Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “active machine learning”

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 145 records · Page 8
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Leveraging gradient weighted class activation mapping to improve classification effectiveness: Case study in transportation infrastructure characterization

Roadway “corners†are common for pedestrian use, whether designated with markings or not. Different types of markings have been deployed, ranging from simple parallel lines to more complex designs. Understanding the impact of different types of crosswalks is important for public safety. In this work we explore methods to improve the logging of marked crosswalk types. We used the Roadway Information Database from the Second Strategic Highway Research Project and used active learning methods with transfer learning to identify the crosswalk types (marked or unmarked). Upon completion we found our classifiers were unable to perform above roughly 94% correct classifications. To improve their efficacy, we separated the crosswalks into their “fine grained†types and used Gradient-Weighted Class Activation Mapping to isolate and study the features that classified the crosswalks. We compared this with sampled manually marked crosswalks and present findings. We believe this use case can represent a process to improve the active learning method for some visual machine learning applications.

Karnowski, Thomas↗
DOE OSTI2022

Bayesian Active Learning for Scanning Probe Microscopy: From Gaussian Processes to Hypothesis Learning

Recent progress in machine learning methods and the emerging availability of programmable interfaces for scanning probe microscopes (SPMs) have propelled automated and autonomous microscopies to the forefront of attention of the scientific community. However, enabling automated microscopy requires the development of task-specific machine learning methods, understanding the interplay between physics discovery and machine learning, and fully defined discovery workflows. This, in turn, requires balancing the physical intuition and prior knowledge of the domain scientist with rewards that define experimental goals and machine learning algorithms that can translate these to specific experimental protocols. Here, we discuss the basic principles of Bayesian active learning and illustrate its applications for SPM. We progress from the Gaussian process as a simple data-driven method and Bayesian inference for physical models as an extension of physics-based functional fits to more complex deep kernel learning methods, structured Gaussian processes, and hypothesis learning. These frameworks allow for the use of prior data, the discovery of specific functionalities as encoded in spectral data, and exploration of physical laws manifesting during the experiment. Here, the discussed framework can be universally applied to all techniques combining imaging and spectroscopy, SPM methods, nanoindentation, electron microscopy and spectroscopy, and chemical imaging methods and can be particularly impactful for destructive or irreversible measurements.

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

Low responsiveness of machine learning models to critical or deteriorating health conditions

Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient’s condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model’s ability to recognize critical patient conditions. We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models’ ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks. We identify serious deficiencies in the models’ responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models. Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2021

Machine-Learning-Based High-Resolution Earthquake Catalog Reveals How Complex Fault Structures Were Activated during the 2016–2017 Central Italy Sequence

The 2016–2017 central Italy seismic sequence occurred on an 80 km long normal-fault system. The sequence initiated with the Mw 6.0 Amatrice event on 24 August 2016, followed by the Mw 5.9 Visso event on 26 October and the Mw 6.5 Norcia event on 30 October. We analyze continuous data from a dense network of 139 seismic stations to build a high-precision catalog of ~900,000 earthquakes spanning a 1 yr period, based on arrival times derived using a deep-neural-network-based picker. Our catalog contains an order of magnitude more events than the catalog routinely produced by the local earthquake monitoring agency. Aftershock activity reveals the geometry of complex fault structures activated during the earthquake sequence and provides additional insights into the potential factors controlling the development of the largest events. Activated fault structures in the northern and southern regions appear complementary to faults activated during the 1997 Colfiorito and 2009 L’Aquila sequences, suggesting that earthquake triggering primarily occurs on critically stressed faults. Delineated major fault zones are relatively thick compared to estimated earthquake location uncertainties, and a large number of kilometer-long faults and diffuse seismicity were activated during the sequence. These properties might be related to fault age, roughness, and the complexity of inherited structures. The rich details resolvable in this catalog will facilitate continued investigation of this energetic and well-recorded earthquake sequence.

