Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “network inference”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,153 records · Page 64
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

AtomAI framework for deep learning analysis of image and spectroscopy data in electron and scanning probe microscopy

Over the past several decades, electron and scanning probe microscopes have become critical components of condensed matter physics, materials science and chemistry research. At the same time, the infrastructure for establishing a connection between microscopy observations and materials behaviour over a broader parameter space is lacking. In this work, we introduce AtomAI, an open-source software package bridging instrument-specific Python libraries, deep learning and simulation tools into a single ecosystem. AtomAI allows direct applications of deep neural networks for atomic and mesoscopic image segmentation converting image and spectroscopy data into class-based local descriptors for downstream tasks such as statistical and graph analysis. For atomically resolved imaging data, the output is types and positions of atomic species, with an option for subsequent refinement. AtomAI further allows the implementation of a broad range of image and spectrum analysis functions, including invariant variational autoencoders for disentangling structural factors of variation and im2spec type of encoder–decoder models for mapping structure–property relationships. Finally, our framework allows seamless connection to the first principles modelling with a Python interface on the inferred atomic positions.

36 MATERIALS SCIENCE↗
DOE OSTI2016

MONITORING SHORT-TERM COSMIC-RAY SPECTRAL VARIATIONS USING NEUTRON MONITOR TIME-DELAY MEASUREMENTS

Neutron monitors (NMs) are ground-based detectors of cosmic-ray showers that are widely used for high-precision monitoring of changes in the Galactic cosmic-ray (GCR) flux due to solar storms and solar wind variations. In the present work, we show that a single neutron monitor station can also monitor short-term changes in the GCR spectrum, avoiding the systematic uncertainties in comparing data from different stations, by means of NM time-delay histograms. Using data for 2007–2014 from the Princess Sirindhorn Neutron Monitor, a station at Doi Inthanon, Thailand, with the world’s highest vertical geomagnetic cutoff rigidity of 16.8 GV, we have developed an analysis of time-delay histograms that removes the chance coincidences that can dominate conventional measures of multiplicity. We infer the “leader fraction” L of neutron counts that do not follow a previous neutron count in the same counter from the same atmospheric secondary, which is inversely related to the actual multiplicity and increases for increasing GCR spectral index. After correction for atmospheric pressure and water vapor, we find that L indicates substantial short-term GCR spectral hardening during some but not all Forbush decreases in GCR flux due to solar storms. Such spectral data from Doi Inthanon provide information about cosmic-ray energies beyond the Earth’s maximum geomagnetic cutoff, extending the reach of the worldwide NM network and opening a new avenue in the study of short-term GCR decreases.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRSDate not supplied

Onboard Hyperspectral Image Classification via Transfer Learning for Communication-Limited Spacecraft

Employing deep-learning and artificial-intelligence (AI) techniques onboard spacecraft can dramatically improve priority data selection to ensure more effective use of the available downlink. However, deployment of effective deep-learning models requires significant training on the ground, which may not be feasible, due to limited data available in an unexplored environment. Therefore, this research explores building robust classification models for onboard data processing where training data is highly limited using transfer-learning techniques. In this paper, we focus on the use case of hyperspectral imaging for remote sensing, a domain where the high dimensionality of the data from the sensor can rapidly saturate the downlink bandwidth. With this bottleneck, there is an impending need to autonomously and robustly classify data onboard to optimize downlink of high-impact measurements, thus maximizing the scientific utility per bit transmitted to the ground. This paper examines the use of deep neural networks onboard for hyperspectral image classification in a communication-limited scenario to analyze how the models perform with limited training data. The use of transfer learning can ameliorate the issue of poor generalization by transferring features learned from training on a large source dataset for one classification task to the target classification task with limited training data. For two deep-learning models from literature, we compare the accuracy of the models trained using transfer learning to models trained from scratch using a random weight initialization with varying amounts of training data. We demonstrate the feasibility and performance of running inference of the deep-learning models on representative flight-like hardware.

