Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “sequential”

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 1,189 records · Page 66
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Low-dimensional carbon materials decorated FAPbI 3 for carbon-based perovskite solar cells

Carbon nanomaterials are at the forefront of research in perovskite solar cells (PSCs) due to their exceptional electrical, optical, and stability properties. Their diverse applications include serving as interfacial layers, additives, hole and electron transport materials, and back electrodes. While the influence of various low-dimensional carbon nanomaterial structures on crystallinity, optical and electrical performance, and overall device efficiency has been a topic of interest, it has not been thoroughly explored until now. In this study, we effectively integrated carbon quantum dots (CQDs), multi-walled carbon nanotubes (MWCNTs), and graphene into the FAPbI 3 photoactive layer using a two-step sequential deposition method. Our experiments revealed marked improvements in the photovoltaic performance of PSCs that incorporated all three types of carbon nanomaterials. In particular, the data shows significant enhancements in power conversion efficiency, demonstrating the effectiveness of these materials in optimizing device functionality. Notably, MWCNTs distinguished themselves by exhibiting a remarkable potential for enhancing long-term stability. This finding underscores the importance of selecting the right carbon nanomaterials for future PSC developments, paving the way for more reliable and efficient solar energy solutions. As a result, our research highlights the critical role of carbon nanomaterials in advancing perovskite solar technology.

14 SOLAR ENERGY↗
DOE OSTI2023

Few-cycle atomic layer deposition to nanoengineer polybenzimidazole for H 2 /CO 2 separation

Atomic layer deposition (ALD) creates uniform sub-nanometer films on a variety of surfaces and nanopore walls and has been used to modify polymers to improve surface affinity towards specific molecules, solvent resistance, and barrier properties to gases and vapors. Here, for the first time, we demonstrate that few-cycle ALD can be used to engineer functional polymers at a sub-nanometer scale to improve both molecular size-sieving ability and counterintuitively, gas permeability. Particularly, 1-cycle ALD treatment of polybenzimidazole (PBI) by sequential exposure to trimethylaluminum (TMA) and water vapor remarkably increases H 2 permeability by 120% - 270% and H 2 /CO 2 selectivity by 30% at 35–200 °C. The ALD not only deposits an AlOx layer on the surface but also enables the TMA to infiltrate and react with the bulk PBI to form an AlO x network, disrupting polymer chain packing and increasing chain rigidity. The membrane exhibits excellent stability when challenged with simulated syngas, overcoming the permeability/selectivity tradeoff for H 2 /CO 2 separation. In conclusion, this study showcases a facile and scalable way of engineering polymeric membranes at a sub-nanometer level to improve molecular separation performance.

36 MATERIALS SCIENCE↗
DOE OSTI2023

The role of surface chemistry in the synthesis of supported CuPd bimetallic/intermetallic catalysts for selective hydrogenation reactions

Understanding and controlling the structure of supported bimetallic/intermetallic catalysts are crucial for many heterogeneous catalytic reactions. However, commonly used synthesis techniques such as coimpregnation and coadsorption often yield nonuniform alloying and phase segregation, largely owing to the lack of interactions between different metal precursors. Here we show that by adopting sequential adsorption of complex metal cations and anions with the assistance of a proper ligand, the interaction between different metal precursors is enhanced, thus resulting in uniformly alloyed intermetallic catalysts. Importantly, the supported CuPd intermetallic catalyst exhibits greatly improved selectivity towards monoolefins in the semihydrogenation of acetylene and butadiene compared to those random CuPd alloys synthesized via coimpregnation and coadsorption methods.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

A systematic multicriteria-based approach to support product portfolio selection in microalgae biorefineries

