Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Test Data”

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 919 records · Page 51

Pulse Jet Mixing Tests With Noncohesive Solids (Revised)

This report summarizes results from pulse jet mixing (PJM) tests with noncohesive solids in Newtonian liquid. The tests were conducted during FY 2007 and 2008 to support the design of mixing systems for the Hanford Waste Treatment and Immobilization Plant (WTP). Tests were conducted at three geometric scales using noncohesive simulants, and the test data were used to develop models predicting two measures of mixing performance for full-scale WTP vessels. The models predict the cloud height (the height to which solids will be lifted by the PJM action) and the critical suspension velocity (the minimum velocity needed to ensure all solids are suspended off the floor, though not fully mixed). From the cloud height, the concentration of solids at the pump inlet can be estimated. The predicted critical suspension velocity for lifting all solids is not precisely the same as the mixing requirement for 'disturbing' a sufficient volume of solids, but the values will be similar and closely related. These predictive models were successfully benchmarked against larger scale tests and compared well with results from computational fluid dynamics simulations. The application of the models to assess mixing in WTP vessels is illustrated in examples for 13 distinct designs and selected operational conditions. The values selected for these examples are not final; thus, the estimates of performance should not be interpreted as final conclusions of design adequacy or inadequacy. However, this work does reveal that several vessels may require adjustments to design, operating features, or waste feed properties to ensure confidence in operation. The models described in this report will prove to be valuable engineering tools to evaluate options as designs are finalized for the WTP. Revision 1 refines data sets used for model development and summarizes models developed since the completion of Revision 0.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Combined crystal plasticity and grain boundary modeling of creep in ferritic-martensitic steels: II. The effect of stress and temperature on engineering and microstructural properties

This paper describes a series of physically-based crystal plasticity finite element method (CPFEM) simulations of long-term creep and creep rupture of Grade 91 steel. It is Part 2 of a two part series of papers. Part 1 describes the simulation framework; this part focuses on specific simulations and on how the predicted long-term creep properties of Grade 91 compare to the assumptions used in current high temperature design practices. This work extends the model developed in Part 1 to look at creep properties at different temperatures, principal stresses, and multiaxial stress states. The simulations show that empirically extrapolating creep rupture stresses from short-term experimental data may substantially over predict the actual long-term creep properties of Grade 91. Additionally, the CPFEM calculations predict a transition from notch strengthening creep behavior for high values of maximum principal stress and moderate notch severity to notch weakening behavior for low principal stresses and more severe notches. The latter regime better categorizes conditions in engineering components designed for long term elevated temperature use, which implies Grade 91 may be a notch weakening material in actual service. This would have a significant impact on high temperature design practices, though confirmatory test data on long-life, low stress notched specimens is difficult to obtain. Finally, one advantage of the physically-based modeling approach adopted here is that the simulation results also elucidate the microstructural mechanisms causing the macroscopic trends in engineering properties predicted by the simulations. This paper shows that the detailed micromechanical mechanisms predicted by the CPFEM simulations can be abstracted with a simple micromechanical model that can be used to both explain the detailed results and make improved predictions of engineering properties from experimental data.

36 MATERIALS SCIENCE↗

Classification Analytics of Pu-239 and U-235 Source Signatures Using Gamma Spectral Regions

Machine learning detection methods using gamma signatures from spectral measurements of low-intensity Pu-239 and U-235 sources are studied. NaI detectors located at different distances fromthe source have been used to collect the training and independent testing data sets. The source is introduced via a shielded conduit into the facility where it is surrounded by 21 NaI detectors deployed over 6 x 6 meters area in the formation of two concentric circles and a spiral. The counts in gamma spectral regions associated with these two sources are estimated at 1 second intervals for each NaI detector, and are used as classifier features for detecting the source presence. Eight different classifiers with five basic properties — namely, smooth, non-smooth, statistical, structural, and hyper-parameter tuning — are trained and tested using the background and source measurements collected over multiple experimental runs. While the overall classifier performance improved as detectors closer to the source are used, some identically produced detectors under-performed but differently between two sources. Some classifiers achieved lower training error but their testing error based on independent measurements is higher for both sources. Overall, these results indicate significant over-fitting by these methods, and illustrate the complexity of training and selecting the machine learning methods to solve these detection problems.