Advanced Avionics, Machine Learning, Data Processi↗
DOE OSTI2024

Constraining the Higgs potential with neural simulation-based inference for di-Higgs production

Determining the form of the Higgs potential is one of the most exciting challenges of modern particle physics. Higgs pair production directly probes the Higgs self-coupling and should be observed in the near future at the High-Luminosity LHC. We explore how to improve the sensitivity to physics beyond the Standard Model through per-event kinematics for di-Higgs events. In particular, we employ machine learning through simulation-based inference to estimate per-event likelihood ratios and gauge potential sensitivity gains from including this kinematic information. In terms of the Standard Model Effective Field Theory, we find that adding a limited number of observables can help to remove degeneracies in Wilson coefficient likelihoods and significantly improve the experimental sensitivity.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2021

Measurement of Volatile Compounds for Real-Time Analysis of Soil Microbial Metabolic Response to Simulated Snowmelt

Snowmelt dynamics are a significant determinant of microbial metabolism in soil and regulate global biogeochemical cycles of carbon and nutrients by creating seasonal variations in soil redox and nutrient pools. With an increasing concern that climate change accelerates both snowmelt timing and rate, obtaining an accurate characterization of microbial response to snowmelt is important for understanding biogeochemical cycles intertwined with soil. However, observing microbial metabolism and its dynamics non-destructively remains a major challenge for systems such as soil. Microbial volatile compounds (mVCs) emitted from soil represent information-dense signatures and when assayed non-destructively using state-of-the-art instrumentation such as Proton Transfer Reaction-Time of Flight-Mass Spectrometry (PTR-TOF-MS) provide time resolved insights into the metabolism of active microbiomes. In this study, we used PTR-TOF-MS to investigate the metabolic trajectory of microbiomes from a subalpine forest soil, and their response to a simulated wet-up event akin to snowmelt. Using an information theory approach based on the partitioning of mutual information, we identified mVC metabolite pairs with robust interactions, including those that were non-linear and with time lags. The biological context for these mVC interactions was evaluated by projecting the connections onto the Kyoto Encyclopedia of Genes and Genomes (KEGG) network of known metabolic pathways. Simulated snowmelt resulted in a rapid increase in the production of trimethylamine (TMA) suggesting that anaerobic degradation of quaternary amine osmo/cryoprotectants, such as glycine betaine, may be important contributors to this resource pulse. Unique and synergistic connections between intermediates of methylotrophic pathways such as dimethylamine, formaldehyde and methanol were observed upon wet-up and indicate that the initial pulse of TMA was likely transformed into these intermediates by methylotrophs. Increases in ammonia oxidation signatures (transformation of hydroxylamine to nitrite) were observed in parallel, and while the relative role of nitrifiers or methylotrophs cannot be confirmed, the inferred connection to TMA oxidation suggests either a direct or indirect coupling between these processes. Overall, it appears that such mVC time-series from PTR-TOF-MS combined with causal inference represents an attractive approach to non-destructively observe soil microbial metabolism and its response to environmental perturbation.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS2002

Biogeochemical Cycles of Carbon and Sulfur

The elements carbon (C) and sulfur (S) interact with each other across a network of elemental reservoirs that are interconnected by an array of physical, chemical and biological processes. These networks are termed the biogeochemical C and S cycles. The compounds of C are highly important, not only as organic matter, but also as atmospheric greenhouse gases, pH buffers in seawater, oxidation-reduction buffers virtually everywhere, and key magmatic constituents affecting plutonism and volcanism. The element S assumes important roles as an oxidation-reduction partner with C and Fe in biological systems, as a key constituent in magmas and volcanic gases, and as a major influence upon pH in certain environments. This presentation describes the modern biogeochemical C and S cycles. Measurements are described whereby stable isotopes can help to infer the nature and quantitative significance of biological and geological processes involved in the C and S cycles. This lecture also summarizes the geological and climatologic aspects of the ancient C and S cycles, as well as the planetary and extraterrestrial processes that influenced their evolution over millions to billions of years.

DesMarais, David J.↗
NASA NTRS1983

Spatial reasoning to determine stream network from LANDSAT imagery

In LANDSAT imagery, spectral and spatial information can be used to detect the drainage network as well as the relative elevation model in mountainous terrain. To do this, mixed information of material reflectance in the original LANDSAT imagery must be separated. From the material reflectance information, big visible rivers can be detected. From the topographic modulation information, ridges and valleys can be detected and assigned relative elevations. A complete elevation model can be generated by interpolating values for nonridge and non-valley pixels. The small streams not detectable from material reflectance information can be located in the valleys with flow direction known from the elevation model. Finally, the flow directions of big visible rivers can be inferred by solving a consistent labeling problem based on a set of spatial reasoning constraints.

