Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “optimal input design”

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 199 records · Page 11
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Adrastea: An Efficient FPGA Design Environment for Heterogeneous Scientific Computing and Machine Learning

We present Adrastea, an efficient FPGA design environment for developing scientific machine learning applications. FPGA development is challenging, from deployment, proper toolchain setup, programming methods, interfacing FPGA kernels, and more importantly, the need to explore design space choices to get the best performance and area usage from the FPGA kernel design. Adrastea provides an automated and scalable design flow to parameterize, implement, and optimize complex FPGA kernels and associated interfaces. We show how virtualization of the development environment via virtual machines is leveraged to simplify the setup of the FPGA toolchain while deploying the FPGA boards and while scaling up the automated design space exploration to leverage multiple machines concurrently. Adrastea provides an automated build and test environment of FPGA kernels. By exposing design space hyper-parameters, Adrastea can automatically search the design space in parallel to optimize the FPGA design for a given metric, usually performance or area. Adrastea simplifies the task of interfacing with the FPGA kernels with a simplified interface API. To demonstrate the capabilities of Adrastea, we implement a complex random forest machine learning kernel with 10,000 input features while achieving extremely low computing latency without loss of prediction accuracy, which is required by a scientific edge application at SNS. We also demonstrate Adrastea using an FFT kernel and show that for both applications Adrastea is able to systematically and efficiently evaluate different design options, which reduced the time and effort required to develop the kernel from months of manual work to days of automatic builds.

Young, Aaron↗
DOE OSTI2022

Application-Driven Creation of Building Metadata Models with Semantic Sufficiency

Semantic metadata models such as Brick, RealEstateCore, Project Haystack, and BOT promise to simplify and lower the cost of developing software for smart buildings, enabling the widespread deployment of energy efficiency applications. However, creating these models remains a challenge. Despite recent advances in creating models from existing digital representations like point labels and architectural models, there is still no feedback mechanism to ensure that the human input to these methods results in a model that can actually support the desired software. In this paper, we introduce the notion of semantic sufficiency, a practical principle for semantic metadata model creation that asserts that a model is "finished" when it contains the metadata necessary to support a given set of applications. To support semantic sufficiency, we design a standard representation for capturing application metadata requirements and a templating system for generating common metadata model components with limited user input. We then construct an iterative model creation workflow that integrates metadata requirements to direct the model creation effort, and present several novel optimizations that increase the model utility while minimizing the effort by a human operator. These new abstractions for model creation and validation lower model development costs and ensure the utility of the resulting model, thus facilitating the adoption of intelligent building applications.

applications↗
DOE OSTI2023

An automated and portable method for selecting an optimal GPU frequency

Power consumption poses a significant challenge in current and emerging graphics processing unit (GPU) enabled high-performance computing systems. In modern GPUs, dynamic voltage frequency scaling (DVFS) appears to be a reliable control to regulate power consumption and performance. However, the DVFS design space is large - hence, brute-force approaches are infeasible to select the optimal frequency. Furthermore, no single frequency can be universally optimal for applications with varying computational intensities. Thus, the application's complexity and the availability of a wide range of frequency settings are a challenge in selecting the optimal frequency configuration for a given GPU workload. To that end, this paper proposes a systematic approach that consists of three steps. The feature characterization study identifies the fine-grain GPU utilization metrics that influence the power consumption and execution time of a given workload. To understand the performance, power, and energy consumption behaviors of a workload across GPU's DVFS design space, we derived analytical power and performance models using the identified fine-grain features. Here, it is shown that the same set of GPU utilization metrics can estimate both the power consumption and execution time while being agnostic of changes to frequency and input sizes. Applying a power control with the single objective of reducing power may cause performance degradation, leading to more energy consumption. A multi-objective approach is proposed to select the optimal GPU DVFS configuration for a workload that reduces power consumption with negligible degradation in performance. The evaluation was conducted using SPEC ACCEL benchmarks and three real applications - NAMD LAMMPS, and LSTM on NVIDIA GV100, GA100, and AMD MI210 GPUs. On average, real applications showed 29.6% energy savings with a performance loss of 5.2% on GA100 and 22.6% energy savings with a performance loss of 4.7% on GV100. Moreover, the proposed models are portable to real applications, GPU architectures, and vendors, and require metric collection at only the default frequency rather than all supported DVFS configurations. Additionally, we conducted a comparison between our models and the GPU assembly instructions (PTX)-based static models. The results revealed a significant reduction in the average error rates, with a decrease from 19.7% to 3.1% for power models and from 29.4% to 5.2% for performance models.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Sensitivity Analysis and Effective Parametrization of PEM Fuel Cell Models

