Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “limiting techniques”

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 1,657 records · Page 92

The First Focused Hard X-Ray Images of the Sun with NuSTAR

We present results from the first campaign of dedicated solar observations undertaken by the Nuclear Spectroscopic Telescope ARray (NuSTAR) hard X-ray (HXR) telescope. Designed as an astrophysics mission, NuSTAR nonetheless has the capability of directly imaging the Sun at HXR energies (3 keV) with an increase in sensitivity of at least two magnitude compared to current non-focusing telescopes. In this paper we describe the scientific areas where NuSTAR will make major improvements on existing solar measurements. We report on the techniques used to observe the Sun with NuSTAR, their limitations and complications, and the procedures developed to optimize solar data quality derived from our experience with the initial solar observations. These first observations are briefly described, including the measurement of the Fe K-shell lines in a decaying X-class flare, HXR emission from high in the solar corona, and full-disk HXR images of the Sun.

Grefenstette, Brian W.↗

Real-time Exponential Curve Fits Using Discrete Calculus

This paper presents an improved solution for curve fitting data to an exponential equation (Y = AeBt + C). This improvement is in four areas ? speed, stability, determinant processing time, and the removal of limits. The solution presented in this paper avoids iterative techniques and their stability errors by using three mathematical ideas ? discrete calculus, a special relationship (between exponential curves and the Mean Value Theorem for Derivatives), and a simple linear curve fit algorithm. This method can also be applied to fitting data to the general power law equation Y = AxB + C and the general geometric growth equation Y = AkBt + C.

Rowe, Geoffrey K.↗

Transforming Science Prioritization Processes Using Artificial Intelligence

Artificial Intelligence (AI) and Machine Learning (ML) have potential to augment significantly the current labor-intensive processes of science prioritization, specifically by the National Academies’ Decadal Survey on behalf of NASA and NSF. Here we summarize what we believe to be the first exploratory demonstration-of-concept results from an application of AI/ML to Survey science prioritization. Specifically, we applied Latent Dirichlet Allocation (LDA) and Natural Language Processing (NLP) to reveal trends in published astrophysics research that may indicate science priorities and which could be applied to strategic planning. For the purpose of the work that we summarize here, AI/ML is able to analyze – that is, to “understand,” in a manner of speaking – a vast amount of text to reveal complex relationships among research topics, including the growth or decline of science community activities in those topics over time. We trained ourselves and AI/ML algorithms by using ~400,000 abstracts in the period 1998 to 2010 to “forecast” the Academies’ Astro2010 recommendations and compare with the solicited white papers. Comparing our results with actual Astro2010 recommendations allowed us to identify candidate metrics that better predicted the actual results of the Survey. We found, for example, that Compound Annual Growth Rate (CAGR) of papers published in a topic area is a good proxy measure for importance of this topic area of research. With this training complete, we identified candidate astrophysics astrophysics science priorities for the 2021+ period using the research during 2007 - 2019 . We conclude that appropriate application of AI can potentially significantly reduce the current workload of the Decadal Survey processes and reveal otherwise unrecognized characteristics in the body of astronomical research. We emphasize throughout the exploratory nature of our work, encouraging colleagues to pursue promising results further. Our most critical governing assumption was that increased (or decreased) research activity can be used to identify scientific or technology topic areas worthy of increased (or decreased) future emphasis. We discuss advantages, limitations, and recognize the “black box” nature of our technique. We note ethics issues associated, for example, with using AI/ML to reveal “hidden” meanings and biases in published work. Furthermore, inevitable improvements in AI may soon enable widespread and welcome identification of and advocacy for science and technology priorities by disparate and diverse groups and organizations. Consequently, we continue to urge a near-term, in-depth evaluation of appropriate applications of AI, including implications and consequences, as well as support for multiple follow-on assessments, of which ours is only a beginning.