Haralick, R. M.↗
DOE OSTI2025

Discriminative versus generative approaches to simulation-based inference

Most of the fundamental, emergent, and phenomenological parameters of particle and nuclear physics are determined through parametric template fits. Simulations are used to populate histograms which are then matched to data. This approach is inherently lossy, since histograms are binned and low-dimensional. Deep learning has enabled unbinned and high-dimensional parameter estimation through neural likelihood(-ratio) estimation. We compare two approaches for neural simulation-based inference (NSBI): one based on discriminative learning (classification) and one based on generative modeling. These two approaches are directly evaluated on the same datasets, with a similar level of hyperparameter optimization in both cases. In addition to a Gaussian dataset, we study NSBI using a Higgs boson dataset from the FAIR Universe Challenge. We find that both the direct likelihood and likelihood ratio estimation are able to effectively extract parameters with reasonable uncertainties. For the numerical examples and within the set of hyperparameters studied, we found that the likelihood ratio method is more accurate and/or precise. Both methods have a significant spread from the network training and would require ensembling or other mitigation strategies in practice.

high energy physics↗
DOE OSTI2022

H-GCN: A Graph Convolutional Network Accelerator on Versal ACAP Architecture

Recently Graph Neural Networks (GNNs) have drawn tremendous attentions due to their unique capability to extend the Machine Learning (ML) approaches to broadly defined applications with unstructured data, especially graphs. Comparing with other ML modalities, the acceleration of GNNs is as critical but even more challenging due to the irregularity and heterogeneity from graph typologies that together limit the performance. Existing efforts mainly focus on handling graphs’ irregularity, however, have not studied the heterogeneity. To this end, in this work, we propose H-GCN, a PL-AIE-based hybrid accelerator that leverages the emerging heterogeneity of Xilinx Versal ACAPs to achieve high-performance GNN inference. In particular, H-GCN partitions each graph into three subgraphs based on its inherent heterogeneity and processes them using PL and the newly emerged AIE respectively. To further improve the performance, we explore the sparsity support of AIE and develop an efficient density-aware method to map tiles of SpMM onto the systolic tensor array automatically. Compared with the current state-of-the-art GCN accelerator, HGCN achieves on average 1.5× speedups.

Zhang, Chengming↗
DOE OSTI2021

A Taxonomy for Classification and Comparison of Dataflows for GNN Accelerators

Recently, Graph Neural Networks (GNNs) have received a lot of interest because of their success in learning representations from graph structured data. However, GNNs exhibit different compute and memory characteristics compared to traditional Deep Neural Networks (DNNs). Graph convolutions require feature aggregations from neighboring nodes (known as the aggregation phase), which leads to highly irregular data accesses. GNNs also have a very regular compute phase that can be broken down to matrix multiplications (known as the combination phase). All recently proposed GNN accelerators utilize different dataflows and microarchitecture optimizations for these two phases. Different communication strategies between the two phases have been also used. However, as more custom GNN accelerators are proposed, the harder it is to qualitatively classify them and quantitatively contrast them. In this work, we present a taxonomy to describe several diverse dataflows for running GNN inference on accelerators. This provides a structured way to describe and compare the design-space of GNN accelerators.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1992

Method and apparatus for predicting the direction of movement in machine vision

A computer-simulated cortical network is presented. The network is capable of computing the visibility of shifts in the direction of movement. Additionally, the network can compute the following: (1) the magnitude of the position difference between the test and background patterns; (2) localized contrast differences at different spatial scales analyzed by computing temporal gradients of the difference and sum of the outputs of paired even- and odd-symmetric bandpass filters convolved with the input pattern; and (3) the direction of a test pattern moved relative to a textured background. The direction of movement of an object in the field of view of a robotic vision system is detected in accordance with nonlinear Gabor function algorithms. The movement of objects relative to their background is used to infer the 3-dimensional structure and motion of object surfaces.

