SEARCH · Search NASA
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.
10 hours chronoamperometry (CA) testing under 1 sun illumination and constant bias at -0.6 V vs RHE, the corresponding Faradaic efficiency reveals a self-improving nature of GaN, inset: CA testing on bare Si for 5 hours under 1 sun illumination and -0.6 V vs RHE, this bare Si photocathode rapidly drops down to < 0.05 mA/cm2 within an hour. Both CA testing performed in 0.5 M H2SO4 (pH=0.4).
The ever-growing number of pre-trained large language models (LLMs) across scientific domains presents a challenge for application developers. While these models offer vast potential, fine-tuning them with custom data, aligning them for specific tasks, and evaluating their performance remain crucial steps for effective utilization. However, applying these techniques to models with tens of billions of parameters can take days or even weeks on modern workstations, making the cumulative cost of model comparison and evaluation a significant barrier to LLM-based application development. To address this challenge, we introduce an end-to-end pipeline specifically designed for building conversational and programmable AI agents on high performance computing (HPC) platforms. Our comprehensive pipeline encompasses: model pre-training, fine-tuning, web and API service deployment, along with crucial evaluations for lexical coherence, semantic accuracy, hallucination detection, and privacy considerations. Here, we demonstrate our pipeline through the development of chatHPC, a chatbot for HPC question answering and script generation. Leveraging our scalable pipeline, we achieve end-to-end LLM alignment in under an hour on the Frontier supercomputer. We propose a novel self-improved, self-instruction method for instruction set generation, investigate scaling and fine-tuning strategies, and conduct a systematic evaluation of model performance. The established practices within chatHPC will serve as a valuable guidance for future LLM-based application development on HPC platforms.
Existing machine learning potentials for predicting phonon properties of crystals are typically limited on a material-to-material basis, primarily due to the exponential scaling of model complexity with the number of atomic species. We address this bottleneck with the developed Elemental Spatial Density Neural Network Force Field, namely Elemental-SDNNFF. The effectiveness and precision of our Elemental-SDNNFF approach are demonstrated on 11,866 full, half, and quaternary Heusler structures spanning 55 elements in the periodic table by prediction of complete phonon properties. Self-improvement schemes including active learning and data augmentation techniques provide an abundant 9.4 million atomic data for training. Deep insight into predicted ultralow lattice thermal conductivity (<1 Wm –1 K –1 ) of 774 Heusler structures is gained by p–d orbital hybridization analysis. Additionally, a class of two-band charge-2 Weyl points, referred to as “double Weyl points”, are found in 68% and 87% of 1662 half and 1550 quaternary Heuslers, respectively.
Machine-learned interatomic potentials (ML-IAPs) continue to gain popularity as accurate, computationally efficient replacements for traditional, physics-based interatomic potentials and expensive ab initio methods. Uncertainty quantification (UQ) of ML-IAPs is a growing area of research as UQ is critical in many applications of IAPs, such as developing curated datasets, active learning-based data augmentation, self-improving models, and estimating the uncertainty of molecular dynamics simulations. In this paper, we construct and benchmark a series of different neural network potentials (NNPs) with varying network architectures to determine the performance of these models with respect to both the mean and uncertainty calibration error. Each NNP method is specifically designed to predict either epistemic or aleatoric uncertainty with particular focus on the differences in behavior between the epistemic and aleatoric uncertainty estimates. We benchmark these methods using multiple datasets common in the ML-IAP literature. The results show that the aleatoric uncertainty from single-shot model architectures is a competitive alternative to ensemble-based epistemic uncertainty predictions in regions of sufficient data-density. However, in regions where the representative data is sparse, aleatoric uncertainty models tend to overpredict and epistemic methods tend to underpredict the actual model error. We conclude that the type of UQ is crucial when discussing performance of probabilistic model results as different methods have different performance characteristics depending on the regime in which they are evaluated. Therefore, the type of UQ method should be carefully evaluated against both the data characteristics and requirements for the intended application.
