Chapter Four · failure evidence
What Surrogate & Reduced-Order Modeling got wrong, from 72 dissertations
The records document practical breakdowns and trade-offs encountered when developing surrogate models and reduced-order approximations for complex physical simulations. Practitioners frequently face severe accuracy degradation from dimensional scaling, nonlinear structural kinematics, multi-fidelity discrepancies, and excessive computational overhead. These records come from PhD theses at 17 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Gaussian process surrogates struggle with high-dimensional scaling and hyperparameter optimization
Gaussian process and Kriging models frequently suffer from flat likelihood surfaces, hyperparameter instability, and cubic computational scaling when applied to high-dimensional problems. Smoothness assumptions and poor hyperparameter estimation also lead to washed-out predictions, diffuse posteriors, and failure on non-smooth or step-like response surfaces.
Lost to a baseline
Gaussian Process buckling surrogates without affine transforms produced heavier wingbox designs (709.37 kg vs 665.47 kg at nominal span) than smeared closed-form solutions in initial quasi-coupled optimizations.
Data-Driven and GPU-Accelerated Computational Methods for High-Fidelity Aerostructural Design · Georgia Tech
Lost to a baseline
Kriging surrogate was outperformed in coefficient of determination by ANN across increasing low-fidelity training sample sizes.
Metamodel-based uncertainty quantification for the mechanical behavior of braided composites · Leibniz Universität Hannover Repository
Considered and rejected
Considered and rejected: Rejected surrogate Kriging/Gaussian Process optimization for the bi-level EV problem, opting instead for direct regression or deep neural network surrogates.
Enabling Artificial Intelligence Techniques at Operational Level · IRIS - POLITO - prod
Tried and failed
Biased surrogate model in Gaussian process prior mean applied to High-dimensional simulation-based optimization. Outcome: worse than baseline. Reason: Inaccurate or biased prior mean models severely mislead exploration and degrade optimization performance.
Tried and failed
per-iteration Gaussian process hyperparameter MLE updates applied to high-dimensional Bayesian optimization. Outcome: unstable. Reason: hyperparameter estimates fluctuated heavily in high dimensions, degrading surrogate quality and optimization performance
Tried and failed
Gaussian process as surrogate forward model applied to offline model-based optimization. Outcome: worse than baseline. Reason: underperformed compared to neural network surrogate models on top-1 evaluation tasks
Distributionally Robust Machine Intelligence for Medicine and Scientific Discovery · Penn
Tried and failed
Gaussian approximation of deep Gaussian process posteriors applied to Bayesian optimization uncertainty estimation. Reason: yielded poor uncertainty quantification and inaccurate variance estimates for acquisition functions
Physics-informed Machine Learning for Digital Twins of Metal Additive Manufacturing · Virginia Tech
Tried and failed
multi-output Gaussian process on concatenated PCA representations applied to cyclic curve surrogate modeling. Outcome: did not converge. Reason: numerous local maxima in high-dimensional likelihood space hindered optimization
Bayesian Protocols for the Assessment of Model Uncertainties in Physics-Based Models at Multiple Length Scales · Georgia Tech
Tried and failed
local maximum likelihood lengthscale optimization in high dimensions applied to high-dimensional regression with local Gaussian processes. Outcome: worse than baseline. Reason: Flat likelihood surfaces led to degenerate lengthscales and washed-out surrogate predictions.
Efficient computer experiment designs for Gaussian process surrogates · Virginia Tech
Tried and failed
Gaussian process regression with hyperparameter priors applied to multivariate engineering surrogate modeling. Outcome: worse than baseline. Reason: Hyperprior bias shifted lengthscales toward overly nonlinear regimes, degrading predictive accuracy.
Uncertainty quantification and management in multidisciplinary design optimisation. · Cranfield
Tried and failed
Gaussian process with fully Bayesian hyperparameter integration applied to aeroelastic gust response surrogate modeling. Reason: Hyperparameter posterior integration yielded overly diffuse predictive distributions with excessive variance
Uncertainty quantification and management in multidisciplinary design optimisation. · Cranfield
Tried and failed
Gaussian process surrogates in Bayesian optimization applied to non-smooth step-like response landscapes. Outcome: did not generalise. Reason: smoothness assumptions caused poor fitting and high false-positive rates on step-like data
Improving supervised machine learning for materials science · MIT
Tried and failed
Gaussian process regression hyperparameter optimization with default iterations applied to constitutive parameter surrogate modeling. Outcome: did not converge. Reason: Default maximum iterations in scipy.optimize.minimize were insufficient for marginal likelihood optimization
Time-dependent damage of soft materials with bond breaking and healing kinetics · Cornell
Lost to a baseline
Gaussian Process surrogate underperformed GradientBoosting during Bayesian optimization, dipping below p=0.05 around iteration 50.
