Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “neural operator 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 163 records · Page 9

Deep transfer operator learning for partial differential equations under conditional shift

Transfer learning enables the transfer of knowledge gained while learning to perform one task (source) to a related but different task (target), hence addressing the expense of data acquisition and labelling, potential computational power limitations and dataset distribution mismatches. Here, we propose a new transfer learning framework for task-specific learning (functional regression in partial differential equations) under conditional shift based on the deep operator network (DeepONet). Task-specific operator learning is accomplished by fine-tuning task-specific layers of the target DeepONet using a hybrid loss function that allows for the matching of individual target samples while also preserving the global properties of the conditional distribution of the target data. Inspired by conditional embedding operator theory, we minimize the statistical distance between labelled target data and the surrogate prediction on unlabelled target data by embedding conditional distributions onto a reproducing kernel Hilbert space. We demonstrate the advantages of our approach for various transfer learning scenarios involving nonlinear partial differential equations under diverse conditions due to shifts in the geometric domain and model dynamics. Our transfer learning framework enables fast and efficient learning of heterogeneous tasks despite considerable differences between the source and target domains.

42 ENGINEERING↗

Dynamic Validation of CNN-Based Surrogate Models for Inverter-Based Resources in Open-Source Solvers

Traditionally, distribution system planning has focused on steady-state analyses, with limited consideration of dynamic behavior. However, as large or medium-scale inverter-based resources (IBRs), particularly grid-following (GFL) inverters in commercial or industry buildings, become more prevalent, understanding their dynamic impact is essential for grid planning and operation. This article presents an innovative deep-learning (DL)-approach using convolutional neural networks technique to model the GFL inverters. Developed from real grid-tied commercial IBR transient data, these dynamic DL models overcome proprietary constraints by requiring minimal knowledge of internal converter physics while maintaining high accuracy and flexibility. To demonstrate their applicability, the models were incorporated into GridLAB-D, an open-source, three-phase distribution analysis tool. This integration enables dynamic simulations of large-scale distribution networks with high IBR penetration stability analysis. Rigorous testing and validation, aligned with industry standards, confirmed the reliability and efficiency of this approach, paving the way for enhanced planning and operational assessments of modern power systems.

Deep-learning↗

Deep Learning Experiments for Tropical Cyclone Intensity Forecasts

Reducing tropical cyclone (TC) intensity forecast errors is a challenging task that has interested the operational forecasting and research community for decades. To address this, we developed a deep learning (DL)-based multilayer perceptron (MLP) TC intensity prediction model. The model was trained using the global Statistical Hurricane Intensity Prediction Scheme (SHIPS) predictors to forecast the change in TC maximum wind speed for the Atlantic basin. In the first experiment, a 24-h forecast period was considered. To overcome sample size limitations, we adopted a leave one year out (LOYO) testing scheme, where a model is trained using data from all years except one and then evaluated on the year that is left out. When tested on 2010–18 operational data using the LOYO scheme, the MLP outperformed other statistical–dynamical models by 9%–20%. Additional independent tests in 2019 and 2020 were conducted to simulate real-time operational forecasts, where the MLP model again outperformed the statistical–dynamical models by 5%–22% and achieved comparable results as HWFI. The MLP model also correctly predicted more rapid intensification events than all the four operational TC intensity models compared. In the second experiment, we developed a lightweight MLP for 6-h intensity predictions. When coupled with a synthetic TC track model, the lightweight MLP generated realistic TC intensity distribution in the Atlantic basin. Therefore, the MLP-based approach has the potential to improve operational TC intensity forecasts, and will also be a viable option for generating synthetic TCs for climate studies.