The cost of proton-exchange-membrane fuel cells (PEMFCs) remains a major hurdle in large-scale commercialization of this technology. To improve their performance and reduce cost, novel materials and electrode designs are continuously envisioned, e.g., non-PGM catalyst layers, ultra-thin Pt/Pt-Ni based catalyst layers, structured ionomer arrays or NSTF catalyst layers.1 Understanding the impact of these improvement strategies can be extremely time and cost intensive due to complex physical phenomena and large design space. We have previously developed a PEMFC modeling framework2 which has been a time and cost effective tool for understanding and optimizing the complex multi-physics phenomena within PEMFCs; however, several of the cell parameters used in the modeling have large spread in measured data.3 Furthermore, several transport parameters such as water adsorption kinetics have not been accurately measured and the approximations are spread over several orders of magnitude. These uncertainties cause problems in ascertaining accuracy of the modeling approach and they reduce the predictive power of the numerical models. The aim of this work is to identify the sensitivity of PEFC numerical model outputs to various input parameters. The previously in-house developed MEA modeling framework2 is used for PEMFC modeling. The sensitivity of the model outputs with respect to inputs parameters is obtained by analyzing the condition numbers for different output-input pairs at varying operating conditions. An example of the sensitivity analysis is shown in Figure 1. The condition numbers are obtained for the entire possible range of input parameters at varying operating conditions to identify the most crucial parameters of the PEMFC model. Based on our preliminary analysis, parameters related to kinetics (exchange current density and ECSA) and heat/water management in electrodes and ionomer (ionomer fraction, thermal conductivity) are most crucial. One of the major goals of this work is to identify the most crucial set of parameters towards which the model shows maximum sensitivity. This will guide future experimentalists to measure these properties with higher accuracy. Furthermore, the sensitivity analysis will also enable us to optimize the PEMFC performance by selectively targeting the most sensitive parameters and thereby making the largest impact. Acknowledgements The work is funded under the Fuel Cell Performance and Durability Consortium (FC-PAD), by the Fuel Cell Technologies Office (FCTO), Office of Energy Efficiency and Renewable Energy (EERE), of the U.S. Department of Energy under contract number DE-AC02-05CH11231. The authors would like to thank Nathan Craig at Robert Bosch LLC for his valuable input in designing the sensitivity analysis. The authors would also like to thank Giovanna Bucci and Matthias Hanauer at Robert Bosch for their valuable inputs and discussion. References P. K. Sinha, W. Gu, A. Kongkanand and E. Thompson, J. Electrochem. Soc., 158, B831 (2011). L. M. Pant, M. R. Gerhardt, N. Macauley, R. Mukundan, R. L. Borup and A. Z. Weber, Electrochim. Acta, 326, 134963 (2019). R. Vetter and J. O. Schumacher, ArXiv181110091 Phys. (2018). Figure 1

Pant, Lalit↗
DOE OSTI2021

Train Like a (Var)Pro: Efficient Training of Neural Networks with Variable Projection