Rao, Nageswara↗

Guided construction of single cell reference for human and mouse lung

Accurate cell type identification is a key and rate-limiting step in single-cell data analysis. Single-cell references with comprehensive cell types, reproducible and functionally validated cell identities, and common nomenclatures are much needed by the research community for automated cell type annotation, data integration, and data sharing. Here, we develop a computational pipeline utilizing the LungMAP CellCards as a dictionary to consolidate single-cell transcriptomic datasets of 104 human lungs and 17 mouse lung samples to construct LungMAP single-cell reference (CellRef) for both normal human and mouse lungs. CellRefs define 48 human and 40 mouse lung cell types catalogued from diverse anatomic locations and developmental time points. We demonstrate the accuracy and stability of LungMAP CellRefs and their utility for automated cell type annotation of both normal and diseased lungs using multiple independent methods and testing data. We develop user-friendly web interfaces for easy access and maximal utilization of the LungMAP CellRefs.

59 BASIC BIOLOGICAL SCIENCES↗

Data-driven organic solubility prediction at the limit of aleatoric uncertainty

Abstract Small molecule solubility is a critically important property which affects the efficiency, environmental impact, and phase behavior of synthetic processes. Experimental determination of solubility is a time- and resource-intensive process and existing methods for in silico estimation of solubility are limited by their generality, speed, and accuracy. This work presents two models derived from the FASTPROP and CHEMPROP architectures and trained on BigSolDB which are capable of predicting solubility at arbitrary temperatures for a wide range of small molecules in organic solvent. Both extrapolate to unseen solutes 2–3 times more accurately than the current state-of-the-art model and we demonstrate that they are approaching the aleatoric limit (0.5–1$$\log S$$ log S ) of available test data, suggesting that further improvements in prediction accuracy require more accurate datasets. The FASTPROP-derived model (called FASTSOLV) and the CHEMPROP-based model are open source, freely accessible via a Python package and web interface, highly reproducible, and up to 2 orders of magnitude faster than current alternatives.

Science & Technology - Other Topics↗

Physics-Based Machine Learning Methods for U-235 Forensics Signatures

Signatures of low-intensity U-235 sources have been recently studied by utilizing a variety of machine learning (ML) classifiers using features derived from gamma spectral measurements collectedunder structured campaigns. Several ML classifiers, such as ensemble of tress and classification trees, revealed misleadingly-optimistic training error due to over-fitting, and furthermore,their performance is not directly relatable to the physical properties due to their data-driven, opaque designs. We present a regression-based ML method that first estimates the inverse distanceto the source and then utilizes a threshold to infer its presence, by representing the background as a source located at an infinite distance. For the inverse distance estimation, we study the ensembleof trees and Gaussian process regression methods, and a hyper parameter auto-tuning and selection method that employs five regression estimators. These methods avoid the over-fittingobserved in several ML classifiers, while providing the classification error nearly comparable to them based on independent test data. Their error is directly related to estimates of the inversephysical distance to source, and the precision of error determines the seperability property that determines the false alarm and missed detection rates. The property of monotonic decrease of thesource strength with increasing detector distance combined with Poisson distribution of measurements is utilized to analytically validate these methods by deriving the generalization equations ofunderlying regression methods.

Rao, Nageswara↗

Accurate Machine Learning for Predicting the Viscosities of Deep Eutectic Solvents

