SEARCH · Engineering Papers

Results for “Statistical accuracy”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,387 records · Page 77

All sources NASA NTRS DOE OSTI

NASA NTRS2005

Minimizing Uncertainty in Cryogenic Surface Figure Measurement

A new facility at the Goddard Space Flight Center is designed to measure with unusual accuracy the surface figure of mirrors at cryogenic temperatures down to 12 K. The facility is currently configured for spherical mirrors with a radius of curvature (ROC) of 600 mm, and apertures of about 150 mm or less. The goals of the current experiment were to 1) Obtain the best possible estimate of test mirror surface figure, S(x,y) at 87 K and 20 K; 2) Obtain the best estimate of the cryo-change, Delta (x,y): the change in surface figure between room temperature and the two cryo-temperatures; and 3) Determine the uncertainty of these measurements, using the definitions and guidelines of the ISO Guide to the Expression of Uncertainty in Measurement. A silicon mirror was tested, and the cry-change from room temperature to 20K was found to be 3.7 nm rms, with a standard uncertainty of 0.23 nm in the rms statistic. Both the cryo-change figure and the uncertainty are among the lowest such figures yet published. This report describes the facilities, experimental methods, and uncertainty analysis of the measurements.

Blake, Peter↗

NASA NTRS1973

Reflectance of vegetation, soil, and water

The author has identified the following significant results. The ability to read the 24-channel MSS CCT tapes, select specified agricultural land use areas from the CCT, and perform multivariate statistical and pattern recognition analyses has been demonstrated. The 5 optimum channels chosen for classifying an agricultural scene were, in the order of their selection the far red visible, short reflective IR, visible blue, thermal infrared, and ultraviolet portions of the electromagnetic spectrum, respectively. Although chosen by a training set containing only vegetal categories, the optimum 4 channels discriminated pavement, water, bare soil, and building roofs, as well as the vegetal categories. Among the vegetal categories, sugar cane and cotton had distinctive signatures that distinguished them from grass and citrus. Acreages estimated spectrally by the computer for the test scene were acceptably close to acreages estimated from aerial photographs for cotton, sugar cane, and water. Many nonfarmable land resolution elements representing drainage ditch, field road, and highway right-of-way as well as farm headquarters area fell into the grass, bare soil plus weeds, and citrus categories and lessened the accuracy of the farmable acreage estimates in these categories. The expertise developed using the 24-channel data will be applied to the ERTS-1 data.

Wiegand, C. L.↗

NASA NTRS1981

A statistical analysis of the low-energy geosynchronous plasma environment. I - Electrons. II - Ions

Data on the geosynchronous plasma environment between approximately 30 eV and 80 keV are analyzed statistically. Nearly 50 days of 10 minute averages of the first four moments of the distribution function from low-energy plasma detectors on the ATS-5 and ATS-6 geosynchronous satellites were used. The data were studied in terms of occurrence frequency, local time variations, and response to geomagnetic activity. These techniques revealed marked differences in the ATS-5 and ATS-6 data bases. When translated into a 2-Maxwellian representation, it was found that (1) the ATS-6 data covered an energy range not covered by ATS-5 (between 1 and 50 eV) and (2) there was a definite change in the ion plasma between the ATS-5 and ATS-6 measurements. Simple expressions are derived to simulate the relationship between the four moments, and a model is presented, which takes into account the ATS-5 and ATS-6 plasma variations. Despite the differences in the ion data, the accuracy of the four-moment representation was found valid for characterizing the geosynchronous electron population.

Garrett, H. B.↗

NASA NTRS1985

Application of closed-form solutions to a mesh point field in silicon solar cells

A computer simulation method is discussed that provides for equivalent simulation accuracy, but that exhibits significantly lower CPU running time per bias point compared to other techniques. This new method is applied to a mesh point field as is customary in numerical integration (NI) techniques. The assumption of a linear approximation for the dependent variable, which is typically used in the finite difference and finite element NI methods, is not required. Instead, the set of device transport equations is applied to, and the closed-form solutions obtained for, each mesh point. The mesh point field is generated so that the coefficients in the set of transport equations exhibit small changes between adjacent mesh points. Application of this method to high-efficiency silicon solar cells is described; and the method by which Auger recombination, ambipolar considerations, built-in and induced electric fields, bandgap narrowing, carrier confinement, and carrier diffusivities are treated. Bandgap narrowing has been investigated using Fermi-Dirac statistics, and these results show that bandgap narrowing is more pronounced and that it is temperature-dependent in contrast to the results based on Boltzmann statistics.