Lawton, Teri B.↗
NASA NTRS2010

Physics-Based Methods of Failure Analysis and Diagnostics in Human Space Flight

The Integrated Health Management (IHM) for the future aerospace systems requires to interface models of multiple subsystems in an efficient and accurate information environment at the earlier stages of system design. The complexity of modern aeronautic and aircraft systems (including e.g. the power distribution, flight control, solid and liquid motors) dictates employment of hybrid models and high-level reasoners for analysing mixed continuous and discrete information flow involving multiple modes of operation in uncertain environments, unknown state variables, heterogeneous software and hardware components. To provide the information link between key design/performance parameters and high-level reasoners we rely on development of multi-physics performance models, distributed sensors networks, and fault diagnostic and prognostic (FD&P) technologies in close collaboration with system designers. The main challenges of our research are related to the in-flight assessment of the structural stability, engine performance, and trajectory control. The main goal is to develop an intelligent IHM that not only enhances components and system reliability, but also provides a post-flight feedback helping to optimize design of the next generation of aerospace systems. Our efforts are concentrated on several directions of the research. One of the key components of our strategy is an innovative approach to the diagnostics/prognostics based on the real time dynamical inference (DI) technologies extended to encompass hybrid systems with hidden state trajectories. The major investments are into the multiphysics performance modelling that provides an access of the FD&P technologies to the main performance parameters of e.g. solid and liquid rocket motors and composite materials of the nozzle and case. Some of the recent results of our research are discussed in this chapter. We begin by introducing the problem of dynamical inference of stochastic nonlinear models and reviewing earlier results. Next, we present our analytical approach to the solution of this problem based on the path integral formulation. The resulting algorithm does not require an extensive global search for the model parameters, provides optimal compensation for the effects of dynamical noise, and is robust for a broad range of dynamical models. In the following Section the strengths of the algorithm are illustrated illustrated by inferring the parameters of the stochastic Lorenz system and comparing the results with those of earlier research. Next, we discuss a number of recent results in application to the development of the IHM for aerospace system. Firstly, we apply dynamical inference approach to a solution of classical three tank problems with mixed unknown continuous and binary parameters. The problem is considered in the context of ground support system for filling fuel tanks of liquid rocket motors. It is shown that the DI algorithm is well suited for successful solution of a hybrid version of this benchmark problem even in the presence of additional periodic and stochastic perturbation of unknown strength. Secondly, we illustrate our approach by its application to an analysis of the nozzle fault in a solid rocket motor (SRM). The internal ballistics of the SRM is modelled as a set of one-dimensional partial differential equations coupled to the dynamics of the propellant regression. In this example we are specifically focussed on the inference of discrete and continuous parameters of the nozzle blocking fault and on the possibility of an application of the DI algorithm to reducing the probability of "misses" of an on-board FD&P for SRM. In the next section re-contact problem caused by first stage/upper stage separation failure is discussed. The reaction forces imposed on the nozzle of the upper stage during the re-contact and their connection to the nozzle damage and to the thrust vector control (TVC) signal are obtained. It is shown that transient impact induced torquean be modelled as a response of an effective damped oscillator. A possible application of the DI algorithm to the inference of damage parameters and predicting fault dynamics ahead of time using the actuator signal is discussed. Finally, we formulate Bayesian inferential framework for development of the IHM system for in-flight structural health monitoring (SHM) of composite materials. We consider the signal generated by piezoelectric actuator mounted on composite structure generating elastic waves in it. The signal received by the sensor is than compared with the baseline signal. The possibility of damage inference is discussed in the context of development of the SHM.

Smelyanskiy, Vadim N.↗
NASA NTRS2007

Interactions with Virtual People: Do Avatars Dream of Digital Sheep?

This paper explores another form of artificial entity, ones without physical embodiment. We refer to virtual characters as the name for a type of interactive object that have become familiar in computer games and within virtual reality applications. We refer to these as avatars: three-dimensional graphical objects that are in more-or-less human form which can interact with humans. Sometimes such avatars will be representations of real-humans who are interacting together within a shared networked virtual environment, other times the representations will be of entirely computer generated characters. Unlike other authors, who reserve the term agent for entirely computer generated characters and avatars for virtual embodiments of real people; the same term here is used for both. This is because avatars and agents are on a continuum. The question is where does their behaviour originate? At the extremes the behaviour is either completely computer generated or comes only from tracking of a real person. However, not every aspect of a real person can be tracked every eyebrow move, every blink, every breath rather real tracking data would be supplemented by inferred behaviours which are programmed based on the available information as to what the real human is doing and her/his underlying emotional and psychological state. Hence there is always some programmed behaviour it is only a matter of how much. In any case the same underlying problem remains how can the human character be portrayed in such a manner that its actions are believable and have an impact on the real people with whom it interacts? This paper has three main parts. In the first part we will review some evidence that suggests that humans react with appropriate affect in their interactions with virtual human characters, or with other humans who are represented as avatars. This is so in spite of the fact that the representational fidelity is relatively low. Our evidence will be from the realm of psychotherapy, where virtual social situations are created that do test whether people react appropriately within these situations. We will also consider some experiments on face-to-face virtual communications between people in the same shared virtual environments. The second part will try to give some clues about why this might happen, taking into account modern theories of perception from neuroscience. The third part will include some speculations about the future developments of the relationship between people and virtual people. We will suggest that a more likely scenario than the world becoming populated by physically embodied virtual people (robots, androids) is that in the relatively near future we will interact more and more in our everyday lives with virtual people- bank managers, shop assistants, instructors, and so on. What is happening in the movies with computer graphic generated individuals and entire crowds may move into the space of everyday life.

