Publications

Blind geothermal systems are believed to be common in the Basin and Range province and represent an underutilized source of renewable green energy. Their discovery has historically been by chance but more methodological strategies for exploration of these resources are being developed. One characteristic of blind systems is that they are often overlain by near-surface zones of low-resistivity caused by alteration of the overlying sediments to swelling clays. These zones can be imaged by resistivity-based geophysical techniques to facilitate their discovery and characterization. Here we present a side-by-side comparison of resistivity models produced from helicopter transient electromagnetic (HTEM) and ground-based broadband magnetotelluric (MT) surveys over a previously discovered blind geothermal system with measured shallow temperatures of ~100°C in East Hawthorne, NV. The HTEM and MT data were collected as part of the BRIDGE project, an initiative for improving methodologies for discovering blind geothermal systems. HTEM data were collected and modelled along profiles, and the results suggest the method can resolve the resistivity structure 300 - 500 m deep. A 61-station MT survey was collected on an irregular grid with ~800 m station spacing and modelled in 3D on a rotated mesh aligned with HTEM flight directions. Resistivity models are compared with results from potential fields datasets, shallow temperature surveys, and available temperature gradient data in the area of interest. We find that the superior resolution of the HTEM can reveal near-surface details often missed by MT. However, MT is sensitive to several km deep, can resolve 3D structures, and is thus better suited for single-prospect characterization. We conclude that HTEM is a more practical subregional prospecting tool than is MT, because it is highly scalable and can rapidly discover shallow zones of low resistivity that may indicate the presence of a blind geothermal system. Other factors such as land access and ground disturbance considerations may also be decisive in choosing the best method for a particular prospect. Resistivity methods in general cannot fully characterize the structural setting of a geothermal system, and so we used potential fields and other datasets to guide the creation of a diagrammatic structural model at East Hawthorne.

Battery systems are typically equipped with state of charge (SoC) estimation algorithms. Sensor measurements used to estimate SoC are susceptible to false data injection attacks (FDIAs) that aim to disturb state estimation and, consequently, damage the system. In this paper, SoC estimation methods are re-purposed to detect FDIAs targeting the current and voltage sensors of a battery stack using a combination of an improved input noise aware unscented Kalman filter (INAUKF) and a cumulative sum detector. The root mean squared error of the states estimated by the INAUKF was at least 85% lower than the traditional unscented Kalman filter for all noise levels tested. The proposed method was able to detect FDIA in the current and voltage sensors of a series-connected battery stack in 99.55% of the simulations.

The MACCS code was created by Sandia National Laboratories for the U.S. Nuclear Regulatory Commission and has been used for emergency planning, level 3 probabilistic risk assessments, consequence analyses and other scientific and regulatory research for over half a century. Specializing in modeling the transport of nuclear material into the environment, MACCS accounts for atmospheric transport and dispersion, wet and dry deposition, probabilistic treatment of meteorology, exposure pathways, varying protective actions for the emergency, intermediate and long-term phases, dosimetry, health effects (including but not limited to population dose, acute radiation injury and increased cancer risk), and economic impacts. Routine updates and recent enhancements to the MACCS code, such as the inclusion of a higher fidelity atmospheric transport and dispersion model, the addition of a new economic impact model, and the application of nearfield modeling, have continuously increased the codes capabilities in consequence analysis. Additionally, investigations of MACCS capabilities for advanced reactor applications have shown that MACCS can provide realistic and informative risk assessments for the new generation of reactor designs. Even so, areas of improvement as well as gaps have been identified that if resolved can increase the usefulness of MACCS in any application regarding a release of nuclear material into the environment.