58 GEOSCIENCES↗

Deep spectral CNN for laser induced breakdown spectroscopy

This work proposes a spectral convolutional neural network (CNN) operating on laser induced breakdown spectroscopy (LIBS) signals to learn to (1) disentangle spectral signals from the sources of sensor uncertainty (i.e., pre-process) and (2) get qualitative and quantitative measures of chemical content of a sample given a spectral signal (i.e., calibrate). Once the spectral CNN is trained, it can accomplish either task through a single feed-forward pass, with real-time benefits and without any additional side information requirements including dark current, system response, temperature and detector-to-target range. Our experiments demonstrate that the proposed method outperforms the existing approaches used by the Mars Science Lab for pre-processing and calibration for remote sensing observations from the Mars rover, ‘Curiosity’.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Enhancing Fluid Flow Pressure and Saturation Prediction Accuracy and Reducing Uncertainty with Committee Machine – Illinois Basin Decatur Project (IBDP) as a Case Study

Presentation at the 17th International Conference on Greenhouse Gas Control Technologies GHGT-17 held in Calgary, Canada, October 20-24, 2024. Carbon capture and storage (CCS) is a way to play a critical role in the global transition to a low-emission economy. Current progress is hampered by a number of factors, among which the lack of risk-informed design tools and decision support frameworks is seen as a major roadblock. Significant interest exists in using artificial intelligence to accelerate CCS site feasibility studies, as well as to facilitate the permit application process. Existing works commonly train a single deep learning model. This work investigates the feasibility of using a conventional ensemble learning (committee machine) technique to further improve prediction accuracy. Ensemble-based algorithms generally improve over individual base learners in terms of robustness and accuracy. Deep ensembles, however, are time-consuming to create and train. A pragmatic question is whether small-sized ensembles may lead to prediction improvement. Here we evaluated the efficacy of an ensemble learning technique using the latent spectral model (LSM), an efficient deep neural operator algorithm, as base learners. Preliminary results, obtained using the Illinois Basin-Decatur Project (IBDP) carbon sequestration data/model, show that small-sized ensembles can improve prediction over the base learners, achieving prediction accuracy of ~1.6 psi root mean square error (RMSE) on pressure (relative the average reservoir pressure of 3150 psi), and less than 1.3% for saturation.

Sun, Alexander↗

Enhancing Fluid Flow Pressure and Saturation Prediction Accuracy and Reducing Uncertainty with Committee Machine – Illinois Basin Decatur Project (IBDP) as a Case Study

This is the conference paper accompanying an oral presentation at the 17th International Conference on Greenhouse Gas Control Technologies GHGT-17 held in Calgary, Canada, October 20-24, 2024. Carbon capture and storage (CCS) is a way to play a critical role in the global transition to a low-emission economy. Current progress is hampered by a number of factors, among which the lack of risk-informed design tools and decision support frameworks is seen as a major roadblock. Significant interest exists in using artificial intelligence to accelerate CCS site feasibility studies, as well as to facilitate the permit application process. Existing works commonly train a single deep learning model. This work investigates the feasibility of using a conventional ensemble learning (committee machine) technique to further improve prediction accuracy. Ensemble-based algorithms generally improve over individual base learners in terms of robustness and accuracy. Deep ensembles, however, are time-consuming to create and train. A pragmatic question is whether small-sized ensembles may lead to prediction improvement. Here we evaluated the efficacy of an ensemble learning technique using the latent spectral model (LSM), an efficient deep neural operator algorithm, as base learners. Preliminary results, obtained using the Illinois Basin-Decatur Project (IBDP) carbon sequestration data/model, show that small-sized ensembles can improve prediction over the base learners, achieving prediction accuracy of ~1.6 psi root mean square error (RMSE) on pressure (relative the average reservoir pressure of 3150 psi), and less than 1.3% for saturation.

Sun, Alexander↗

Learning a trajectory using adjoint functions and teacher forcing

A new methodology for faster supervised temporal learning in nonlinear neural networks is presented which builds upon the concept of adjoint operators to allow fast computation of the gradients of an error functional with respect to all parameters of the neural architecture, and exploits the concept of teacher forcing to incorporate information on the desired output into the activation dynamics. The importance of the initial or final time conditions for the adjoint equations is discussed. A new algorithm is presented in which the adjoint equations are solved simultaneously (i.e., forward in time) with the activation dynamics of the neural network. We also indicate how teacher forcing can be modulated in time as learning proceeds. The results obtained show that the learning time is reduced by one to two orders of magnitude with respect to previously published results, while trajectory tracking is significantly improved. The proposed methodology makes hardware implementation of temporal learning attractive for real-time applications.