Data-Driven Design of Recycling-Friendly Aluminium Alloys · MIT
Lost to a baseline
Radial basis function (RBF) surrogate outperformed Gaussian process regression on the 6-dimensional OTL circuit benchmark function
RETROSPECTIVE AND EXPLORATORY ANALYSES FOR ENHANCING THE SAFETY OF ROTORCRAFT OPERATIONS · Georgia Tech
Lost to a baseline
Physics-informed Bayesian neural network surrogate models outperformed Gaussian processes with handcrafted kernels on Bayesian optimization tasks for scientific problems.
Interpretable Physics-informed Machine Learning Methods for Scientific Modeling and Data Analysis · MIT
Lost to a baseline
On the 2D Schaffer function at n = 1000, the scaled stationary Vecchia GP (GP SVEC) surpassed the 2-layer DGP with Hamiltonian Monte Carlo (DGP HMC) on RMSE/CRPS due to blurry inducing point approximations in DGP HMC.
Deep Gaussian Process Surrogates for Computer Experiments · Virginia Tech
Considered and rejected
Considered and rejected: Rejected Gaussian Process regression as the Bayesian optimization surrogate due to high dimensionality (208 features) and lack of prior distribution knowledge, replacing it with deep ensemble MLPs.
Machine learning powered insights into metamaterial prediction and design · OpenBU
Considered and rejected
Considered and rejected: Rejected Gaussian Process Bayesian Optimization due to O(n^3) matrix inversion scaling, adopting neural network heteroscedastic surrogate models instead.
Surrogate Modeling for Semiconductor Packaging and Systems Using Machine Learning · Georgia Tech
Considered and rejected
Considered and rejected: Rejected non-modular unified GP surrogate model for multi-agent planning due to curse of dimensionality causing the model to learn only subset of constraints.
Multi-fidelity Optimal Trajectory Generation: Optimal Experiment Design for Robot Learning · MIT
Considered and rejected
Considered and rejected: Rejected non-parametric RBF kernels with tunable hyperparameters for low-fidelity surrogate modeling in MF-MBAS because hyperparameter optimization scales poorly with increasing input dimensionality.
Reduced-Order Modeling Techniques for Aircraft Design in High-Dimensional Spaces · Georgia Tech
Reduced-order models fail to capture nonlinear kinematics and complex physical structures
Projection-based and reduced-order formulations fail to capture sharp shock fronts, high-wavenumber modes, membrane stretching, and preloaded contact forces. Simplifying physical models by omitting time delays, structural properties, or higher modes produces artificial stiffening, numerical singularities, and constraint violations.
Tried and failed
enforcing dissipativity constraints in reduced-order models applied to turbulent flow simulation. Outcome: worse than baseline. Reason: the constraints caused excessive overdamping in higher-dimensional model subspaces
Data-Driven Variational Multiscale Reduced Order Modeling of Turbulent Flows · Virginia Tech
Tried and failed
spatial commutation error closure modeling applied to reduced order models of turbulent flow. Outcome: no signal. Reason: inclusion of the commutation error model did not noticeably change or improve model accuracy
Data-Driven Variational Multiscale Reduced Order Modeling of Turbulent Flows · Virginia Tech
Tried and failed
first-order finite element reduced order modeling applied to stress-constrained topology optimization. Outcome: did not generalise. Reason: First-order elements underestimated peak stresses, leading to designs that violated constraints under higher-order validation.
Topology optimization and uncertainty quantification using component-wise reduced order modeling · UT Austin
Tried and failed
adding spanwise dimensions to reduced-order model applied to shock boundary layer interaction modeling. Outcome: worse than baseline. Reason: introduced chaotic noise without improving prediction accuracy
Simulation of Coupled Conjugate Heat Transfer and Nonequilibrium Boundary Layer Dynamics in High-Speed Flow Environments · Georgia Tech
Tried and failed
Proper orthogonal decomposition reduced order modeling applied to multiphase fluid flow in porous media. Outcome: did not generalise. Reason: POD fails to capture non-smooth behaviors and sharp propagating shock fronts in advection-dominated problems
Simulation of geothermal reservoirs with data assimilation and reduced order modelling · Imperial
Tried and failed
vorticity-streamfunction reduced-order modeling via SVD applied to distorted internal duct flows. Outcome: did not generalise. Reason: Inadequate modeling of viscous effects and increased vorticity causes failure to capture high-wavenumber flow modes.