Here this work proposes and applies a sequential approach of objective methods to aid the decision-making process for the deployment of microalgae biorefineries. The strategy combines Multicriteria Decision Analysis (MCDA) and weight assignment methods to simultaneously consider technical, economic, and environmental criteria to (1) outrank the best bioproduct options from different biomass fractions present in microalgae biomass at different ratios (namely carbohydrates, lipids, and protein) and (2) define the most suitable biorefining pathways associated with specific pairings of microalgae strains and cultivation conditions. The first part of the assessment identified succinic acid, acrylic acid, and citric acid as the top-ranked bioproducts from carbohydrates, polyurethane from lipids, and thermoplastic extrusion co-feed from protein. The second step of the analysis determined that, when production of a hydrocarbon fuel is desired, the compositional profile of a strain is paramount in defining the biorefining setup that should be pursued. In summary, microalgae lipids should be sent to the production of hydrocarbon fuels if the ratio between neutral lipids and fermentable carbohydrates is higher than roughly 1, with carbohydrates and protein being converted to the higher-value products noted above. Finally, this result was corroborated through process simulations, which indicated superior economic and environmental metrics when strains are paired with suitable conversion pathways identified through MCDA based on their compositional profiles. The outcomes of this work provide clear, objective, guidelines for establishing the best biorefining approach for a large suite of biochemical compositions as a screening method prior to employing detailed process simulations alongside rigorous techno-economic and life-cycle assessments.

09 BIOMASS FUELS↗
DOE OSTI2020

Intricate Genetic Programs Controlling Dormancy in Mycobacterium tuberculosis

Mycobacterium tuberculosis (MTB) displays the remarkable ability to transition in and out of dormancy, a hallmark of the pathogen’s capacity to evade the immune system and exploit susceptible individuals. Uncovering the gene regulatory programs that underlie the phenotypic shifts in MTB during disease latency and reactivation has posed a challenge. We develop an experimental system to precisely control dissolved oxygen levels in MTB cultures in order to capture the transcriptional events that unfold as MTB transitions into and out of hypoxia-induced dormancy. Using a comprehensive genome-wide transcription factor binding map and insights from network topology analysis, we identify regulatory circuits that deterministically drive sequential transitions across six transcriptionally and functionally distinct states encompassing more than three-fifths of the MTB genome. The architecture of the genetic programs explains the transcriptional dynamics underlying synchronous entry of cells into a dormant state that is primed to infect the host upon encountering favorable conditions.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

Atomistic thermodynamics and kinetics of dicalcium silicate dissolution

Low-temperature cement manufacturing has garnered academic and industrial attention for its low environmental footprints. However, the sluggish hydration kinetics of the resultant cement affects their early-age strength development. This motivates fundamental studies to unravel the mechanistic picture of the dissolution process and discover science-informed pathways to accelerate hydration. Standard atomistic simulations seldomly exceed a microsecond making them impractical to study slow dissolution processes. We are using rare event sampling techniques, we provide the mechanistic picture of Ca 2+ ion dissolution from a kink site on the dicalcium silicate surface. The Ca 2+ ion dissolution is comprised of two sequential stages: breaking restraints from the kink sites to form a ledge adatom and detaching from the ledge/terrace adatom sites into the solution. The first and second stages feature free energy barriers of ~63 kJ/mol and ~ 29 kJ/mol respectively, making the first stage the rate-limiting step of the entire Ca 2+ dissolution kinetics. Using the reactive flux method, the rate and equilibrium constants for each reaction step are calculated, which yield the Ca 2+ ion activity of ~1.03 × 10 -5 . The diffusion calculations indicate that the surface effects lower the self-diffusion coefficient of Ca 2+ ions at the solid-water interface.

36 MATERIALS SCIENCE↗
DOE OSTI2021

A generalized distillation network synthesis model

In this report we propose a generalized superstructure-based distillation network synthesis model with improved modeling capabilities resulting in an extended solution space. The model can assign multiple mixtures to be separated (e.g., reactor network effluents) to different columns of the network while considering the interactions among different separation steps. In terms of outlets, products with general specifications, including pure components and multi-component mixtures, as well as streams without strict specifications (e.g., reactor recycle streams) can be readily handled. Furthermore, stream bypass is considered to avoid unnecessary separations, and thermal coupling is considered to yield novel solutions. The proposed model enables seamless integration with reactor network synthesis models thereby leading to solutions that are superior to the ones obtained by conventional sequential approaches.

