstract Detecting substructure within strongly lensed images is a promising route to shed light on the nature of dark matter. However, it is a challenging task, which traditionally requires detailed lens modeling and source reconstruction, taking weeks to analyze each system. We use machine learning to circumvent the need for lens and source modeling and develop a neural network to both locate subhalos in an image as well as determine their mass using the technique of image segmentation. The network is trained on images with a single subhalo located near the Einstein ring across a wide range of apparent source magnitudes. The network is then able to resolve subhalos with masses m ≳ 10 8.5 M ⊙ . Training in this way allows the network to learn the gravitational lensing of light, and, remarkably, it is then able to detect entire populations of substructure, even for locations further away from the Einstein ring than those used in training. Over a wide range of the apparent source magnitude, the false-positive rate is around three false subhalos per 100 images, coming mostly from the lightest detectable subhalo for that signal-to-noise ratio. With good accuracy and a low false-positive rate, counting the number of pixels assigned to each subhalo class over multiple images allows for a measurement of the subhalo mass function (SMF). When measured over three mass bins from 10 9 –10 10 M ⊙ the SMF slope is recovered with an error of 36% for 50 images, and this improves to 10% for 1000 images with Hubble Space Telescope-like noise.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

A Catalog of Candidate Double and Lensed Quasars from Gaia and WISE Data

Making use of strong correlations between closely separated multiple or double sources and photometric and astrometric metadata in Gaia Early Data Release 3 (EDR3), we generate a catalog of candidate double- and multiply imaged lensed quasars and active galactic nuclei (AGNs), comprising 3140 systems. It includes two partially overlapping parts: a sample of distant (redshifts mostly greater than 1) sources with perturbed data; and systems that have been resolved into separate components by Gaia at separations less than 2''. For the first part, which is roughly one-third of the published catalog, we synthesized 0.617 million redshifts using multiple machine-learning prediction and classification methods, using independent photometric and astrometric data from Gaia EDR3 and the Wide-field Infrared Survey Explorer, with accurate spectroscopic redshifts from the Sloan Digital Sky Survey (SDSS) as a training set. Using these synthetic redshifts, we estimate a 4.9% rate of interlopers with spectroscopic redshifts below 1 in this part of the catalog. Unresolved candidate double and dual AGNs and quasars are selected as sources with a marginally high BP/RP excess factor (phot_bp_rp_excess_factor), which is sensitive to source extent, limiting our search to high-redshift quasars. For the second part of the catalog, additional filters on measured parallax and near-neighbor statistics are applied to diminish the propagation of the remaining stellar contaminants. The estimated rate of the positives (double or multiple sources) is 98%, and the estimated rate of dual (physically related) quasars is greater than 54%. A few dozen serendipitously found objects of interest are discussed in more detail, including known and new lensed images, planetary nebulae, young IR stars of peculiar morphology, and quasars with catastrophic redshift errors in SDSS.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2021

Challenging Practices of Algebraic Battery Life Models through Statistical Validation and Model Identification via Machine-Learning

Various modeling techniques are used to predict the capacity fade of Li-ion batteries. Algebraic reduced-order models, which are inherently interpretable and computationally fast, are ideal for use in battery controllers, technoeconomic models, and multi-objective optimizations. For Li-ion batteries with graphite anodes, solid-electrolyte-interphase (SEI) growth on the graphite surface dominates fade. This fade is often modeled using physically informed equations, such as square-root of time for predicting solvent-diffusion limited SEI growth, and Arrhenius and Tafel-like equations predicting the temperature and state-of-charge rate dependencies. In some cases, completely empirical relationships are proposed. However, statistical validation is rarely conducted to evaluate model optimality, and only a handful of possible models are usually investigated. This article demonstrates a novel procedure for automatically identifying reduced-order degradation models from millions of algorithmically generated equations via bi-level optimization and symbolic regression. Identified models are statistically validated using cross-validation, sensitivity analysis, and uncertainty quantification via bootstrapping. On a LiFePO 4 /Graphite cell calendar aging data set, automatically identified models utilizing square-root, power law, stretched exponential, and sigmoidal functions result in greater accuracy and lower uncertainty than models identified by human experts, and demonstrate that previously known physical relationships can be empirically "rediscovered" using machine learning.

25 ENERGY STORAGE↗
DOE OSTI2023

Integrated deep learning framework for unstable event identification and disruption prediction of tokamak plasmas

Abstract The ability to identify underlying disruption precursors is key to disruption avoidance. In this paper, we present an integrated deep learning (DL) based model that combines disruption prediction with the identification of several disruption precursors like rotating modes, locked modes, H-to-L back transitions and radiative collapses. The first part of our study demonstrates that the DL-based unstable event identifier trained on 160 manually labeled DIII-D shots can achieve, on average, 84% event identification rate of various frequent unstable events (like H-L back transition, locked mode, radiative collapse, rotating MHD mode, large sawtooth crash), and the trained identifier can be adapted to label unseen discharges, thus expanding the original manually labeled database. Based on these results, the integrated DL-based framework is developed using a combined database of manually labeled and automatically labeled DIII-D data, and it shows state-of-the-art (AUC = 0.940) disruption prediction and event identification abilities on DIII-D. Through cross-machine numerical disruption prediction studies using this new integrated model and leveraging the C-Mod, DIII-D, and EAST disruption warning databases, we demonstrate the improved cross-machine disruption prediction ability and extended warning time of the new model compared with a baseline predictor. In addition, the trained integrated model shows qualitatively good cross-machine event identification ability. Given a labeled dataset, the strategy presented in this paper, i.e. one that combines a disruption predictor with an event identifier module, can be applied to upgrade any neural network based disruption predictor. The results presented here inform possible development strategies of machine learning based disruption avoidance algorithms for future tokamaks and highlight the importance of building comprehensive databases with unstable event information on current machines.

