Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Model error”

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 1,783 records · Page 99
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Entropy-Infused Deep Learning Loss Function for Capturing Extreme Values in Wind Power Forecasting

Extreme scenarios in wind power generation occur with higher frequency and larger magnitude in the recent years due to the ever-increasing extreme meteorological factors. Accurate forecasting of the occurrence of extreme values in wind power generation is of great concern to ensure reliable power system operation. Recently, deep learning models have surged in popularity for wind power forecasting, with the mean squared error (MSE) loss function being commonly used. However, the MSE loss function, being sensitive to extreme values, disproportionately penalizes larger errors, cannot adequately capture the extreme values present in wind energy data, and novel loss functions have seldom been tailored for wind power forecasting. To this end, in this paper, we introduce a novel loss function specifically crafted to capture extreme values in wind power forecasting. The experimental results with four fundamental deep learning methods on open source wind power dataset validate that the new loss function is efficient and superior in all cases compared to MSE in capturing extreme values while maintaining forecasting performance.

17 WIND ENERGY↗
DOE OSTI2023

Multi-layered Energy Management Framework for Extreme Fast Charging Stations Considering Demand Charges, Battery Degradation, and Forecast Uncertainties

To achieve a cost-effective and expeditious charging experience for extreme fast charging station (XFCS) owners and electric vehicle (EV) users, the optimal operation of XFCS is crucial. It is however challenging to simultaneously manage the profit from energy arbitrage, the cost of demand charges, and the degradation of a battery energy storage system (BESS) under uncertainties. This paper, therefore, proposes a multi-layered multi-time scale energy flow management framework for an XFCS by considering long- and short-term forecast uncertainties, monthly demand charges reduction, and BESS life degradation. In the proposed approach, an upper scheduling layer (USL) ensures the overall operation economy and yields optimal scheduling of the energy resources on a rolling horizon basis, thereby considering the long-term forecast errors. A lower dispatch layer (LDL) takes the short-term forecast errors into account during the real-time operation of the XFCS. Per the latest research, monthly demand charges can be as high as 90% of the total monthly bills for EV fast charging stations; to this end, this paper takes the first attempt at the reduction of demand charges cost by considering the trade-off between the energy cost and monthly demand charges. Contrasting literature, this work allocates an energy reserve in the BESS stored energy to deal with the impact of short-term forecast errors on the optimized real-time operation of the XFCS. Moreover, degradation modeling considers the trade-off between short-term benefits and long-term BESS life degradation. As a result, case studies and a comparative analysis prove the efficacy of the proposed framework.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2026

Leveraging Inequality-Constrained Data for Enhanced Liquidus Temperature Prediction in Nuclear Waste Glass Melts

Inequality-constrained data are frequently discarded in engineering, leading to significant information loss in data-scarce domains like glass characterization in nuclear waste vitrification. This paper presents a nonparametric censored-data regression framework based on an l1-norm optimization criterion that leverages slack variables to integrate left-, right-, and interval-constrained observations into training without distributional assumptions. Validated on synthetic data and a Physics-Informed Neural Network (PINN) for predicting liquidus temperature (TL), the method improved R2 from 0.60 to 0.89 and reduced Mean Absolute Error (MAE) by 48% (51.46 to 26.89?rC) on deterministic values. The traditional models failed to satisfy any inequality constraints while the proposed l1-norm PINN satisfies 81.25% of the constraints. The proposed framework effectively extracts actionable information from previously unusable data to enhance predictive accuracy, reduce epistemic uncertainty, and ensure physical consistency in complex industrial applications.

Garcia-Morado, Erick↗
DOE OSTI2021

Measuring Nd(III) Solution Concentration in the Presence of Interfering Er(III) and Cu(II) Ions: A Partial Least Squares Analysis of Ultraviolet–Visible Spectra

Optical spectroscopy is a powerful characterization tool with applications ranging from fundamental studies to real-time process monitoring. However, it can be difficult to apply to complex samples that contain interfering analytes which are common in processing streams. Multivariate (chemometric) analysis has been examined for providing selectivity and accuracy to the analysis of optical spectra and expanding its potential applications. Here we will discuss chemometric modeling with an in-depth comparison to more simplistic analysis approaches and outline how chemometric modeling works while exploring the limits on modeling accuracy. Understanding the limitations of the chemometric model can provide better analytical assessment regarding the accuracy and precision of the analytical result. This will be explored in the context of UV–Vis absorbance of neodymium (Nd 3+ ) in the presence of interferents, erbium (Er 3+ ) and copper (Cu 2+ ) under conditions simulating the liquid–liquid extraction approach used to recycle plutonium (Pu) and uranium (U) in used nuclear fuel worldwide. Finally, the selected chemometric model, partial least squares regression, accurately quantifies Nd 3+ with a low percentage error in the presence of interfering analytes and even under conditions that the training set does not describe.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

