Search NASA⌕ Search

SEARCH · Search NASA

Results for “deep generative models”

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 1,297 records · Page 72

yProv4ML: Effortless provenance tracking for machine learning systems

The rapid growth in interest in deep learning and foundation models (FMs) in particular, has attracted the attention of a diverse range of researchers thanks to their generalization ability. However, the advent of these techniques has also brought to light the lack of transparency and rigor in the way development is pursued. In particular, the inability to determine the number of epochs and other hyperparameters in advance presents challenges in identifying the best model. To address this challenge, machine learning frameworks such as MLFlow can automate the collection of this type of information. However, these tools capture data using proprietary formats and pose little attention to lineage. This paper proposes yProv4ML, a framework that captures provenance information generated during machine learning processes in PROV-JSON format, with minimal code modification.

Machine learning↗

Detailed investigations of the dimmest bursts from two magnetars, SGR J0501+4516 and SGR J1550–5418

We applied the Bayesian blocks representation technique to search for the dimmest bursts from two magnetars: we identified 320 events from SGR J0501+4516 using a deep XMM-Newton observation and 404 bursts from SGR J1550–5418 using two Swift/X-Ray Telescope pointings. The fluence level of our sample for both sources is about one to two orders of magnitude lower than earlier studies. We systematically investigated the morphological characteristics and duration distributions of these bursts, as these properties are directly obtained from their Bayesian blocks profiles. We also studied the spectral behavior of the dimmest bursts, which were grouped based on the morphological types and fluences. Our results helped us further differentiate the spectral nature of very dim bursts from that of the persistent emission, both fitted with physically motivated continuum emission models. Moreover, we generated the differential burst fluence distribution for these two magnetars in the lowest fluence regime.

79 ASTRONOMY AND ASTROPHYSICS↗

Water Management Strategies for Improved Coalbed Methane Production in the Black Warrior Basin

The modern coalbed methane industry was born in the Black Warrior Basin of Alabama and has to date produced more than 2.6 trillion cubic feet of gas and 1.6 billion barrels of water. The coalbed gas industry in this area is dependent on instream disposal of co-produced water, which ranges from nearly potable sodium-bicarbonate water to hypersaline sodium-chloride water. This study employed diverse analytical methods to characterize water chemistry in light of the regional geologic framework and to evaluate the full range of water management options for the Black Warrior coalbed methane industry. Results reveal strong interrelationships among regional geology, water chemistry, and gas chemistry. Coalbed methane is produced from multiple coal seams in Pennsylvanian-age strata of the Pottsville Coal Interval, in which water chemistry is influenced by a structurally controlled meteoric recharge area along the southeastern margin of the basin. The most important constituents of concern in the produced water include chlorides, ammonia compounds, and organic substances. Regional mapping and statistical analysis indicate that the concentrations of most ionic compounds, metallic substances, and nonmetallic substances correlate with total dissolved solids and chlorides. Gas is effectively produced at pipeline quality, and the only significant impurity is N{sub 2}. Geochemical analysis indicates that the gas is of mixed thermogenic-biogenic origin. Stable isotopic analysis of produced gas and calcite vein fills indicates that widespread late-stage microbial methanogenesis occurred primarily along a CO{sub 2} reduction metabolic pathway. Organic compounds in the produced water appear to have helped sustain microbial communities. Ammonia and ammonium levels increase with total dissolved solids content and appear to have played a role in late-stage microbial methanogenesis and the generation of N{sub 2}. Gas production tends to decline exponentially, whereas water production tends to decline hyperbolically. Hyperbolic decline indicates that water volume is of greatest concern early in the life of a coalbed methane project. Regional mapping indicates that gas production is controlled primarily by the ability to depressurize permeable coal seams that are natively within the steep part of the adsorption isotherm. Water production is greatest within the freshwater intrusion and below thick Cretaceous cover strata and is least in areas of underpressure. Water management strategies include instream disposal, which can be applied effectively in most parts of the basin. Deep disposal may be applicable locally, particularly where high salinity limits the ability to dispose into streams. Artificial wetlands show promise for the management of saline water, especially where the reservoir yield is limited. Beneficial use options include municipal water supply, agricultural use, and industrial use. The water may be of use to an inland shrimp farming industry, which is active around the southwestern coalbed methane fields. The best opportunities for beneficial use are reuse of water by the coalbed methane industry for drilling and hydraulic fracturing. This research has further highlighted opportunities for additional research on treatment efficiency, the origin of nitrogen compounds, organic geochemistry, biogenic gas generation, flow modeling, and computer simulation. Results of this study are being disseminated through a vigorous technology transfer program that includes web resources, numerous presentations to stakeholders, and a variety of technical publications.

