Engineering Papers⌕ Search

Engineering topics

Lin, Sen

Publications and source records attributed to Lin, Sen.

11 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Continual Learning for Particle Accelerators

Particle accelerators operate under dynamically changing conditions, which often lead to data distribution drifts. These drifts pose significant challenges for Machine Learning (ML) models, which typically fail to maintain performance when faced with such non-stationary data. In particle accelerators, the primary sources of these data drifts include changes in accelerator settings and non-measured parameters such as machine degradation and environmental factors. Previous research has proposed conditional models to handle multiple beam configurations effectively; however, it is challenging to train the ML models on all possible configuration settings. Additionally, conditional models alone can not address performance degradation caused by drifts due to non-measured factors. These limitations contribute to a significant gap between ML development and its deployment in real-world operational settings. To bridge this gap, in this paper, we identify some of the key areas within particle accelerators where continual learning can help mitigate drift-induced performance degradation. In addition, we present a practical use case where a conditional Auto-Encoder model coupled with memory-based continual learning has been employed to demonstrate stable performance even when underlying data drifts.

Schram, Malachi [Thomas Jefferson National Acceler↗
DOE OSTI2026

Continual Learning for Particle Accelerators

Particle accelerators operate under dynamically changing conditions, which often lead to data distribution drifts. These drifts pose significant challenges for Machine Learning (ML) models, which typically fail to maintain performance when faced with such non-stationary data. In particle accelerators, the primary sources of these data drifts include changes in accelerator settings and non-measured parameters such as machine degradation and environmental factors. Previous research has proposed conditional models to handle multiple beam configurations effectively; however, it is challenging to train the ML models on all possible configuration settings. Additionally, conditional models alone can not address performance degradation caused by drifts due to non-measured factors. These limitations contribute to a significant gap between ML development and its deployment in real-world operational settings. To bridge this gap, in this paper, we identify some of the key areas within particle accelerators where continual learning can help mitigate drift-induced performance degradation. In addition, we present a practical use case where a conditional Auto-Encoder model coupled with memory-based continual learning has been employed to demonstrate stable performance even when underlying data drifts

Rajput, Kishansingh [Thomas Jefferson National Acc↗
DOE OSTI2025

Uncertainty based Online Ensemble on Non-Stationary Data for Fusion Science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior due to drifts in the data. The drifts can arise from both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with non-stationary data streams.Online learning can be used to continuously adapt the models with new data as it is acquired. However, traditional online learning can suffer from short-term performance degradation, as ground truth are not available before making the prediction. To address this challenge, we propose uncertainty aware ensemble approach for online learning. We use Deep Gaussian Process Approximation (DGPA) technique for calibrated uncertainty estimation and use the uncertainty values to guide a meta-algorithm that produces predictions based on ensemble of learners. Moreover, DGPA also provides uncertainty estimation along with the predictions for decision makers. This paper demonstrates that the proposed method outperforms traditional online learning approach, and a naive ensemble without uncertainty guidance by about 7% and 6%, respectively, on B-coil deflection prediction at DIII-D Fusion Facility.

Rajput, Kishansingh [Thomas Jefferson National Acc↗
DOE OSTI2025

Uncertainty based Online Ensemble on Non-Stationary Data for Fusion Science

Machine Learning (ML) is poised to play a pivotal role in the development and operation of next-generation fusion devices. Fusion data shows non-stationary behavior due to drifts in the data. The drifts can arise from both experimental evolution and machine wear-and-tear. ML models assume stationary distribution and fail to maintain performance when encountered with non-stationary data streams.Online learning can be used to continuously adapt the models with new data as it is acquired. However, traditional online learning can suffer from short-term performance degradation, as ground truth are not available before making the prediction. To address this challenge, we propose uncertainty aware ensemble approach for online learning. We use Deep Gaussian Process Approximation (DGPA) technique for calibrated uncertainty estimation and use the uncertainty values to guide a meta-algorithm that produces predictions based on ensemble of learners. Moreover, DGPA also provides uncertainty estimation along with the predictions for decision makers. This paper demonstrates that the proposed method outperforms traditional online learning approach, and a naive ensemble without uncertainty guidance by about 7% and 6%, respectively, on B-coil deflection prediction at DIII-D Fusion Facility.

Rajput, Kishansingh [Thomas Jefferson National Acc↗
DOE OSTI2023

Online data-driven changepoint detection for high-dimensional dynamical systems

In this study, the detection of anomalies or transitions in complex dynamical systems is of critical importance to various applications. In this study, we propose the use of machine learning to detect changepoints for high-dimensional dynamical systems. Here, changepoints indicate instances in time when the underlying dynamical system has a fundamentally different characteristic—which may be due to a change in the model parameters or due to intermittent phenomena arising from the same model. We propose two complementary approaches to achieve this, with the first devised using arguments from probabilistic unsupervised learning and the latter devised using supervised deep learning. To accelerate the deployment of transition detection algorithms in high-dimensional dynamical systems, we introduce dimensionality reduction techniques. Our experiments demonstrate that transitions can be detected efficiently, in real-time, for the two-dimensional forced Kolmogorov flow and the Rössler dynamical system, which are characterized by anomalous regimes in phase space where dynamics are perturbed off the attractor at potentially uneven intervals. Finally, we also demonstrate how variations in the frequency of detected changepoints may be utilized to detect a significant modification to the underlying model parameters by utilizing the Lorenz-63 dynamical system.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Assessing density functionals for describing methane dissociative chemisorption on Pt(110)-(2×1) surface

In this work, we explore the suitability of several density functionals with the generalized gradient approximation (GGA) and beyond for describing the dissociative chemisorption of methane on the reconstructed Pt(110)-(2×1) surface. The bulk and surface structures of the metal, methane adsorption energy, and dissociation barrier are used to assess the functionals. A van der Waals corrected GGA functional (optPBE-vdW) and a metaGGA functional with van der Waals correction (MS PBEl-rVV10) are selected for ab initio molecular dynamics calculations of the sticking probability. Our results suggest that the use of these two functionals may lead to a better agreement with existing experimental results, thus serving as a good starting point for future development of reliable machine-learned potential energy surfaces for the dissociation of methane on the Pt(110)-(2×1) surface.