Deep eutectic solvents (DESs) are emerging as environmentally friendly designer solvents for mass transport and heat transfer processes in industrial applications; however, the lack of accurate tools to predict and thus control their viscosities under both a range of environmental factors and formulations hinders their general application. While DESs may serve as designer solvents, with nearly unlimited combinations, this unfortunately makes it experimentally infeasible to comprehensively measure the viscosities of all DESs of potential industrial interest. To assist in the design of DESs, we have developed several new machine learning (ML) models that accurately and rapidly predict the viscosities of a diverse group of DESs at different temperatures and molar ratios using, to date, one of the most comprehensive data sets containing the properties of over 670 DESs over a wide range of temperatures (278.15–385.25 K). Three ML models, including support vector regression (SVR), feed forward neural networks (FFNNs), and categorical boosting (CatBoost), were developed to predict DES viscosity as a function of temperature and molar ratio and contrasted with multilinear and two-factor polynomial regression baselines. Further, quantum chemistry-based, COSMO-RS-derived sigma profile (σ-profile) features were used as inputs for the ML models. The CatBoost model is excellent at externally predicting DES viscosity, as indicated by high R 2 (0.99) and low root-mean-square-error (RMSE) and average absolute relative deviations (AARD) (5.22%) values for the testing data sets, and 98% of the data points lie within the 15% of AARD deviations. Furthermore, SHapley additive explanation (SHAP) analysis was employed to interpret the ML results and rationalize the viscosity predictions. The result is an ML approach that accurately predicts viscosity and will aid in accelerating the design of appropriate DESs for industrial applications.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Summary of Savannah River Site FY23 Salt Waste Qualification Data

The Savannah River National Laboratory (SRNL) analyzed samples from Savannah River Site (SRS) Waste Tanks 41H and 21H to support qualification of Salt Waste Processing Facility (SWPF) Waste Batches 8 and 9 for processing (the FY23 Salt Batch Qualification samples). These Tanks (i.e. 41H and 21H) are blend tanks for feed to SWPF. None of the samples displayed any unusual or unexpected characteristics such as large amounts of solids, floating solids, or unusual color. Characterization of these samples confirmed similar chemical composition and characteristics to previous salt waste batches. The results for Batches 8 and 9 were provided by SRNL to Savannah River Mission Completion (SRMC), the Liquid Waste Operations subcontractor at SRS, as External Sample Results (Laboratory Information Management System (LIMS)) Reports. Additionally, a separate technical memo was issued by SRNL to report re-test data for Batch 8 for Cs-137 for filtered samples only which were run at the request of SWPF. For Batch 9, a set of samples was also analyzed in parallel by the SWPF-Analytical Laboratory (SWPF-AL). The SWPF-AL data is included herein for comparison with the SRNL data where applicable. The analytical results (both rapid, typically 4 weeks, and long term, typically 8 weeks) for Batches 8 and 9 are now summarized and discussed in this technical report.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Selection of site specific vibration equation by using analytic hierarchy process in a quarry

This paper presents a new approach for the selection of the most accurate SSVA (Site Specific Vibration Attenuation) equation for blasting processes in a quarry located near settlements in Istanbul, Turkey. In this context, the SSVA equations obtained from the same study area in the literature were considered in terms of distance between the shot points and buildings and the amount of explosive charge. In this purpose, 11 different SSVA equations obtained from the study area in the past 12 years, forecasting capabilities according to designated new conditions, using 102 vibration records as test data obtained from the study area was investigated. In this study, AHP (Analytic Hierarchy Process) was selected as an analysis method in order to determine the most accurate equation among 11 SSAV equations, and the parameters such as year, distance, charge, and r{sup 2} of the equations were used as criteria for AHP. Finally, the most appropriate equation was selected among the existing ones, and the process of selecting according to different target criteria was presented. Furthermore, it was noted that the forecasting results of the selected equation is more accurate than that formed using the test results. - Highlights: • The optimum Site Specific Vibration Attenuation equation for blasting in a quarry located near settlements was determined. • It is indicated that SSVA equations changing over the years don’t give always accurate estimates at changing conditions. • Selection of the blast induced SSVA equation was made using AHP. • Equation selection method was highlighted based on parameters such as charge, distance, and quarry geometry changes (year).

54 ENVIRONMENTAL SCIENCES↗

Physics-Based Machine Learning Methods for U-235 Forensics Signatures

Signatures of low-intensity U-235 sources have been recently studied by utilizing a variety of machine learning (ML) classifiers using features derived from gamma spectral measurements collected under structured campaigns. Several ML classifiers, such as ensemble of tress and classification trees, revealed misleadingly-optimistic training error due to over-fitting, and furthermore, their performance is not directly relatable to the physical properties due to their data-driven, opaque designs. We present a regression-based ML method that first estimates the inverse distance to the source and then utilizes a threshold to infer its presence, by representing the background as a source located at an infinite distance. For the inverse distance estimation, we study the ensemble of trees and Gaussian process regression methods, and a hyper parameter auto-tuning and selection method that employs five regression estimators. These methods avoid the over-fitting observed in several ML classifiers, while providing the classification error nearly comparable to them based on independent test data. Their error is directly related to estimates of the inverse physical distance to source, and the precision of error determines the seperability property that determines the false alarm and missed detection rates. The property of monotonic decrease of the source strength with increasing detector distance combined with Poisson distribution of measurements is utilized to analytically validate these methods by deriving the generalization equations of underlying regression methods.