Toomarian, Nikzad B.↗

Demonstration of acoustic monitoring for structural health of microreactors: Through use of neural networks and resonant ultrasound spectroscopy

Nuclear microreactors prioritize modularity and portability and are intended to be a cost-effective technology for non-conventional nuclear markets. As such, the development of microreactors into a safe and feasible solution for energy security applications will necessitate the development of non-destructive technologies to monitor the integrity of inaccessible reactors components during operation. This demonstration applies linear and nonlinear acoustic techniques, in combination with machine learning, to detect and classify mechanical changes (stress and damage) in a test article which are broadly representative of potential operating challenges within a functioning microreactor. All necessary data has been collected for this demonstration, with minor experimental issues identified that can be addressed in follow-on work. Motivated by the expected conditions within a functioning microreactor, we have demonstrated our monitoring techniques on a core-block-like test article using an unstructured excitation source that approximates the noisy acoustic environment expected during reactor operation. At all stress states mechanically applied to the test article, a machine learning model using an artificial neural network was able to classify with 100% accuracy whether a 3D laser vibrometry point measurement was made on an intact or artificially defective test article. Further, model predictions about whether the defect interface was rough or smooth were 95% accurate, indicating the ability of acoustic techniques to recover defect characteristics. Resonant ultrasound spectroscopy (RUS) was also applied to the dataset to provide further quantitative insights about material properties. RUS analysis was ultimately hampered by several minor experimental and data issues, limiting results to certain cases for this demonstration. Last, analysis using nonlinear RUS exhibited sensitivity to changing levels of applied stresses for each intact and defective state. As presented in this demonstration, acoustic monitoring exhibits sensitivity to stress changes, which are of concern due to high thermal gradients expected during startup and operation. Further, our techniques distinguish between measurements made on intact and damaged test articles.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Randomized probe imaging through deep k-learning

Randomized probe imaging (RPI) is a single-frame diffractive imaging method that uses highly randomized light to reconstruct the spatial features of a scattering object. The reconstruction process, known as phase retrieval, aims to recover a unique solution for the object without measuring the far-field phase information. Typically, reconstruction is done via time-consuming iterative algorithms. In this work, we propose a fast and efficient deep learning based method to reconstruct phase objects from RPI data. The method, which we call deep k-learning, applies the physical propagation operator to generate an approximation of the object as an input to the neural network. This way, the network no longer needs to parametrize the far-field diffraction physics, dramatically improving the results. Deep k-learning is shown to be computationally efficient and robust to Poisson noise. The advantages provided by our method may enable the analysis of far larger datasets in photon starved conditions, with important applications to the study of dynamic phenomena in physical science and biological engineering.

Guo, Zhen (ORCID:0000000213473451)↗

Abisko: Deep codesign of an architecture for spiking neural networks using novel neuromorphic materials

The Abisko project aims to develop an energy-efficient spiking neural network (SNN) computing architecture and software system capable of autonomous learning and operation. The SNN architecture explores novel neuromorphic devices that are based on resistive-switching materials, such as memristors and electrochemical RAM. Equally important, Abisko uses a deep codesign approach to pursue this goal by engaging experts from across the entire range of disciplines: materials, devices and circuits, architectures and integration, software, and algorithms. Here, the key objectives of our Abisko project are threefold. First, we are designing an energy-optimized high-performance neuromorphic accelerator based on SNNs. This architecture is being designed as a chiplet that can be deployed in contemporary computer architectures and we are investigating novel neuromorphic materials to improve its design. Second, we are concurrently developing a productive software stack for the neuromorphic accelerator that will also be portable to other architectures, such as field-programmable gate arrays and GPUs. Third, we are creating a new deep codesign methodology and framework for developing clear interfaces, requirements, and metrics between each level of abstraction to enable the system design to be explored and implemented interchangeably with execution, measurement, a model, or simulation. As a motivating application for this codesign effort, we target the use of SNNs for an analog event detector for a high-energy physics sensor.

97 MATHEMATICS AND COMPUTING↗

Neural network with dynamically adaptable neurons

This invention is an adaptive neuron for use in neural network processors. The adaptive neuron participates in the supervised learning phase of operation on a co-equal basis with the synapse matrix elements by adaptively changing its gain in a similar manner to the change of weights in the synapse IO elements. In this manner, training time is decreased by as much as three orders of magnitude.