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

Vapor-phase self-assembly for generating thermally stable single-atom catalysts

Preparation of thermally stable metal single-atom catalysts (SACs) is a challenge in heterogeneous catalysis, especially on conventional supports that provide a weak metal-support interaction. In this work, we report that a modified support MgAl 2 O 4 can stabilize Pt single atoms by a mechanism of vapor-phase self-assembly in a high-temperature treatment (800°C, air). The experimental results on the formation mechanism and the structure are validated by DFT and ab initio molecular dynamics simulations. We infer that stable triangular K 3 O 3 structures help stabilize Pt single atoms at high temperatures in oxidizing conditions, exhibiting excellent reactivity for methane oxidation. The obtained Pt/K/MgAl 2 O 4 SAC presents excellent stability in methane oxidation after steam treatment at elevated temperatures, whereas the Pt/MgAl 2 O 4 nanocatalyst suffers from rapid deactivation due to Pt nanoparticle growth. Now this work paves the way for preparing thermally stable and highly active SACs using conventional high-surface-area supports, despite the weak metal-support interaction.

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

Engineering catalyst supports to stabilize PdO x two-dimensional rafts for water-tolerant methane oxidation

The treatment of emissions from natural gas engines is an important area of research since methane is a potent greenhouse gas. The benchmark catalysts, based on Pd, still face challenges such as water poisoning and long-term stability. In this work, we report an approach for catalyst synthesis that relies on the trapping of metal single atoms on the support surface, in thermally stable form, to modify the nature of further deposited metal/metal oxide. By anchoring Pt ions on a catalyst support we can tailor the morphology of the deposited phase. In particular, two-dimensional (2D) rafts of PdO x are formed, resulting in higher reaction rates and improved water tolerance during methane oxidation. The results show that modifying the support by trapping single atoms could provide an important addition to the toolkit of catalyst designers for controlling the nucleation and growth of metal and metal oxide clusters in heterogeneous catalysts.

2D raft↗
DOE OSTI2021

Unraveling the Intermediate Reaction Complexes and Critical Role of Support-Derived Oxygen Atoms in CO Oxidation on Single-Atom Pt/CeO 2

CeO 2 -supported Pt single-atom catalysts have been extensively studied due to their relevance in automobile emission control and for the fundamental understanding of CeO 2 -based catalysts. Though CeO 2 -supported Pt nanoparticles are often more active than their single-atom counterparts, the former could easily redisperse to Pt single atom under oxidizing diesel conditions. Therefore, to maximize the reactivity of every Pt atom, it is important to fully understand the reaction mechanism of CeO 2 -supported Pt single atoms. Here, we report a CO oxidation study on a Pt/CeO 2 single-atom catalyst, where we can account for all of the neighbors using in situ and operando spectroscopy techniques and microcalorimetric measurements. Coupled with density functional theory calculations, we present a comprehensive picture of the dynamics of the surface species, the role of surface intermediates, and explain the observed reaction kinetics. We started with a catalyst containing exclusively single atoms and used in situ/operando spectroscopy to provide evidence for their stability during the reaction and to identify the Pt 1 complexes before and during the reaction and their binding to CO. The results reveal that in the precatalyst, Pt is present as Pt(O) 4 on the CeO 2 (111) step edge sites, but during CO oxidation, we find that two Pt 1 complexes coexist, representing two states of the same active site in the reaction cycle. The dominant state/complex remains Pt(O) 4 , which adsorbs CO very weakly as shown by CO microcalorimetry. The second, minority state/complex, Pt(CO)(O) 3 is generated through the reaction of Pt(O) 4 with CO, and CO is bound strongly to Pt 1 . Labile oxygen adatoms from the CeO 2 surface play a major role in the regeneration of Pt(O) 4 either directly from Pt(O) 3 or by reaction with the strongly adsorbed CO in Pt(CO)(O) 3 . We show that the formation of an oxygen vacancy and generation of a labile O* are not barrierless, which explains the long lifetime of Pt(CO)(O) 3 and its detectability despite being a minority complex. The results help to develop a comprehensive view of the dynamic evolution of Pt 1 complexes along the reaction cycle and provide mechanistic insights to guide the design of Pt-based single-atom catalysts.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2019

Environmentally benign synthesis of a PGM-free catalyst for low temperature CO oxidation

Dopants enhance the catalytic properties of ceria. However, conventional techniques for synthesizing doped ceria have limitations in terms of structural homogeneity, surface area, and catalytic activity of the resulting oxide. Use of toxic and corrosive chemicals presents further challenges. The sol-gel method described in this work provides a facile approach for incorporating high concentrations of dopants in a uniform, high surface area structure, yielding excellent catalytic activity. Addition of polyvinylpyrrolidone (PVP) complexing agent prevents the segregation of cerium and dopant atoms during synthesis. Surface areas up to 179 m 2 /g are achieved, which represents a substantial improvement over doped ceria produced through coprecipitation, solution combustion, or melt-synthesis methods. The resulting powders exhibit dramatically improved CO oxidation activity (T 90 = 132 °C for 3.2 wt% Cu-CeO 2 compared to 274 °C for a 2 wt% Pt-Al 2 O 3 reference catalyst). First principles calculations suggest a Mars Van Krevelen mechanism, which is facilitated by dopants causing oxygen vacancies.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