Deep neural networks (DNNs) have achieved state-of-the-art performance across a variety of traditional machine learning tasks, e.g., speech recognition, image classification, and segmentation. The ability of DNNs to efficiently approximate high-dimensional functions has also motivated their use in scientific applications, e.g., to solve partial differential equations and to generate surrogate models. In this paper, we consider the supervised training of DNNs, which arises in many of the above applications. We focus on the central problem of optimizing the weights of the given DNN such that it accurately approximates the relation between observed input and target data. Devising effective solvers for this optimization problem is notoriously challenging due to the large number of weights, nonconvexity, data sparsity, and nontrivial choice of hyperparameters. To solve the optimization problem more efficiently, we propose the use of variable projection (VarPro), a method originally designed for separable nonlinear least-squares problems. Our main contribution is the Gauss--Newton VarPro method (GNvpro) that extends the reach of the VarPro idea to nonquadratic objective functions, most notably cross-entropy loss functions arising in classification. These extensions make GNvpro applicable to all training problems that involve a DNN whose last layer is an affine mapping, which is common in many state-of-the-art architectures. In our four numerical experiments from surrogate modeling, segmentation, and classification, GNvpro solves the optimization problem more efficiently than commonly used stochastic gradient descent (SGD) schemes. Finally, GNvpro finds solutions that generalize well, and in all but one example better than well-tuned SGD methods, to unseen data points.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Thermo-Hydrological Modeling of Thermal Energy Storage in a Depleted Oil Reservoir

Thermal energy storage in oil and gas reservoirs leverages the existing surface and subsurface infrastructure, which can pave the way for economic production of geothermal energy. Existing studies on geothermal energy storage are focused mostly on the use of aquifers with more homogeneous rock and fluid properties. Coupling of heat and fluid flow in a multiphase-multicomponent system, such as an oil reservoir, is imperative especially if existing oil field assets need to be repurposed as required for a sustainable energy transition. The objective is to model the subsurface thermo-hydrological processes associated with reservoir performance and operational sustainability. The model evaluates formation pressure and temperature within the reservoir and at the injection/production wells during multiple charge and discharge cycles. Hot water (approximately 200 degrees C) heated by Concentrating Solar Power (CSP) at high pressure is injected into the existing oil reservoir for storage and produced as thermal energy for power generation, which will be accompanied by enhanced oil recovery. To demonstrate the coupled fluid and heat flow during the injection/production cycle in the subsurface reservoir, TOUGH3 (developed by Berkeley Lab) is used to simulate the thermo-hydrological (TH) processes in a multiphase, multicomponent system. Two well geometries are considered within the reservoir grid: 1) a single-well huff-n-puff system (same well is used for injection and production), and 2) an isolated injection-production well doublet. Seasonal charge and discharge cycling are implemented based on the scheduling specified in the model input file. The model reports pressure, temperature, enthalpy, liquid fluxes, heat fluxes, pore velocities, and changes in porosity & permeability due to temperature and pressure variations during the cyclic Reservoir Thermal Energy Storage (RTES) operations. The results from the simulations can be used to optimize the operational parameters (such as well spacing and injection/production rates) and round-trip efficiency for surface power-plants coupled with thermal energy storage over time. They can also serve as important inputs for levelized cost of storage estimations. The research will help to design and integrate surface renewable energy sources, such as concentrating solar power (CSP), with RTES to help balance out power supply and demand on the grid.

