Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “learned”

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 487 records · Page 27

Efficient Active Learning for New Domains

The promise of active learning is to reduce the number of labeled examples required by supervised machine learning algorithms. The largest potential benefits lie in entirely new domains, for which no labeled examples yet exist. Yet to date, most active learning studies are retroactive and demonstrate the benefits that could have been gained if active learning had been used. What are the barriers to true adoption and utilization of active learning? We focus on two: (1) the cold start or class discovery problem, in which active learning methods may struggle to make progress with zero labeled examples, and (2) the cost of having the classifier in the loop to select the next example to be labeled. We assess different active learning approaches in the context of these two barriers and conclude with recommendations for how to employ active learning in new domains. As an example, we report on the use of active learning on a large, novel data set of Mars surface images.

Lu, Steven↗

Reinforcement learning based automated history matching for improved hydrocarbon production forecast

History matching aims to find a numerical reservoir model that can be used to predict the reservoir performance. An engineer and model calibration (data inversion) method are required to adjust various parameters/properties of the numerical model in order to match the reservoir production history. In this study, we develop deep neural networks within the reinforcement learning framework to achieve automated history matching that will reduce engineers’ efforts, human bias, automatically and intelligently explore the parameter space, and remove the need of large set of labeled training data. To that end, a fast-marching-based reservoir simulator is encapsulated as an environment for the proposed reinforcement learning. The deep neural-network-based learning agent interacts with the reservoir simulator within reinforcement learning framework to achieve the automated history matching. Reinforcement learning techniques, such as discrete Deep Q Network and continuous Deep Deterministic Policy Gradients, are used toth, used to train the learning agents. The continuous actions enable the Deep Deterministic Policy Gradients to explore more states at each iteration in a a learning episode; consequently, a better history matching is achieved using this algorithm as compared to Deep Q Network. For simplified dual-target composite reservoir models, the best history-matching performances of the discrete and continuous learning methods in terms of normalized root mean square errors are 0.0447 and 0.0038, respectively. Furthermore, our study shows that continuous action space achieved by the deep deterministic policy gradient drastically outperforms deep Q network.

42 ENGINEERING↗

Inverse reinforcement learning control for building energy management

Reinforcement learning (RL) based control is widely considered a promising approach in building automation and control as it has demonstrated the potential to deal with complex objectives in adjacent domains like robotics, autonomous vehicles, gaming applications, and advertisement recommendations. When applied to any environment, model-free RL learns to improve its control performance over time without requiring a control model, by receiving and then analyzing feedback from the building environment after each control action. Operational objectives are becoming increasingly complex through the simultaneous consideration of thermal comfort, carbon emissions, grid services, and indoor air quality. In this context, conventional rule-based control approaches are proving sub-optimal, mostly heuristic, and inadequate. The model-free and self learning nature of RL appears promising and attractive as it may address the scalability issues associated with advanced control approaches. However, it suffers from long training times and unstable control behavior during the early stages of its learning process, which makes it unsuitable to be applied directly to buildings. This paper addresses these issues using an inverse reinforcement learning approach (IRL), a technique utilized to learn the objective of a controller agent which is considered an expert in its respective domain. Here, we consider a rule-based control as the expert demonstrator. IRL is different from a direct imitation (i.e., direct mapping of states to actions) of control actions as it tries to find the underlying intent of an expert's policy, providing the controller with a better-generalized policy for unseen states or environments with slightly different dynamics. This approach propels the RL controller's policy to levels similar to or better than that of a rule-based policy before it starts learning by interacting with the building. This makes RL for building energy management applications more practical as it prevents the erratic and exploratory behavior in the initial training period, simultaneously speeding up the learning process when compared to applying an untrained RL agent directly to a building environment.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Bridging Hydrological Ensemble Simulation and Learning Using Deep Neural Operators