58 GEOSCIENCES↗
DOE OSTI2023

Exploring Classification of Topological Priors With Machine Learning for Feature Extraction

In many scientific endeavors, increasingly abstract representations of data allow for new interpretive methodologies and conceptualization of phenomena. For example, moving from raw imaged pixels to segmented and reconstructed objects allows researchers new insights and means to direct their studies toward relevant areas. Thus, the development of new and improved methods for segmentation remains an active area of research. With advances in machine learning and neural networks, scientists have been focused on employing deep neural networks such as U-Net to obtain pixel-level segmentations, namely, defining associations between pixels and corresponding/referent objects and gathering those objects afterward. Topological analysis, such as the use of the Morse-Smale complex to encode regions of uniform gradient flow behavior, offers an alternative approach: first, create geometric priors, and then apply machine learning to classify. This approach is empirically motivated since phenomena of interest often appear as subsets of topological priors in many applications. Using topological elements not only reduces the learning space but also introduces the ability to use learnable geometries and connectivity to aid the classification of the segmentation target. Here, in this article, we describe an approach to creating learnable topological elements, explore the application of ML techniques to classification tasks in a number of areas, and demonstrate this approach as a viable alternative to pixel-level classification, with similar accuracy, improved execution time, and requiring marginal training data.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

On-the-fly autonomous control of neutron diffraction via physics-informed Bayesian active learning

We demonstrate the first live, autonomous control over neutron diffraction experiments by developing and deploying ANDiE: the autonomous neutron diffraction explorer. Neutron scattering is a unique and versatile characterization technique for probing the magnetic structure and behavior of materials. However, instruments at neutron scattering facilities in the world is limited, and instruments at such facilities are perennially oversubscribed. We demonstrate a significant reduction in experimental time required for neutron diffraction experiments by implementation of autonomous navigation of measurement parameter space through machine learning. Prior scientific knowledge and Bayesian active learning are used to dynamically steer the sequence of measurements. We show that ANDiE can experimentally determine the magnetic ordering transition of both MnO and Fe 1.09 Te all while providing a fivefold enhancement in measurement efficiency. Furthermore, in a hypothesis testing post-processing step, ANDiE can determine transition behavior from a set of possible physical models. ANDiE's active learning approach is broadly applicable to a variety of neutron-based experiments and can open the door for neutron scattering as a tool of accelerated materials discovery.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Coupling High-Throughput Experiments and Regression Algorithms to Optimize PGM-Free ORR Electrocatalyst Synthesis

Over the past decades, significant improvement has been achieved in the performance of platinum group metal-free (PGM-free) materials as an alternative to Pt-based electrocatalysts for oxygen reduction reaction (ORR). However, further progress in ORR activity requires evaluation of precursors and synthesis approaches. In response to this challenge, we generated a first of its kind experimental data set of 36 samples using high-throughput synthesis and activity measurements. Several control parameters (e.g., Fe precursor identity, the precursor content, and pyrolysis temperature) were varied. We then developed several state-of-the-art machine learning (ML) based regression models to predict ORR activity, dependent on selected synthesis variables. Through an iterative algorithm, higher prediction accuracy (smaller root-mean-square error) was achieved. We identified that gradient boosting regression (GBR) and support vector regression (SVR), among several methods, work best for this data set. Aided by our ML-based surrogate models, we decided to alter catalyst synthesis conditions, which resulted in a 36% increase in measured ORR activity in comparison to the maximum ORR mass activity value of 21.9 A/g catalyst in the original data set. Overall, this combined experiment and machine learning approach represents a promising path forward toward developing highly efficient next-generation ORR electrocatalysts and, more generally, functional materials.