CSP↗
DOE OSTI2020

Catalytic Bio-crude Production in a Novel, Short-Contact Time Reactor

RTI has developed a catalytic fast pyrolysis (CFP) process to produce biomass-based hydrocarbon replacement fuels. From a technology perspective, this advanced biofuels process produces liquid transportation fuels that can leverage capital expenditures in the existing petroleum distribution infrastructure. From a business perspective, this technology does not face potential market limitations caused by oxygenate blend limitations for light duty vehicle applications. A promising catalyst has been developed with excellent potential for deoxygenation to produce a hydrocarbon-rich biocrude that is more suitable for upgrading with traditional hydroprocessing technology. The role of the catalyst is to control the chemistry during biomass pyrolysis to produce a biocrude in a single step that has lower oxygen content and is more thermally stable than conventional biomass fast pyrolysis oil. Catalyst properties are optimized to minimize gas and coke production and improve catalytic deoxygenation and biocrude yields. Results from this project have demonstrated that biocrudes from catalytic fast pyrolysis contain <20 wt% oxygen, with the organic fraction as low as 12% oxygen. These oxygen numbers are > 50% lower compared to traditional fast pyrolysis oils. The biocrude produced from our CFP process can be upgraded to gasoline range hydrocarbons. A very conservative approach for upgrading included a preliminary stabilization step followed by a hydroprocessing step. The oxygen content in the upgraded product was decreased by 97% (to less than 3wt %) from the oxygen content of the biocrude sample. Additionally, the acid content was reduced by 99% from the acid content of the biocrude sample. In an optimized integrated process, the CFP step and the hydroprocessing (upgrading) step need to be carefully balanced to maximize the input biomass energy that is recovered in the finished biofuels. Hydrogen utilization and carbon recovery during both catalytic biomass pyrolysis and biocrude upgrading need to be optimized to maximize energy recovery in the finished biofuel. A nominal 1 ton per day (TPD) catalytic biomass pyrolysis system was designed based on the laboratory experimental data. The system was designed to achieve the short residence times (0.5–2 sec) and high heat transfer rates required for maximum liquid biocrude yields while optimizing process integration to maintain catalyst activity by continuous regeneration. The catalytic biomass pyrolysis reactor is a continuously circulating single-loop transport reactor design that is flexible enough to allow sensitivity studies around temperature, residence time, biomass feed rate, and catalyst-to-biomass ratio (i.e., catalyst circulation rate) for process optimization. For a nominal 100 lb/hr feed rate of dry biomass, the products include ~25 lb/hr char and coke, ~20 lb/hr permanent gases, and ~55 lb/hr condensable vapors. Pyrolysis temperatures are 350–600°C, and the system is operated close to ambient pressure. Production of the catalyst developed within the project was scaled up to provide enough material (200kg) for bench-scale testing. The 1TPD bench-scale unit was fabricated, installed and commissioned with this project. Biocrude samples from wood were produced in this unit with the CFP catalyst with material balance > 90%. The intermediate biocrude yields and chemical composition compare well with previous laboratory results and a hydroprocessing strategy for upgrading the resulting biocrude was developed. The experimental results obtained within this project were used to develop a process model for a 2000 TPD commercial facility. Detailed flow sheets were developed for all major unit operations in the process and provided a basis for an economic analysis to develop capital and operating costs. Estimates of the biofuel production costs were used to develop internal rates of return for the yields and compositions measured for the current state of technology at the end of this project and for the yields anticipated for an optimized (nth plant) scenario as additional process improvements are made.