Slater, Mel↗
DOE OSTI2024

Application of Modified Meshgraphnets for Subsurface Prediction during CO2 Sequestration

In the face of the increasingly dire consequences of anthropogenic climate change, capturing and storing carbon dioxide is paramount. However, several impediments exist to the safe and effective subsurface storage of CO2, such as cost of transport, identification of suitable sites for subsurface storage, and assessment of long-term risk from storage in subsurface aquifers. Accurate subsurface modeling is necessary to ensure that CO2 storage is both safe and effective. Still, such modeling has traditionally required either substantial time and computational power (numerical simulation) or a substantial amount of pre-existing data for training (machine learning models). Additionally, these models lack flexibility in dealing with both changes in discretization of the input data and generalizability beyond the data on which they are trained. In order to address these issues, this research applies graph neural networks (GNNs) to predict subsurface saturation and pressure during CO₂ injection in a model of the Illinois Basin-Decatur Project (IBDP). GNNs provide a flexible, intuitive method for representing and manipulating complex unstructured data, which is often found in many practical domain problems such as fluid flow and subsurface characterization. These unstructured grids are easily represented in GNNs by representing spatially-localized features such as permeability, porosity, saturation, and pressure as nodes in a graph and relationships between these properties as edges connecting these nodes. This research applies a specific GNN model called MeshGraphNets (MGN) to model the change in CO2 saturation and pressure over a 50-month time period (36 months of injection, 14 months post-injection). The MGN model leverages a message passing process that allows the network to learn both the spatial and temporal dynamics of this system simultaneously. Additionally, training on a limited dataset (64 realizations, 20 time points each) resulted in a high degree of accuracy in saturation prediction both within the same timeframe as the training (20 months, 0.039 average RMSE) and when projecting out to the end of injection (36 months, 0.053 average RMSE). Temporal predictions such as those generated by MGNs and other similar models are prone to accumulated error over time; in order to address this, a multi-step rollout (MSR) training process was applied to calculate training loss. This method mimics the forward prediction during inference by “rolling out” multiple time points in a single training step using the previous prediction as input to the MGN model. By calculating the loss several time steps forward from the current prediction, the model is forced to find a more stable state over time. Application of MSR to the MGN model resulted in an average 15% reduction in inference error over time during forward prediction. This study showcases the immense potential of GNNs as a game-changing methodology for predicting pressure and saturation evolution in CCS projects, ultimately paving the way for more sustainable and effective carbon storage solutions. Presentation prepared for the 2024 AiChE Annual Meeting, October 27 to November 1 2024, San Diego, CA.

Holcomb, Paul↗
NASA NTRS1988

Geodetic Measurement of Deformation East of the San Andreas Fault in Central California

Triangulation and trilateration data from two geodetic networks located between the western edge of the Great Valley and the San Andreas fault have been used to calculate shear strain rates in the Diablo Range and to estimate the slip rate along the Calaveras and Paicines faults in Central California. Within the Diablo Range the average shear strain rate was determined for the time period between 1962 and 1982 to be 0.15 + or - 0.08 microrad/yr, with the orientation of the most compressive strain at N 16 deg E + or - 14 deg. The orientation of the principal compressive strain predicted from the azimuth of the major structures in the region is N 25 deg E. It is inferred that the measured strain is due to compression across the folds of this area: the average shear straining corresponds to a relative shortening rate of 4.5 + or - 2.4 mm/yr. From an examination of wellbore breakout orientations and the azimuths of P-axes from earthquake focal mechanisms the inferred orientation of maximum compressive stress was found to be similar to the direction of maximum compressive strain implied by the trend of local fold structures. Results do not support the hypothesis of uniform fault-normal compression within the Coast Ranges. From trilateration measurements made between 1972 and 1987 on lines that are within 10 km of the San Andreas fault, a slip rate of 10 to 12 mm/yr was calculated for the Calaveras-Paicines fault south of Hollister. The slip rate of the Paicines fault decreases to 4 mm/yr near Bitter.