Lamorte, M. F.↗

DOE OSTI2022

A new benchmark of soft X-ray transition energies of $$\mathrm {Ne}$$, $$\mathrm {CO}_2$$, and $$\mathrm {SF}_6$$: paving a pathway towards ppm accuracy

Abstract A key requirement for the correct interpretation of high-resolution X-ray spectra is that transition energies are known with high accuracy and precision. We investigate the K-shell features of $$\mathrm {Ne}$$ Ne , $$\mathrm {CO}_2$$ CO 2 , and $$\mathrm {SF}_6$$ SF 6 gases, by measuring their photo ion-yield spectra at the BESSY II synchrotron facility simultaneously with the 1s– n p fluorescence emission of He-like ions produced in the Polar-X EBIT. Accurate ab initio calculations of transitions in these ions provide the basis of the calibration. While the $$\mathrm {CO}_2$$ CO 2 result agrees well with previous measurements, the $$\mathrm {SF}_6$$ SF 6 spectrum appears shifted by $$\sim $$ ∼ 0.5 eV, about twice the uncertainty of the earlier results. Our result for $$\mathrm {Ne}$$ Ne shows a large departure from earlier results, but may suffer from larger systematic effects than our other measurements. The molecular spectra agree well with our results of time-dependent density functional theory. We find that the statistical uncertainty allows calibrations in the desired range of 1–10 meV, however, systematic contributions still limit the uncertainty to $${\sim }$$ ∼ 40–100 meV, mainly due to the temporal stability of the monochromator energy scale. Combining our absolute calibration technique with a relative energy calibration technique such as photoelectron energy spectroscopy will be necessary to realize its full potential of achieving uncertainties as low as 1–10 meV. Graphical abstract

74 ATOMIC AND MOLECULAR PHYSICS↗

DOE OSTI2024

Precise cosmological constraints from BOSS galaxy clustering with a simulation-based emulator of the wavelet scattering transform

For this study, we perform a reanalysis of the BOSS CMASS DR12 galaxy dataset using a simulation-based emulator for the wavelet scattering transform (WST) coefficients. Moving beyond our previous works, which laid the foundation for the first galaxy clustering application of this estimator, we construct a neural net-based emulator for the cosmological dependence of the WST coefficients and the 2-point correlation function multipoles, trained from the state-of-the-art suite of abacussummit simulations combined with a flexible halo occupation distribution (HOD) galaxy model. In order to confirm the accuracy of our pipeline, we subject it to a series of thorough internal and external mock parameter recovery tests, before applying it to reanalyze the CMASS observations in the redshift range 0.46 < z < 0.57. We find that a joint WST+2-point correlation function likelihood analysis allows us to obtain marginalized 1⁢σ errors on the Λ⁢ CDM parameters that are tighter by a factor of 2.5–6, compared to the 2-point correlation function, and by a factor of 1.4–2.5 compared to the WST-only results. This corresponds to a competitive 0.9%, 2.3% and 1% level of determination for parameters ω c , ⁢σ 8 &n s , respectively, and also to a 0.7% and 2.5% constraint on derived parameters h and ƒ⁡(z)⁢⁢σ 8 ⁡(z), in agreement with the Planck 2018 results. Our results reaffirm the constraining power of the WST and highlight the exciting prospect of employing higher-order statistics in order to fully exploit the power of upcoming stage-IV spectroscopic observations.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2024

Precise relative magnitude measurement improves fracture characterization during hydraulic fracturing