VALID Past, Present, and Future [Slides]

This presentation discusses the Verified, Archived Library of Inputs and Data (VALID) procedure which is managed per computational procedure under the overall SCALE quality assurance plan. The principle behind VALID is that two independent, qualified people prepare and review the model, while a dual, independent check acts as a primary barrier to prevent errors in the library. A final check by VALID QAC of procedural compliance is then performed. The future for VALID includes plans to add more experiments, a review for all ICSBEP cases containing Deuterium, ZEUS intermediate and fast spectrum evaluations, intermediate enrichment experiments, other cases of interest to sponsors, and mixed/intermediate spectrum experiments.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

AI-Driven Crack Detection for Remanufacturing Cylinder Heads Using Deep Learning and Engineering-Informed Data Augmentation

Detecting cracks in cylinder heads traditionally relies on manual inspection, which is time-consuming and susceptible to human error. As an alternative, automated object detection utilizing computer vision and machine learning models has been explored. However, these methods often face challenges due to a lack of sufficiently annotated training data, limited image diversity, and the inherently small size of cracks. Addressing these constraints, this paper introduces a novel automated crack-detection method that enhances data availability through a synthetic data generation technique. Unlike general data augmentation practices, our method involves copying cracks from one location to another, guided by both random and informed engineering decisions about likely crack formations due to cyclic thermomechanical loads. The innovative aspect of our approach lies in the integration of domain-specific engineering knowledge into the synthetic generation process, which substantially improves detection accuracy. We evaluate our method’s effectiveness using two metrics: the F2 score, which emphasizes recall to prioritize detecting all potential cracks, and mean average precision (MAP), a standard measure in object detection. Experimental results demonstrate that, without engineering insights, our method increases the F2 score from 0.40 to 0.65, while maintaining a stable MAP. Incorporating detailed engineering knowledge further enhances the F2 score to 0.70 and improves MAP to 0.57, representing increases of 63% and 43%, respectively. These results confirm that our approach not only mitigates the limitations of traditional data augmentation but also significantly advances the reliability and precision of crack detection in industrial settings.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2022

Towards a neutron multiplicity measurement with the Accelerator Neutrino Neutron Interaction Experiment