01 COAL, LIGNITE, AND PEAT↗

MSD CoP Webinar: AI and Extreme Events - Overcoming Data Challenges for Improved Characterization of Climate Extremes

Context: This webinar was hosted by the MultiSector Dynamics Community of Practice (MSD CoP; https://multisectordynamics.org). Abstract: Artificial Intelligence (AI) models require large volumes of data for training and testing. Data requirements present challenges for using AI to explore extreme events with limited observational data. This webinar will showcase two innovative methods developed by part of the European Climate Intelligence (CLINT) project to overcome data challenges and harness AI to improve our understanding of climate extremes. Dr. Ascenso will present his research on data augmentation methods to improve estimates of tropical cyclones using satellite data. His presentation will review established methods for data augmentation and explore opportunities and challenges for using generative AI to generate images of extreme, life-threatening tropical cyclones. Next, Dr. Plesiat will present his research on deep learning techniques to overcome limited observational data sets. His presentation will illustrate deep learning methods to develop AI reconstructions of four climate indices across Europe. Presenters : Dr. Guido Ascenso (post-doctoral researcher, Politecnico di Milano); Dr. Étienne Plésiat (German Climate Computing Centre - DKRZ) Moderator(s): Stefano Galelli (MSD CoP WG Co-Lead), David Gold (MSD CoP WG Co-Lead), Jillian Sturtevant (MSD CoP WG Communications Officer), Matteo Giuliani (Politecnico di Milano, MSD CoP WG Member, Moderator and Organizer) This webinar was held on: October 11, 2024 from 11AM - 1PM ET

AI↗

Spectral and Temporal Characteristics of X-Ray-Bright Stars in the Pleiades

We follow up our deep ROSAT imaging survey of the Pleiades (Stauffer et al. 1994) with an analysis of the spectral and temporal characteristics of the X-ray-bright stars in the Pleiades. Raymond & Smith (1977) one and two-temperature models have been used to fit the position-sensitive proportional counter (PSPC) pulse-height spectra of the dozen or so brightest sources associated with late-type Pleiades members. The best-fit temperatures suggest hot coronal temperatures for K, M, and rapidly rotating G stars, and cooler temperatures for F and slowly rotating G stars. In order to probe the many less X-ray-luminous stars, we have generated composite spectra by combining net counts from all Pleiades members according to spectral type and rotational velocity. Model fits to the composite spectra confirm the trend seen in the individual spectral fits. Particularly interesting is the apparent dependence of coronal temperature on L(sub x)/L(sub bol). A hardness-ratio analysis also confirms some of these trends. The PSPC data have also revealed a dozen or so strong X-ray flares with peak X-ray luminosities in excess of approx. 10(exp 30) ergs/sec. We have modeled the brightest of these flares with a simple quasi-static cooling loop model. The peak temperature and emission measure and the inferred electron density and plasma volume suggest a very large scale flaring event. The PSPC data were collected over a period of approx. 18 months, allowing us to search for source variability on timescales ranging from less than a day (in the case of flares) to more than a year between individual exposures. On approximately year-long timescales, roughly 25% of the late-type stars are variable. Since the Pleiades was also intensively monitored by the imaging instruments on the Einstein Observatory, we have examined X-ray luminosity variations on the 10 yr timescale between Einstein and ROSAT and find that up to 40% of the late-type stars are X-ray variable. Since there is only marginal evidence for increased variability on decade-long timescales, the variability observed on long and short timescales may have a common physical origin.