25 ENERGY STORAGE↗
DOE OSTI2026

Effect of Solvent on the Local Structure, Dynamics, and Vibrational Density of States in Sn-BEA Zeolite

Lewis acid zeolites are attractive catalysts for epoxidation and biomass valorization, as they are highly active and selective in the liquid phase and can operate at or near ambient conditions. While a rich experimental literature exists on liquid-phase Lewis acid zeolite catalysis, our understanding of the molecular organization and solvent dynamics in the vicinity of Lewis acid sites with differing metal site speciation remains limited. In this work, we investigate the molecular coordination and diffusion of two common solvents (methanol and water) around the closed and open Sn-BEA zeolite active sites using molecular dynamics simulations with a machine-learned interatomic potential trained on ab initio molecular dynamics trajectories. Molecular dynamics simulations reveal that introducing active sites significantly enhances local order in the first and second solvation shells compared to the pure silica case. For methanol, both closed and open active sites are singly coordinated, while more than two water molecules coordinate the open site. In contrast to methanol, we observed that water molecules dissociate, leading to the formation of additional Sn-OH and silanol groups away from the active site. The diffusion coefficients of water and methanol are functions of the solvent population in the pore. Here, our work provides insights into how active site speciation in Lewis acid zeolites affects solvent coordination, diffusion, and vibrational signature. This information is foundational for catalyst design and optimization of liquid-phase catalytic processes in zeolites. It also demonstrates the suitability of machine-learned interatomic potentials for modeling reactive systems, enabling sufficiently long trajectories for appropriate statistical averaging.

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

Uncovering the True Active Sites in Ni–N–C Catalysts for CO 2 Electroreduction

Understanding and designing active sites in single-atom catalysts (SACs) requires going beyond static models to capture their dynamic evolution under realistic electrochemical conditions. Here, in this work, we develop an integrated theoretical framework that accounts for operational conditions, by combining grand canonical density functional theory (GC-DFT) with machine-learning-accelerated sampling, to uncover structure–activity–stability relationships in Ni–N–C SACs for the CO 2 reduction reaction (CO 2 RR). A library of NiN x C 4–x (x = 0–4) motifs─representing coordination defects likely formed during high-temperature synthesis─was systematically evaluated. Under working conditions, these sites were found to undergo hydrogenation, and NiN 3 C 1_ H 1 was identified as the most probable active site. At reducing potentials, hydrogen adsorbs spontaneously at C–Ni bridge sites rather than Ni top sites, while subsurface hydrogen facilitates bent CO 2 adsorption crucial for activation. High CO 2 RR selectivity toward CO arises from site separation: Ni centers drive CO2RR, while the hydrogen evolution reaction (HER) occurs at the C–Ni bridge or N sites and from thermodynamic suppression of HER at moderate hydrogen coverage. At more negative potentials, a shift in the CO 2 RR rate-determining process (RDP) and Ni out-of-surface displacement induced by coadsorption of H and H 2 O jointly reduce activity and selectivity. Thus, both the high CO2RR selectivity of Ni–N–C catalysts and its reversal with more negative potentials can be rationalized by accounting for hydrogenated surfaces. This highlights the necessity of modeling realistic; in situ conditions. This framework provides generalizable insights into the dynamic behavior of active sites in SACs, offering guidance for the rational design of active and robust catalysts for a wide range of electrochemical reactions.

25 ENERGY STORAGE↗
DOE OSTI2025

A Weakly Supervised Machine Learning Procedure for Magnet Quench Diagnostics

Voltage taps remain the standard and reliable diagnostic tool for detecting quenches in superconducting magnets. However, they identify a quench only at the time of voltage rise and do not provide information on earlier physical precursors. In this work, we investigate whether acoustic emission data can reveal precursor activity that occurs before conventional voltage detection using machine learning techniques. We introduce an event selection method and a weakly supervised machine learning procedure to learn data-driven criteria for identifying potential acoustic precursors to quenches. Two Convolutional Neural Network (CNN) architectures are trained: one on acoustic sensor events from our selection procedure and one on the Fast Fourier Transforms (FFTs) of these events. Both networks are trained iteratively using confidence-weighted loss functions to associate certain subsets of training data with a precursor label. We evaluate the performance of these models by examining the time distribution of events classified as potential precursors relative to the quench onset. Results indicate that the proposed approach can possibly distinguish acoustic emission events occurring closer to the quench from earlier acoustic activity during ramping, suggesting the potential for flagging quench precursors in acoustic data.