SUMMARY Microseismic monitoring is an important technique to obtain detailed knowledge of in-situ fracture size and orientation during stimulation to maximize fluid flow throughout the rock volume and optimize production. Furthermore, considering that the frequency of earthquake magnitudes empirically follows a power law (i.e. Gutenberg–Richter), the accuracy of microseismic event magnitude distributions is potentially crucial for seismic risk management. In this study, we analyse microseismicity observed during four hydraulic fracture treatments of the legacy Cotton Valley experiment in 1997 at the Carthage gas field of East Texas, where fractures were activated at the base of the sand-shale Upper Cotton Valley formation. We perform waveform cross-correlation to detect similar event clusters, measure relative amplitude from aligned waveform pairs with a principal component analysis, then measure precise relative magnitudes. The new magnitudes significantly reduce the deviations between magnitude differences and relative amplitudes of event pairs. This subsequently reduces the magnitude differences between clusters located at different depths. Reduction in magnitude differences between clusters suggests that some attenuation-related biases could be effectively mitigated with relative magnitude measurements. The maximum likelihood method is applied to understand the magnitude frequency distributions and quantify the seismogenic index of the clusters. Statistical analyses with new magnitudes suggest that fractures that are more favourably oriented for shear failure have lower b-value and higher seismogenic index, suggesting higher potential for relatively larger earthquakes, rather than fractures subparallel to maximum horizontal principal stress orientation.

58 GEOSCIENCES↗

DOE OSTI2020

A Bayesian Approach for Statistical–Physical Bulk Parameterization of Rain Microphysics. Part II: Idealized Markov Chain Monte Carlo Experiments

Observationally informed development of a new framework for bulk rain microphysics, the Bayesian Observationally Constrained Statistical–Physical Scheme (BOSS; described in Part I of this study), is demonstrated. This scheme’s development is motivated by large uncertainties in cloud and weather simulations associated with approximations and assumptions in existing microphysics schemes. Here, a proof-of-concept study is introduced using a Markov chain Monte Carlo sampling algorithm with BOSS to probabilistically estimate microphysical process rates and parameters directly from a set of synthetically generated rain observations. The framework utilized is an idealized steady-state one-dimensional column rainshaft model with specified column-top rain properties and a fixed thermodynamical profile. Different configurations of BOSS—flexibility being a key feature of this approach—are constrained via synthetic observations generated from a traditional three-moment bulk microphysics scheme. The ability to retrieve correct parameter values when the true parameter values are known is illustrated. For cases when there is no set of true parameter values, the accuracy of configurations of BOSS that have different levels of complexity is compared. It is found that addition of the sixth moment as a prognostic variable improves prediction of the third moment (proportional to bulk rain mass) and rain rate. In contrast, increasing process rate formulation complexity by adding more power terms has little benefit—a result that is explained using further-idealized experiments. BOSS rainshaft simulations are shown to well estimate the true process rates from constraint by bulk rain observations, with the further benefit of rigorously quantified uncertainty of these estimates.

58 GEOSCIENCES↗

NASA NTRS1993

What determines the stellar mass functions in globular clusters?

We analyze the dependence of stellar mass function slopes for a sample of 17 globular clusters on a variety of cluster parameters. The principal novelty of our approach is the use of appropriate multivariate statistical methods to disentangle the complex situation which is present in this problem: the slopes depend simultaneously on more than one variable, and many cluster parameters are mutually correlated. We find that the mass function slopes in the range M/M(solar) = 0.5-0.8 are largely determined by the position in the Galaxy and to a lesser extent by the cluster metallicity. Clusters closer to the Galactic center or plane have shallower mass function slopes. At a given distance to the Galactic center, clusters closer to the Galactic plane have shallower mass function slopes. At a given R(GC) and/or Z(GP), more metal-rich clusters have shallower mass function slopes. Thus, the monovariate correlations with the position or metallicity are both correct, but only partial, and in terms of slopes, biased descriptions of the situation. We present trivariate least-squares solutions where the mass function slopes can be predicted within the measurement accuracy. This relation can serve as a powerful observational constraint for theories of globular cluster formation and evolution, and it is one of the tightest correlations between globular cluster properties now known.