09 BIOMASS FUELS↗
DOE OSTI2020

Sensitivity and Uncertainty Quantification of Transition Scenario Simulations

This report documents the first collective attempt at developing and applying capabilities to quantify uncertainties, assess parametric sensitivities, and optimize multiple parameters and metrics in fuel cycle simulations generated by the SA&I Campaign. To do this, external codes that were designed to perform sensitivity analysis and uncertainty quantification (SA&UQ) needed to be coupled to the SA&I Campaign’s nuclear fuel cycle simulators (NFCS). In FY20, two approaches were pursued: 1) coupling Cyclus to an ORNL-internal code called MOT (Metaheuristic Optimization Tool) and 2) coupling DYMOND to the opensource SA&UQ tool kit Dakota. The primary objective of having these NFCS/SA&UQ coupled capabilities is to better inform DOE-NE and other stakeholders on the results generated from the NFCS. For a given set of fuel cycle strategies, policies, and technology assumptions that make up a fuel cycle scenario, these NFCS have traditionally been used by the SA&I Campaign to provide quantitative answers in terms of year-by-year mass flows, infrastructure requirements, costs, etc. With these newly developed coupled capabilities, the SA&I Campaign can now efficiently simulate hundreds or thousands of these scenarios, sample large ranges of parameters and assumptions, and use the unique features of the SA&UQ tools to process the data. This enables providing answers with known and propagated uncertainties, determining the sensitivity of important metrics to different parameters and assumptions, quantifying how much fuel cycle and technology parameters impact each other, and producing optimized fuel cycle strategies for single and multiple variables. To demonstrate these new capabilities, the Cyclus/MOT was used to model several scenarios ranging from simple fleet retirements to transitions to advanced reactors. Specifically, for a transition scenario from LWRs to SFRs and advanced LWRs, uncertainty quantification, sensitivity analysis, and optimization studies were applied to cases involving single and multiple parameter (input) and single and multiple metric (output) variations. In addition, a similar transition scenario was modeled to demonstrate how to optimize the reprocessing capacity parameter to minimize two performance metrics while taking into account uncertainties from two other parameters. Lastly, a depletion module based on SCALE/ORIGEN was added in Cyclus to simulate the third scenario that was designed to quantify the impact of the modeling assumption that all LWR used nuclear fuel have the same burnup. The newly developed DYMOND/Dakota capability was also applied to a transition scenario from the existing fleet to small modular reactors and fast reactors. This particular scenario involves not only explicit isotopic depletion via ORIGEN-2, but also includes multirecycling and utilizing the criticality search feature to determine the fresh fuel composition of recycled fuel, a feature unique to the DYMOND NFCS. A large database of simulations were run with 4 main parameters that were sampled: start date of reprocessing, reprocessing capacity, energy demand growth rate, and advanced reactor share of the fleet. The 4 main metrics were uranium consumption, enrichment requirements, waste generation, and levelized cost of electricity using data from the Cost Basis Report. The demonstrated SA&UQ results include those that inform on how to choose parameters to avoid “failed” scenarios, Sobol’ indices that inform on the importance of various parameters individually and synergistically, and Analysis of Variance (ANOVA) studies that decompose parameter ranges into groups and informs on whether variations are statistically significant.

Feng, B.↗
DOE OSTI2022

Maximum power point tracking for a multi-layered piezoelectric heel charger with a levered mechanism toward impact-based energy harvesting

The piezoelectric footstep energy harvester does not always work at its maximum power point when the external load is fixed, as the optimal load changes when the walking excitation alters. Thus, the harvesting efficiency is downgraded largely in real-life scenarios compared to in-lab experiments and theoretical or numerical predictions due to the mismatch between the actual load and the optimal load. To address this issue, the concept of Maximum Power Point Tracking (MPPT) is investigated in this paper and the circuit design is implemented for a multi-layered levered piezoelectric footstep energy harvester (heel charger). The proposed event-driven MPPT circuit interface with a customized buck converter aims to maximize the power gained from daily walking using the heel charger to power a fixed load, such as smart insole or shoes. The MPPT circuit design is conceptually simulated and then tested with the heel charger to further validate if it works at its maximum power point when the frequency of the input excitation alters. Results show that the extracted power from the heel charger connected to a fixed resistance load with MPPT implementation is improved up to 300% compared to the one without MPPT implementation in simulation and up to 180% in the experiment when connected to a fixed load. The difference between simulation and experimental results is due to the optimization of using voltage sources as the heel charger and the control signals (pulse width modulation) from the microcontroller in the simulation.

47 OTHER INSTRUMENTATION↗
DOE OSTI2022

Calibrating hypersonic turbulence flow models with the HIFiRE-1 experiment using data-driven machine-learned models.