Tawel, Raoul↗

Solving Newton’s equations of motion with large timesteps using recurrent neural networks based operators

Classical molecular dynamics simulations are based on solving Newton’s equations of motion. Using a small timestep, numerical integrators such as Verlet generate trajectories of particles as solutions to Newton’s equations. We introduce operators derived using recurrent neural networks that accurately solve Newton’s equations utilizing sequences of past trajectory data, and produce energy-conserving dynamics of particles using timesteps up to 4000 times larger compared to the Verlet timestep. We demonstrate significant speedup in many example problems including 3D systems of up to 16 particles.

Newton’s equations↗

Probing Electron Beam Induced Transformations on a Single-Defect Level via Automated Scanning Transmission Electron Microscopy

A robust approach for real-time analysis of the scanning transmission electron microscopy (STEM) data streams, based on ensemble learning and iterative training (ELIT) of deep convolutional neural networks, is implemented on an operational microscope, enabling the exploration of the dynamics of specific atomic configurations under electron beam irradiation via an automated experiment in STEM. Combined with beam control, this approach allows studying beam effects on selected atomic groups and chemical bonds in a fully automated mode. Here, in this study, we demonstrate atomically precise engineering of single vacancy lines in transition metal dichalcogenides and the creation and identification of topological defects in graphene. The ELIT-based approach facilitates direct on-the-fly analysis of the STEM data and engenders real-time feedback schemes for probing electron beam chemistry, atomic manipulation, and atom by atom assembly.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Semisupervised Learning for Seismic Monitoring Applications

The impressive performance that deep neural networks demonstrate on a range of seismic monitoring tasks depends largely on the availability of event catalogs that have been manually curated over many years or decades. However, the quality, duration, and availability of seismic event catalogs vary significantly across the range of monitoring operations, regions, and objectives. Semisupervised learning (SSL) enables learning from both labeled and unlabeled data and provides a framework to leverage the abundance of unreviewed seismic data for training deep neural networks on a variety of target tasks. We apply two SSL algorithms (mean-teacher and virtual adversarial training) as well as a novel hybrid technique (exponential average adversarial training) to seismic event classification to examine how unlabeled data with SSL can enhance model performance. In general, we find that SSL can perform as well as supervised learning with fewer labels. We also observe in some scenarios that almost half of the benefits of SSL are the result of the meaningful regularization enforced through SSL techniques and may not be attributable to unlabeled data directly. Lastly, the benefits from unlabeled data scale with the difficulty of the predictive task when we evaluate the use of unlabeled data to characterize sources in new geographic regions. Finally, in geographic areas where supervised model performance is low, SSL significantly increases the accuracy of source-type classification using unlabeled data.

58 GEOSCIENCES↗

Integrated Cyber/Physical Grid Resiliency Modeling

This project explored coupling modeling and analysis methods from multiple domains to address complex hybrid (cyber and physical) attacks on mission critical infrastructure. Robust methods to integrate these complex systems are necessary to enable large trade-space exploration including dynamic and evolving cyber threats and mitigations. Reinforcement learning employing deep neural networks, as in the AlphaGo Zero solution, was used to identify "best" (or approximately optimal) resilience strategies for operation of a cyber/physical grid model. A prototype platform was developed and the machine learning (ML) algorithm was made to play itself in a game of 'Hurt the Grid'. This proof of concept shows that machine learning optimization can help us understand and control complex, multi-dimensional grid space. A simple, yet high-fidelity model proves that the data have spatial correlation which is necessary for any optimization or control. Our prototype analysis showed that the reinforcement learning successfully improved adversary and defender knowledge to manipulate the grid. When expanded to more representative models, this exact type of machine learning will inform grid operations and defense - supporting mitigation development to defend the grid from complex cyber attacks! This same research can be expanded to similar complex domains.