Artificial Intelligence↗

Enabling Single Cell Research in Hollow Hydrogel Microparticles

Studying biology at the single cell level enables novel discoveries through ultra-high throughput identification and isolation of rare cells and interesting subpopulations. Encapsulation of single cells in hollow hydrogel microparticles, or PicoShells, enables easy visualization, manipulation, and assaying of single cells. The PicoShell’s porous hydrogel shell allows diffusion of nutrients and wastes into and out of the hollow liquid core, allowing uninhibited cell growth. The porous shell also allows diffusion of assay reagents, enabling solution changes, and multi-step assays. Finally, single cells and their progeny can be tracked through time, for example to measure single cell growth rates. However, PicoShells are fabricated using microfluidic techniques, which require specialized expertise, equipment, and facilities. This limits PicoShell’s widespread utility for single cell biology. For Saccharomyces cerevisiae and other desiccation-tolerant microbes, desiccating PicoShells containing cells of interest offers a solution, enabling those without microfluidic capabilities to obtain dried, shelf-stable PicoShells containing their cells of interest from a specialized microfluidics facility. Researchers could then rehydrate the PicoShells and perform their single cell assays, bypassing the microfluidic steps. Desiccated PicoShells could also be flown in space, where desiccation is already a common practice for microbes (i.e. BioSentinel and BioNutrients), bringing single cell analysis capabilities to space. In this work, we develop a method to desiccate PicoShells containing S. cerevisiae while maintaining characteristics similar to fresh, non-desiccated PicoShells. We show good microparticle morphology and hydrogel pore size as well as high yeast viability in line with previous studies. We also demonstrate sorting of a fast-growing yeast population using size-based filtration, as S. cerevisiae in PicoShells are able to physically stretch the microparticle as they grow, increasing the PicoShell diameter. These efforts contribute to the feasibility of leveraging microparticle PicoShells for single cell space biology research on Earth and in space.

Simon Ng↗

Development of New Lunar Highland Regolith Simulant, NUW-LHT-5M

Introduction: NASA has a need for large quantities of lunar simulants that closely match future manned lunar missions at landing sites near the Lunar South Pole. The current simulant was designed to approximate NU-LHT-2M, and -4M, except using natural minerals and a fully synthetic, non-basaltic, high calcium glass. Washington Mills in Niagara Falls, NY was contracted to produce this new simulant due to their electric arc furnace (EAF) fusion technology and capabilities for crushing and sizing ceramics. The primary objectives of this work were: (1) to create a commercially produced simulant that contains a high fidelity, high calcium (An90+), low Mg/Fe glass without basaltic constituents; (2) to create a simulant that uses existing, terrestrially mined, and readily available minerals blended with a synthetic glass, and (3) accomplish objectives (1) and (2) using conventional processing methods capable of production levels that can meet substantial current and future demands of NASA projects. Another consideration was that highland simulants closest to meeting these requirements, NU-LHT-2M and -4M types, are out of production and largely out of stock. Design: The design chosen for development and production for this work is NU-LHT-2M, de-signed by Doug Stoeser and Douglas Rickman, which uses Stillwater anorthosite and norite rocks and synthetic glass. The composition targeted the average composition, glass content and particle size distribution of Apollo 16 samples [1]. The design utilizes two Stillwater minerals, anorthosite (37.7 wt.%), norite (17.6 wt.%), a commercially sourced olivine (4.7 wt.%), and a synthetic glass (40 wt.%). Processing: Stillwater rocks were hand collected and initially crushed by the USGS. Further crushing and milling was performed at Washing-ton Mills. For the glass, oxides were blended in the ap-propriate ratios, blended in a v-blender, and placed in a graphite lined, water cooled pot. The mixture was fused in Washington Mills’ pilot scale, 500 kW EAF. The molten glass was then poured into water and quenched to inhibit crystallization. Several pour/quench cycles were required to produce the total quantity of glass needed. Anorthosite, norite and olivine were milled together to the target particle size in a ball mill. The glass was ball milled separately. The complete simulant was produced by blending the milled materials. Particle size distribution, phase and chemical analyses of the materials was were per-formed at Washington Mills. Results: The glass composition was found to match the Apollo 16 oxide content average [1] quite well, with the exception of SiO2 and Fe2O3. The deviation of SiO2 and Fe2O3 from batched content was believed to be due to limitations inherent in the use of the arc furnace technique for these types of materials. A particle size distribution of the full simulant formulation closely matching the average for the Apollo 16 samples was also achieved. Fig. 1 shows the morphology of the final simulant particles. The size distribution is compared to the target in Fig. 2. Conclusions: A new lunar highlands type simulant containing a high-quality synthetic glass, closely matching an average composition of the Apollo 16 regolith was produced. Processing utilized easily scalable mineral and glass making methods.

simulant↗

Vapor Growth of ZnSe-based Compound Semiconductors in Low Gravity environment on International Space Station