Ensemble-based simulation and learning (ESnL) has long been used in hydrology for parameter inference, but computational demands of process-based ESnL can be quite high. To address this issue, we propose a deep neural operator learning approach. Neural operators are generic machine learning algorithms that can learn functional mappings between infinite-dimensional spaces, providing a highly flexible tool for scientific machine learning. Our approach is built upon DeepONet, a specific deep neural operator, and is designed to address several common problems in hydrology, namely, model parameter estimation, prediction at ungaged locations, and uncertainty quantification. Here we demonstrate the effectiveness of our DeepONet-based workflow using an existing large model ensemble created for an eastern U.S. watershed that is instrumented with 10 streamflow gages. Results suggest DeepONet achieves high efficiency in learning an ML surrogate model from the model ensemble, with the modified Kling-Gupta Efficiency exceeding 0.9 on holdout test sets. Parameter inference, carried out using the trained DeepONet surrogate model and genetic algorithm, also yields robust results. Additionally, we formulate and train a separate DeepONet model for physics-informed, seq-to-seq streamflow forecasting, which further reduces biases in the pre-trained DeepONet surrogate model. While this study focuses primarily on a single watershed, our approach is general and may be extended to enable learning from model ensembles across multiple basins or models. Thus, this research represents a significant contribution to the application of hybrid machine learning in hydrology.

54 ENVIRONMENTAL SCIENCES↗

Deep transfer operator learning for partial differential equations under conditional shift

Transfer learning enables the transfer of knowledge gained while learning to perform one task (source) to a related but different task (target), hence addressing the expense of data acquisition and labelling, potential computational power limitations and dataset distribution mismatches. Here, we propose a new transfer learning framework for task-specific learning (functional regression in partial differential equations) under conditional shift based on the deep operator network (DeepONet). Task-specific operator learning is accomplished by fine-tuning task-specific layers of the target DeepONet using a hybrid loss function that allows for the matching of individual target samples while also preserving the global properties of the conditional distribution of the target data. Inspired by conditional embedding operator theory, we minimize the statistical distance between labelled target data and the surrogate prediction on unlabelled target data by embedding conditional distributions onto a reproducing kernel Hilbert space. We demonstrate the advantages of our approach for various transfer learning scenarios involving nonlinear partial differential equations under diverse conditions due to shifts in the geometric domain and model dynamics. Our transfer learning framework enables fast and efficient learning of heterogeneous tasks despite considerable differences between the source and target domains.

42 ENGINEERING↗

Machine learning assisted modeling of mixing timescale for LES/PDF of high-Karlovitz turbulent premixed combustion

Accurate modeling of mixing in the transported probability density function (PDF) method remains a great challenge, especially for turbulent premixed combustion under extreme conditions such as high Karlovitz number Ka. Recently, a power-law based mixing timescale model was developed for the large-eddy simulations (LES)/PDF modeling of high-Ka number turbulent premixed flames. It is found in this work that the power-law mixing timescale model is highly sensitive to the model parameters. It is thus critically needed to develop accurate calibration of these model parameters. The empirical specification of the model parameters developed in Zhang et. al. is found to be inadequate for accurate modeling of the mixing timescale. Here, machine learning is introduced as an attractive alternative in this work for the specification of the model parameters. A high-Ka number DNS jet flame is used as the training and validation of the machine learning models. The choices of the input parameters are discussed and compared for the machine learning models. The effect of differential molecular diffusion on mixing is examined by including the effect of the Lewis number in the training of the machine learning models. The performance of different machine learning algorithms is compared for the specification of the mixing model parameters. Overall, excellent performance of the machine learning models is observed for assisting the mixing modeling. The feasibility, interpretability, applicability, generality, and portability of using machine learning are discussed in general to provide a perspective on applying data-driven machine learning for turbulent combustion modeling studies.