Comparative Evaluation of Vorticity Transport Modeled Distortions and High-Fidelity ANSYS Solutions Using Modal Assurance Criterion · Virginia Tech
Considered and rejected
Considered and rejected: Rejected intrusive POD-Galerkin reduced order modeling for two-phase water/air flows due to insurmountable numerical instabilities and 4th-order tensor construction bottlenecks
Multi-scale multi-fidelity numerical modelling of wave energy converter farms · IRIS - POLITO - prod
Tried and failed
Proper orthogonal decomposition without difference quotients applied to time-dependent reduced order modeling. Outcome: worse than baseline. Reason: Pointwise-in-time projection and error bounds degrade with temporal discretization refinement factor
Numerical Analysis for Data-Driven Reduced Order Model Closures · Virginia Tech
Tried and failed
unstructured model reduction on second-order systems applied to second-order mechanical systems. Outcome: worse than baseline. Reason: ignoring internal physical structure missed critical frequency response peaks and reduced accuracy
Dimension Reduction in Structured Dynamical Systems: Optimal-𝓗<sub>2</sub> Approximation, Data-Driven Balancing, and Real-Time Monitoring · Virginia Tech
Tried and failed
modal derivatives in reduced order modelling applied to geometrically nonlinear structural dynamics. Outcome: unstable. Reason: produces unphysical singularities at 1:1 modal frequency ratios without genuine internal resonance
Tried and failed
recursive least squares reduced-order system identification applied to time-varying higher-order dynamic systems. Outcome: did not converge. Reason: model order mismatch prevented accurate parameter tracking during dynamic parameter variations despite working for static parameters
An improved algorithm for identification of time varying parameters using recursive digital techniques · Virginia Tech
Tried and failed
subspace state-space system identification with high model order applied to reduced-order modal dynamics modeling. Outcome: overfit. Reason: setting model order much higher than true mode count caused parameter explosion and negative fit metrics
Tried and failed
component mode synthesis without preloaded contact forces applied to constrained non-linear structural dynamics. Outcome: worse than baseline. Reason: omitting static preload contact forces creates an inaccurate reduced stiffness matrix and inaccurate displacement predictions
Structural Dynamics Design of Steam Turbine Blades with Friction Contacts for Renewable Energy · IRIS - POLITO - prod
Tried and failed
linear modal projection reduced-order modeling applied to geometrically nonlinear structural dynamics. Outcome: did not converge. Reason: linear eigenmodes miss membrane stretching and Poisson contraction, causing artificial over-stiffening without high-frequency modes
Lost to a baseline
Lumped Element Model (LEM) overestimates effective stiffness and resonant frequency compared to the Stiffness Matrix Method (SMM) due to perturbation approximations of mass loading.
STL NEMS fabrication, design, and inspection · Imperial
Lost to a baseline
Single-mode approximation underestimates the nonlinear reduction/stiffening in beam response compared to multi-mode formulations.
Nonlinear stochastic vibration in geometrically varying beams · Virginia Tech
Considered and rejected
Considered and rejected: Balanced truncation for model order reduction: rejected because resulting reduced states represent energy coordinates rather than physically meaningful rotorcraft states
Model-Based Life Extending Control for Rotorcraft · Georgia Tech
Considered and rejected
Considered and rejected: Eliminating time-delays in the reduced model by setting Tξ={0} and Tu={0}, rejected because reducing infinite-dimensional systems to finite-dimensional ones alters specific dynamical stability properties.
Interconnection-based model order reduction for quadratic-bilinear systems · Imperial
Surrogate training and inference overheads eliminate expected computational savings
Complex machine learning surrogates and active learning schemes often incur high training, inference, and retraining runtimes that negate any speedup relative to full numerical solvers. Global surrogate construction also becomes computationally intractable in high-dimensional design spaces due to exponential data requirements.