Aircraft icing refers to the ice buildup on the surface of an aircraft flying in icing conditions. The ice accretion on the aircraft alters the original aerodynamic configuration and degrades the aerodynamic performances and may lead to unsafe flight conditions. Evaluating the flow structure, icing mechanism and consequences is of great importance to the development of an anti/deicing technique. Studies have shown computational fluid dynamics (CFD) and machine learning (ML) to be effective in predicting the ice shape and icing severity under different flight conditions. CFD solves a set of partial differential equations to obtain the air flow fields, water droplets trajectories and ice shape. ML is a branch of artificial intelligence and, based on the data, the self-improved computer algorithms can be effective in finding the nonlinear mapping relationship between the input flight conditions and the output aircraft icing severity features.
This viewgraph presentation provides information on research undertaken by NASA to facilitate the development of information technologies. Specific ideas covered here include: 1) Bio/nano technologies: biomolecular and nanoscale systems and tools for assembly and computing; 2) Evolvable hardware: autonomous self-improving, self-repairing hardware and software for survivable space systems in extreme environments; 3) High Confidence Software Technologies: formal methods, high-assurance software design, and program synthesis; 4) Intelligent Controls and Diagnostics: Next generation machine learning, adaptive control, and health management technologies; 5) Revolutionary computing: New computational models to increase capability and robustness to enable future NASA space missions.
Technologies that facilitate the design and control of complex, hybrid, and resource-constrained systems are examined. This paper focuses on design methodologies, and system architectures, not on specific control methods that may be applied to life support subsystems. Honeywell and Boeing have estimated that 60-80Y0 of the effort in developing complex control systems is software development, and only 20-40% is control system development. It has also been shown that large software projects have failure rates of as high as 50-65%. Concepts discussed include the Unified Modeling Language (UML) and design patterns with the goal of creating a self-improving, self-documenting system design process. Successful architectures for control must not only facilitate hardware to software integration, but must also reconcile continuously changing software with much less frequently changing hardware. These architectures rely on software modules or components to facilitate change. Architecting such systems for change leverages the interfaces between these modules or components.
This is an abstract for an invited paper at the AIAA Aviation Conference, June 2021. The work described here is part of an effort of coordinated experiments in the Unitary Plan Wind Tunnel (UPWT) facility at the NASA Langley Research Center (LaRC) and matching CFD simulations. The primary goal of the work is to assess the productivity and true predictive accuracy of CFD, absent any guidance from experiment, in the high supersonic speed range as compared to experiments performed at the NASA LaRC’s UPWT facility. This report concerns CFD simulation of the primary flow-path in the empty wind tunnel, including the settling chamber, nozzle, test section, and some of the tunnel downstream of the test section. The Mach number in the test section ranges from M~2.4 to M~4.6, and the required area ratio variation is achieved by translation of a nozzle block which constricts the area of a loosely S-shaped throat. Flow past protuberances in the settling chamber and into this S-bend throat are predicted by CFD to generate streamwise vorticity that may, or may not, persist through the throat and into the test section as coherent vortices. Some flow conditions are notably unsteady at frequencies well below those of turbulence, due to unsteady separated flow ahead of the nozzle block. The bulk flow moves at velocities ranging from 'walking speed' in the settling chamber to M~4.6 in the test section. Heat transfer to the settling chamber walls and buoyancy are significant at high Mach number. Subtle variations in surface curvature in the nozzle generate Mach waves that propagate into the test section. CFD of the empty tunnel serves two purposes. Firstly, the full-tunnel simulations are used to provide upstream boundary conditions for CFD of vehicle aerodynamics which are generally performed in a domain confined to the wind tunnel test section; these companion studies are addressed in other papers. Secondly, it is a challenging test for CFD to resolve all of the empty tunnel flow phenomena relevant to flow in the test section. It requires a more complete definition of geometry than was originally anticipated. In addition, it requires good spatial and temporal accuracy, and turbulence modeling that performs well on specific phenomena such as corner flows. The boundary conditions and solution algorithms must perform well from incompressible to almost hypersonic speeds. The final state of the pre-test CFD was a product of an iterative self-improvement process. The initial simulations of the empty tunnel were deficient in many respects, but hints to those deficiencies were recognized in the solutions, and