Tabulated chemistry models are widely used to simulate large-scale turbulent fires in applications including energy generation and fire safety. Tabulation via piecewise Cartesian interpolation suffers from the curse-of-dimensionality, leading to a prohibitive exponential growth in parameters and memory usage as more dimensions are considered. Artificial neural networks (ANNs) have attracted attention for constructing surrogates for chemistry models due to their ability to perform high-dimensional approximation. However, due to well-known pathologies regarding the realization of suboptimal local minima during training, in practice they do not converge and provide unreliable accuracy. Partition of unity networks (POUnets) are a recently introduced family of ANNs which preserve notions of convergence while performing high-dimensional approximation, discovering a mesh-free partition of space which may be used to perform optimal polynomial approximation. We assess their performance with respect to accuracy and model complexity in reconstructing unstructured flamelet data representative of nonadiabatic pool fire models. Our results show that POUnets can provide the desirable accuracy of classical spline-based interpolants with the low memory footprint of traditional ANNs while converging faster to significantly lower errors than ANNs. For example, we observe POUnets obtaining target accuracies in two dimensions with 40 to 50 times less memory and roughly double the compression in three dimensions. We also address the practical matter of efficiently training accurate POUnets by studying convergence over key hyperparameters, the impact of partition/basis formulation, and the sensitivity to initialization.

I experimentally investigated and modeled a proposed frequency-domain method for detecting and tracking cislunar spacecraft and near-earth asteroids using heliostat fields at night. Unlike imaging, which detects spacecraft and asteroids by their streak in star-fixed long-exposure photographs, the proposed detection method oscillates the orientation of heliostats concentrating light from the stellar field and measures the light's photocurrent power spectrum at sub-milliHertz resolution. If heliostat oscillation repetitively traces out a closed loop fixed to the stars, light from spacecraft or asteroids moving along that loop produce photocurrent at a frequency slightly shifted from starlight. The frequency shift is proportional to the spacecraft or asteroid's apparent angular rate relative to sidereal. Relative phase corresponds to relative angular position, enabling tracking. Since heliostats are inexpensive compared to an astronomical observatory and otherwise unused at night, the proposed method may cost-effectively augment observatory systems such as NASA's Asteroid Terrestrial-impact Last Alert System (ATLAS).

Sandia National Laboratories (SNL) has completed a comparative evaluation of three design assessment approaches for a 2-liter (2L) capacity containment vessel (CV) of a novel plutonium air transport (PAT) package designed to survive the hypothetical accident condition (HAC) test sequence defined in Title 10 of the United States (US) Code of Federal Regulations (CFR) Part 71.74(a), which includes a 129 meter per second (m/s) impact of the package into an essentially unyielding target. CVs for hazardous materials transportation packages certified in the US are typically designed per the requirements defined in the American Society of Mechanical Engineers (ASME) Boiler and Pressure Vessel Code (B&PVC) Section III Division 3 Subsection WB “Class TC Transportation Containments.” For accident conditions, the level D service limits and analysis approaches specified in paragraph WB-3224 are applicable. Data derived from finite element analyses of the 129 m/s impact of the 2L-PAT package were utilized to assess the adequacy of the CV design. Three different CV assessment approaches were investigated and compared, one based on stress intensity limits defined in subparagraph WB-3224.2 for plastic analyses (the stress-based approach), a second based on strain limits defined in subparagraph WB-3224.3, subarticle WB-3700, and Section III Nonmandatory Appendix FF for the alternate strain-based acceptance criteria approach (the strain-based approach), and a third based on failure strain limits derived from a ductile fracture model with dependencies on the stress and strain state of the material, and their histories (the Xue-Wierzbicki (X-W) failure-integral-based approach). This paper gives a brief overview of the 2L-PAT package design, describes the finite element model used to determine stresses and strains in the CV generated by the 129 m/s impact HAC, summarizes the three assessment approaches investigated, discusses the analyses that were performed and the results of those analyses, and provides a comparison between the outcomes of the three assessment approaches.

More than 90% of utility-scale photovoltaic (PV) power plants in the US use single-axis trackers (SATs) due to their potential for substantially higher power production over fixed-array systems. However, they are subject to software misconfigurations and mechanical failures, leading to suboptimal tracking accuracy. If failures are left undetected, the overall power yield of the PV power plant is reduced significantly. Robust detection and diagnosis of SAT faults is needed to minimize downtime and ensure continuous and efficient operation. This work presents analytic tools based on machine learning to detect deviations in SAT tracking performance and classify SAT faults.