Rao, Nageswara↗

Science based stockpile stewardship, uncertainty quantification, and fission fragment beams

Stewardship of this nation's nuclear weapons is predicated on developing a fundamental scientific understanding of the physics and chemistry required to describe weapon performance without the need to resort to underground nuclear testing and to predict expected future performance as a result of intended or unintended modifications. In order to construct more reliable models, underground nuclear test data is being reanalyzed in novel ways. The extent to which underground experimental data can be matched with simulations is one measure of the credibility of our capability to predict weapon performance. To improve the interpretation of these experiments with quantified uncertainties, improved nuclear data is required. As an example, the fission yield of a device was often determined by measuring fission products. Conversion of the measured fission products to yield was accomplished through explosion code calculations (models) and a good set of nuclear reaction cross-sections. Because of the unique high-fluence environment of an exploding nuclear weapon, many reactions occurred on radioactive nuclides, for which only theoretically calculated cross-sections are available. Inverse kinematics reactions at CARIBU offer the opportunity to measure cross-sections on unstable neutron-rich fission fragments and thus improve the quality of the nuclear reaction cross-section sets. One of the fission products measured was {sup 95}Zr, the accumulation of all mass 95 fission products of Y, Sr, Rb and Kr (see Fig. 1). Subsequent neutron-induced reactions on these short lived fission products were assumed to cancel out - in other words, the destruction of mass 95 nuclides was more or less equal to the production of mass 95 nuclides. If a {sup 95}Sr was destroyed by an (n,2n) reaction it was also produced by (n,2n) reactions on {sup 96}Sr, for example. However, since these nuclides all have fairly short half-lives (seconds to minutes or even less), no experimental nuclear reaction cross-sections exist, and only theoretically modeled cross-sections are available. Inverse kinematics reactions at CARIBU offer the opportunity, should the beam intensity be sufficient, to measure cross-sections on a few important nuclides in order to benchmark the theoretical calculations and significantly improve the nuclear data. The nuclides in Fig. 1 are prioritized by importance factor and displayed in stoplight colors, green the highest and red the lowest priority.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Risk assessment of turbine rotor failure using probabilistic ultrasonic non-destructive evaluations

The study presents a method and application of risk assessment methodology for turbine rotor fatigue failure using probabilistic ultrasonic nondestructive evaluations. A rigorous probabilistic modeling for ultrasonic flaw sizing is developed by incorporating the model-assisted probability of detection, and the probability density function (PDF) of the actual flaw size is derived. Two general scenarios, namely the ultrasonic inspection with an identified flaw indication and the ultrasonic inspection without flaw indication, are considered in the derivation. To perform estimations for fatigue reliability and remaining useful life, uncertainties from ultrasonic flaw sizing and fatigue model parameters are systematically included and quantified. The model parameter PDF is estimated using Bayesian parameter estimation and actual fatigue testing data. The overall method is demonstrated using a realistic application of steam turbine rotor, and the risk analysis under given safety criteria is provided to support maintenance planning.

42 ENGINEERING↗

Inferring safety critical events from vehicle kinematics in naturalistic driving environment: Application of deep learning Algorithms