42 ENGINEERING↗
DOE OSTI2022

Iron precipitation under controlled oxygen flow: Mineralogical implications for BIF precursors in the Archean ocean

Banded iron formation (BIF) deposition in the Archean ocean is thought to have been initiated by the oxidation of dissolved iron into ferric primary precursor phases that descended through the water column and were deposited on the seafloor. Effective interpretation of the trace element and isotopic composition of BIFs in the geologic record requires an understanding of the identity and longevity of those precursor phases. Temporal and spatial variation in oxidant concentration and ocean chemistry may have driven precursor mineralogy. Furthermore, precursor phases may undergo transitions during descent and burial. This experimental study follows the evolution of iron mineralogy, speciation, and redox state during the precipitation of iron under variable oxygen fluxes and in fluids containing different iron-complexing anions (chloride, sulfate, and phosphate). Additionally, suspensions collected after incomplete oxidation were anoxically aged to simulate changes to precursor mineralogy during descent below the seawater redoxcline. Results from X-ray diffraction, sequential dissolution, colorimetric determination of Fe redox state, and Fe K-edge X-ray absorption fine structure spectroscopy showed that under all experimental conditions, intermediate precipitates collected before complete iron uptake contained mineralized ferrous iron. Purely ferric mineral assemblages were not observed until iron was removed from solution. In low-oxygen experiments, intermediate precipitates contained phases with stoichiometric Fe 2+ , including magnetite, vivianite, and green rust (GR). Chloride green rust (GR1) had more ferric iron than sulfate green rust (GR2) and more readily transitioned to magnetite via disproportionation. GR2 was stable over a broader range of iron uptake; magnetite appeared mainly as an additional oxidation product in sulfate solution. In high oxygen experiments, mineralized ferrous iron occurred in amorphous phases or as a non-stoichiometric component in ferric oxides; final ferric assemblages were less crystalline than in low-oxygen equivalents. Low concentrations of phosphate increased the total iron oxidation rate, but also increased the ferrous content of intermediate precipitates. The incorporation of phosphate stabilized GR2 and facilitated GR2 precipitation over a larger range of iron uptake. In higher concentrations of phosphate, vivianite and poorly crystalline ferric hydroxides formed in place of GR. Here, these findings suggest that over a range of possible ocean chemistries, mixed-valence phases may have been short lived but relevant precursors to BIF. Additionally, anionic chemistry and oxidant concentration are shown to influence the crystallinity and chemical resistivity of final ferric assemblages, affecting their reactivity during later anoxic burial.

58 GEOSCIENCES↗
DOE OSTI2021

Copper transformation, speciation, and detoxification in anoxic and suboxic freshwater sediments

The complex chemistry of copper (Cu) in freshwater sediments at low concentrations is not well understood. We evaluated the transformation processes of Cu added to freshwater sediments under suboxic and anoxic conditions. Freshwater sediments from three sources in Michigan with different characteristics (Spring Creek, River Raisin, and Maple Lake) were spiked with 30 or 60 mg kg -1 Cu and incubated under a nitrogen atmosphere. After 28-d, each treatment subset was amended with organic matter (OM) to promote anoxic conditions and evaluate its effects on Cu speciation. OM addition triggered a shift from suboxic to anoxic conditions, and sequential extractions showed that Cu accordingly shifted from acid-soluble to oxidizable fractions. Extended X-ray absorption fine-structure (EXAFS) spectroscopy revealed that Cu sulfides dominated all anoxic samples except for Spring Creek 30 mg kg -1 , where Cu(I) was predominantly complexed to thiol groups of OM. Covellite and chalcopyrite (CuFeS 2 ) were the predominant Cu species in nearly all anoxic samples, as determined by Raman spectroscopy, scanning electron microscopy, and X-ray absorption near-edge structure (XANES) spectroscopy. Copper reduction also occurred under suboxic conditions: for two of three sediments, around 80% had been reduced to Cu(I), while the remaining 20% persisted as Cu(II) complexed to OM. However, in the third coarsest (i.e., Spring Creek), around 50% of the Cu had been reduced, forming Cu(I)-OM complexes, while the remainder was Cu(II)-OM complexes. Toxicity tests showed that survival of H. azteca and D. magna were significantly lower in suboxic treatments. Anoxic sediments triggered a near-complete transformation of Cu to sulfide minerals, reducing its toxicity.