Tried and failed
global generative surrogate optimization with neural networks applied to high-dimensional simulation-based design optimization. Outcome: infeasible cost. Reason: exponentially increasing simulation data requirements in higher dimensions caused by the curse of dimensionality
Optimisation of the SHiP Beam Dump Facility with generative surrogate models · Imperial
Considered and rejected
Considered and rejected: Decided against direct global surrogate modeling (e.g., Gaussian processes, kriging, or deep neural networks) across the full input space due to the curse of dimensionality over thousands of stochastic healthcare parameters.
Data-Driven Decision Analytics in Complex Systems: Information Infrastructure, Problem Decomposition, and Algorithmic Design · Georgia Tech
Tried and failed
surrogate-based multirate partitioning with projection applied to reaction-diffusion partial differential equations. Outcome: too slow. Reason: Surrogate evaluations and projection overheads exceeded full model evaluation costs, eliminating error reduction gains.
Multimethods for the Efficient Solution of Multiscale Differential Equations · Virginia Tech
Tried and failed
neural network surrogates in constrained Bayesian optimization applied to benchmark function optimization. Outcome: too slow. Reason: Deep surrogate models incur substantially higher training and inference overhead compared to standard Gaussian processes
Bayesian Optimization for Engineering Design and Quality Control of Manufacturing Systems · Virginia Tech
Tried and failed
surrogate guidance in both exploration and exploitation applied to high-dimensional simulation-based optimization. Outcome: infeasible cost. Reason: Negligible performance improvement over exploration-only surrogate guidance despite an extreme increase in computational runtime.
Tried and failed
Pre-trained convolutional neural networks applied to spatial grid risk estimation surrogate modeling. Outcome: too slow. Reason: Excessive computational overhead without improving prediction accuracy compared to simpler architectures
An Approach for Risk-Informed UAS Mission Planning in Urban Environments to Support First Responders · Georgia Tech
Tried and failed
active learning with variance-based acquisition applied to Gaussian process surrogate modeling. Outcome: too slow. Reason: frequent retraining overhead negated sampling efficiency, underperforming random sampling for fast simulations
Time-dependent damage of soft materials with bond breaking and healing kinetics · Cornell
Tried and failed
surrogate modeling of localized failure criteria applied to multi-fidelity composite structural optimization. Outcome: unstable. Reason: mesh-dependent local response discontinuities and excessive computational post-processing time prevented effective surrogate training
Multi-fidelity probabilistic optimisation of composite structures · Imperial
Lost to a baseline
For simple uniaxial deformation paths, the DL surrogate model provided little to no computational speedup compared to standard numerical constitutive integration.
Deep learning informed multiphysics material modeling of fiber reinforced polymer nanocomposites · Leibniz Universität Hannover Repository
Lost to a baseline
1D-CNN surrogate in ISOP+ incurred higher inference runtime compared to earlier MLP/XGBoost baseline in ISOP due to CNN computational complexity
Machine learning-driven design automation for high-frequency circuits and packaging · UT Austin
Considered and rejected
Considered and rejected: Rejected purely data-driven interpolation-based surrogates for high-resolution (1km) climate modeling in favor of a hierarchical parametrization approach due to computational expense and failure to generalize/respect physical constraints.
Deep Learning Emulators for Accessible Climate Projections · MIT
Neural networks and complex surrogates degrade under training instability and scarce data
Neural networks and data-driven surrogates experience severe accuracy loss and chaotic divergence when trained on scarce simulation datasets or across long time horizons. Complex nonlinear surrogate formulations and field models frequently underperform simpler scalar baselines or fail to translate predictive accuracy into optimization gains.
Tried and failed
approximate deconvolution reduced order modeling applied to lid-driven cavity flow. Outcome: did not converge. Reason: chaotic trajectory divergence over long training time intervals deteriorated verifiability rate
Filtering and Domain Decomposition Techniques for Intrusive and Non-intrusive Reduced Order Models of Convection-Dominated Problems · Virginia Tech
Tried and failed
support vector regression surrogate modeling applied to reactor multiphysics simulation dataset. Outcome: worse than baseline. Reason: systematically underperformed and unable to accurately capture complex variations in the dataset
Development of Coupled Machine Learning and Optimization Framework for Comprehensive Molten Salt Reactor Design · Georgia Tech
Tried and failed
Augmenting surrogate training data with line search evaluations applied to global surrogate model training. Outcome: worse than baseline. Reason: Adding intermediate line search points degraded surrogate model accuracy compared to using only structured gradient samples
Optimal sample set selection for the simplex gradient · Imperial
Tried and failed
field surrogate using reduced-order modeling applied to spatial dynamic state surrogate estimation. Outcome: worse than baseline. Reason: lower goodness-of-fit compared to scalar surrogate counterparts
Uncertainty-Based Methodology for the Development of Space Domain Awareness Architectures in Three-Body Regimes · Georgia Tech
Tried and failed
recurrent and time-delay neural networks applied to aerodynamic time-series surrogate modeling. Outcome: worse than baseline. Reason: training sensitivity caused recurrent architectures to underperform simpler feedforward networks on time-series data
Tried and failed
Artificial neural networks for surrogate modeling applied to aerodynamic design optimization. Outcome: data insufficient. Reason: Severe accuracy degradation and unreliable gradients when trained on scarce high-fidelity simulation datasets
Nacelle aerodynamic design and optimisation · Cranfield
Lost to a baseline
SIMO (using simple linear Theodorsen model) outperformed IMO (using the complex nonlinear modified Goman-Khrabrov surrogate model) in LDVM lift regulation.