Djorgovski, S.↗

NASA NTRS2014

A Method for Obtaining High Time and Spatial Resolution Convective Cloud Top Data for the TTL

A method for obtaining high time and spatial resolution convective cloud top data for the TTL Leonhard Pfister, Eric Jensen, Rei Ueyama, Eliot Atlas, and Maria Navarro Convective systems in the tropics have a maximum in the cloud top altitude distribution of about 13.5 km. However, there is a significant tail to this distribution -- a few percent reach the cold point tropopause (CPT) at 16.5 km, and there has been clear evidence of convective mass deposited as high as 19 km in the tropics. The region between 13.5 km and the cold point tropopause is transitional, between the free tropical troposphere where convective mixing dominates, and the stratosphere where slow upward ascent dominates. In this region (the Tropical Tropopause Layer), convective injection, slow ascent, and mixing from midlatitudes all have similar time scales. So, even though only a few percent of convective systems reach the CPT, convection is important. Space Based Lidar and cloud radar measurements have yielded information about long term average statistical distributions of cloud altitude as a function of location. However, we also need time-dependent cloud top altitude and cloud top potential temperature information, primarily to understand the water vapor and TTL cloud distributions. This is because the effect of convection depends on the local temperature, and on the subsequent temperature history. Time dependent cloud top information is also needed to understand short-lived tracers because cross-isentropic flow is time and space dependent. This paper presents a method of obtaining time and space dependent convective cloud top theta (and altitude) information using 3-hourly geostationary brightness temperature data, coupled with global 3 -hourly rainfall estimates and temperature analyses. We explore different mixing algorithms to obtain the most reasonable agreement with near-simultaneous observations by cloudsat and calipso. Observations of short-lived tracers from ATTREX, coupled with short-term trajectories are used to test the method's accuracy. An important caveat is the ambiguity of evaluating convective cloud top altitudes under from combined cloudsat and calipso measurements.

cold point tropopause (CPT)↗

DOE OSTI2024

Upsampling Monte Carlo Reactor Simulation Tallies in Depleted Sodium-Cooled Fast Reactor Assemblies Using a Convolutional Neural Network

The computational demand of neutron Monte Carlo transport simulations can increase rapidly with the spatial and energy resolution of tallied physical quantities. Convolutional neural networks have been used to increase the resolution of Monte Carlo simulations of light water reactor assemblies while preserving accuracy with negligible additional computational cost. Here, we show that a convolutional neural network can also be used to upsample tally results from Monte Carlo simulations of sodium-cooled fast reactor assemblies, thereby extending the applicability beyond thermal systems. The convolutional neural network model is trained using neutron flux tallies from 300 procedurally generated nuclear reactor assemblies simulated using OpenMC. Validation and test datasets included 16 simulations of procedurally generated assemblies, and a realistic simulation of a European sodium-cooled fast reactor assembly was included in the test dataset. We show the residuals between the high-resolution flux tallies predicted by the neural network and high-resolution Monte Carlo tallies on relative and absolute bases. The network can upsample tallies from simulations of fast reactor assemblies with diverse and heterogeneous materials and geometries by a factor of two in each spatial and energy dimension. The network’s predictions are within the statistical uncertainty of the Monte Carlo tallies in almost all cases. This includes test assemblies for which burnup values and geometric parameters were well outside the ranges of those in assemblies used to train the network.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

NASA NTRS2015

Retrieval of Aerosol Microphysical Properties from AERONET Photo-Polarimetric Measurements. 2: A New Research Algorithm and Case Demonstration

A new research algorithm is presented here as the second part of a two-part study to retrieve aerosol microphysical properties from the multispectral and multiangular photopolarimetric measurements taken by Aerosol Robotic Network's (AERONET's) new-generation Sun photometer. The algorithm uses an advanced UNified and Linearized Vector Radiative Transfer Model and incorporates a statistical optimization approach.While the new algorithmhas heritage from AERONET operational inversion algorithm in constraining a priori and retrieval smoothness, it has two new features. First, the new algorithmretrieves the effective radius, effective variance, and total volume of aerosols associated with a continuous bimodal particle size distribution (PSD) function, while the AERONET operational algorithm retrieves aerosol volume over 22 size bins. Second, our algorithm retrieves complex refractive indices for both fine and coarsemodes,while the AERONET operational algorithm assumes a size-independent aerosol refractive index. Mode-resolved refractive indices can improve the estimate of the single-scattering albedo (SSA) for each aerosol mode and thus facilitate the validation of satellite products and chemistry transport models. We applied the algorithm to a suite of real cases over Beijing_RADI site and found that our retrievals are overall consistent with AERONET operational inversions but can offer mode-resolved refractive index and SSA with acceptable accuracy for the aerosol composed by spherical particles. Along with the retrieval using both radiance and polarization, we also performed radiance-only retrieval to demonstrate the improvements by adding polarization in the inversion. Contrast analysis indicates that with polarization, retrieval error can be reduced by over 50% in PSD parameters, 10-30% in the refractive index, and 10-40% in SSA, which is consistent with theoretical analysis presented in the companion paper of this two-part study.