In this paper we study the efficacy of combining machine-learning methods with projection-based model reduction techniques for creating data-driven surrogate models of computationally expensive, high-fidelity physics models. Such surrogate models are essential for many-query applications e.g., engineering design optimization and parameter estimation, where it is necessary to invoke the high-fidelity model sequentially, many times. Surrogate models are usually constructed for individual scalar quantities. However there are scenarios where a spatially varying field needs to be modeled as a function of the model’s input parameters. Here we develop a method to do so, using projections to represent spatial variability while a machine-learned model captures the dependence of the model’s response on the inputs. The method is demonstrated on modeling the heat flux and pressure on the surface of the HIFiRE-1 geometry in a Mach 7.16 turbulent flow. The surrogate model is then used to perform Bayesian estimation of freestream conditions and parameters of the SST (Shear Stress Transport) turbulence model embedded in the high-fidelity (Reynolds-Averaged Navier–Stokes) flow simulator, using shock-tunnel data. The paper provides the first-ever Bayesian calibration of a turbulence model for complex hypersonic turbulent flows. We find that the primary issues in estimating the SST model parameters are the limited information content of the heat flux and pressure measurements and the large model-form error encountered in a certain part of the flow.

42 ENGINEERING↗
DOE OSTI2022

Issues in Bi-Directional Interoperability between BIM and BEM

The state-of-the-art interoperability between Building Information Modeling (BIM) and Building Energy Modeling (BEM) is mainly one-way. BEM software like EnergyPlus takes design input from BIM and simulates the energy consumption of the building. BIM, however, is unable to leverage energy simulation results from BEM smoothly. This lack of bi-directional interoperability between BIM and BEM prevents buildings from seeking more energy-efficient design options and limits BIM applications in optimal generative design. In this research, the authors have conducted the following activities related to two-way interoperability between BIM and BEM: (1) surveyed its current status through literature review, (2) investigated barriers to the bi-directional interoperability, and (3) suggested possible approaches to enable bi-directional interoperability and research outlook in the field. As a result, the authors have identified three main barriers to BIM BEM interoperability, namely, inefficient collaboration, limited scope in information transformation, and lack of BEM to BIM transformation. Future directions are suggested to address these barriers.

building design↗
DOE OSTI2025

SymbolNet: neural symbolic regression with adaptive dynamic pruning for compression

Abstract Compact symbolic expressions have been shown to be more efficient than neural network (NN) models in terms of resource consumption and inference speed when implemented on custom hardware such as field-programmable gate arrays (FPGAs), while maintaining comparable accuracy (Tsoi et al 2024 EPJ Web Conf. 295 09036). These capabilities are highly valuable in environments with stringent computational resource constraints, such as high-energy physics experiments at the CERN Large Hadron Collider. However, finding compact expressions for high-dimensional datasets remains challenging due to the inherent limitations of genetic programming (GP), the search algorithm of most symbolic regression (SR) methods. Contrary to GP, the NN approach to SR offers scalability to high-dimensional inputs and leverages gradient methods for faster equation searching. Common ways of constraining expression complexity often involve multistage pruning with fine-tuning, which can result in significant performance loss. In this work, we propose S y m b o l N e t , a NN approach to SR specifically designed as a model compression technique, aimed at enabling low-latency inference for high-dimensional inputs on custom hardware such as FPGAs. This framework allows dynamic pruning of model weights, input features, and mathematical operators in a single training process, where both training loss and expression complexity are optimized simultaneously. We introduce a sparsity regularization term for each pruning type, which can adaptively adjust its strength, leading to convergence at a target sparsity ratio. Unlike most existing SR methods that struggle with datasets containing more than O ( 10 ) inputs, we demonstrate the effectiveness of our model on the LHC jet tagging task (16 inputs), MNIST (784 inputs), and SVHN (3072 inputs).