plasma instabilities↗
DOE OSTI2025

Domain Shift Analysis in Chest Radiographs Classification in a Veterans Healthcare Administration Population

This study aims to assess the impact of domain shift on chest X-ray classification accuracy and to analyze the influence of ground truth label quality and demographic factors such as age group, sex, and study year. We used a DenseNet121 model pre-trained MIMIC-CXR dataset for deep learning-based multi-label classification using ground truth labels from radiology reports extracted using the CheXpert and CheXbert Labeler. We compared the performance of the 14 chest X-ray labels on the MIMIC-CXR and Veterans Healthcare Administration chest X-ray dataset (VA-CXR). The validation of ground truth and the assessment of multi-label classification performance across various NLP extraction tools revealed that the VA-CXR dataset exhibited lower disagreement rates than the MIMIC-CXR datasets. Additionally, there were notable differences in AUC scores between models utilizing CheXpert and CheXbert. When evaluating multi-label classification performance across different datasets, minimal domain shift was observed in the unseen VA dataset, except for the label “Enlarged Cardiomediastinum.” The subgroup with the most significant variations in multi-label classification performance was study year. These findings underscore the importance of considering domain shift in chest X-ray classification tasks, paying particular attention to the temporality of the exam. Our study reveals the significant impact of domain shift and demographic factors on chest X-ray classification, emphasizing the need for improved transfer learning and robust model development. Addressing these challenges is crucial for advancing medical imaging research and improving patient care.

chest X-ray image classification↗
DOE OSTI2026

Variable rate neural compression for sparse detector data

Particle colliders produce data at extraordinary rates, posing major challenges for transmission and storage. High-throughput compression algorithms are therefore essential. In the sPHENIX experiment taking data at the Relativistic Heavy Ion Collider, a time projection chamber records three-dimensional (3D) particle trajectories that are highly sparse, making conventional learning-free lossy compression ineffective. Convolutional neural networks have surpassed traditional methods in compression ratio and accuracy. However, they fail to exploit sparsity for efficiency. To address these gaps, we present BCAE-VS, a bicephalous convolutional autoencoder with variable compression ratio for sparse data, which adapts compression to input complexity through key-point identification and sparse convolution. BCAE-VS achieves higher accuracy and compression ratios than prior neural approaches while being orders of magnitude smaller. Moreover, its throughput increases with sparsity—a property not observed in other methods. Although it was developed for collider experiments, BCAE-VS readily extends to other sparse data domains, such as light detection and ranging (LiDAR) sensing and 3D microscopy.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2022

Sub-10 second fly-scan nano-tomography using machine learning

Abstract X-ray computed tomography is a versatile technique for 3D structure characterization. However, conventional reconstruction algorithms require that the sample not change throughout the scan, and the timescale of sample dynamics must be longer than the data acquisition time to fulfill the stable sample requirement. Meanwhile, concerns about X-ray-induced parasite reaction and sample damage have driven research efforts to reduce beam dosage. Here, we report a machine-learning-based image processing method that can significantly reduce data acquisition time and X-ray dose, outperforming conventional approaches like Filtered-Back Projection, maximum-likelihood, and model-based maximum-a-posteriori probability. Applying machine learning, we achieve ultrafast nano-tomography with sub-10 second data acquisition time and sub-50 nm pixel resolution in a transmission X-ray microscope. We apply our algorithm to study dynamic morphology changes in a lithium-ion battery cathode under a heating rate of 50 o C min −1 , revealing crack self-healing during thermal annealing. The proposed method can be applied to various tomography modalities.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Generative models for simulation of KamLAND-Zen

Abstract The next generation of searches for neutrinoless double beta decay ($$0 \nu \beta \beta $$ 0 ν β β ) are poised to answer deep questions on the nature of neutrinos and the source of the Universe’s matter–antimatter asymmetry. They will be looking for event rates of less than one event per ton of instrumented isotope per year. To claim discovery, accurate and efficient simulations of detector events that mimic$$0 \nu \beta \beta $$ 0 ν β β is critical. Traditional Monte Carlo (MC) simulations can be supplemented by machine-learning-based generative models. This work describes the performance of generative models that we designed for monolithic liquid scintillator detectors like KamLAND to produce accurate simulation data without a predefined physics model. We present their current ability to recover low-level features and perform interpolation. In the future, the results of these generative models can be used to improve event classification and background rejection by providing high-quality abundant generated data.

Physics↗
DOE OSTI2026

Optimization and stabilization of Fermilab Booster using hybrid Bayesian/RL framework

PIPII project will raise Fermilab Booster intensity and ramp rate. Beam losses will limit average power and are hard to simulate. Presently, Booster uses operator-guided empirical tuning. This task is challenging due to high dimensionality, multiple objectives, critical safety constraints, and drifts. We developed a synergistic suite of Bayesian optimization (BO) and reinforcement learning (RL) tools to optimize and stabilize beam losses. First, active learning was used to build a rough model. Data was collected parasitically using two novel safety constraint types – nonlinear input space restrictions (based on optics model), and uncertainty constraints (to stop bad steps/beam aborts). We then applied online multi-objective BO with scalarized objectives and fitting to improve/rebalance losses, increasing safety margins by 25%. Using BO model as a safety veto, we tried several on/off-policy RL agents for long term stabilization; SAC had best performance. We found that adding contextual (state) information further improved performance, eventually integrating key knobs like linac phase and temperature into the parameter space. Long term testing is ongoing to enable operational use.

Kuklev, Nikita [Fermilab]↗