Khan, Maira [Fermilab] (ORCID:0009000891602387)↗
DOE OSTI2025

MOOSE ProbML: Parallelizable Probabilistic Machine Learning and Uncertainty Quantification Capabilities

The Multiphysics Object Oriented Simulation Environment (MOOSE) is a widely used open- source finite element software for performing multiphysics multiscale simulations in a massively parallel fashion. Recently, the computational team at Idaho National Laboratory (INL) has implemented Probabilistic Machine Learning (ProbML) capabilities in MOOSE—in a parallelized fashion—and enable active learning with large-scale computational models for tasks such as surrogate model development, scale bridging, forward/inverse uncertainty quantification (UQ), Bayesian optimization, etc. This presentation summarizes these developments in MOOSE along with demonstrations on several real applications relevant to nuclear energy. At the fundamental level, samplers like Monte Carlo/Latin Hypercube, variance reduction, parallelized Markov Chain Monte Carlo (MCMC) support uncertainty propagation in both forward and inverse settings. These samplers can be integrated with the Gaussian processes (GP) suite in MOOSE, which offer several variants like scalar GPs, multi-output GPs, and deep GPs, to enable active learning. These GPs can be tuned using gradient-based optimization methods like Adam and its variants or gradient-free methods like the elliptical slice sampler (a variant of MCMC adept under Gaussian settings) for more complex covariance kernels or likelihoods whose gradient computations can be cumbersome. A variety of batch acquisition functions permit parallelized evaluation of the computational model and support different learning objectives with high efficiency like Bayesian inference, global surrogate development, optimization, etc. Furthermore, libtorch integration supports training, evaluation, and re-training of neural networks and other complex machine learning models in active learning settings. The impacts of these developments are shown on several real applications: (1) nuclear fuel inverse UQ and model inadequacy assessment using the Kennedy O’Hagan framework; (2) uncertainty aware surrogate modeling for additive manufacturing to predict field quantities; (3) nuclear reactor rare events analysis; and (4) complex fluid flow prediction using a global surrogate with quantified prediction uncertainty. Finally, the outlook of MOOSE ProbML is discussed for both outer-loop and inner-loop computations in the broad view to accelerate fuels and materials qualification, address gaps in knowledge and data, and assess new reactor/fuel systems.

11 - NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2020

Scalable Wind Turbine Generator Bearing Fault Prediction Using Machine Learning: A Case Study

Operation and maintenance (O&M) costs for wind turbines pose a risk to competitiveness and asset owners. With machine-learning technologies and digitalization rapidly maturing, the wind industry is actively investigating these new technologies to optimize O&M practices and reduce costs. This paper reviews recent work on machine-learning approaches to generator bearing failure prediction and presents a relevant real-world case study through a collaboration between the National Renewable Energy Laboratory and Envision Digital Corporation. In the case study, we evaluate the performance of representative machine-learning algorithms for predicting wind turbine generator bearing failures. Operational supervisory control and data acquisition data from one wind power plant was used to train and test the machine-learning models. The investigated data channels are chosen based on whether physically they reflect the failed generator bearing conditions and the component historical usage, including both environmental and operational conditions. Benefits and drawbacks of different methods are identified.

generator bearing failures↗
DOE OSTI2023

Studying the hadron structure with PANDA and CLAS using machine learning techniques