The Accelerator Neutrino Neutron Interaction Experiment (ANNIE) is a 26 ton Gadolinium (Gd)-loaded water Cherenkov detector located on the Booster Neutrino Beam line at Fermilab. Its main goals are the measurement of the neutron multiplicity in neutrino-nucleus interactions as well as the cross-section of Charged Current Quasi-Elastic (CCQE) neutrino interactions on water. Besides the physics goals, the experiment also aims to be a testbed for new technologies such as Large Area Picosecond Photodetectors (LAPPDs) and Water-based Liquid Scintillators (WbLS). This thesis presents a preliminary measurement of the neutron multiplicity with {ANNIE}, using an analysis conducted on a fraction of the 2021 beam year. As preparatory measures, the efficiency of ANNIE's Front Muon Veto (FMV) was determined to be {$\bar{\varepsilon}_{\mathrm{FMV}} = (95.6 \pm 1.6)\%$} while the average efficiency for active scintillator paddles in the Muon Range Detector (MRD) was found to be {$\bar{\varepsilon}_ {\mathrm{MRD}} = (92.1 \pm 7.9)\%$}. Furthermore, the simulation framework used for ANNIE was validated and adapted to reproduce the experimental data by comparing the detector response for samples of Michel electrons, Americium Beryllium neutrons, and through-going muons. The analysis finds average neutron yields of {$\bar{n}_{\mathrm{data}} (\mathrm{beam}) = (0.272 \pm 0.010_{\mathrm{stat}})$} for an inclusive set of all identified muon neutrino candidates and {$\bar{n}_{\mathrm{data}} (\mathrm{beam,FV}) = (0.287 \pm 0.044_{\mathrm{stat}})$ for interactions which happened inside of the Fiducial Volume of ANNIE, which was optimized to increase the neutron detection acceptance. The presented neutron multiplicity values represent the number of detected neutrons after all event selection cuts and are not yet corrected for the neutron detection efficiency. An equivalent analysis on a simulated beam sample predicts neutron yields of $\bar{n}_{\mathrm{MC}}(\mathrm{beam}) = (0.515 \pm 0.0 07_{\mathrm{stat}})$ and $\bar{n}_{\mathrm{MC}}(\mathrm{beam,FV}) = (0.627 \pm 0.031_{\mathrm{stat}})$, indicating that the models tend to overpredict the number of neutrons produced in such interactions. Systematic errors have been briefly considered to contribute {$\sigma_{\mathrm{sys,FMV}} \sim 0.01\,$neutrons/$\nu$-interaction} due to the slight FMV inefficiency and {$\sigma_{\mathrm{sys,n}} \sim 0.05\,$neutrons/$\nu$-interaction} due to the neutron detection efficiency. Simulation studies further highlighted the importance of neutron detection in Diffuse Supernova Background (DSNB) searches. A combination of neutron tagging and Convolutional Neural Networks was found to reduce the most relevant Neutral Current Quasi-Elastic (NCQE) interaction background below the signal level, achieving a Signal-to-Background ratio of 4:1. In a further study, we investigated the positive impact of a deployment of a WbLS target on the energy reconstruction in ANNIE. WbLS provides a scintillation signal from hadronic recoils in addition to the charged lepton that can be included in neutrino energy reconstruction. It was found that a deployed WbLS volume in ANNIE improves the neutrino energy reconstruction from 14\% to 12\%, with the potential of going beyond this if more sophisticated reconstruction algorithms are developed in the future.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Machine Learning Based Prediction of Airflow Maldistribution in A-Type Heat Exchangers

Airflow maldistribution is one of the primary causes of performance degradation in air-to refrigerant heat exchangers (HX) and has been shown to decrease heat transfer by as much as 35%. As a result, many units are oversized to meet the target capacity, resulting in increased system cost and refrigerant charge. Several studies have explored how characteristics like package type and HX geometry impact the flow profile, but results are restricted to a limited range of parameters and cannot be extrapolated to new designs. In this work, a machine learning (ML) model is trained to predict the inlet flow profile of dry air entering A-type HXs across a broad range of geometries and conditions. Flow profiles are generated using a porous media CFD model and used to train an Artificial Neural Network (ANN) which exhibits maximum and average relative L2 norm errors of 0.48 and 0.05. Additionally, these predictions take less than a second to generate resulting in a speed up factor of 2.42E5 compared to CFD. Component-level simulations are conducted to determine the performance degradation resulting from the predicted airflow maldistribution profiles. The new ML model will enable rapid and accurate prediction of performance degradation resulting from airflow maldistribution in A-type HXs, allowing for more accurate and cost-effective HX design.

42 ENGINEERING↗
DOE OSTI2025

RSPECT: A PROSPECT-based model incorporating the real structure of rice leaves

Radiative transfer models (RTMs) describe how light is absorbed, scattered, and transmitted within leaves by simulating mechanistic light propagation processes. The PROSPECT model is based on measurable parameters (the leaf biochemical content) and a non-measurable parameter (the leaf anatomical structure represented by the leaf structure parameter (N)). The effect of N on the optical properties of leaves has been investigated through a number of local and global sensitivity analyses. Other studies have directly evaluated the effect of the leaf anatomical structure on spectral reflectance, particularly in the near infrared region. However, the relationship between N and the anatomical structure is unclear. Here, in this study, we leveraged eLeaf, a ray tracing-based 3D rice leaf simulator, to establish relationships between leaf anatomical features and spectral properties, enabling us to replace N in the PROSPECT-4 model with measurable leaf anatomical parameters and develop the RSPECT model. The leaf thickness at minor vein, leaf thickness at bulliform cells, mesophyll thickness at minor vein, and distance between two minor veins could be used to predict N effectively. The RSPECT model achieved spectral simulation accuracy comparable to PROSPECT-4 and was more suitable for parameter inversion of the physical and chemical properties of rice leaves, with relative root mean square errors of 7.4% for chlorophyll content, 5.6% for equivalent water thickness, and 7.5% for dry matter content. In conclusion, RSPECT improves radiative transfer modeling by integrating measurable anatomical features and provides a framework for extending this approach to other vegetation types.