The diesel-piloted dual-fuel compression ignition combustion strategy is well-suited to accelerate the decarbonization of transportation by adopting hydrogen as a renewable energy carrier into the existing internal combustion engine with minimal engine modifications. Despite the simplicity of engine modification, many questions remain unanswered regarding the optimal pilot injection strategy for reliable ignition with minimum pilot fuel consumption. The present study uses a single-cylinder heavy-duty optical engine to explore the phenomenology and underlying mechanisms governing the pilot fuel ignition and the subsequent combustion of a premixed hydrogen-air charge. The engine is operated in a dual-fuel mode with hydrogen premixed into the engine intake charge with a direct pilot injection of n-heptane as a diesel pilot fuel surrogate. Optical diagnostics used to visualize in-cylinder combustion phenomena include high-speed IR imaging of the pilot fuel spray evolution as well as high-speed HCHO* and OH* chemiluminescence as indicators of low-temperature and high-temperature heat release, respectively. Three pilot injection strategies are compared to explore the effects of pilot fuel mass, injection pressure, and injection duration on the probability and repeatability of successful ignition. The thermodynamic and imaging data analysis supported by zero-dimensional chemical kinetics simulations revealed a complex interplay between the physical and chemical processes governing the pilot fuel ignition process in a hydrogen containing charge. Hydrogen strongly inhibits the ignition of pilot fuel mixtures and therefore requires longer injection duration to create zones with sufficiently high pilot fuel concentration for successful ignition. Results show that ignition typically tends to rely on stochastic pockets with high pilot fuel concentration, which results in poor repeatability of combustion and frequent misfiring. This work has improved the understanding on how the unique chemical properties of hydrogen pose a challenge for maximization of hydrogen's energy share in hydrogen dual-fuel engines and highlights a potential mitigation pathway.

Multiple scattering is a common phenomenon in acoustic media that arises from the interaction of the acoustic field with a network of scatterers. This mechanism is dominant in problems such as the design and simulation of acoustic metamaterial structures often used to achieve acoustic control for sound isolation, and remote sensing. In this study, we present a physics-informed neural network (PINN) capable of simulating the propagation of acoustic waves in an infinite domain in the presence of multiple rigid scatterers. This approach integrates a deep neural network architecture with the mathematical description of the physical problem in order to obtain predictions of the acoustic field that are consistent with both governing equations and boundary conditions. The predictions from the PINN are compared with those from a commercial finite element software model in order to assess the performance of the method.

Disposal of commercial spent nuclear fuel in a geologic repository is studied. In situ heater experiments in underground research laboratories provide a realistic representation of subsurface behavior under disposal conditions. This study describes process model development and modeling analysis for a full-scale heater experiment in opalinus clay host rock. The results of thermal-hydrology simulation, solving coupled nonisothermal multiphase flow, and comparison with experimental data are presented. The modeling results closely match the experimental data.

Operation and control of a galvanically isolated three-phase AC-AC converter for solid state transformer applications is described. The converter regulates bidirectional power transfer by phase shifting voltages applied on either side of a high-frequency transformer. The circuit structure and control system are symmetrical around the transformer. Each side operates independently, enabling conversion between AC systems with differing voltage magnitude, phase angle, and frequency. This is achieved in a single conversion stage with low component count and high efficiency. The modulation strategy is discussed in detail and expressions describing the relationship between phase shift and power transfer are presented. Converter operation is demonstrated in a 3 kW hardware prototype.

Spatial navigation involves the formation of coherent representations of a map-like space, while simultaneously tracking current location in a primarily unsupervised manner. Despite a plethora of neurophysiological experiments revealing spatially-tuned neurons across the mammalian neocortex and subcortical structures, it remains unclear how such representations are acquired in the absence of explicit allocentric targets. Drawing upon the concept of predictive learning, we utilize a biologically plausible learning rule which utilizes sensory-driven observations with internally-driven expectations and learns through a contrastive manner to better predict sensory information. The local and online nature of this approach is ideal for deployment to neuromorphic hardware for edge-applications. We implement this learning rule in a network with the feedforward and feedback pathways known to be necessary for spatial navigation. After training, we find that the receptive fields of the modeled units resemble experimental findings, with allocentric and egocentric representations in the expected order along processing streams. These findings illustrate how a local and self-supervised learning method for predicting sensory information can extract latent structure from the environment.