Six low-gravity experiments of crystal growth of ZnSe-based compound semiconductors by physical vapor transport (PVT) are scheduled to be processed inside Low Gradient Furnace (LGF) on Materials Science Research Rack (MSRR) in Material Science Laboratory (MSL) on International Space Station (ISS) starting from February 27th, 2023. The flight samples consist of three unseeded growths of ZnSe, Cr-doped ZnSe and ZnSeTe as well as growth of ZnSe on top of three single crystal ZnSe seeds with orientation of (100), (110) and (111). The main scientific objective of the flight experiments is to establish the effects of gravity-driven fluid flows on Earth to (i) the non-uniform incorporation of impurities and defects and (ii) the deviation from stoichiometry observed in the grown crystals caused by growth interface fluctuations and evolution as the results of buoyancy-driven convection and irregular fluid-flows. The investigation will also evaluate the effects of gravity on the PVT process by examining (i) the growth kinetics on various seed orientations, (ii) the dopant segregation and distribution in the Cr doped ZnSe, and (iii) the compositional segregation and distribution in the ternary ZnSeTe crystal grown by PVT. Additionally, from X-ray diffraction/topography and by measuring the density and distribution of various crystalline defects it will assess the self-induced strain effects developed during processing at elevated temperatures caused by the weight of the crystals. During the ground-based preparations, the following tasks have been accomplished: • Establish the thermodynamic properties of the ZnSe binary and related ternary systems by measuring the partial pressures of the vapor phase coexisting with the condensed phases and perform phase diagram calculations for comparison. • Evaluate the effects of gravity-driven convection in the growth process by performing experiments under various vapor transport orientations relative to gravity vector, i.e., horizontal, vertically stabilized and destabilized configurations. • Establish a quantitative correlation between growth parameters and the characteristics and properties of the grown crystals by characterizing the crystals using a variety of techniques. • Evaluate the fundamentals of the current theories on vapor transport and crystal growth kinetics by performing in-situ and real-time monitoring techniques during growth. • Develop a one-dimensional diffusion limited theory of mass transport and two- and three-dimensional numerical simulations of both mass transport and heat transfer and thus establish the microgravity requirements for the flight experiments and to predict the characteristics of the grown crystals such as the solid-vapor interface shapes… etc. and establish a fundamental understanding of the crystal growth process from the comparison with experimental results.

ZnSe↗

PREEMPT: Scalable Epidemic Interventions Using Submodular Optimization on Multi-GPU Systems

Preventing and slowing the spread of epidemics is achieved through techniques such as vaccination and social distancing. Given practical limitations on the number of vaccines and cost of administration, optimization becomes a necessity. Previous approaches using mathematical programming methods have shown to be effective but are limited by computational costs. In this work, we make several contributions: First, we present a new approach for intervention via maximizing the influence of vaccinated nodes on the network. We call this method \preempt. Next, we prove submodular properties associated with the objective function of our method so that it aids in construction of an efficient greedy approximation strategy. Consequently, we present a new parallel algorithm based on greedy hill climbing for \preempt, and present an efficient parallel implementation for distributed CPU-GPU heterogeneous platforms. Our results demonstrate that \preempt{} is able to achieve a significant reduction (up to 6.75$\times$) in the percentage of people infected on a city-scale network. We also show strong scaling results of \preempt{} on 128 nodes of the Summit supercomputer. Our parallel implementation is able to significantly reduce time to solution, from hours to minutes on large networks. This work represents a first-of-its-kind effort in parallelizing greedy hill climbing and applying it toward devising effective interventions for epidemics.

Minutoli, Marco↗

Constraints on the electron-hole pair creation energy and Fano factor below 150 eV from Compton scattering in a Skipper-CCD

Fully-depleted thick silicon Skipper-charge-coupled devices (Skipper-CCDs) have achieved sub-electron read-out noise and are an important technology to probe neutrino and light dark matter interactions. However, the successful search for rare neutrino or dark-matter events requires the signal and all backgrounds to be fully characterized. In particular, a measurement of the electron-hole pair creation energy below 150 eV and the Fano factor are necessary for characterizing the dark matter and neutrino signals. Moreover, photons from background radiation may Compton scatter in the silicon bulk, producing events that can mimic a dark matter or neutrino signal. We present a measurement of the Compton spectrum using a Skipper-CCD and a $^{241}$Am source. With these data, we measure the electron-hole pair-creation energy to be $\left(3.71 \pm 0.08\right)$ eV at 130 K in the energy range between 99.3 eV and 150 eV. By measuring the widths of the steps at 99.3 eV and 150 eV in the Compton spectrum, we introduce a novel technique to measure the Fano factor, setting an upper limit of 0.31 at 90% C.L. These results prove the potential of Skipper-CCDs to characterize the Compton spectrum and to measure precisely the Fano factor and electron-hole pair creation energy below 150 eV.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Nb3Sn Superconducting Radiofrequency Cavity Fabricated by Sputter Deposition from a Stoichiometric Target