aerosol retrieval algorithm↗

NASA NTRS1991

Global ocean tide mapping using TOPEX/Poseidon altimetry

The investigation's main goals are to produce accurate tidal maps of the main diurnal, semidiurnal, and long-period tidal components in the world's deep oceans. This will be done by the application of statistical estimation techniques to long time series of altimeter data provided by the TOPEX/POSEIDON mission, with additional information provided by satellite tracking data. In the prelaunch phase, we will use in our simulations and preliminary work data supplied by previous oceanographic missions, such as Seasat and Geosat. These results will be of scientific interest in themselves. The investigation will also be concerned with the estimation of new values, and their uncertainties, for tidal currents and for the physical parameters appearing in the Laplace tidal equations, such as bottom friction coefficients and eddy viscosity coefficients. This will be done by incorporating the altimetry-derived charts of vertical tides as boundary conditions in the integration of those equations. The methodology of the tidal representation will include the use of appropriate series expansions such as ocean-basin normal modes and spherical harmonics. The results of the investigation will be space-determined tidal models of coverage and accuracy superior to that of the present numerical models of the ocean tides, with the concomitant benefits to oceanography and associated disciplinary fields.

Sanchez, Braulio V.↗

NASA NTRS1977

Alteration mapping at Goldfield, Nevada, by cluster and discriminant analysis of LANDSAT digital data

The ability of Landsat multispectral digital data to differentiate among 62 combinations of rock and alteration types at the Goldfield mining district of Western Nevada was investigated by using statistical techniques of cluster and discriminant analysis. Multivariate discriminant analysis was not effective in classifying each of the 62 groups, with classification results essentially the same whether data of four channels alone or combined with six ratios of channels were used. Bivariate plots of group means revealed a cluster of three groups including mill tailings, basalt and all other rock and alteration types. Automatic hierarchical clustering based on the fourth dimensional Mahalanobis distance between group means of 30 groups having five or more samples was performed. The results of the cluster analysis revealed hierarchies of mill tailings vs. natural materials, basalt vs. non-basalt, highly reflectant rocks vs. other rocks and exclusively unaltered rocks vs. predominantly altered rocks. The hierarchies were used to determine the order in which sets of multiple discriminant analyses were to be performed and the resulting discriminant functions were used to produce a map of geology and alteration which has an overall accuracy of 70 percent for discriminating exclusively altered rocks from predominantly altered rocks.

Ballew, G.↗

NASA NTRS1994

A passive microwave technique for estimating rainfall and vertical structure information from space. Part 1: Algorithm description

This paper describes a multichannel physical approach for retrieving rainfall and vertical structure information from satellite-based passive microwave observations. The algorithm makes use of statistical inversion techniques based upon theoretically calculated relations between rainfall rates and brightness temperatures. Potential errors introduced into the theoretical calculations by the unknown vertical distribution of hydrometeors are overcome by explicity accounting for diverse hydrometeor profiles. This is accomplished by allowing for a number of different vertical distributions in the theoretical brightness temperature calculations and requiring consistency between the observed and calculated brightness temperatures. This paper will focus primarily on the theoretical aspects of the retrieval algorithm, which includes a procedure used to account for inhomogeneities of the rainfall within the satellite field of view as well as a detailed description of the algorithm as it is applied over both ocean and land surfaces. The residual error between observed and calculated brightness temperatures is found to be an important quantity in assessing the uniqueness of the solution. It is further found that the residual error is a meaningful quantity that can be used to derive expected accuracies from this retrieval technique. Examples comparing the retrieved results as well as the detailed analysis of the algorithm performance under various circumstances are the subject of a companion paper.