hyperspectral↗
DOE OSTI2020

Development of aerosol activation in the double-moment Unified Model and evaluation with CLARIFY measurements

Representing the number and mass of cloud and aerosol particles independently in a climate, weather prediction or air quality model is important in order to simulate aerosol direct and indirect effects on radiation balance. Here we introduce the first configuration of the UK Met Office Unified Model in which both cloud and aerosol particles have “double-moment” representations with prognostic number and mass. The GLObal Model of Aerosol Processes (GLOMAP) aerosol microphysics scheme, already used in the Hadley Centre Global Environmental Model version 3 (HadGEM3) climate configuration, is coupled to the Cloud AeroSol Interacting Microphysics (CASIM) cloud microphysics scheme. We demonstrate the performance of the new configuration in high-resolution simulations of a case study defined from the CLARIFY aircraft campaign in 2017 near Ascension Island in the tropical southern Atlantic. We improve the physical basis of the activation scheme by representing the effect of existing cloud droplets on the activation of new aerosol, and we also discuss the effect of unresolved vertical velocities. We show that neglect of these two competing effects in previous studies led to compensating errors but realistic droplet concentrations. While these changes lead only to a modest improvement in model performance, they reinforce our confidence in the ability of the model microphysics code to simulate the aerosol–cloud microphysical interactions it was designed to represent. Capturing these interactions accurately is critical to simulating aerosol effects on climate.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Detection of False Data Injection Attacks in Battery Stacks Using Input Noise-Aware Nonlinear State Estimation and Cumulative Sum Algorithms

Grid-scale battery energy storage systems (BESSs) are vulnerable to false data injection attacks (FDIAs), which could be used to disrupt state of charge (SoC) estimation. Inaccurate SoC estimation has negative impacts on system availability, reliability, safety, and the cost of operation. In this article a combination of a Cumulative Sum (CUSUM) algorithm and an improved input noise-aware extended Kalman filter (INAEKF) is proposed for the detection and identification of FDIAs in the voltage and current sensors of a battery stack. The series-connected stack is represented by equivalent circuit models, the SoC is modeled with a charge reservoir model and the states are estimated using the INAEKF. Further, the root mean squared error of the states’ estimation by the modified INAEKF was found to be superior to the traditional EKF. By employing the INAEKF, this article addresses the research gap that many state estimators make asymmetrical assumptions about the noise corrupting the system. Additionally, the INAEKF estimates the input allowing for the identification of FDIA, which many alternative methods are unable to achieve. The proposed algorithm was able to detect attacks in the voltage and current sensors in 99.16% of test cases, with no false positives. Utilizing the INAEKF compared to the standard EKF allowed for the identification of FDIA in the input of the system in 98.43% of test cases.

25 ENERGY STORAGE↗
DOE OSTI2021

Assessing boundary condition and parametric uncertainty in numerical-weather-prediction-modeled, long-term offshore wind speed through machine learning and analog ensemble

To accurately plan and manage wind power plants, not only does the time-varying wind resource at the site of interest need to be assessed but also the uncertainty connected to this estimate. Numerical weather prediction (NWP) models at the mesoscale represent a valuable way to characterize the wind resource offshore, given the challenges connected with measuring hub-height wind speed. The boundary condition and parametric uncertainty associated with modeled wind speed is often estimated by running a model ensemble. However, creating an NWP ensemble of long-term wind resource data over a large region represents a computational challenge. Here, we propose two approaches to temporally extrapolate wind speed boundary condition and parametric uncertainty using a more convenient setup in which a mesoscale ensemble is run over a short-term period (1 year), and only a single model covers the desired long-term period (20 year). We quantify hub-height wind speed boundary condition and parametric uncertainty from the short-term model ensemble as its normalized across-ensemble standard deviation. Then, we develop and apply a gradient-boosting model and an analog ensemble approach to temporally extrapolate such uncertainty to the full 20-year period, for which only a single model run is available. As a test case, we consider offshore wind resource characterization in the California Outer Continental Shelf. Both of the proposed approaches provide accurate estimates of the long-term wind speed boundary condition and parametric uncertainty across the region (R 2 >0.75), with the gradient-boosting model slightly outperforming the analog ensemble in terms of bias and centered root-mean-square error. At the three offshore wind energy lease areas in the region, we find a long-term median hourly uncertainty between 10 % and 14 % of the mean hub-height wind speed values. Finally, we assess the physical variability in the uncertainty estimates. In general, we find that the wind speed uncertainty increases closer to land. Also, neutral conditions have smaller uncertainty than the stable and unstable cases, and the modeled wind speed in winter has less boundary condition and parametric sensitivity than summer.