Nb$_3$Sn is a promising coating material for superconducting radiofrequency cavities in next-generation accelerators due to its higher superconducting transition temperature (~18.3 K) and superheating field (~400 mT), offering the potential to surpass the intrinsic performance limits of bulk niobium. We developed a sputter coating technique employing a stoichiometric target to fabricate Nb$_3$Sn films on the inner surfaces of niobium cavities. A Nb$_3$Sn tube target, prepared by growing a >20 m thick Nb$_3$Sn coating on the surface of a Nb tube via a Sn vapor diffusion process, was used in a DC cylindrical magnetron sputtering system. Approximately 170 nm thick Nb-Sn films were sputter-coated onto flat Nb substrates positioned to replicate the beam tube and equator regions of a 2.6 GHz cavity, followed by annealing at 950 °C for 3 hours to form Nb$_3$Sn. The composition, morphology, and structure of the annealed Nb$_3$Sn films were examined. Then, a ~1.2 m-thick film was sputter-coated onto the interior of a 2.6 GHz Nb cavity following the procedure developed for the flat samples. The cavity was subsequently annealed under the same conditions used for the flat samples. Cryogenic RF testing of the sputter-coated cavity demonstrated a T$_c$ of approximately 17.8 K, consistent with Nb$_3$Sn layer. Subsequently, the cavity underwent a light Sn recoating followed by a vapor diffusion Nb$_3$Sn coating process, achieving a quality factor of approximately 8.5 × 10$^9$ and an accelerating gradient of up to 11.3 MV/m at 4.2 K.

Shakel, Md Sharifuzzaman [Old Dominion U.; Jeffers↗

Imaging Analysis of the Hard X-Ray Telescope ProtoEXIST2 and New Techniques for High-Resolution Coded-Aperture Telescopes

Wide-field (greater than or approximately equal to 100 degrees squared) hard X-ray coded-aperture telescopes with high angular resolution (greater than or approximately equal to 2 minutes) will enable a wide range of time domain astrophysics. For instance, transient sources such as gamma-ray bursts can be precisely localized without the assistance of secondary focusing X-ray telescopes to enable rapid followup studies. On the other hand, high angular resolution in coded-aperture imaging introduces a new challenge in handling the systematic uncertainty: the average photon count per pixel is often too small to establish a proper background pattern or model the systematic uncertainty in a timescale where the model remains invariant. We introduce two new techniques to improve detection sensitivity, which are designed for, but not limited to, a high-resolution coded-aperture system: a self-background modeling scheme which utilizes continuous scan or dithering operations, and a Poisson-statistics based probabilistic approach to evaluate the significance of source detection without subtraction in handling the background. We illustrate these new imaging analysis techniques in high resolution coded-aperture telescope using the data acquired by the wide-field hard X-ray telescope ProtoEXIST2 during a high-altitude balloon flight in fall 2012. We review the imaging sensitivity of ProtoEXIST2 during the flight, and demonstrate the performance of the new techniques using our balloon flight data in comparison with a simulated ideal Poisson background.

balloons - instrumentation: detectors - techniques↗

Optimal Control Allocation with Load Sensor Feedback for Active Load Suppression, Experiment Development

The problem of control command and maneuver induced structural loads is an important aspect of any control system design. The aircraft structure and the control architecture must be designed to achieve desired piloted control responses while limiting the imparted structural loads. The classical approach is to utilize high structural margins, restrict control surface commands to a limited set of analyzed combinations, and train pilots to follow procedural maneuvering limitations. With recent advances in structural sensing and the continued desire to improve safety and vehicle fuel efficiency, it is both possible and desirable to develop control architectures that enable lighter vehicle weights while maintaining and improving protection against structural damage. An optimal control technique has been explored and shown to achieve desirable vehicle control performance while limiting sensed structural loads. The subject of this paper is the design of the optimal control architecture, and provides the reader with some techniques for tailoring the architecture, along with detailed simulation results.