Fault location, isolation, and service restoration of a self-healing, self-Assembling microgrid operating off-grid from distributed inverter-based resources (IBRs) can be a unique challenge because of the fault current limitations and uncertainties regarding which sources are operational at any given time. The situation can become even more challenging if data sharing between the various microgrid controllers, relays, and sources is not available. This paper presents an innovative robust partitioning approach, which is used as part of a larger self-Assembling microgrid concept utilizing local measurements only. This robust partitioning approach splits a microgrid into sub-microgrids to isolate the fault to just one of the sub-microgrids, allowing the others to continue normal operation. A case study is implemented in the IEEE 123-bus distribution test system in Simulink to show the effectiveness of this approach. The results indicate that including the robust partitions leads to less loss of load and shorter overall restoration times.

The ability to accurately predict the structure and dynamics of pool fires using computational simulations is of great interest in a wide variety of applications, including accidental and wildland fires. However, the presence of physical processes spanning a broad range of spatial and temporal scales poses a significant challenge for simulations of such fires, particularly at conditions near the transition between laminar and turbulent flow. In this study, we examine the transition to turbulence in methane pool fires using high-resolution simulations with multi-step finite rate chemistry, where adaptive mesh refinement (AMR) is used to directly resolve small-scale flow phenomena. We perform three simulations of methane pool fires, each with increasing diameter, corresponding to increasing inlet Reynolds and Richardson numbers. As the diameter increases, the flow transitions from organized vortex roll-up via the puffing instability to much more chaotic mixing associated with finger formation along the shear layer and core collapse near the inlet. These effects combine to create additional mixing close to the inlet, thereby enhancing fuel consumption and causing more rapid acceleration of the fluid above the pool. We also make comparisons between the transition to turbulence and core collapse in the present pool fires and in inert helium plumes, which are often used as surrogates for the study of buoyant reacting flows.

Gallium nitride (GaN)-based nanoscale vacuum electron devices, which offer advantages of both traditional vacuum tube operation and modern solid-state technology, are attractive for radiation-hard applications due to the inherent radiation hardness of vacuum electron devices and the high radiation tolerance of GaN. Here, we investigate the radiation hardness of top-down fabricated n-GaN nanoscale vacuum electron diodes (NVEDs) irradiated with 2.5-MeV protons (p) at various doses. We observe a slight decrease in forward current and a slight increase in reverse leakage current as a function of cumulative protons fluence due to a dopant compensation effect. The NVEDs overall show excellent radiation hardness with no major change in electrical characteristics up to a cumulative fluence of 5E14 p/cm2, which is significantly higher than the existing state-of-the-art radiation-hardened devices to our knowledge. The results show promise for a new class of GaN-based nanoscale vacuum electron devices for use in harsh radiation environments and space applications.

This experimental work investigates the flow field associated with a tandem expansion-compression geometry in the Sandia Trisonic Wind Tunnel. PIV measurements of the boundary layer before the expansion characterized properties of the incoming boundary layer. Schlieren and oil-flow experiments were conducted at Mach 1.5 and 2 for a range of stagnation pressures. Further PIV experiments were conducted across the expansion and compression corners to observe the post-expansion changes in the boundary layer and its influence on the shock/boundary-layer interaction at the compression corner. Velocity profiles qualitatively matched well with RANS simulations and showed rapid growth of the boundary layer following the expansion, tapering off to a slower rate of growth with distance. Turbulence intensity, exemplified by the streamwise component of turbulent normal stress, diminished substantially after the expansion corner leading to the belief that relaminarization processes were occurring. Comparison with analysis from the literature suggests that the distortion of the boundary layer due to expansion leads to a separation length 30% larger than for an equilibrium turbulent boundary layer.