remedies were implemented. Possible further improvements will be studied in the post-test phase when comparisons with experimental data are possible. The CFD was performed by five separate collaborative teams using four different flow solvers: FUN3D, Overflow, Star-CCM+ and USM3D. The level of effort of these teams varied significantly, but each made important contributions to the goals of the work. A concerted effort to use uncertainty quantification methods (UQ) in CFD is also a goal of this work. To this end, variations in CFD results due to grid refinement, turbulence modeling, and boundary conditions have been characterized. Code-to-code variation is another means of assessing CFD uncertainty. All CFD solvers predict similarity among the primary flow features; these include the variation of Mach number due to changes in Reynolds number, and the bulk flow angularity due to tunnel-wall curvature. All CFD solvers also predict similar trends in secondary flows, such as the downwash in the side-wall boundary layers. Three of the high-spatial resolution simulations give similar predictions of a complex secondary flow phenomena, streamwise vortices generated in the S-bend throat that persist into the side-wall boundary layers of the test section. Two of the highest-resolution simulations, run with the same turbulence model in different CFD solvers, gave encouragingly similar predictions of a complex tertiary flow phenomenon, small transient "sprites" of upwash flows, resulting from vortices that presumably originate in the separated flow near the leading edge of the nozzle block. While the CFD was done in a "blind pre-test" mode, requests from the experimental team for CFD results pertaining to unsteadiness and total temperature variations in the test section resulted in CFD runs that included a cooled wall in settling chamber. This then led to a change in the standard practice for running the Overflow results, which would not have occurred without this "release" of this experimental information. CFD was also used to guide some measurements. The paper will focus on establishing the consensus among CFD results and understanding differences among those results. Initial findings from the efforts to characterize CFD uncertainty have been done and will be included in the paper.
As particle accelerators grow in complexity, traditional control methods face increasing challenges in achieving optimal performance. This paper envisions a paradigm shift: a decentralized multi-agent framework for accelerator control, powered by Large Language Models (LLMs) and distributed among autonomous agents. We present a proposition of a self-improving decentralized system where intelligent agents handle high-level tasks and communication and each agent is specialized control individual accelerator components. This approach raises some questions: What are the future applications of AI in particle accelerators? How can we implement an autonomous complex system such as a particle accelerator where agents gradually improve through experience and human feedback? What are the implications of integrating a human-in-the-loop component for labeling operational data and providing expert guidance? We show two examples, where we demonstrate viability of such architecture.
Accelerated CA testing on bare epilayer Si/GaN photocathode at constant bias of -0.6 V vs RHE under 3.5 sun illumination for 150 hours in 0.5 M H2SO4 (pH=0.4)
Intermittent chronoamperometry (CA) testing was performed on Si/GaN photocathode in hourly manner. The LSV scan at 0 hour, 1 hour CA, 2 hours CA, 3 hours CA, 4 hours CA, 6 hours CA, 8 hours CA and 10 hours CA were recorded to track the changes of the photoelectrochemica (PEC) performance of the photocathode. The scan range was set to be 0V vs open circuit voltage (Eoc) to -0.7V vs RHE. Test was performed under one sun illumination in 0.5M H2SO4 (pH=0.4).
Fig1d results were obtained using a 3-electrode configuration with IrOx as the counter electrode and saturated calomel electrode as the reference electrode. Two Si/GaN photocathodes were compared. Sample #1 (named as S#1) was the as-received sample, and sample #2 (named as S#2) was CA tested for 4 hours under same testing conditions described in Fig.1a. The PEC performance (J-V curve) was recorded for both samples. Then a photodeposition of Pt was performed on both samples, each loading time was set to be 5 minutes, then the J-V curve was recorded to track the changes of the performance. The Pt photodeposition was stopped until the J-V curve won't shift (the maximum onset potential was achieved). The results show that, S#1 needs at least 20 minutes of Pt photodeposition to achieve its best PEC performance while S#2, which is after CA testing, only needs 5 minutes of Pt photodeposition to achieve its best PEC performance.
Photoconductive atomic force microscopy (PC-AFM) measurements were conducted on Si/GaN samples. PeakForce TUNA mode was used to acquire the morphology and current simultaneously. A PtIr conductive probe with spring constant of 2.8 Nm-1 was used for the scanning. A white light source was used to illuminate the surface during the acquisition. Line scan extracting the topography and photocurrent from a as-received Si/GaN sample. The topography mapping is 2 × 2 µm2. The sample bias was 0.4V and this photoconductive AFM measurement was performed under white light front illumination with an angle.