Advances in sensing technology has enabled the collection of countless terabytes of second-by-second kinematics data. Such data provides opportunities for real-time monitoring of driving behavior and identification of safety critical events (SCEs) including crashes and near crashes. The concept of volatility is relevant in this context, which identifies instability and erratic variations in driving behavior prior to involvement in SCEs. This study utilized vehicle kinematics from a large-scale naturalistic driving data to develop a deep learning approach based on 1D convolutional neural networks (CNN) for inferring SCEs. The data are unique in the sense that such accurate pre-crash data at high fidelity are not available in traditional crash repositories. This study contributes to the literature by providing a first attempt at predicting responses to SCEs by developing deep learning-based CNN architectures using novel driving volatility based kinematic thresholds for a sample of 9553 events. The key contribution lies in developing a volatility-based CNN input layout that is acceptable to CNN schemes and represents the motion kinematics such as speed, acceleration and volatility measures. Several 1D-CNN architectures were developed using layers, numbers of convolutions, layer patterns, and kernels. Shallow and deep architectures were tested, revealing higher accuracy of shallow architectures in detecting SCEs. The optimal number of epochs were identified using an early stopping method while the CNN performance was improved by increasing the number of epochs. The ensemble CNN had the highest predictive accuracy of 95.6% for detection of crashes and near crashes, which was 2.5% higher than the optimal CNN using 20% hold out test data. The ensemble CNN also outperformed classical machine learning models and model performance reported in past studies on detection of SCEs. Finally, these results have implications for identification of safety hotspots and providing real-time alerts and warnings in connected and highly automated vehicle environment including society of automotive engineers levels 3–5.

42 ENGINEERING↗

Integral and Separate Effects Test Facilities To Support Water Cooled Small Modular Reactors: A Review

This study reviews previous experimental facilities and test programs relevant to water-cooled reactor system design and analysis to meet regulatory compliances. This study aims to find the best solution for designing the required experiments, obtaining necessary test data, and verifying the developed computer code/models to support the new reactor design and development while minimizing cost and time while leveraging experiences from previous facilities to minimize. Nuclear reactor licensing requires supportive design, analysis, and experimental results to ensure the safety of the full-scale prototype reactor in regular operation, as well as during postulated accident scenarios. These reactor design analyses are generally performed using system codes and other associated simulation tools that require assessment, verification and validation using an appropriate experimental dataset. Experimental facilities used for reactor system safety analysis and system code assessments are categorized into integral effects test (IET) and separate effects test (SET) facilities. Further, these IET and SET experiments and studies use geometrically scaled systems to reproduce the prototype system behavior at a reasonable cost, albeit with some scaling-related distortions. The design challenge of these model facilities is to identify and minimize scaling distortions while reproducing the most important operational phenomena in steady-state operation and in postulated accident scenarios. Lessons learned from previous experimental facilities, models, and correlations can support the development of new multipurpose, scaled, hybrid, integrated, and modular experimental facilities for advanced light water-cooled small modular reactors (SMRs). Successful operation of these facilities can significantly reduce upfront reactor development and demonstration costs and time to deployment.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Novel Low-Cost Process for the Gasification of Biomass and Low-Rank Coals

Farm Energy envisaged a phased demonstration program, in which a pilot-scale straw gasifier will be installed on a farm. The synthesis gas product will be used to initially (i) generate electricity in a 300 kW diesel generator, and subsequently (ii) used as a feedstock to produce ethanol or mixed alcohols. They were seeking straw gasification and alcohol synthesis technologies that may be implemented on farm-scale. The consortium, along with the USDA ARS station in Corvallis, OR, expressed interest in the dual-bed gasification concept promoted by WRI and Taylor Energy, LLC. This process operated at atmospheric pressure and employed a solids-circulation type oxidation/reduction cycle significantly different from traditional fluidized-bed or up-draft type gasification reactors. The objectives of this project were to perform bench-scale testing to determine technical feasibility of gasifier concept, to characterize the syngas product, and to determine the optimal operating conditions and configuration. We used the bench-scale test data to complete a preliminary design and cost estimate for a 1-2 ton per hour pilot-scale unit that is also appropriate for on-farm scale applications. The gasifier configuration with the 0.375-inch stainless steel balls recirculating media worked consistently and for periods up to six hours of grass feed. The other principle systems like the boiler, the air pump, and feeder device also worked consistently during all feeding operations. Minor hiccups during operation tended to come from secondary systems like the flare or flammable material buildup in the exit piping. Although we did not complete the extended hour tests to 24 or 48 hours due to time and budget constraints, we developed the confidence that the gasifier in its current configuration could handle those tests. At the modest temperatures we operated the gasifier, slagging was not a problem. The solid wastes were dry and low density. The majority of the fixed carbon from the grass ended up in the solid waste collected in the external cyclone. The volatiles were almost all removed in the gasifier. While the average gas heating value of the collected gas products was 50 BTUs/scf or less, addition a of the second gas exit for combustion gases would increase that value by a factor of two or three. Other changes to the current design such as shortening the gasifier body and draft tube would lead to lower air use and shorter heating times. There was no evidence of steam reforming at the current operating temperature. Likewise there was no indication of significant tar production. Reconfiguration of the gasifier at the on farm site may yet yield more significant results that would better qualify this gasifier for small scale biomass operations.