Neuromorphic computing platforms hold the promise to dramatically reduce power requirements for calculations that are computationally intensive. One such application space is scientific machine learning (SciML). Techniques in this space use neural networks to approximate solutions of scientific problems. For instance, the popular physics-informed neural network (PINN) approximates the solution to a partial differential equation by using a trained feed-forward neural network, and injecting the knowledge of the physics through the loss function. Recent efforts have demonstrated how to convert a trained PINN to a spiking network architecture. In this work, we discuss our approach to quantization and implementation required to migrate these spiking PINNs to Intel's Loihi 2 neuromorphic hardware. We explore the effect of quantization on the model accuracy, as well as the energy and throughput characteristics of the implementation. It is our intent that this serve as a starting point for additional SciML implementations on neuromorphic hardware.

The primary goal of any laboratory test is to expose the unit-under-test to conservative realistic representations of a field environment. Satisfying this objective is not always straightforward due to laboratory equipment constraints. For vibration and shock tests performed on shakers over-testing and unrealistic failures can result because the control is a base acceleration and mechanical shakers have nearly infinite impedance. Force limiting and response limiting are relatively standard practices to reduce over-test risks in random-vibration testing. Shaker controller software generally has response limiting as a built-in capability and it is done without much user intervention since vibration control is a closed loop process. Limiting in shaker shocks is done for the same reasons, but because the duration of a shock is only a few milliseconds, limiting is a pre-planned user in the loop process. Shaker shock response limiting has been used for at least 30 years at Sandia National Laboratories, but it seems to be little known or used in industry. This objective of this paper is to re-introduce response limiting for shaker shocks to the aerospace community. The process is demonstrated on the BARBECUE testbed.

Point-source transient events (PSTEs) - optical events that are both extremely fast and extremely small - pose several challenges to an imaging system. Due to their speed, accurately characterizing such events often requires detectors with very high frame rates. Due to their size, accurately detecting such events requires maintaining coverage over an extended field-of-view, often through the use of imaging focal plane arrays (FPA) with a global shutter readout. Traditional imaging systems that meet these requirements are costly in terms of price, size, weight, power consumption, and data bandwidth, and there is a need for cheaper solutions with adequate temporal and spatial coverage. To address these issues, we develop a novel compressed sensing algorithm adapted to the rolling shutter readout of an imaging system. This approach enables reconstruction of a PSTE signature at the sampling rate of the rolling shutter, offering a 1-2 order of magnitude temporal speedup and a proportional reduction in data bandwidth. We present empirical results demonstrating accurate recovery of PSTEs using measurements that are spatially undersampled by a factor of 25, and our simulations show that, relative to other compressed sensing algorithms, our algorithm is both faster and yields higher quality reconstructions. We also present theoretical results characterizing our algorithm and corroborating simulations. The potential impact of our work includes the development of much faster, cheaper sensor solutions for PSTE detection and characterization.

Dendrites enable neurons to perform nonlinear operations. Existing silicon dendrite circuits sufficiently model passive and active characteristics, but do not exploit shunting inhibition as an active mechanism. We present a dendrite circuit implemented on a reconfigurable analog platform that uses active inhibitory conductance signals to modulate the circuit's membrane potential. We explore the potential use of this circuit for direction selectivity by emulating recent observations demonstrating a role for shunting inhibition in a directionally-selective Drosophila (Fruit Fly) neuron.

Anelastic strain recovery, the process of measuring the time dependent recovered strain after a core is cut at depth was utilized to make a measure of the in-situ properties stresses at depth at the FORGE (Frontier Observatory for Research in Geothermal Energy) site in Milford Utah. Core was collected from a region of well 16B at approximately 4860-4870 ft. Core was instrumented with strain gages within 10 hours of the core being cut. The relaxation of the cores was measured for approximately one month, and the results analyzed, which showed that the principal stresses were slightly off vertical, and magnitudes are close to equal.