42 ENGINEERING↗

Promoting system-level learning from project-level lessons

A growing number of low and middle income nations (LMCs) have adopted some sort of system for environmental impact assessment (EIA). However, generally many of these EIA systems are characterised by a low performance in terms of timely information dissemination, monitoring and enforcement after licencing. Donor actors (such as the World Bank) have attempted to contribute to a higher performance of EIA systems in LMCs by intervening at two levels: the project level (e.g. by providing scoping advice or EIS quality review) and the system level (e.g. by advising on EIA legislation or by capacity building). The aims of these interventions are environmental protection in concrete cases and enforcing the institutionalisation of environmental protection, respectively. Learning by actors involved is an important condition for realising these aims. A relatively underexplored form of learning concerns learning at EIA system-level via project level donor interventions. This 'indirect' learning potentially results in system changes that better fit the specific context(s) and hence contribute to higher performances. Our exploratory research in Ghana and the Maldives shows that thus far, 'indirect' learning only occurs incidentally and that donors play a modest role in promoting it. Barriers to indirect learning are related to the institutional context rather than to individual characteristics. Moreover, 'indirect' learning seems to flourish best in large projects where donors achieved a position of influence that they can use to evoke reflection upon system malfunctions. In order to enhance learning at all levels donors should thereby present the outcomes of the intervention elaborately (i.e. discuss the outcomes with a large audience), include practical suggestions about post-EIS activities such as monitoring procedures and enforcement options and stimulate the use of their advisory reports to generate organisational memory and ensure a better information dissemination.

54 ENVIRONMENTAL SCIENCES↗

Stream Temperature Prediction in a Shifting Environment: Explaining the Influence of Deep Learning Architecture

Stream temperature is a fundamental control on ecosystem health. Recent efforts incorporating process guidance into deep learning models for predicting stream temperature have been shown to outperform existing statistical and physical models. This performance is in part because deep learning architectures can actively learn spatiotemporal relationships that govern how water and energy propagate through a river network. However, exploration of how spatiotemporal awareness and process guidance influence a model's generalizability under shifting environmental conditions such as climate change is limited. Here, we use Explainable Artificial Intelligence (XAI) to interrogate how differing deep learning architectures affect a model's learned spatial and temporal dependencies, and how those learned dependencies affect a model's ability to maintain high accuracy when applied to unseen environmental conditions. Using the Delaware River Basin in the northeastern United States as a test case, we compare two spatiotemporally aware process–guided deep learning models for predicting stream temperature (a recurrent graph convolution network—RGCN, and a temporal convolution graph model—Graph WaveNet). Both models achieve equally high predictive performance when testing data are well represented in the training data (test root mean squared errors of 1.64°C and 1.65°C); however, Graph WaveNet significantly outperforms RGCN in 4 out of 5 experiments where test partitions represent different types of unseen environmental conditions. XAI results show that the architecture of Graph WaveNet leads to learned spatial relationships with greater fidelity to physical processes, and that this fidelity improves the generalizability of the model when applied to shifting and/or unseen environmental conditions.

54 ENVIRONMENTAL SCIENCES↗

Design principles for lifelong learning AI accelerators

Lifelong learning-an agent's ability to learn throughout its lifetime-is a hallmark of biological learning systems and a central challenge for artificial intelligence (AI). The development of lifelong learning algorithms could lead to a range of novel AI applications, but this will also require the development of appropriate hardware accelerators, particularly if the models are to be deployed on edge platforms, which have strict size, weight and power constraints. Here we explore the design of lifelong learning AI accelerators that are intended for deployment in untethered environments. We identify key desirable capabilities for lifelong learning accelerators and highlight metrics to evaluate such accelerators. We then discuss current edge AI accelerators and explore the future design of lifelong learning accelerators, considering the role that different emerging technologies could play. The Perspective explores the future design of lifelong learning artificial intelligence (AI) accelerators that are intended for deployment in untethered environments, identifying key desirable capabilities for such edge AI accelerators and guidance on metrics to evaluate them.