The hadron spectroscopy and structure are currently very active fields of research to study the non-perturbative regime of quantum chronodynamics. The first one studies the complex structure of excited hadrons by looking at their decay products, while the latter uses lepton scattering on nucleons. Both methods require reconstruction algorithms with great efficiency and good particle identification and background rejection rates. This work aims to provide these by either improving the existing methods or developing new ones. The first part of this document presents a feasibility study of a predicted hybrid charmonium state for the PANDA experiment. Lattice QCD calculations predict the ground state hybrid charmonium to be a spin exotic with quantum numbers of JP C = 1?+ at a mass of around 4.3 GeV with a width to be around 20 MeV. A machine learning based data analysis scheme is proposed to further improve the signal efficiency and the background reduction, alongside with improvements of the analysis software (PandaRoot), that are vital for this study. These improvements include a reworked clustering algorithm for the electromagnetic calorimeter (EMC) and an optimized monte carlo matching for neutral particles. The second part of this document is about studying the proton structure. A multidimensional study of the structure function ratio Fsin(?)LU /FUU has been performed for K±, based on the measurement of beam-spin asymmetries. It uses the high statistics data recorded with the CLAS12 spectrometer at Jefferson Laboratory. Fsin(?)LU is a twist-3 quantity that provides information about the quark gluon correlations in the proton. This document will present for the first time a simultaneous analysis of two kaon channels over a large kinematic range of z, xB , PT and Q2 with virtualities Q2 ranging from 1 GeV2 up to 8 GeV2 using machine learning techniques for improved particle identification.

Kripko, Aron↗
DOE OSTI2020

Clostridioides difficile Toxin A Remodels Membranes and Mediates DNA Entry Into Cells to Activate Toll-Like Receptor 9 Signaling

Background & Aims: Clostridioides difficile toxin A (TcdA) activates the innate immune response. TcdA co-purifies with DNA. Toll-like receptor 9 (TLR9) recognizes bacterial DNA to initiate inflammation. We investigated whether DNA bound to TcdA activates an inflammatory response in murine models of C difficile infection via activation of TLR9. Methods: We performed studies with human colonocytes and monocytes and macrophages from wild-type and TLR9 knockout mice incubated with TcdA or its antagonist (ODN TTAGGG) or transduced with vectors encoding TLR9 or small-interfering RNAs. Cytokine production was measured with enzyme-linked immunosorbent assay. We studied a transduction domain of TcdA (TcdA 57-80 ), which was predicted by machine learning to have cell-penetrating activity and confirmed by synchrotron small-angle X-ray scattering. Intestines of CD1 mice, C57BL6J mice, and mice that express a form of TLR9 that is not activated by CpG DNA were injected with TcdA, TLR9 antagonist, or both. Enterotoxicity was estimated based on loop weight to length ratios. A TLR9 antagonist was tested in mice infected with C difficile. We incubated human colon explants with an antagonist of TLR9 and measured TcdA-induced production of cytokines. Results: The TcdA 57-80 protein transduction domain had membrane remodeling activity that allowed TcdA to enter endosomes. TcdA-bound DNA entered human colonocytes. TLR9 was required for production of cytokines by cultured cells and in human colon explants incubated with TcdA. TLR9 was required in TcdA-induced mice intestinal secretions and in the survival of mice infected by C difficile. Even in a protease-rich environment, in which only fragments of TcdA exist, the TcdA 57-80 domain organized DNA into a geometrically ordered structure that activated TLR9. Conclusions: TcdA from C difficile can bind and organize bacterial DNA to activate TLR9. TcdA and TcdA fragments remodel membranes, which allows them to access endosomes and present bacterial DNA to and activate TLR9. Rather than inactivating the ability of DNA to bind TLR9, TcdA appears to chaperone and organize DNA into an inflammatory, spatially periodic structure.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2023

Machine-Learning-Based Rotating Detonation Engine Diagnostics: Evaluation for Application in Experimental Facilities

Real-time monitoring of combustion behavior is a crucial step toward actively controlled rotating detonation engine (RDE) operation in laboratory and industrial environments. Various machine learning methods have been developed to advance diagnostic efficiencies from conventional postprocessing efforts to real-time methods. Here this work evaluates and compares conventional techniques alongside convolutional neural network (CNN) architectures trained in previous studies, including image classification, object detection, and time series classification, according to metrics affecting diagnostic feasibility, external applicability, and performance. Real-time, capable diagnostics are deployed and evaluated using an altered experimental setup. Image-based CNNs are applied to externally provided images to approximate dataset restrictions. Image classification using high-speed chemiluminescence images and time series classification using high-speed flame ionization and pressure measurements achieve classification speeds enabling real-time diagnostic capabilities, averaging laboratory-deployed diagnostic feedback rates of 4–5 Hz. Object detection achieves the most refined resolution of 20 μs in postprocessing. Image and time series classification require the additional correlation of sensor data, extending their time-step resolutions to 80 ms. Comparisons show that no single diagnostic approach outperforms its competitors across all metrics. This finding justifies the need for a machine learning portfolio containing a host of networks to address specific needs throughout the RDE research community.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2026