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

A Tchebycheff-based multi-objective combined with a PSO–SQP dynamic real-time optimization framework for cycling energy systems

The aim in this work is to develop an optimization framework to address cycling of baseload energy systems due to the penetration of renewables into the grid. The developed strategy corresponds to a multi-objective and dynamic real-time optimization (MOO-DRTO) framework applied to a postcombustion MEA-based CO 2 capture process from a baseload coal-fired power plant under cycling conditions. A Tchebycheff-based method is used for the multi-objective optimization (MOO) component. Also, a hybrid approach consisting of Particle Swarm Optimization (PSO) and Sequential Quadratic Programming (SQP) is implemented for the first time in process systems engineering to solve the dynamic real-time optimization (DRTO) component. The objectives considered in the MOO-DRTO framework are economic and environmental. The proposed MOO-DRTO strategy is successfully implemented and 24-h optimal output trajectories for the carbon capture system under cycling are generated. Also, the optimal compromise is chosen from the Pareto front according to a set of selected weights for the objectives with minimal interaction between the framework and decision maker. The results indicated that the developed framework has potential to be extended to plant-wide optimization applications under cycling.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Anion exchange and extraction chromatography tandem column isolation of zirconium-89 ( 89 Zr) from cyclotron bombarded targets using an automated fluidic platform

The long-lived positron emitter 89 Zr is a highly promising nuclide employed in diagnostic Positron Emission Tomography (PET) imaging. Methods of radiochemical processing to obtain 89 Zr for clinical use are traditionally performed with a single hydroxamate resin column. In this work, we present a tandem column purification method for the preparation of high-purity 89 Zr from cyclotron bombarded Y metal foils. The primary column is a macroporous, strongly basic anion exchange resin on styrene divinylbenzene co-polymer. The secondary microcolumn, with an internal volume of 33 µL, is packed with an extraction chromatography resin (ExCR) loaded with HDEHP. A condition of “inverted selectivity” is presented, wherein the 89 Zr elution from the primary column is synonymous with the load condition on the secondary column. The ability to transfer 89 Zr from one column to the next allows two sequential column clean-up methods to be performed prior to the final elution of the 89 Zr product. This approach assures delivery of very high purity 89 Zr . The tandem column purification process has been implemented into a prototype automated fluidic system. Optimization of the method is presented, followed by evaluation of the process using seven cyclotron bombarded Y metal foil targets. Once optimized, we found that 93.7±2.3% of the 89 Zr present in the foils was recovered in the secondary column elution fraction (0.8 M oxalic acid). Radiochromatograms of the product elution peaks enabled determination of full width at half-maximum (FWHM) and 89 Zr collection yields as a function of volume. Because of the small size of the secondary microcolumn, a 89 Zr product volume of ~0.28 mL is reported, which provides a substantially increased nuclide concentration over traditional methods. Finally, we evaluated the transchelation of the resulting 89 Zr oxalate product to deferoxamine mesylate (DFOM) salt. We observed effective specific activities (ESA) and bindable metals concentrations ([M B ]) that exceed those reported by the traditional single hydroxamate column method.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
DOE OSTI2023

Predictive model for real-time energy disaggregation using long short-term memory