controls↗

Techniques for Measuring Low Earth Orbital Atomic Oxygen Erosion of Polymers

Polymers such as polyimide Kapton and Teflon FEP (fluorinated ethylene propylene) are commonly used spacecraft materials due to their desirable properties such as flexibility, low density, and in the case of FEP, a low solar absorptance and high thermal emittance. Polymers on the exterior of spacecraft in the low Earth orbit (LEO) environment are exposed to energetic atomic oxygen. Atomic oxygen reaction with polymers causes erosion, which is a threat to spacecraft durability. It is therefore important to understand the atomic oxygen erosion yield (E, the volume loss per incident oxygen atom) of polymers being considered in spacecraft design. The most common technique for determining E is through mass loss measurements. For limited duration exposure experiments, such as shuttle experiments, where the atomic oxygen fluence is often so low that mass loss measurements can not produce acceptable uncertainties, recession measurements based on atomic force microscopy analyses can be used. Equally necessary to knowing the mass loss or recession depth for determining the erosion yield of polymers is the knowledge of the atomic oxygen fluence that the polymers were exposed to in space. This paper discusses the procedures and relevant issues for mass loss and recession depth measurements for passive atomic oxygen erosion yield characterization of polymers, along with techniques for active atomic oxygen fluence and erosion characterization. One active atomic oxygen erosion technique discussed is a new technique based on optical measurements. Details including the use of both semi-transparent and opaque polymers for active erosion measurement are reviewed.

deGroh, Kim K.↗

An analysis of short pulse and dual frequency radar techniques for measuring ocean wave spectra from satellites

Scanning beam microwave radars were used to measure ocean wave directional spectra from satellites. In principle, surface wave spectral resolution in wave number can be obtained using either short pulse (SP) or dual frequency (DF) techniques; in either case, directional resolution obtains naturally as a consequence of a Bragg-like wave front matching. A four frequency moment characterization of backscatter from the near vertical using physical optics in the high frequency limit was applied to an analysis of the SP and DF measurement techniques. The intrinsic electromagnetic modulation spectrum was to the first order in wave steepness proportional to the large wave directional slope spectrum. Harmonic distortion was small and was a minimum near 10 deg incidence. NonGaussian wave statistics can have an effect comparable to that in the second order of scattering from a normally distributed sea surface. The SP technique is superior to the DF technique in terms of measurement signal to noise ratio and contrast ratio.

Jackson, F. C.↗

End-to-end protocol for high-quality quantum approximate optimization algorithm parameters with few shots

The quantum approximate optimization algorithm (QAOA) is a quantum heuristic for combinatorial optimization that has been demonstrated to scale better than state-of-the-art classical solvers for some problems. For a given problem instance, QAOA performance depends crucially on the choice of the parameters. While average-case optimal parameters are available in many cases, meaningful performance gains can be obtained by fine-tuning these parameters for a given instance. This task is especially challenging, however, when the number of circuit executions (shots) is limited. In this work, we develop an end-to-end protocol that combines multiple parameter settings and fine-tuning techniques. We use large-scale numerical experiments to optimize the protocol for the shot-limited setting and observe that optimizers with the simplest internal model (linear) perform best. We implement the optimized pipeline on a trapped-ion processor using up to 32 qubits and 5 QAOA layers, and we demonstrate that the pipeline is robust to small amounts of hardware noise. To the best of our knowledge, these are the largest demonstrations of QAOA parameter fine-tuning on a trapped-ion processor in terms of two-qubit gate count.