Developing a Deep Learning-Computer Vision Framework to Monitor Avian Interactions with Solar Energy Facility Infrastructure (Final Technical Report)

The project addressed an inability to monitor avian interactions with photovoltaic (PV) solar energy facilities necessary for understanding PV solar impacts on birds. In the project, machine-vision technology that continuously monitors avian activities at PV solar facilities was developed. The technology includes four machine-learning (ML) models, each of which accomplishes a specific task in detecting birds and classifying their activities in live or recorded videos—detecting and tracking moving objects, differentiating birds from other objects, detecting bird collisions with solar panels, and classifying non-collision bird activities around PV facilities. Major project outcomes include adoption by two of DOE SETO’s SolWEB projects, providing novel observational data on birds to promote co-location of PV solar development and habitat conservation, known as ecovoltaics.

14 SOLAR ENERGY↗
DOE OSTI2021

Deep active learning for classifying cancer pathology reports

Abstract Background Automated text classification has many important applications in the clinical setting; however, obtaining labelled data for training machine learning and deep learning models is often difficult and expensive. Active learning techniques may mitigate this challenge by reducing the amount of labelled data required to effectively train a model. In this study, we analyze the effectiveness of 11 active learning algorithms on classifying subsite and histology from cancer pathology reports using a Convolutional Neural Network as the text classification model. Results We compare the performance of each active learning strategy using two differently sized datasets and two different classification tasks. Our results show that on all tasks and dataset sizes, all active learning strategies except diversity-sampling strategies outperformed random sampling, i.e., no active learning. On our large dataset (15K initial labelled samples, adding 15K additional labelled samples each iteration of active learning), there was no clear winner between the different active learning strategies. On our small dataset (1K initial labelled samples, adding 1K additional labelled samples each iteration of active learning), marginal and ratio uncertainty sampling performed better than all other active learning techniques. We found that compared to random sampling, active learning strongly helps performance on rare classes by focusing on underrepresented classes. Conclusions Active learning can save annotation cost by helping human annotators efficiently and intelligently select which samples to label. Our results show that a dataset constructed using effective active learning techniques requires less than half the amount of labelled data to achieve the same performance as a dataset constructed using random sampling.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2025

Machine Learning-Accelerated First-Principles Molecular Dynamics Reveals C–C Coupling Mechanisms toward Ethylene on Cu(100)

Here, the Cu(100) termination has been identified as the most effective facet for converting CO and CO 2 into ethylene. To enhance both the activity and selectivity of ethylene production, we perform machine-learning-accelerated, first-principles molecular dynamics simulations at 298 K in an explicit solvent at pH 7 to elucidate the C–C coupling mechanism─the critical reaction step in forming C 2+ products. Among the six potential C–C coupling pathways, the most feasible are CO* dimerization and CO – CHO* and CHO* – CHO* couplings. Using the computational hydrogen electrode method, we demonstrate that all three pathways are equally accessible at −0.6 V vs RHE. At a potential below −1.0 V vs RHE, the thermodynamic barriers for the CO – CHO* and CHO* – CHO* pathways become negligible. Our computational findings explain the experimental observations, particularly the absence of C 2+ products above −0.4 V vs RHE and the peaks in ethylene production near −0.6 and −1.0 V vs RHE. Since CHO* acts as a key intermediate common to both C–C coupling and CH 4 formation, we propose that suppressing CHO* hydrogenation would inhibit CH 4 pathways, thereby maximizing ethylene selectivity.

CO2 reduction↗