17 WIND ENERGY↗
DOE OSTI2022

Decoding Golden Eagle Movement Behavior from High-Resolution, Variable-Rate Telemetry Data Through Bayesian Filtering

The recent advances in animal tracking technology have enabled the collection of a vast amount of in situ data regarding the movement of wildlife at high spatiotemporal resolution. These data are usually available at variable time resolutions and contains noise (error) originating from GPS fixes. Decoding movement characteristics, particularly of flying animals, from telemetry data while handling these factors is a challenging yet important task for conservation purposes. Typically, this task is broken into two subtasks: resampling, and model calibration. The resampling subtask converts the variable rate positional data into a constant time interval data, while the model calibration subtask uses the resampled data to tune time-invariant parameters of the proposed models. For telemetry data at high temporal resolutions (order of 1 second), it is very challenging to decouple noise from actual movements using interpolation-based resampling techniques. Any errors introduced during resampling can significantly alter the the calibration and prediction attributes of the movement model. We address this problem through a unified Bayesian state-space framework that can handle both the resampling and calibration tasks in a single step. In addition, we use the speed and heading of the bird from telemetry data to regularize the position information of the bird. We use a Kalman filtering approach to include these nonlinearly related motion parameters within the state space framework. We cross-validated to quantify how this inclusion affects the model performance in estimating true bird movements. The relationship between the true state of the bird and environmental and topographical covariates is then represented parametrically. These parameters are then tuned using stochastic sampling strategies like Markov Chain Monte Carlo (MCMC). We use the telemetry data collected from golden eagles in the western USA to demonstrate the applicability of this approach to build a predictive, probabilistic movement model. Our preliminary results show that this approach provides improved predictive performance in terms of capturing higher-order motion parameters such as angular and horizontal accelerations, which may have simpler and more direct relationships with environmental covariates than corresponding speeds. In this talk, we will demonstrate how this state-space approach benefits the prediction capabilities of a movement model in simulating golden eagle paths through a wind power plant in Wyoming given certain atmospheric conditions. The model outcomes are aimed at informing mitigation strategies that can minimize the potential for collisions of golden eagles with wind turbines.

Bayesian methods↗
DOE OSTI2022

Modeling deformation and failure in AlSi-polyester abradable sealcoating material using microstructure-based finite element simulation

A plasma-sprayed aluminum-silicon (AlSi)/polyester coating is applied in modern gas turbine engines as an abradable sealcoating to maintain tight clearances between the rotating blades and the static casing. While running the engine, the rotating blades “rub” with the abradable coating, which results in extreme strain rate (up to 10 6 ) dynamics and a high-temperature environment. Due to the difficulty of collecting direct measurements, predictive computational models are important for analyzing the deformation and failure of the abradable material, to help meet the design target of avoiding damage to the blade tip and maintaining high fuel efficiency. In this research, a microstructure-based finite element (FE) computational model was developed to capture the complex mechanical behavior of the AlSi/polyester microstructure. The model is based on a virtual representative-volume-element (RVE) of a metal-polymer microstructure, reconstructed from x-ray computed tomography. It models the plastic deformation of, and damage to, each AlSi and polyester constituents, as well as the failure at their interface. The model was calibrated and validated with uniaxial tension and compression experiments, conducted at two temperatures (298 K and 533 K) at an applied strain rate of 10 3 - 10 4 s -1 . The material exhibited strongly asymmetric tension-compression behavior and a sensitivity to temperature, which was well captured by the model. The model was further applied to investigate changes in mechanical behavior due to variations in constituents’ volume fractions, which provides guidance to the microstructural design of AlSi/polyester abradable materials. The model is expected to facilitate the development of improved abradable materials by bypassing the conventional trial-and-error approach and extensive testing requirements.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

Limiting the number of quantisation levels of a harmonic lens as a method for improving the quality of the generated image