42 ENGINEERING↗

Neuromorphic learning with Mott insulator NiO

Habituation and sensitization (nonassociative learning) are among the most fundamental forms of learning and memory behavior present in organisms that enable adaptation and learning in dynamic environments. Emulating such features of intelligence found in nature in the solid state can serve as inspiration for algorithmic simulations in artificial neural networks and potential use in neuromorphic computing. In this work, we demonstrate nonassociative learning with a prototypical Mott insulator, nickel oxide (NiO), under a variety of external stimuli at and above room temperature. Similar to biological species such as Aplysia, habituation and sensitization of NiO possess time-dependent plasticity relying on both strength and time interval between stimuli. A combination of experimental approaches and first-principles calculations reveals that such learning behavior of NiO results from dynamic modulation of its defect and electronic structure. An artificial neural network model inspired by such nonassociative learning is simulated to show advantages for an unsupervised clustering task in accuracy and reducing catastrophic interference, which could help mitigate the stability–plasticity dilemma. Mott insulators can therefore serve as building blocks to examine learning behavior noted in biology and inspire new learning algorithms for artificial intelligence.

42 ENGINEERING↗

Optimizing thermodynamic trajectories using evolutionary and gradient-based reinforcement learning

Here using a model heat engine, we show that neural-network-based reinforcement learning can identify thermodynamic trajectories of maximal efficiency. We consider both gradient and gradient-free reinforcement learning. We use an evolutionary learning algorithm to evolve a population of neural networks, subject to a directive to maximize the efficiency of a trajectory composed of a set of elementary thermodynamic processes; the resulting networks learn to carry out the maximally efficient Carnot, Stirling, or Otto cycles. When given an additional irreversible process, this evolutionary scheme learns a previously unknown thermodynamic cycle. Gradient-based reinforcement learning is able to learn the Stirling cycle, whereas an evolutionary approach achieves the optimal Carnot cycle. Our results show how the reinforcement learning strategies developed for game playing can be applied to solve physical problems conditioned upon path-extensive order parameters.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Machine learning model inputs, outputs, and scripts associated with “Artificial intelligence-guided iterations between observations and modeling significantly improve environmental predictions”