Modeling and Control of Wing Maneuvers in Transverse Gust Encounters · DSpace at SUNY Buffalo
Lost to a baseline
Recursive virial force prediction by the neural network surrogate had an average error (e.g., 344 for Fyy) exceeding the ground truth standard deviation (152), performing worse than a baseline predicting the mean value.
Wear across scales · EPFL
Tried and failed
surrogate model predictive accuracy for optimization applied to configurable software performance optimization. Outcome: no signal. Reason: lower prediction error on surrogate performance models did not correlate with final optimization solution quality
Finding near-optimal configurations in colossal product spaces of highly configurable systems · UT Austin
Tried and failed
surrogate modeling with IDW and RBF applied to scenario-based optimization under uncertainty. Outcome: worse than baseline. Reason: underperformed across metrics and failed beta stability stopping criteria
Decision support methodology for waste-to-value process integration · EPFL
Linear and low-degree polynomial surrogates fail to capture nonlinear simulation responses
Linear and low-order polynomial response surfaces lack the capacity to represent complex nonlinear couplings and parameter variances across physical simulations. These simplified metamodels consistently underestimate extreme outputs, leading to high ranking errors and constraint violations during optimization.
Tried and failed
low-order polynomial surrogate modeling applied to nonlinear aerodynamic force and moment prediction. Reason: insufficient polynomial degree to capture complex nonlinear aeropropulsive coupling
Hybrid Automaton Based Nominal and Contingency Planning for an Over-Actuated Tandem Tiltwing eVTOL Aircraft · Georgia Tech
Tried and failed
linear regression metamodeling of simulation outputs applied to manufacturing assembly time surrogate modeling. Outcome: did not generalise. Reason: linear models could not capture non-linear simulation dynamics, consistently underestimating completion times
Tried and failed
multiple linear regression surrogate modeling applied to building daylight and energy simulation metrics. Outcome: worse than baseline. Reason: severe non-linear relationships in climate-based daylight and operational energy metrics
Tried and failed
simple linear regression surrogate modeling applied to parameter mapping for response spectrum statistics. Outcome: worse than baseline. Reason: failed to capture nonlinear variance across periods accurately compared to Gaussian process regression
Utility of Stochastic and Physics-Based Ground Motion Simulations in Addressing Data Limitations · Texas Tech
Lost to a baseline
Least-squares and sparse least-squares (L1-regularized) polynomial surrogate models underperformed HierGP despite utilizing identical perfectly specified bases.
Considered and rejected
Considered and rejected: Rejected linear regression (LR) and polynomial regression (PR) as surrogate models for the simulation-based optimization problem due to low accuracy and high ranking errors.
On traffic state estimation and control in the world of connected vehicles · UT Austin
Considered and rejected
Considered and rejected: Rejected third-order polynomial surrogate models for estimating optimization variables in favor of second-order response surface equations due to poorer fits
Methodological Improvements for the Integration of Spacecraft Trajectory Optimization into Conceptual Space Mission Design · Georgia Tech
Multi-fidelity surrogates break down due to poor correlation across fidelity levels
Multi-fidelity surrogates fail to improve accuracy or reduce costs when low-fidelity approximations fail to correlate with high-fidelity targets. Linear discrepancy models and empirical corrections in low-fidelity data introduce bias and inaccurate gradient estimates that provide no advantage over single-fidelity baselines.