Tsoi, Ho Fung (ORCID:0000000225502184)↗
DOE OSTI2021

Technoeconomic Analysis of Novel PV Plant Designs for Extreme Cost Reductions

This project sought to gain a deeper understanding of the cost and performance of future solar photovoltaic (PV) plant components, including bifacial PV modules, tandem PV modules, increased plant voltage architectures, and module-level power electronics, and how they may be integrated into new PV plant designs to significantly reduce the levelized cost of electricity (LCOE) of PV. This was done through extensive modeling of current PV plants and future technologies in three different locations, informed by a comprehensive literature review and informational interviews to develop performance and cost assumptions for these technologies. Sensitivities were conducted to understand tradeoffs between different design options, such as the added energy from increasing row spacing versus additional land costs. An optimization tool was then created utilizing an evolutionary algorithm to determine an optimal PV plant configuration for a given set of technologies that resulted in minimized plant LCOE based upon typical performance and cost inputs.

14 SOLAR ENERGY↗
DOE OSTI2021

SMART-COM – Scalable Multi-Agent Adaptive Resolution Tools for Collaborative Outage Management

The purpose of this grant was to conduct scientific research and prototype applications to support NPP outage staff in their adaptive decision-making in efficient scheduling and resource allocation while preventing violation of safety technical specifications. The project contributed to scientific knowledge and engineering methods in (1) user interface design, (2) scheduling optimization and risk estimation, and (2) natural language processing that would benefit the nuclear power plants in minimizing schedule overruns and even unexpected shutdowns. The research team conducted site visits at a test reactor facility and an operating nuclear power plant to gather necessary information and inputs for research and development of a software application to support NPP staff in executing their outages. The final software application consisted of three modules. First, the natural language processing module supports interactive processing of technical documentation to build a database for outage staff to query non-permissible actions on system components. This module can alleviate outage staff from reviewing extensive documentation and minimize violation of technical specifications, especially in time-sensitive situations. Second, the schedule optimization module schedules outage activities and compute risk indices that outperform existing software and current practice. This module can reduce completion time of an outage that typically have too many activities for human to optimize based on current practice that does not apply the latest operations research. Finally, the visualization module presents progress and risk information of the overall outage and individual activities, as well as enabling access to the natural language processing and schedule optimization modules. This module can provide outage staff with situation awareness that are necessary to make risk-informed decisions in response to unexpected events during the execution of an outage.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2023

Efficient low-temperature, catalyst-free dehydrogenation of alkanes

Steam cracking of ethane, a non-catalytic thermochemical process, remains the dominant means of ethylene production. The severe reaction conditions and energy expenditure involved in this process incentivize the search for alternative reaction pathways and reactor designs which maximize ethylene yield while minimizing cost and energy input. According to the present invention, ethylene yields as high as 68% were obtained with a quartz open tube reactor without the use of a catalyst or a cofed stream of oxidizing agents. The open tube reactor design promotes simplicity, low cost, and negligible coke formation. Reactor designs can be optimized to improve the conversion of ethane to ethylene via non-oxidative dehydrogenation, an approach which shows promise for decentralized production of ethylene from natural gas deposits.