NOTE: The manuscript associated with this data package is currently in review. The data may be revised based on reviewer feedback. Upon manuscript acceptance, this data package will be updated with the final dataset and additional metadata. This data package is associated with the manuscript “Artificial intelligence-guided iterations between observations and modeling significantly improve environmental predictions” (Malhotra et al., in prep). This effort was designed following ICON (integrated, coordinated, open, and networked) principles to facilitate a model-experiment (ModEx) iteration approach, leveraging crowdsourced sampling across the contiguous United States (CONUS). New machine learning models were created every month to guide sampling locations. Data from the resulting samples were used to test and rebuild the machine learning models for the next round of sampling guidance. Associated sediment and water geochemistry and in situ sensor data can be found at https://data.ess-dive.lbl.gov/datasets/doi:10.15485/1923689, https://data.ess-dive.lbl.gov/datasets/doi:10.15485/1729719, and https://data.ess-dive.lbl.gov/datasets/doi:10.15485/1603775. This data package is associated with two GitHub repositories found at https://github.com/parallelworks/dynamic-learning-rivers and https://github.com/WHONDRS-Hub/ICON-ModEx_Open_Manuscript. In addition to this readme, this data package also includes two file-level metadata (FLMD) files that describes each file and two data dictionaries (DD) that describe all column/row headers and variable definitions. This data package consists of two main folders (1) dynamic-learning-rivers and (2) ICON-ModEx_Open_Manuscript which contain snapshots of the associated GitHub repositories. The input data, output data, and machine learning models used to guide sampling locations are within dynamic-learning-rivers. The folder is organized into five top-level directories: (1) “input_data” holds the training data for the ML models; (2) “ml_models” holds machine learning (ML) models trained on the data in “input_data”; (3) “examples” contains files for direct experimentation with the machine learning model, including scripts for setting up “hindcast” run; (4) “scripts” contains data preprocessing and postprocessing scripts and intermediate results specific to this data set that bookend the ML workflow; and (5) “output_data” holds the overall results of the ML model on that branch. Each trained ML model resides on its own branch in the repository; this means that inputs and outputs can be different branch-to-branch. There is also one hidden directory “.github/workflows”. This hidden directory contains information for how to run the ML workflow as an end-to-end automated GitHub Action but it is not needed for reusing the ML models archived here. Please see the top-level README.md in the GitHub repository for more details on the automation. The scripts and data used to create figures in the manuscript are within ICON-ModEx_Open_Manuscript. The folder is organized into four folders which contain the scripts, data, and pdf for each figure. Within the “fig-model-score-evolution” folder, there is a folder called “intermediate_branch_data” which contains some intermediate files pulled from dynamic-learning-rivers and reorganized to easily integrate into the workflows. NOTE: THIS FOLDER INCLUDES THE FILES AT THE POINT OF PAPER SUBMISSION. IT WILL BE UPDATED ONCE THE PAPER IS ACCEPTED WITH ANY REVISIONS AND WILL INCLUDE A DD/FLMD AT THAT POINT. We thank the United States Forest Service, Washington Department of Fish and Wildlife, Washington Department of Natural Resources, Cowiche Canyon Conservatory, Washington State Parks and Recreation Commission (Scientific Research Permit #210901), and the Confederated Tribes and Bands of the Yakama Nation for access to field locations where the samples labeled “SSS” were collected. We also thank the Yakama Nation Tribal Council and Yakama Nation Fisheries for working with us to facilitate sample collection and optimization of data usage according to their values and worldview. WHONDRS consortium members were asked to provide any acknowledgments for the collection of samples labeled “CM” and the following is a list of acknowledgments that were submitted with their corresponding Site IDs: (MART) Research activities were conducted in part on the Wind River Experimental Forest within the Gifford Pinchot National Forest; (MP- 100379) Philadelphia is part of Lenapehoking, the ancestral homelands of the Lenape peoples; (MP-102398) Land surveyed is the ancestral homelands of the Nookhose'iinenno (Arapaho), Tsis tsis'tas (Cheyenne), and Nuuchu (Ute); (MP-100749 and MP- 100747) Georgia Coastal Ecosystem LTER, OCE-1832178; (SP-70 and SP-72) Eastern Shoshone, Shoshone-Bannock; (MP- 102944) Funded by Oregon Watershed Enhancement Board. On the traditional lands of the Confederated Tribes of the Siletz, Confederated Tribes of the Grand Rhonde, and the Clatsop-Nehalem Confederated Tribe; (MP- 100607) Holiday Creek is located on the traditional territory of the Monacan Indian Nation; (SP-45) Lafayette Blue Springs State Park; (MP-102420) NSF DEB-2016749; (MP-100019) New Hampshire Agriculture Experiment Station; (SP-35) Rayonier (land owner; https://www.rayonier.com/); (MP- 101276) US Department of Energy, Office of Science, Biological and Environmental Research, Subsurface Biogeochemical Research, Watershed Dynamics and Evolution SFA at ORNL; (MP- 103224) Watershed Dynamics and Evolution SFA at ORNL; (MP- 101584) Traditional lands of the Oceti Sakowin (Dakota, Lakota, Nakoda) and Anishinaabe Peoples.

54 ENVIRONMENTAL SCIENCES↗

Rapid motor learning in the translational vestibulo-ocular reflex

Motor learning was induced in the translational vestibulo-ocular reflex (TVOR) when monkeys were repeatedly subjected to a brief (0.5 sec) head translation while they tried to maintain binocular fixation on a visual target for juice rewards. If the target was world-fixed, the initial eye speed of the TVOR gradually increased; if the target was head-fixed, the initial eye speed of the TVOR gradually decreased. The rate of learning acquisition was very rapid, with a time constant of approximately 100 trials, which was equivalent to <1 min of accumulated stimulation. These learned changes were consolidated over >or=1 d without any reinforcement, indicating induction of long-term synaptic plasticity. Although the learning generalized to targets with different viewing distances and to head translations with different accelerations, it was highly specific for the particular combination of head motion and evoked eye movement associated with the training. For example, it was specific to the modality of the stimulus (translation vs rotation) and the direction of the evoked eye movement in the training. Furthermore, when one eye was aligned with the heading direction so that it remained motionless during training, learning was not expressed in this eye, but only in the other nonaligned eye. These specificities show that the learning sites are neither in the sensory nor the motor limb of the reflex but in the sensory-motor transformation stage of the reflex. The dependence of the learning on both head motion and evoked eye movement suggests that Hebbian learning may be one of the underlying cellular mechanisms.