Tried and failed
multi-fidelity kriging with semi-empirical low-fidelity models applied to aerodynamic drag polar surrogate modeling. Outcome: worse than baseline. Reason: linearized low-fidelity source with empirical corrections failed to provide correlated information to improve high-fidelity emulation
A Methodology for Design Space Exploration of Novel Supersonic Aircraft Using High-Fidelity Aerodynamic Analysis · Georgia Tech
Tried and failed
multifidelity multi-objective optimization with additive Kriging applied to aerodynamic high-lift device design. Outcome: worse than baseline. Reason: poor correlation between low- and high-fidelity models for lift-to-drag ratio undermined surrogate accuracy
Multifidelity multiobjective trust-region-based optimisation for high-lift devices. PhD in Aerospace · Cranfield
Tried and failed
multi-fidelity active subspaces with RBF surrogates applied to high-dimensional aerodynamic field prediction. Outcome: worse than baseline. Reason: linear discrepancy and low-fidelity surrogates failed at accurate gradient estimation in high dimensions
Reduced-Order Modeling Techniques for Aircraft Design in High-Dimensional Spaces · Georgia Tech
Tried and failed
multi-fidelity surrogate modeling applied to aerodynamic design optimization. Outcome: worse than baseline. Reason: provided no performance gain over single-fidelity surrogate models under fixed computational budget constraints
Nacelle aerodynamic design and optimisation · Cranfield
Tried and failed
surrogate modeling coupled with spatial regression applied to aerodynamic uncertainty quantification. Outcome: did not generalise. Reason: Sparse training sets caused severe bias and poor conservativeness without multi-fidelity corrections.
Development and Use of a Spatially Accurate Polynomial Chaos Method for Aerospace Applications · Virginia Tech
Left open by the authors
Problems the authors named and did not get to.
Left open
Implement the reduced-order polymer model within CFD solvers to simulate non-uniform flows and flow-induced crystallization. Blocker: Lack of specific benchmark geometries, flow conditions, and precise coupling equations for crystallization
Model reduction for driven PDEs: Application to polymer constitutive equations · University of Nottingham Repository
Left open
Implement a machine learning surrogate model (CFD/ANN) for pedestrian wind comfort analysis in early architectural design. Blocker: Lacks specific implementation details, target CFD simulation parameters, and dataset definition beyond a conceptual thought experiment
Left open
Apply space-time refinement jump snapshot-selection to classical POD reduced-order models for multiphase flow and compare performance against deep-learning ROMs. Blocker: None
Development and optimization of a deep-learning reduced-order model for multiphase flow · UT Austin
Left open
Evaluate the vorticity transport reduced order model on flow cases with distinct, characterizable viscous contributions against high-fidelity simulations. Blocker: None
Comparative Evaluation of Vorticity Transport Modeled Distortions and High-Fidelity ANSYS Solutions Using Modal Assurance Criterion · Virginia Tech
Left open
Develop a unified reduced order modeling framework combining geometric and contact nonlinearities for structural dynamics simulations. Blocker: The task is described only as a general direction without a defined technical formulation or concrete target.
Left open
Investigate reduced-order parameter identification during parameter variations using recursive weighted least squares conditioning algorithms. Blocker: Lack of specific criteria, reduced-order models, or clear experimental targets.
An improved algorithm for identification of time varying parameters using recursive digital techniques · Virginia Tech
Left open
Integrate aerodynamic and mechanical performance metrics, including kinematic wing membrane constraints, into the graphic statics surrogate model. Blocker: Lack of specific mathematical formulation, targets, or methodology for modeling membrane kinematics with graphic statics
Geometry And Topology: Building Machine Learning Surrogate Models With Graphic Statics Method · Penn
Left open
Extend surrogate uncertainty quantification to include input, process, and material uncertainties, scaling to multi-physics process-structure-property-performance modeling in additive manufacturing. Blocker: Lacks specific implementation details, mathematical formulation, and proprietary training data/multi-physics simulation models from the thesis
Smart Quality Assurance System for Additive Manufacturing using Data-driven based Parameter-Signature-Quality Framework · Virginia Tech
Left open
Generate surrogate training data using parabolized stability equations to incorporate nonparallel and nonlinear effects into boundary-layer transition models. Blocker: None
Machine Learning Approaches to Data-Driven Transition Modeling · Virginia Tech
Left open
Develop discontinuous surrogate models to fit cohesive envelope properties directly from empirical data. Blocker: Lacks specific target envelope definitions and empirical dataset specifications
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.