Gagne, Marc↗

Next-Generation Ion Propulsion Being Developed

The NASA Glenn Research Center ion-propulsion program addresses the need for high specific-impulse systems and technology across a broad range of mission applications and power levels. One activity is the development of the next-generation ion-propulsion system as a follow-on to the successful Deep Space 1 system. The system is envisioned to incorporate a lightweight ion engine that can operate over 1 to 10 kW, with a 550-kg propellant throughput capacity. The engine concept under development has a 40-cm beam diameter, twice the effective area of the Deep Space 1 engine. It incorporates mechanical features and operating conditions to maximize the design heritage established by the Deep Space 1 engine, while incorporating new technology where warranted to extend the power and throughput capability. Prototype versions of the engine have been fabricated and are under test at NASA, with an engineering model version in manufacturing. Preliminary performance data for the prototype engine have been documented over 1.1- to 7.3-kW input power. At 7.3 kW, the engine efficiency is 0.68, at 3615-sec specific impulse. Critical component temperatures, including those of the discharge cathode assembly and magnets, have been documented and are within established limits, with significant margins relative to the Deep Space 1 engine. The 1- to 10-kW ion thruster approach described here was found to provide the needed power and performance improvement to enable important NASA missions. The Integrated In-Space Transportation Planning (IISTP) studies compared many potential technologies for various NASA, Government, and commercial missions. These studies indicated that a high-power ion propulsion system is the most important technology for development because of its outstanding performance versus perceived development and recurring costs for interplanetary solar electric propulsion missions. One of the best applications of a highpower electric propulsion system was as an integral part of a solar electric propulsion (SEP) stage to send a payload to outer planet targets. The IISTP studies showed that either trip time or launch vehicle class could be significantly reduced when compared with state-of-the-art systems.

Patterson, Michael J.↗

The corrosion of UK vitrified waste during disposal

The UK's vitrified high-level-waste (HLW) is destined to be disposed of in a deep geological disposal facility (GDF). Contact with groundwater will be the process by which radionuclides are released during 100,000+ years of disposal. Understanding the mechanisms and kinetics of glass dissolution allow models and predictions to be made on the long-term fate of the disposed HLW. Current standard laboratory scale experimental methodologies may be unrepresentative of the geological disposal environment. Aim: Provide insight into the dissolution behaviour of alkali-alumino-borosilicate UK vitrified waste in geological disposal environments. A 'cyclic vapor-static' dissolution methodology is proposed, capturing expected temporal behaviour. This will elucidate dissolution kinetics and mechanisms, optimize the base glass composition, waste loading content and generate significant, pragmatic disposal performance assessment data. Variants of three glasses to be investigated: waste-loaded MW25 and ZnCa MW28, and the ZnCa MW base. Preliminary Results: Suggests drying process decreases the chemical durability of the glass. Addition of ZnCa to UK HLW improves chemical durability. Increased waste-loading improves chemical durability. Banded structure in altered glass possibly attributed to dissolution/reprecipitation of secondary phases). Conclusion - Further Work: Cyclic vapor-static dissolution methodology may elucidate dissolution behaviour expected within a UK GDF. Potential to provide beneficial, pragmatic data for the performance assessment to implement geological disposal. Fundamental mechanisms to be understood.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Key Drivers for Offshore Wind Deployment in the Western United States

The western coast of the United States has abundant natural resources, including wind, sun, and water. Despite strong ocean winds, offshore wind energy (OSW) in the Western United States is in the early stages of technology development due largely to the deep water off the coast, which requires floating platforms rather than fixed‐bottom turbines. Floating OSW increases the technical difficulty of installation and thus the cost relative to both fixed‐bottom OSW and land‐based turbines. OSW is a potential generation option to help meet increasing demand on the west coast of the United States because it likely has fewer land‐use conflicts than other technologies and complements sources in the existing electricity supply by providing energy during times of high system stress. This paper examines the possible drivers and barriers to OSW deployment on the West Coast using the National Laboratory of the Rockies' state‐of‐the‐art capacity expansion model and the Regional Energy Deployment System (ReEDS) model. We use ReEDS to explore a multitude of future scenarios looking at key drivers for OSW deployment, including variations on the cost of OSW, electricity demand growth, and the availability of competing technologies to examine the factors that may play a role in OSW growth. Assuming coastal state policies such as renewable portfolio and clean energy standards remain in place, we find that OSW can play a role in meeting electricity demand and provide energy during stressful grid conditions and find that deployment from the least‐cost investment model ranges from 7.6 to 38 GW by 2045.

17 WIND ENERGY↗

Tuning Catalytic Performance through a Single or Sequential Post-Synthesis Reaction(s) in a Gas Phase