To provide affordable energy-saving solutions for the small and medium-sized manufacturers (SMMs), we propose a unified framework for generating predictive models that support real-time disaggregation of power consumption from combined inputs, enabling automatic machine state identification simultaneously for joint analysis of energy usage patterns. Further, the proposed framework transforms raw power consumption into a time series with look-back and bootstrap capabilities for historical pattern detection, while a learning architecture utilizes the stacked long short-term memory (LSTM) layers as encoders for embedding generation with sequential awareness. Experimental results demonstrate 93.65% minimum accuracy in ideal case of real-time energy usage and machine state prediction.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2020

A non-cooperative meta-modeling game for automated third-party calibrating, validating and falsifying constitutive laws with parallelized adversarial attacks

The evaluation of constitutive models, especially for high-risk and high-regret engineering applications, requires efficient and rigorous third-party calibration, validation and falsification. While there are numerous efforts to develop paradigms and standard procedures to validate models, difficulties may arise due to the sequential, manual, and often biased nature of the commonly adopted calibration and validation processes, thus slowing down data collections, hampering the progress towards discovering new physics, increasing expenses and possibly leading to misinterpretations of the credibility and application ranges of proposed models. This work attempts to introduce concepts from game theory and machine learning techniques to overcome many of these existing difficulties. Here, we introduce an automated meta-modeling game where two competing AI agents systematically generate experimental data to calibrate a given constitutive model and to explore its weakness such that the experiment design and model robustness can be improved through competitions. The two agents automatically search for the Nash equilibrium of the meta-modeling game in an adversarial reinforcement learning framework without human intervention. In particular, a protagonist agent seeks to find the more effective ways to generate data for model calibrations, while an adversary agent tries to find the most devastating test scenarios that expose the weaknesses of the constitutive model calibrated by the protagonist. By capturing all possible design options of the laboratory experiments into a single decision tree, we recast the design of experiments as a game of combinatorial moves that can be resolved through deep reinforcement learning by the two competing players. Our adversarial framework emulates idealized scientific collaborations and competitions among researchers to achieve a better understanding of the application range of the learned material laws and prevent misinterpretations caused by conventional AI-based third-party validation. Numerical examples are given to demonstrate the wide applicability of the proposed meta-modeling game with adversarial attacks on both human-crafted constitutive models and machine learning models.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

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

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

42 ENGINEERING↗
DOE OSTI2022

Learning high-dimensional parametric maps via reduced basis adaptive residual networks

We propose a scalable framework for the learning of high-dimensional parametric maps via adaptively constructed residual network (ResNet) maps between reduced bases of the inputs and outputs. When just few training data are available, it is beneficial to have a compact parametrization in order to ameliorate the ill-posedness of the neural network training problem. By linearly restricting high-dimensional maps to informed reduced bases of the inputs, one can compress high-dimensional maps in a constructive way that can be used to detect appropriate basis ranks, equipped with rigorous error estimates. A scalable neural network learning framework is thus to learn the nonlinear compressed reduced basis mapping. Unlike the reduced basis construction, however, neural network constructions are not guaranteed to reduce errors by adding representation power, making it difficult to achieve good practical performance. Inspired by recent approximation theory that connects ResNets to sequential minimizing flows, we present an adaptive ResNet construction algorithm. This algorithm allows for depth-wise enrichment of the neural network approximation, in a manner that can achieve good practical performance by first training a shallow network and then adapting. We prove universal approximation of the associated neural network class for $L^2_v$ functions on compact sets. Our overall framework allows for constructive means to detect appropriate breadth and depth, and related compact parametrizations of neural networks, significantly reducing the need for architectural hyperparameter tuning. Numerical experiments for parametric PDE problems and a 3D CFD wing design optimization parametric map demonstrate that the proposed methodology can achieve remarkably high accuracy for limited training data, and outperformed other neural network strategies we compared against.