After 10 hrs of chronoamperometry (CA) photoconductive atomic force microscopy (PC-AFM) measurements were conducted on Si/GaN samples using a commercial AFM system (Bruker Dimension Icon). PeakForce TUNA mode was used to acquire the morphology and current simultaneously. A PtIr conductive probe with spring constant of 2.8 Nm-1 was used for the scanning. A white light source was used to illuminate the surface during the acquisition. Line scan extracting the topography and photocurrent from this 10 hour CA tested Si/GaN sample. The topography mapping is 2 × 2 µm2. No sample bias was applied and this photoconductive AFM measurement was performed under white light front illumination with an angle.
XPS after 0hr of chronoampometry. Surface chemical composition and valence band structure of GaN were obtained by X-ray photoemission spectroscopy (XPS) on a Kratos Axis Ultra DLD system at a takeoff angle of 0° relative to the surface normal. An Al Kα source (hν = 1486.6 eV) was used to excite the core level electrons. Pass energy of 20 eV was used for the narrow scan of core levels and valence band spectra, and step size of 0.05 eV and 0.025 eV, respectively. The Spectral fitting was conducted using CasaXPS analysis software. The binding energy scales of all core levels were corrected to the N 1s of Ga – N bond at 397.8 eV. XPS O1s core level spectra from as-received Si/GaN sample, and deconvolution shows O - Ga bond and OH - H2O bond.
XPS of Si/GaN photocathode after 1 hour chronoamperometry (CA) testing. Surface chemical composition and valence band structure of GaN were obtained by X-ray photoemission spectroscopy (XPS) on a Kratos Axis Ultra DLD system at a takeoff angle of 0° relative to the surface normal. An Al Kα source (hν = 1486.6 eV) was used to excite the core level electrons. Pass energy of 20 eV was used for the narrow scan of core levels and valence band spectra, and step size of 0.05 eV and 0.025 eV, respectively. The Spectral fitting was conducted using CasaXPS analysis software. The binding energy scales of all core levels were corrected to the N 1s of Ga – N bond at 397.8 eV. X-ray photoemission spectroscopy (XPS) O1s core level spectra from 1 hour chronoamperometry (CA) tested Si/GaN sample, and deconvolution shows O - Ga bond, O - N - Ga bond, and OH - H2O bond.
XPS of Si/GaN photocathode after 10 hour chronoamperometry (CA) testing. Surface chemical composition and valence band structure of GaN were obtained by X-ray photoemission spectroscopy (XPS) on a Kratos Axis Ultra DLD system at a takeoff angle of 0° relative to the surface normal. An Al Kα source (hν = 1486.6 eV) was used to excite the core level electrons. Pass energy of 20 eV was used for the narrow scan of core levels and valence band spectra, and step size of 0.05 eV and 0.025 eV, respectively. The Spectral fitting was conducted using CasaXPS analysis software. The binding energy scales of all core levels were corrected to the N 1s of Ga – N bond at 397.8 eV. XPS O1s core level spectra from 10 hours CA tested Si/GaN sample, and deconvolution shows O - Ga bond, O - N - Ga bond and OH - H2O bond.
XPS of Si/GaN photocathode after 4 hour chronoamperometry (CA) testing. Surface chemical composition and valence band structure of GaN were obtained by X-ray photoemission spectroscopy (XPS) on a Kratos Axis Ultra DLD system at a takeoff angle of 0° relative to the surface normal. An Al Kα source (hν = 1486.6 eV) was used to excite the core level electrons. Pass energy of 20 eV was used for the narrow scan of core levels and valence band spectra, and step size of 0.05 eV and 0.025 eV, respectively. The Spectral fitting was conducted using CasaXPS analysis software. The binding energy scales of all core levels were corrected to the N 1s of Ga – N bond at 397.8 eV. XPS O1s core level spectra from 4 hour CA tested Si/GaN sample, and deconvolution shows O - Ga bond, O - N - Ga bond and OH - H2O bond.