quantum algorithms & computation↗

A Review of Existing Test Methods for Occupancy Sensors

One of the key new features of connected lighting systems (CLS) is their ability to collect data from various types of integral sensors and share that data with other lighting or building systems. Occupancy and vacancy sensors have been widely adopted as an energy-saving strategy in buildings, yet published test methods for reproducibly characterizing their performance remain few and limited in their sophistication. As a result, it has been difficult to predict their performance in a specific application, and in practice, they frequently do not meet energy-savings expectations. Occupants at times remove or otherwise bypass occupancy sensors that hinder their work or otherwise do not perform as expected, thereby compromising the sensors’ potential to reduce energy consumption. Poor performance can result from multiple causes – ranging from fundamental limitations of the sensor technology, to misconfiguration, to poor placement in the room or space. Innovative occupancy sensors, some of them combining multiple sensing technologies, have come on the market over the years, with claims of improved performance compared to their predecessors; yet in practice, their performance has led to neither high deployment nor high user satisfaction in any significant application, nor has it differed enough from the performance of previous products to necessitate a test method that facilitated comparison between them. While the performance of both common and novel occupancy sensors has been the subject of many published research articles, the test methods that have been employed for them have typically been loosely described and incorporated custom equipment or techniques that render them difficult to reproduce, or have been limited in their ability to fairly characterize devices that utilize varying sensor technology. The lack of a well-described, technology-agnostic test method that yields reproducible results across different implementations has been a barrier to the commercial success of new occupancy sensor products, as users and specifiers who have been disappointed with previous products are often unwilling to take a chance with new ones. Motivated by a desire to fairly characterize new technologies that continue to enter the market and claim not only improved occupancy detection but, in some cases, also additional capabilities (e.g., the ability to measure traffic or discern between different object types), this report presents the results of a literature review of recent published, well-described test methods for characterizing occupancy sensor performance, as well as research articles containing ad-hoc test methods. The review also identifies and consolidates the best techniques for characterizing sensor performance in indoor spaces and identifies test method gaps that need to be filled in order to evaluate emerging technologies and products.

47 OTHER INSTRUMENTATION↗

Advanced cooling techniques for high-pressure, hydrocarbon-fueled rocket engines

The regenerative cooling limits (maximum chamber pressure) are defined for oxygen/hydrocarbon (Methane, Propane, and RP-1) rocket engines over a thrust range of 20,000 to 600,000 lbf for a reusable life of 250 missions. Chamber pressure limits are first defined without a hot-gas wall carbon layer (unenhanced designs). Cooling enhancement chamber pressure limits are then established for seven thermal barriers (carbon layer, ceramic coating, graphite liner, film cooling, zoned combustion, transpiration cooling, and a combination of two of the above). The maximum regenerative-cooled chamber pressure is attained with the oxygen/methane propellant combination.

Cook, R. T.↗

A Full-Envelope Air Data Calibration and Three-Dimensional Wind Estimation Method Using Global Output-Error Optimization and Flight-Test Techniques

A novel, efficient air data calibration method is proposed for aircraft with limited envelopes. This method uses output-error optimization on three-dimensional inertial velocities to estimate calibration and wind parameters. Calibration parameters are based on assumed calibration models for static pressure, angle of attack, and flank angle. Estimated wind parameters are the north, east, and down components. The only assumptions needed for this method are that the inertial velocities and Euler angles are accurate, the calibration models are correct, and that the steady-state component of wind is constant throughout the maneuver. A two-minute maneuver was designed to excite the aircraft over the range of air data calibration parameters and de-correlate the angle-of-attack bias from the vertical component of wind. Simulation of the X-48B (The Boeing Company, Chicago, Illinois) aircraft was used to validate the method, ultimately using data derived from wind-tunnel testing to simulate the un-calibrated air data measurements. Results from the simulation were accurate and robust to turbulence levels comparable to those observed in flight. Future experiments are planned to evaluate the proposed air data calibration in a flight environment.

Taylor, Brian R.↗

Assessment of fine-tuned large language models for real-world chemistry and material science applications

The current generation of large language models (LLMs) has limited chemical knowledge. Recently, it has been shown that these LLMs can learn and predict chemical properties through fine-tuning. Using natural language to train machine learning models opens doors to a wider chemical audience, as field-specific featurization techniques can be omitted. In this work, we explore the potential and limitations of this approach. We studied the performance of fine-tuning three open-source LLMs (GPT-J-6B, Llama-3.1-8B, and Mistral-7B) for a range of different chemical questions. We benchmark their performances against “traditional” machine learning models and find that, in most cases, the fine-tuning approach is superior for a simple classification problem. Depending on the size of the dataset and the type of questions, we also successfully address more sophisticated problems. The most important conclusions of this work are that, for all datasets considered, their conversion into an LLM fine-tuning training set is straightforward and that fine-tuning with even relatively small datasets leads to predictive models. These results suggest that the systematic use of LLMs to guide experiments and simulations will be a powerful technique in any research study, significantly reducing unnecessary experiments or computations.

Van Herck, Joren↗