42 ENGINEERING↗
DOE OSTI2024

Simultaneous shape and topology optimization of inflatable soft robots

Simultaneous shape and topology optimization is used to design pressure-activated inflatable soft robots. The pressure loaded boundary is meshed conformingly and shape optimized, while the morphology of the robot is topology optimized. The design objective is to exert maximum force on an object, i.e. to produce soft “grippers”. The robot’s motion is modeled using nearly incompressible finite deformation hyperelasticity. To ensure stability of the robot, the buckling load factors obtained via linearized buckling analyses are constrained. The finite element method is used to evaluate the optimization cost and constraint functions and the adjoint method is employed to compute their sensitivities. The numerical examples produce pressure-driven soft robots with varying complexity. We also compare our simultaneous optimization results to those obtained via sequential topology and then shape optimization.

42 ENGINEERING↗
DOE OSTI2024

A fast and accurate domain decomposition nonlinear manifold reduced order model

Here, this paper integrates nonlinear-manifold reduced order models (NM-ROMs) with domain decomposition (DD). NM ROMs approximate the full order model (FOM) state in a nonlinear-manifold by training a shallow, sparse autoencoder using FOM snapshot data. These NM-ROMs can be advantageous over linear-subspace ROMs (LS-ROMs) for problems with slowly decaying Kolmogorov n-width. However, the number of NM-ROM parameters that need to be trained scales with the size of the FOM. Moreover, for “extreme-scale” problems, the storage of high-dimensional FOM snapshots alone can make ROM training expensive. To alleviate the training cost, this paper applies DD to the FOM, computes NM-ROMs on each subdomain, and couples them to obtain a global NM-ROM. This approach has several advantages: Subdomain NM-ROMs can be trained in parallel, involve fewer parameters to be trained than global NM-ROMs, require smaller subdomain FOM dimensional training data, and can be tailored to subdomain specific features of the FOM. The shallow, sparse architecture of the autoencoder used in each subdomain NM-ROM allows application of hyper-reduction (HR), reducing the complexity caused by nonlinearity and yielding computational speedup of the NM-ROM. This paper provides the first application of NM-ROM (with HR) to a DD problem. In particular, this paper details an algebraic DD reformulation of the FOM, training a NM-ROM with HR for each sub domain, and a sequential quadratic programming (SQP) solver to evaluate the coupled global NM-ROM. Theoretical convergence results for the SQP method and a priori and a posteriori error estimates for the DD NM-ROM with HR are provided. The proposed DD NM-ROM with HR approach is numerically compared to a DD LS-ROM with HR on the 2D steady-state Burgers’ equation, showing an order of magnitude improvement in accuracy of the proposed DD NM-ROM over the DD LS-ROM.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Solving high-dimensional inverse problems using amortized likelihood-free inference with noisy and incomplete data

Here, we present a likelihood-free probabilistic inversion method based on normalizing flows for high-dimensional inverse problems. The proposed method is composed of two complementary networks: a summary network for data compression and an inference network for parameter estimation. The summary network encodes raw observations into a fixed-size vector of summary features, while the inference network generates samples of the approximate posterior distribution of the model parameters based on these summary features. The posterior samples are produced in a deep generative fashion by sampling from a latent Gaussian distribution and passing these samples through an invertible transformation. We construct this invertible transformation by sequentially alternating conditional invertible neural network and conditional neural spline flow layers. The summary and inference networks are trained simultaneously. We apply the proposed method to an inversion problem in groundwater hydrology to estimate the posterior distribution of the log-conductivity field conditioned on spatially sparse time-series observations of the system’s hydraulic head responses. The conductivity field is represented with 706 degrees of freedom in the considered problem. Comparison with the likelihood-based iterative ensemble smoother PEST-IES method demonstrates that the proposed method accurately estimates the parameter posterior distribution and the observations’ predictive posterior distribution at a fraction of the inference time of PEST-IES.

conditional invertible neural network↗