Kummerow, Christian↗

NASA NTRS1997

Atomic, Molecular, and Optical Physics: Optical Excitation Function of H(1s-2p) Produced by electron Impact from Threshold to 1.8 keV

The optical excitation function of prompt Lyman-Alpha radiation, produced by electron impact on atomic hydrogen, has been measured over the extended energy range from threshold to 1.8 keV. Measurements were obtained in a crossed-beams experiment using both magnetically confined and electrostatically focused electrons in collision with atomic hydrogen produced by an intense discharge source. A vacuum-ultraviolet mono- chromator system was used to measure the emitted Lyman-Alpha radiation. The absolute H(1s-2p) electron impact excitation cross section was obtained from the experimental optical excitation function by normalizing to the accepted optical oscillator strength, with corrections for polarization and cascade. Statistical and known systematic uncertainties in our data range from +/- 4% near threshold to +/- 2% at 1.8 keV. Multistate coupling affecting the shape of the excitation function up to 1 keV impact energy is apparent in both the present experimental data and present theoretical results obtained with convergent close- coupling (CCC) theory. This shape function effect leads to an uncertainty in absolute cross sections at the 10% level in the analysis of the experimental data. The derived optimized absolute cross sections are within 7% of the CCC calculations over the 14 eV-1.8 keV range. The present CCC calculations converge on the Bethe- Fano profile for H(1s-2p) excitation at high energy. For this reason agreement with the CCC values to within 3% is achieved in a nonoptimal normalization of the experimental data to the Bethe-Fano profile. The fundamental H(1s-2p) electron impact cross section is thereby determined to an unprecedented accuracy over the 14 eV - 1.8 keV energy range.

James, G. K.↗

NASA NTRS2004

Physical Validation of TRMM TMI and PR Monthly Rain Products Over Oklahoma

The Tropical Rainfall Measuring Mission (TRMM) provides monthly rainfall estimates using data collected by the TRMM satellite. These estimates cover a substantial fraction of the earth's surface. The physical validation of TRMM estimates involves corroborating the accuracy of spaceborne estimates of areal rainfall by inferring errors and biases from ground-based rain estimates. The TRMM error budget consists of two major sources of error: retrieval and sampling. Sampling errors are intrinsic to the process of estimating monthly rainfall and occur because the satellite extrapolates monthly rainfall from a small subset of measurements collected only during satellite overpasses. Retrieval errors, on the other hand, are related to the process of collecting measurements while the satellite is overhead. One of the big challenges confronting the TRMM validation effort is how to best estimate these two main components of the TRMM error budget, which are not easily decoupled. This four-year study computed bulk sampling and retrieval errors for the TRMM microwave imager (TMI) and the precipitation radar (PR) by applying a technique that sub-samples gauge data at TRMM overpass times. Gridded monthly rain estimates are then computed from the monthly bulk statistics of the collected samples, providing a sensor-dependent gauge rain estimate that is assumed to include a TRMM equivalent sampling error. The sub-sampled gauge rain estimates are then used in conjunction with the monthly satellite and gauge (without sub- sampling) estimates to decouple retrieval and sampling errors. The computed mean sampling errors for the TMI and PR were 5.9% and 7.796, respectively, in good agreement with theoretical predictions. The PR year-to-year retrieval biases exceeded corresponding TMI biases, but it was found that these differences were partially due to negative TMI biases during cold months and positive TMI biases during warm months.

Fisher, Brad L.↗

DOE OSTI2020

Fouling modeling and prediction approach for heat exchangers using deep learning

In this article, we develop a generalized and scalable statistical model for accurate prediction of fouling resistance using commonly measured parameters of industrial heat exchangers. This prediction model is based on deep learning where a scalable algorithmic architecture learns non-linear functional relationships between a set of target and predictor variables from large number of training samples. Here, the efficacy of this modeling approach is demonstrated for predicting fouling in an analytically modeled cross-flow heat exchanger, designed for waste heat recovery from flue-gas using room temperature water. The performance results of the trained models demonstrate that the mean absolute prediction errors are under 10 –4 KW –1 for flue-gas side, water side and overall fouling resistances. The coefficients of determination (R 2 ), which characterize the goodness of fit between the predictions and observed data, are over 99%. Even under varying levels of measurement noise in the inputs, we demonstrate that predictions over an ensemble of multiple neural networks achieves better accuracy and robustness to noise. We find that the proposed deep-learning fouling prediction framework learns to follow heat exchanger flow and heat transfer physics, which we confirm using locally interpretable model agnostic explanations around randomly selected operating points. Overall, we provide a robust algorithmic framework for fouling prediction that can be generalized and scaled to various types of industrial heat exchangers.

42 ENGINEERING↗