Catalytic performance of a bimetallic catalyst is determined by geometric structure and electronic state of the surface or even the near-surface region of the catalyst. Here we report that single and sequential postsynthesis reactions of an as-synthesized bimetallic nanoparticle catalyst in one or more gas phases can tailor surface chemistry and structure of the catalyst in a gas phase, by which catalytic performance of this bimetallic catalyst can be tuned. Pt–Cu regular nanocube (Pt–Cu RNC) and concave nanocube (Pt–Cu CNC) are chosen as models of bimetallic catalysts. Surface chemistry and catalyst structure under different reaction conditions and during catalysis were explored in gas phase of one or two reactants with ambient-pressure X-ray photoelectron spectroscopy (AP-XPS) and extended X-ray absorption fine structure (EXAFS) spectroscopy. The newly formed surface structures of Pt–Cu RNC and Pt–Cu CNC catalysts strongly depend on the reactive gas(es) used in the postsynthesis reaction(s). A reaction of Pt–Cu RNC-as synthesized with H2 at 200 °C generates a near-surface alloy consisting of a Pt skin layer, a Cu-rich subsurface, and a Pt-rich deep layer. This near-surface alloy of Pt–Cu RNC-as synthesized-H2 exhibits a much higher catalytic activity in CO oxidation in terms of a low activation barrier of 39 ± 4 kJ/mol in contrast to 128 ± 7 kJ/mol of Pt–Cu RNC-as synthesized. Here the significant decrease of activation barrier demonstrates a method to tune catalytic performances of as-synthesized bimetallic catalysts. A further reaction of Pt–Cu RNC-as synthesized-H2 with CO forms a Pt–Cu alloy surface, which exhibits quite different catalytic performance in CO oxidation. It suggests the capability of generating a different surface by using another gas. The capability of tuning surface chemistry and structure of bimetallic catalysts was also demonstrated in restructuring of Pt–Cu CNC-as synthesized.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Reactive Transport Simulations of High-Tempertature Geologic Thermal Energy Storage (GeoTES) in Deep Saline Formations - I/O Files

Simulation input and output files, post-processed figures and excel tables, and tecplot layout files for generating figures. These simulations were run with TOUGHREACT V4.12 by Lawrence Berkeley National Laboratory in 2021. This work was completed as part of the geologic thermal energy storage (GeoTES) research project reported in the final report for Phase I of this work, which is linked below.

15 GEOTHERMAL ENERGY↗

Performance of a Ka-band transponder breadboard for deep-space applications

This article summarizes the design concepts applied in the development of and advanced Ka-band (34.4 GHz/32 GHz) transponder breadboard for the next generation of space communications systems applications. The selected architecture upgrades the X-band (7.2 GHz/8.4 GHz) deep-space transponder (DST) to provide Da-band up/Ka- and X-band down capability. The Ka-band transponder breadboard incorporates several state-of-the-art components, including sampling mixers, a Ka-band dielectric resonator oscillator, and microwave monolithic integrated circuits (MMICs). The MMICs that were tested in the breadboard include upconverters, downconverters, automatic gain control circuits, mixers, phase modulators, and amplifiers. The measured receiver dynamic range, tracking range, acquisition rate, static phase error, and phase jitter characteristics of the Ka-band breadboard interfaced to the advanced engineering model X-band DST are in good agreement with the expected performance. The results show a receiver tracking threshold of -149 dBm with a dynamic range of 80 dB and a downlink phase jitter of 7 deg rms. The analytical results of phase noise and Allan standard deviation are in good agreement with the experimental results.

Mysoor, N. R.↗

The development of thermodynamically consistent and physics-informed equation-of-state model through machine learning

Ab initio molecular dynamics (AIMD) simulations have become an important tool used in the construction of equations of state (EOS) tables for warm dense matter. Due to computational costs, only a limited number of system state conditions can be simulated, and the remaining EOS surface must be interpolated for use in radiation-hydrodynamic simulations of experiments. In this work, we develop a thermodynamically consistent EOS model that utilizes a physics-informed machine learning approach to implicitly learn the underlying Helmholtz free-energy from AIMD generated energies and pressures. The model, referred to as PIML-EOS, was trained and tested on warm dense polystyrene producing a fit within a 1% relative error for both energy and pressure and is shown to satisfy both the Maxwell and Gibbs–Duhem relations. In addition, we provide a path toward obtaining thermodynamic quantities, such as the total entropy and chemical potential (containing both ionic and electronic contributions), which are not available from current AIMD simulations.

74 ATOMIC AND MOLECULAR PHYSICS↗

Sampling Subjective Polygons for Patch-Based Deep Learning Land-Use Classification in Satellite Images

Model generalization remains a key challenge in the analysis of large amounts of heterogeneous satellite image data. One major limiting factor in developing generalizable models, in the context of supervised learning, is the lack of high quality training datasets. A model's capacity to perform well on new data is often inhibited by imbalance and bias in the data that was used for training. This is especially a problem when using convolutional neural networks to classify urban land-use in satellite images. Notable dataset imbalance issues in this application include land-use type imbalance and image scene imbalance. To begin understanding these dataset imbalance problems in more detail, we develop and test a number of sampling methods for generating training image datasets from subjective training polygons for urban land-use classification. We investigate sampling at different point densities as a means to reduce content repetition and therefore content imbalance and bias in the training image dataset.