The characteristics (modulation transfer functions) of harmonic lenses made by direct laser patterning of resist are experimentally studied. The influence of technological errors in the manufacture of lenses on the formation of the point spread function is modelled for harmonic lenses with 32 and 256 quantisation levels. Direct laser writing is used to produce lenses with these numbers of levels. It is experimentally shown that the modulation transfer function of the harmonic lens with 32 quantisation levels is significantly greater at high spatial frequencies (50 mm{sup −1}). (paper)

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

Cognitive simulation models for inertial confinement fusion: Combining simulation and experimental data

The design space for inertial confinement fusion (ICF) experiments is vast, and experiments are extremely expensive. Researchers rely heavily on computer simulations to explore the design space in search of high-performing implosions. However, ICF multiphysics codes must make simplifying assumptions, and thus deviate from experimental measurements for complex implosions. For more effective design and investigation, simulations require input from past experimental data to better predict future performance. In this work, we describe a cognitive simulation method for combining simulation and experimental data into a common, predictive model. This method leverages a machine learning technique called “transfer learning,” the process of taking a model trained to solve one task, and partially retraining it on a sparse dataset to solve a different, but related task. In the context of ICF design, neural network models are trained on large simulation databases and partially retrained on experimental data, producing models that are far more accurate than simulations alone. Here, we demonstrate improved model performance for a range of ICF experiments at the National Ignition Facility and predict the outcome of recent experiments with less than 10% error for several key observables. We discuss how the methods might be used to carry out a data-driven experimental campaign to optimize performance, illustrating the key product—models that become increasingly accurate as data are acquired.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2020

Prediction of organic homolytic bond dissociation enthalpies at near chemical accuracy with sub-second computational cost

Bond dissociation enthalpies (BDEs) of organic molecules play a fundamental role in determining chemical reactivity and selectivity. However, BDE computations at sufficiently high levels of quantum mechanical theory require substantial computing resources. In this paper, we develop a machine learning model capable of accurately predicting BDEs for organic molecules in a fraction of a second. We perform automated density functional theory (DFT) calculations at the M06-2X/def2-TZVP level of theory for 42,577 small organic molecules, resulting in 290,664 BDEs. A graph neural network trained on a subset of these results achieves a mean absolute error of 0.58 kcal mol-1 (vs DFT) for BDEs of unseen molecules. We further demonstrate the model on two applications: first, we rapidly and accurately predict major sites of hydrogen abstraction in the metabolism of drug-like molecules, and second, we determine the dominant molecular fragmentation pathways during soot formation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

An Investigation of LES Wall Modeling for Rayleigh–Bénard Convection via Interpretable and Physics-Aware Feedforward Neural Networks with DNS

Abstract The traditional approach of using the Monin–Obukhov similarity theory (MOST) to model near-surface processes in large-eddy simulations (LESs) can lead to significant errors in natural convection. In this study, we propose an alternative approach based on feedforward neural networks (FNNs) trained on output from direct numerical simulation (DNS). To evaluate the performance, we conduct both a priori and a posteriori tests. In the a priori (offline) tests, we compare the statistics of the surface shear stress and heat flux, computed from filtered DNS input variables, to the stress and flux obtained from the filtered DNS. Additionally, we investigate the importance of various input features using the Shapley additive explanations value and the conditional average of the filter grid cells. In the a posteriori (online) tests, we implement the trained models in the System for Atmospheric Modeling (SAM) LES and compare the LES-generated surface shear stress and heat flux with those in the DNS. Our findings reveal that vertical velocity, a traditionally overlooked flow quantity, is one of the most important input features for determining the wall fluxes. Increasing the number of input features improves the a priori test results but does not always improve the model performance in the a posteriori tests because of the differences in input variables between the LES and DNS. Last, we show that physics-aware FNN models trained with logarithmic and scaled parameters can well extrapolate to more intense convection scenarios than in the training dataset, whereas those trained with primitive flow quantities cannot. Significance Statement The traditional near-surface turbulence model, based on a shear-dominated boundary layer flow, does not represent near-surface turbulence in natural convection. Using a feedforward neural network (FNN), we can construct a more accurate model that better represents the near-surface turbulence in various flows and reveals previously overlooked controlling factors and process interactions. Our study shows that the FNN-generated models outperform the traditional model and highlight the importance of the near-surface vertical velocity. Furthermore, the physics-aware FNN models exhibit the potential to extrapolate to convective flows of various intensities beyond the range of the training dataset, suggesting their broader applicability for more accurate modeling of near-surface turbulence.

54 ENVIRONMENTAL SCIENCES↗