Non-NASA Center↗

Onboard Hyperspectral Image Classification via Transfer Learning for Communication-Limited Spacecraft

Employing deep-learning and artificial-intelligence (AI) techniques onboard spacecraft can dramatically improve priority data selection to ensure more effective use of the available downlink. However, deployment of effective deep-learning models requires significant training on the ground, which may not be feasible, due to limited data available in an unexplored environment. Therefore, this research explores building robust classification models for onboard data processing where training data is highly limited using transfer-learning techniques. In this paper, we focus on the use case of hyperspectral imaging for remote sensing, a domain where the high dimensionality of the data from the sensor can rapidly saturate the downlink bandwidth. With this bottleneck, there is an impending need to autonomously and robustly classify data onboard to optimize downlink of high-impact measurements, thus maximizing the scientific utility per bit transmitted to the ground. This paper examines the use of deep neural networks onboard for hyperspectral image classification in a communication-limited scenario to analyze how the models perform with limited training data. The use of transfer learning can ameliorate the issue of poor generalization by transferring features learned from training on a large source dataset for one classification task to the target classification task with limited training data. For two deep-learning models from literature, we compare the accuracy of the models trained using transfer learning to models trained from scratch using a random weight initialization with varying amounts of training data. We demonstrate the feasibility and performance of running inference of the deep-learning models on representative flight-like hardware.

Advanced Avionics, Machine Learning, Data Processi↗

Deep Learning for Scientific Discovery

Deep learning has positively impacted fields were perceptive tasks are paramount, such as computer vision and natural language processing. Tremendous strides have also been made in deep reinforcement learning towards robotics and gaming. Given the success in these fields, it is apparent that deep learning may also accelerate scientific discovery. Herein, we discuss several topic areas where modern representation learning is driving innovation. This article presents a) learning non-linear models from data, b) learning scientific laws from models, c) deep learning on graphs, and d) approaches to scalable deep learning.

deep learning, scientific computing, koopman, mate↗

A progressive deep transfer learning approach to cycle-skipping mitigation in FWI, In SEG Technical Program Expanded Abstracts 2019, pp. 2348-2352. Society of Exploration Geophysicists, 2019.

To effectively mitigate the cycle-skipping phenomenon in full waveform inversion (FWI), we developed an inception-based deep learning neural network to reconstruct the absent low frequency data by exploiting the subsurface low wavenumber information buried in the acquired high frequency data. Two unique features of our deep learning network for the low frequency data reconstruction are 1) Dual Data Feed; 2) Progressive Transfer Learning. With the Dual Data Feed feature, not only the high frequency data, but also the corresponding Beat Tone data are fed into the neural network, significantly reducing the network complexity and the network training cost. The Progressive Transfer Learning technique enables us to initiate the deep learning network with only one training dataset. Unlike other deep learning approaches in this category, our training dataset is not fixed. Instead, the training dataset and the network accuracy are progressively enhanced in an iterative manner within the Progressive Transfer Learning workflow. The numerical experiments validated that the low frequency data reconstructed by this progressive transfer learning network are sufficiently accurate for a successful FWI inversion without suffering from the cycle-skipping without any priori geological information.