Riley, Christopher Ryan↗
DOE OSTI2021

Bayesian learning of orthogonal embeddings for multi-fidelity Gaussian Processes

Uncertainty propagation in complex engineering systems often poses significant computational challenges related to modeling and quantifying probability distributions of model outputs, as those emerge as the result of various sources of uncertainty that are inherent in the system under investigation. Gaussian Processes regression (GPs) is a robust meta-modeling technique that allows for fast model prediction and exploration of response surfaces. Multi-fidelity variations of GPs further leverage information from cheap and low fidelity model simulations in order to improve their predictive performance on the high fidelity model. In order to cope with the high volume of data required to train GPs in high dimensional design spaces, a common practice is to introduce latent design variables that are typically projections of the original input space to a lower dimensional subspace, and therefore substitute the problem of learning the initial high dimensional mapping, with that of training a GP on a low dimensional space. Here in this paper, we present a Bayesian approach to identify optimal transformations that map the input points to low dimensional latent variables. The \projection" mapping consists of an orthonormal matrix that is considered a priori unknown and needs to be inferred jointly with the GP parameters, conditioned on the available training data. The proposed Bayesian inference scheme relies on a two-step iterative algorithm that samples from the marginal posteriors of the GP parameters and the projection matrix respectively, both using Markov Chain Monte Carlo (MCMC) sampling. In order to take into account the orthogonality constraints imposed on the orthonormal projection matrix, a Geodesic Monte Carlo sampling algorithm is employed, that is suitable for exploiting probability measures on manifolds. We extend the proposed framework to multi-fidelity models using GPs including the scenarios of training multiple outputs together. We validate our framework on three synthetic problems with a known lower-dimensional subspace. The benefits of our proposed framework, are illustrated on the computationally challenging aerodynamic optimization of a last-stage blade for an industrial gas turbine, where we study the effect of an 85-dimensional shape parameterization of a three-dimensional airfoil on two output quantities of interest, specifically on the aerodynamic efficiency and the degree of reaction

42 ENGINEERING↗
DOE OSTI2023

Understanding and control of Zener pinning via phase field and ensemble learning

Zener pinning refers to the dispersion of fine particles which influences grain size distribution via movement of grain boundaries in a polycrystalline material. Grain size distribution in polycrystals has a significant impact on their properties including physical, chemical, mechanical, and optical to name a few. We explore the use of Phase-field modeling and machine-learning techniques to understand and improve the control of grain size distribution via Zener pinning in polycrystalline materials. We develop a machine learning model that determines the relative importance of various parameters to exercise microstructure control via Zener pinning. Our workflow combines high-throughput phase-field simulations and machine learning to address the computational bottlenecks associated with large-scale simulations as well as identify features necessary for microstructure control in polycrystals. A random forest (RF) regression model was developed to predict grain sizes based on five Phase-field model parameters, achieving an average prediction error of 0.72 nm for the training data and 1.44 nm for the test data. The importance of the input parameters is analyzed using the SHapley Additive exPlanations (SHAP) approach which reveals that diffusivity, volume fraction, and particle diameter are the most important parameters in determining the final grain size. These findings will allow us to select the best second-phase particles, optimize grain size distributions and thus design microstructures with the desired properties. The developed method is a highly versatile and generalizable approach that can be used to assess the combined effects of individual features in the presence of multiple variables.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Numerical Simulation Studies of Ultrasonic De-Icing for Heating, Ventilation, Air Conditioning, and Refrigeration Structures

Ice accumulation on heating, ventilation, air conditioning, and refrigeration (HVACR) structures presents significant operational challenges. These challenges include reduced efficiency, increased energy consumption, and potential damage to equipment. Traditional de-icing methods, such as chemical treatments, mechanical scraping, or heating-based techniques, are often labor-intensive, costly, and environmentally harmful. Here, this study uniquely investigates ultrasonic de-icing as an energy-efficient alternative for HVACR applications, focusing on the specific structural geometries found in these systems. A comprehensive numerical simulation framework was developed using finite element analysis to explore ultrasonic wave propagation across four distinct HVACR structures. Key parameters such as ultrasonic frequency, power levels, and the number and placement of actuators were examined for their impact on ice detachment efficiency. Results from simulations on a plate structure reveal that ultrasonic excitation can propagate effectively across large areas (at least 150 × 150 mm), enhancing the de-icing coverage. Lower frequency (e.g., 30 to 45 kHz) excitation results in greater displacement, improving de-icing performance, while increased actuator numbers with the same total power input also enhance effectiveness. Two actuators seem sufficient for the de-icing of a 300 × 300 mm plate. For tube-and-fin structures, specific high-power ultrasonic frequencies selectively excite the fin plates, demonstrating efficient ice removal when actuated on the tube. However, optimal performance requires careful design of actuator placement and vibration modes to accommodate the irregular shapes of these structures.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