09 BIOMASS FUELS↗

Comparison of algorithms for determination of rotation measure and Faraday structure. I. 1100–1400 MHz

Faraday rotation measures (RMs) and more general Faraday structures are key parameters for studying cosmic magnetism and are also sensitive probes of faint ionized thermal gas. A definition of which derived quantities are required for various scientific studies is needed, as well as addressing the challenges in determining Faraday structures. A wide variety of algorithms has been proposed to reconstruct these structures. In preparation for the Polarization Sky Survey of the Universe's Magnetism (POSSUM) to be conducted with the Australian Square Kilometre Array Pathfinder and the ongoing Galactic Arecibo L-band Feeds Array Continuum Transit Survey (GALFACTS), we run a Faraday structure determination data challenge to benchmark the currently available algorithms, including Faraday synthesis (previously called RM synthesis in the literature), wavelet, compressive sampling, and QU-fitting. The input models include sources with one Faraday thin component, two Faraday thin components, and one Faraday thick component. The frequency set is similar to POSSUM/GALFACTS with a 300 MHz bandwidth from 1.1 to 1.4 GHz. We define three figures of merit motivated by the underlying science: (1) an average RM weighted by polarized intensity, RM{sub wtd}, (2) the separation Δϕ of two Faraday components, and (3) the reduced chi-squared χ{sub r}{sup 2}. Based on the current test data with a signal-to-noise ratio of about 32, we find the following. (1) When only one Faraday thin component is present, most methods perform as expected, with occasional failures where two components are incorrectly found. (2) For two Faraday thin components, QU-fitting routines perform the best, with errors close to the theoretical ones for RM{sub wtd} but with significantly higher errors for Δϕ. All other methods, including standard Faraday synthesis, frequently identify only one component when Δϕ is below or near the width of the Faraday point-spread function. (3) No methods as currently implemented work well for Faraday thick components due to the narrow bandwidth. (4) There exist combinations of two Faraday components that produce a large range of acceptable fits and hence large uncertainties in the derived single RMs; in these cases, different RMs lead to the same Q, U behavior, so no method can recover a unique input model. Further exploration of all these issues is required before upcoming surveys will be able to provide reliable results on Faraday structures.

79 ASTRONOMY AND ASTROPHYSICS↗

Efficient mapping between void shapes and stress fields using Deep Convolutional Neural Networks with sparse data

Establishing fast and accurate structure-to-property relationships is an important component in the design and discovery of advanced materials. Physics-based simulation models like the finite element method (FEM) are often used to predict deformation, stress, and strain fields as a function of material microstructure in material and structural systems. Such models may be computationally expensive and time intensive if the underlying physics of the system is complex. This limits their application to solve inverse design problems and identify structures that maximize performance. In such scenarios, surrogate models are employed to make the forward mapping computationally efficient to evaluate. However, the high dimensionality of the input microstructure and the output field of interest often renders such surrogate models inefficient, especially when dealing with sparse data. Deep convolutional neural network (CNN) based surrogate models have shown great promise in handling such high-dimensional problems. In this paper, a single ellipsoidal void structure under a uniaxial tensile load represented by a linear elastic, high-dimensional and expensive-to-query, FEM model. We consider two deep CNN architectures, a modified convolutional autoencoder framework with a fully connected bottleneck and a UNet CNN, and compare their accuracy in predicting the von Mises stress field for any given input void shape in the FEM model. Additionally, a sensitivity analysis study is performed using the two approaches, where the variation in the prediction accuracy on unseen test data is studied through numerical experiments by varying the number of training samples from 20 to 100.

surrogate modeling; convolutional neural networks;↗