97 MATHEMATICS AND COMPUTING↗

Speeding up and reducing memory usage for scientific machine learning via mixed precision

Scientific machine learning (SciML) has emerged as a versatile approach to address complex computational science and engineering problems. Within this field, physics-informed neural networks (PINNs) and deep operator networks (DeepONets) stand out as the leading techniques for solving partial differential equations by incorporating both physical equations and experimental data. However, training PINNs and DeepONets require significant computational resources, including long computational times and large amounts of memory. In search of computational efficiency, training neural networks using half precision (float16) rather than the conventional single (float32) or double (float64) precision has gained substantial interest, given the inherent benefits of reduced computational time and memory consumed. However, we find that float16 cannot be applied to SciML methods, because of gradient divergence at the start of training, weight updates going to zero, and the inability to converge to a local minima. To overcome these limitations, we explore mixed precision, which is an approach that combines the float16 and float32 numerical formats to reduce memory usage and increase computational speed. Our experiments showcase that mixed precision training not only substantially decreases training times and memory demands but also maintains model accuracy. Here, we also reinforce our empirical observations with a theoretical analysis. The research has broad implications for SciML in various computational applications.

97 MATHEMATICS AND COMPUTING↗

Real-time deep-learning inversion of seismic full waveform data for CO 2 saturation and uncertainty in geological carbon storage monitoring

Deep-learning inversion has recently drawn attention in geological carbon storage research due to its potential of imaging and monitoring carbon storage in real time, significantly improving efficiency and safety of carbon storage operations. We present a deep-learning full waveform inversion method that after the neural network has been trained can image CO 2 saturation and its uncertainty in real time. Our deep-learning inversion method is based on the U-Net architecture with the neural network trained on pairs of synthetic seismic data and CO 2 saturation models. Accordingly, our training establishes a mapping relationship between seismic data and CO 2 saturation models and once fully trained directly estimates CO 2 saturation as a function of subsurface location. We further quantify uncertainties of CO 2 saturation estimates using the Monte Carlo dropout method and a bootstrap aggregating method. For this proof-of-concept study, the CO 2 training models and data are derived from the Kimberlina 1.2 model, a hypothetical 3D geological carbon storage model that is constructed based on various geological and hydrological data from the Southern San Joaquin Basin, California. We perform deep-learning inversion experiments using noise-free and noisy training and test data sets and compare the results. Our modelling experiments show that (1) the deep-learning inversion can estimate 2D distributions of CO 2 fairly well even in the presence of Gaussian random noise and (2) both CO 2 saturation imaging and uncertainty quantification can be done in real time. Our results suggest that the deep-learning inversion method can serve as a robust real-time monitoring tool for geological carbon storage and/or other time-varying reservoir/aquifer properties that result from injection, extraction, and/or other subsurface transport phenomena.

58 GEOSCIENCES↗

Huge ensembles – Part 1: Design of ensemble weather forecasts using spherical Fourier neural operators

Abstract. Simulating low-likelihood high-impact extreme weather events in a warming world is a significant and challenging task for current ensemble forecasting systems. While these systems presently use up to 100 members, larger ensembles could enrich the sampling of internal variability. They may capture the long tails associated with climate hazards better than traditional ensemble sizes. Due to computational constraints, it is infeasible to generate huge ensembles (comprised of 1000–10 000 members) with traditional, physics-based numerical models. In this two-part paper, we replace traditional numerical simulations with machine learning (ML) to generate hindcasts of huge ensembles. In Part 1, we construct an ensemble weather forecasting system based on spherical Fourier neural operators (SFNOs), and we discuss important design decisions for constructing such an ensemble. The ensemble represents model uncertainty through perturbed-parameter techniques, and it represents initial condition uncertainty through bred vectors, which sample the fastest-growing modes of the forecast. Using the European Centre for Medium-Range Weather Forecasts Integrated Forecasting System (IFS) as a baseline, we develop an evaluation pipeline composed of mean, spectral, and extreme diagnostics. With large-scale, distributed SFNOs with 1.1 billion learned parameters, we achieve calibrated probabilistic forecasts. As the trajectories of the individual members diverge, the ML ensemble mean spectra degrade with lead time, consistent with physical expectations. However, the individual ensemble members' spectra stay constant with lead time. Therefore, these members simulate realistic weather states during the rollout, and the ML ensemble passes a crucial spectral test in the literature. The IFS and ML ensembles have similar extreme forecast indices, and we show that the ML extreme weather forecasts are reliable and discriminating. These diagnostics ensure that the ensemble can reliably simulate the time evolution of the atmosphere, including low-likelihood high-impact extremes. In Part 2, we generate a huge ensemble initialized each day in summer 2023, and we characterize the simulations of extremes.

Mahesh, Ankur↗