Sauber, Jeanne M.↗
NASA NTRS2002

Ionospheric Convection in the Postnoon Auroral Oval: SuperDARN and Polar UVI Observations

Super Dual Auroral Radar Network (SuperDARN) observations, ultraviolet imaging from the Polar satellite (UVI), and particle precipitation data from DMSP satellites have been used to investigate the electrodynamics of the postnoon auroral oval in the Northern hemisphere. We show that: (1) For negative IMF By, the convection reversal (CR) was co-located with the maximum of auroral luminosity, but during positive IMF By the convection reversal was poleward of the auroral oval up to several degrees in latitude; (2) Postnoon auroral oval was associated with a large-scale upward field-aligned current (FAC) of the order of 6x10(exp -7). A m(exp -2) in magnitude (the FAC was inferred from the SuperDARN and UVI data). For negative IMF By, maximum of the auroral intensity coincides in latitude with the maximum of the upward field-aligned current. However, for positive IMF By. the maximum of the upward FAC was shifted to the poleward edge of the auroral oval; (3) In response to the IMF By turning from positive to negative, the maximum of the auroral luminosity did not change its position noticeably, but the position of the convection reversal changed considerably from 80-81 degs to about 76 degs MLAT, and the maximum of FAC moved from 77-78 degs to about 76 degs MLAT. Thus, after IMF By turns negative, both the FAC maximum and CR tend to coincide with the auroral maximum; (4) The IMF Bz positive deflection was followed by a decrease in both field-aligned current intensity and auroral luminosity. However, the decrease in the auroral luminosity lags behind the FAC decrease by about 12 min. Firstly, these observations allow us to suggest that the IMF By-related electric field can penetrate into the closed magnetosphere and produce convection and FAC changes in the region of the postnoon auroral oval. Secondly, we suggest that the interchange instability is a promising mechanism for the postnoon auroras.

Kozlovsky, A.↗
NASA NTRS2020

Intensive disc-reverberation mapping of Fairall 9: first year of Swift and LCO monitoring

We present results of time-series analysis of the first year of the Fairall 9 intensive disc-reverberation campaign. We used Swift and the Las Cumbres Observatory global telescope network to continuously monitor Fairall 9 from X-rays to near-infrared at a daily to subdaily cadence. The cross-correlation function between bands provides evidence for a lag spectrum consistent with the τ ∝ λ^(4/3) scaling expected for an optically thick, geometrically thin blackbody accretion disc. Decomposing the flux into constant and variable components, the variable component’s spectral energy distribution is slightly steeper than the standard accretion disc prediction. We find evidence at the Balmer edge in both the lag and flux spectra for an additional bound-free continuum contribution that may arise from reprocessing in the broad-line region. The inferred driving light curve suggests two distinct components, a rapidly variable (<4 d) component arising from X-ray reprocessing, and a more slowly varying (>100 d) component with an opposite lag to the reverberation signal.

J V Hernández Santisteban↗
DOE OSTI2024

Machine Learning‐Assisted Microearthquake Location Workflow for Monitoring the Newberry Enhanced Geothermal System

Abstract Enhanced geothermal systems (EGS) offer a sustainable energy source but face challenges in accurately locating microearthquakes induced during reservoir stimulation. Locating these microearthquakes provides reliable feedback on the stimulation progress. Current deep learning methods for locating earthquakes require extensive data sets for training, which is problematic as detected microearthquakes are often limited. To address the scarcity of training data, we propose a practical workflow using probabilistic multilayer perceptron (PMLP) which predicts microearthquake locations from cross‐correlation time lags in waveforms. Utilizing a 3D velocity model of Newberry site derived from ambient noise interferometry, we generate numerous synthetic microearthquakes and 3D acoustic waveforms for PMLP training. Accurate synthetic tests prompt us to apply the trained network to the 2012 and 2014 stimulation field waveforms. To enhance the accuracy of source localization, we carefully handpick the P‐arrival times. Predictions on the 2012 stimulation data set show major microseismic activity at depths of 0.5–1.2 km, correlating with a known casing leakage scenario. In the 2014 data set, the majority of predictions concentrate at 2.0–2.9 km depths, consistent with results obtained from conventional physics‐based inversion, and align with the presence of natural fractures from 2.0 to 2.7 km. We validate our findings by comparing the synthetic and field picks, demonstrating a satisfactory match for the first arrivals. By combining the benefits of quick inference speeds and accurate location predictions, we demonstrate the feasibility of using realistic synthetic data set to locate microseismicity for EGS monitoring.

15 GEOTHERMAL ENERGY↗