In general, multiple-input/multiple-output (MIMO) vibration testing utilizes a response-controlled test methodology where specifications are in the form of response quantities at various locations distributed on the device under test (DUT). There are some advantages to this approach, namely that DUT response could be measured in some field environment and directly used as MIMO specifications for subsequent MIMO vibration tests on similar DUTs. However, in some cases it may be advantageous to control the MIMO vibration test at the inputs rather than the responses. One such case is free-flight environments, where the DUT is unconstrained, and all loads come from aerodynamic pressures. In this case, the force-controlled test method is much more robust to system changes such as unit-to-unit variability as compared to a response-controlled test method. This could make force-controlled MIMO test specifications more generalizable and easier to derive. This is exactly akin to transfer path analysis, where pseudo-forces are applicable in special circumstances. This paper will explore the force-controlled test concept and demonstrate it with a numerical example, comparing performance under various conditions vs. the traditional response-controlled test method.

In low inertia grids, significant frequency deviations can occur as a result of changes in power (load, generation, etc.), These deviations may activate various protection schemes designed to safeguard the system, potentially leading to blackouts. Therefore, assessing the frequency stability of the power system is crucial. The Frequency Security Index (FSI) serves as a metric for evaluating system stability. However, computing the FSI for a specific load change necessitates actual load changes on the system, which is often impractical. This paper introduces a method for calculating the FSI without requiring load changes for all values. A mathematical expression for the FSI is derived, which uses the values of microgrid parameters (such as inertia and damping constant) to compute the FSI for any load change. Subsequently, the parameters that most significantly affect the FSI are identified. Then, the paper introduces a Moving Horizon Estimation (MHE)-based parameter estimation approach, which leverages small perturbations from an energy storage system to estimate the most influential parameters for the FSI. The results show that the FSI calculation with the estimated parameters is more accurate (compared to COI averaged parameters), enabling a more effective state of health monitoring of the microgrid.

Multifidelity emulators have found wide-ranging applications in both forward and inverse problems within the computational sciences. Thanks to recent advancements in neural architectures, they provide significant flexibility for integrating information from multiple models, all while retaining substantial efficiency advantages over single-fidelity methods. In this context, existing neural multifidelity emulators operate by separately resolving the linear and nonlinear correlation between equally parameterized high-and low-fidelity approximants. However, many complex models ensembles in science and engineering applications only exhibit a limited degree of linear correlation between models. In such a case, the effectiveness of these approaches is impeded, i.e., larger datasets are needed to obtain satisfactory predictions. In this work, we present a general strategy that seeks to maximize the linear correlation between two models through input encoding. We showcase the effectiveness of our approach through six numerical test problems, and we show the ability of the proposed multifidelity emulator to accurately recover the high-fidelity model response under an increasing number of quasi-random samples. In our experiments, we show that input encoding produces in many cases emulators with significantly simpler nonlinear correlations. Finally, we demonstrate how the input encoding can be leveraged to facilitate the fusion of information between low-and high-fidelity models with dissimilar parametrization, i.e., situations in which the number of inputs is different between low-and high-fidelity models.

Uncertainty quantification (UQ) plays a vital role in addressing the challenges and limitations encountered in full-waveform inversion (FWI). Most UQ methods require parameter sampling which requires many forward and adjoint solves. This often results in very high computational overhead compared to traditional FWI, which hinders the practicality of the UQ for FWI. In this work, we develop an efficient UQ-FWI framework based on unsupervised variational autoencoder (VAE) to assess the uncertainty of single and multi-parameter FWI. The inversion operator is modeled using an encoder-decoder network. The input to the network is seismic shot gathers and the output are samples (distribution) of model parameters. We then use these samples to estimate the mean and standard deviation of each parameter population, which provide insights on the uncertainty in the inversion process. To speed up the UQ process, we carried out the reconstruction in an unsupervised learning approach. Moreover, we physics-constrained the network by injecting the FWI gradients during the backpropagation process, leading to better reconstruction. The computational cost of the proposed approach is comparable to the traditional autoencoder full-waveform inversion (AE-FWI), which is encouraging to be used to get further insight on the quality of the inversion. We apply this idea for synthetic data to show its potential in assessing uncertainty in multi-parameter FWI.