Arndt, Jacob↗

Probing Interfacial Electronic States in CdSe Quantum Dots using Second Harmonic Generation Spectroscopy

Understanding and rationally controlling the properties of nanomaterial surfaces is a rapidly expanding field of research due to the dramatic role they play on the optical and electronic properties vital to light harvesting, emitting and detection technologies. This information is essential to the continued development of synthetic approaches designed to tailor interfaces for optimal nanomaterial based device performance. In this work, closely spaced electronic excited states in model CdSe quantum dots (QDs) are resolved using second harmonic generation (SHG) spectroscopy, and the corresponding contributions from surface species to these states are assessed. Two distinct spectral features are observed in the SHG spectra, which are not readily identified in linear absorption and photoluminescence excitation spectra. These features include a weak band at 395 6 nm, which coincides with transitions to the 2S1/2 1Se state, and a much more pronounced band at 423 4 nm arising from electronic transitions to the 1P3/2 1Pe state. Chemical modification of the QD surfaces through oxidation resulted in disappearance of the SHG band corresponding to the 1P3/2 1Pe state, indicating prominent surface contributions. Signatures of deep trap states localized on the surfaces of the QDs are also observed. We further find that the SHG signal intensities depend strongly on the electronic states being probed and their relative surface contributions, thereby offering additional insight into the surface specificity of SHG signals from QDs.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Applying Machine Learning to the Classification of DC-DC Converters. NA-22 Final Report

Tools are now available that enable measurement electromagnetic radiation (EMR) from active electronics in an item. This radiation may be intended WIFI or cellular network links, for example or unintended such as the switching noise generated by DC-to-DC converters. It would be extremely valuable to have the capability to discriminate between the low-voltage DC-to-DC converters or other digital noise prevalent in most modern electronics, versus the high-voltage DC-to-DC converters used in utility firesets. Previous work performed under a Sandia Laboratory Directed Research and Development (LDRD) project on Charge State Detection using a deep neural network has been continued in this effort. A state-of-the-art supervised machine learning algorithm has not only been extended to discriminate between low and high voltage converters but has been validated in determining a converters make and model.

42 ENGINEERING↗

Hyperspectral remote sensing-based plant community map for region around NGEE-Arctic intensive research watersheds at Seward Peninsula, Alaska, 2017-2019

Using airborne hyperspectral remote sensing data from NASA Airborne Visible-Infrared Imaging Spectrometer- Next Generation (AVIRIS-NG) platforms in a region near NGEE-Arctic intensive watersheds at Seward peninsula of Alaska, high resolution (5m) maps of plant community distribution were developed and included in this data collected. AVIRIS-NG data collected over 2017-2019 period were used to develop deep neural networks, trained using vegetation plot observations collected at NGEE-Arctic watersheds at Kougarok, Council and Teller. A hierarchical vegetation classification scheme consisting of six classes at Level I, and 16 classes at Level II contained in two .txt files were used to developed the plant community maps for the region. Two geospatial raster data files (.tif) at both thematic levels are shared in this data collection. Data files in this collection use Alaska Albers Equal Area projection. Readme files available in three formats (*.html, *.md, *.pdf) and one *.png visualization map.The Next-Generation Ecosystem Experiments: Arctic (NGEE Arctic), was a research effort to reduce uncertainty in Earth System Models by developing a predictive understanding of carbon-rich Arctic ecosystems and feedbacks to climate. NGEE Arctic was supported by the Department of Energy's Office of Biological and Environmental Research.The NGEE Arctic project had two field research sites: 1) located within the Arctic polygonal tundra coastal region on the Barrow Environmental Observatory (BEO) and the North Slope near Utqiagvik (Barrow), Alaska and 2) multiple areas on the discontinuous permafrost region of the Seward Peninsula north of Nome, Alaska.Through observations, experiments, and synthesis with existing datasets, NGEE Arctic provided an enhanced knowledge base for multi-scale modeling and contributed to improved process representation at global pan-Arctic scales within the Department of Energy's Earth system Model (the Energy Exascale Earth System Model, or E3SM), and specifically within the E3SM Land Model component (ELM).