58 GEOSCIENCES↗

E-Learning in Neutron Activation Analysis

Since the middle 1990's electronic technology (E learning) has slowly infiltrated and now dominates education at all levels. E-learning encompasses a wide array of deliveries from simple introductions of animations in a teaching class to Massive Open-On-Line Courses (MOOCs) and Modular Object-Oriented Dynamic Learning Environment (MOODLEs). Even the definition of e-learning can be allusive. E-learning can or cannot have the following aspects: internet availability, supervised exams, a teacher to advise the students by electronic means such as e-mail, telephone, SKYPE, GOOGLE HANGOUT, FACETIME, etc., and multimedia designs. E-learning has transformed education from the Traditional Classroom Course to a) Synchronous Distributed Course in which web-based technologies are used to extend classroom lectures and discussions to students at remote sites in real time; b) Web-Enhanced Course in which online course activity complements class sessions without reducing the number of required class meetings; c). 'Emporium' Course where this model is designed for on-campus use, eliminates all class meetings and replaces them with a learning resource center featuring online materials and on-demand personalized assistance; d) Hybrid Course where online activity is mixed with classroom meetings, replacing at least 20 percent, but not all required face-to-face meetings; and e) Online Course where all course activity is done online and there are no required face-to-face sessions within the course and no requirements for on-campus activity. E-learning takes an extra challenge where the subject matter is highly scientific and engineering oriented encompassing multiple complex mathematical notations and concepts. A straight forward presentation of equations in Power Point is a recipe for an unsuccessful delivery of lectures. But how does one deliver intricate concepts without the use of a traditional blackboard to write on and how does one reach a worldwide diverse community in a fast moving arena of technological changes in education delivery?. The International Atomic Energy Agency (IAEA) and other international organizations as well, have undertaken large efforts to secure nuclear knowledge management in a wide area of nuclear science and engineering including research reactors. (authors)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

A Study of Physics-Informed Deep Learning for System Fluid Dynamics Closures

The sub-grid-scale (SGS) physics models, or so-called closure relations (CR), are essential in thermal-fluid modeling and simulation codes, ranging from large-eddy simulation to system-level thermo-fluid dynamic problems. While details vary, all thermo-fluid codes solve coupled mass-momentum-energy conservation partial differential equations (PDE). Traditionally, the analytical form of closure models has to be known, and then we calibrate the model using data from a set of applicable experiments. Characteristically, it takes long effort to gain insights and mechanistic understanding needed for developing closure models. Such a model is a compact form of data. The long time required for new model development limits the pace of model application while dealing with new geometries, new coolants or system conditions, particularly in newly designed systems. Alternative to mechanistic and semi-analytical models, it is possible to discover the pattern of closure relations by applying the modern machine learning methods to the body of relevant experimental data. Deep learning (DL) is a relatively new branch in the machine learning family, and a major breakthrough has been achieved since Hinton introduces a fast algorithm to efficiently train a feed-forward neural network (NN) with multilayer perceptrons. Generally, any NN with more than two layers is considered as deep learning, and it contains a large number of hyper-parameters to achieve pattern recognition or regression for complex datasets. Hornik shows that multilayer NN is compatible with universal approximation theorem. There is no constraint for a multilayer NN to capture the properties of any measurable information. This capability makes DL attractive for closure models in a system thermal-hydraulic code. As system geometry and/or operating conditions change, new physical experiments (increasingly sophisticated) are conducted to support evaluation and revision of closure models. Such a so-called data-driven modeling relaxes the structural form of system codes to directly assimilate data instead of hard-wiring the model to pre-calibrated correlations. The objective of this research is to enable cost-effective development of a DL-based closure that reflects the data. The previous work on applying machine learning techniques to fluid dynamics focused on statistical learning with basic NN from direct numerical simulations (DNS) dataset. In the present work, we show that DL-based closure models can cause the fluid calculation to diverge if multilayer NN is used. To address the issue, we investigate conditions by which the DL-based closure models work compatibly, stably, and effectively with PDE-constrained forward prediction problems. Optimally meeting the compatibility conditions constitutes a physics-informed deep learning (PIDL) strategy. We refer to this process as PDE-constrained machine learning or PDE-ML. To develop the technical basis for PDE-ML using DL-based fluid dynamics closure, numerical experiments are performed for manufactured problems of increasing complexity. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