54 ENVIRONMENTAL SCIENCES↗

Generating An Advanced Cross-section Library For HTGR Pebble Bed Depletion Calculations Using Reduced-Order Model Generation Techniques

For code development, Advanced Reactor Technologies - Gas Cooled Reactors Program (ART-GCR) rely on a collaboration with the Nuclear Energy Advanced Modeling and Simulation (NEAMS) program, but the cross sections generation and the methodology definition is part of this program area goals. Based on previous studies in FY23, the size of microscopic cross section libraries increases rapidly with the number of tabulations, requiring significant amount of memory and drastically slowing down the Griffin calculations when evaluating cross sections via the multivariate linear interpolation approach. Rising to these challenges, this work investigates constructing Reduced-order Models (ROMs) for the multi-group microscopic cross sections to accelerate the cross section evaluation in Griffin. A database of multigroup cross sections is first collected considering all possible parameters that a designer could change for optimization. Down-selection of the ROM techniques afterward shows Deep Neural Network (DNN) as the best candidate when jointly consider memory efficiency, predictive accuracy, computational cost, scalability, flexibility and ease of implementation of the algorithms in comparison to the multidimensional interpolation. This work develops a specific interface that enables the cross section predictions using pre-trained DNN models into Griffin leveraging the existing ROM capabilities. DNNs have been trained for all isotopes for use in Griffin. Preliminary Griffin testing shows that DNNs exhibit exceptional predictive accuracy and the use of DNNs provides orders of magnitude improvement in memory efficiency compared to conventional interpolation techniques. With such ROM techniques, it holds great promise to further increase the fidelity of the Pebble Bed Reactor (PBR) simulation by increasing the number of tabulations/state variables during cross section evaluation, while maintaining the computational cost affordable in Griffin.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Pulsed Thermal Tomography Nondestructive Examination of Additively Manufactured Reactor Materials and Components (Final Technical Report)

Metal Additive Manufacturing (AM) is a promising method for cost-efficient fabrication of complex shape structures for applications in harsh environment, such as in a nuclear reactor. However, internal defects (pores) occur in high-strength AM alloys, which are manufactured with Laser Powder Bed Fusion (LPBF) AM method. Pulsed Infrared Thermography (PIT) is an efficient nondestructive evaluation (NDE) method to examine actual structures, because this method offers one-sided non-contact measurements, and fast processing of large sample areas. However, imaging of material defects, particularly defects with sizes at microscopic level, is challenging. In this report, we benchmark the performance of several Unsupervised Learning (UL) algorithms designed to enhance imaging of microscopic defects in metals with PIT. UL aims to learn the latent principal patterns (dictionaries) in PIT data to detect defects with minimal human supervision. Performance of Independent Component Analysis (ICA), Sparse Coding (SC), Principal Component Analysis (PCA) and Exploratory Factor Analysis (EFA) was compared using F-score, UL model training time and defects reconstruction time. We obtained the average F-score of 0.75, and a highest F-score of 0.89 for the EFA algorithm. Overall, EFA outperforms other UL algorithms considered in this study. In another approach, we investigate Thermal Tomography (TT), which is a computational method for reconstruction of depth profile of internal material defects from PIT nondestructive evaluation (NDE). TT algorithm obtains depth reconstructions of thermal effusivity, which has been shown to provide visualization of subsurface internals defects in metals. In many applications, one needs to determine the defect shape and orientation from reconstructed effusivity images. Interpretation of TT images is non-trivial because of blurring, which increases with depth due to heat diffusion-based nature of image formation. We have developed a deep learning convolutional neural network (CNN) to classify size and orientation of subsurface material defects in TT images. CNN was trained with TT images produced with computer simulations of 2D metallic structures (thin plates) containing elliptical subsurface voids. Performance of CNN was investigated using test TT images developed with computer simulations of plates containing elliptical defects, and defects with shape imported from scanning electron microscopy (SEM) images. CNN demonstrated the ability to classify radii and angular orientation of elliptical defects in previously unseen test TT images. We have also demonstrated that CNN trained on TT images of elliptical defects is capable of classifying shape and orientation of irregular defects. Training the CNN on irregular defect shapes instead of on elliptical shapes would make the resulting classifications more descriptive of actual defect shapes. However, this requires a much higher volume of SEM images of material defects, which are difficult to obtain because of random occurrence of defects in LPBF. To address this challenge, we developed a generative adversarial network (GAN) to augment the existing dataset of SEM defect images. The GAN model is demonstrated to create novel yet realistic defect shapes that can be used as input for simulated PTT images to train CNN. We also investigate several approaches based on Gaussian Random Circle and Bezier Curves for constructing parametric models of irregular-shape defects.

36 MATERIALS SCIENCE↗