5. Capabilities

Version 3.3 of the quoFEM app was released in March 30. The following lists the functionality available in this current version. (Note: New features and fixes in this release are marked blue in the following list of features.)

5.1. UQ (Uncertainty Quantification and Optimization Options)

  1. Forward Uncertainty Propagation

    1. Dakota

      1. Monte Carlo Sampling (MCS)

      2. Latin Hypercube Sampling (LHS)

      3. Gaussian Process Regression

      4. Polynomial Chaos Expansion

    2. SimCenterUQ

      1. Monte Carlo Sampling (MCS)

        1. Resample from existing correlated dataset of samples

  2. Global Sensitivity Analysis

    1. Dakota

      1. MCS

      2. LHS

    2. SimCenterUQ

      1. Probability Model-based Global Sensitivity Analysis (PM-GSA)

        1. First-order Sobol indices

        2. Total-effect Sobol indices

        3. Group-wise Sobol indices

        4. Principal component analysis and probabilistic model-based GSA (PCA-PSA) for high-dimensional QoIs

        5. Aggregated Sobol indices for field QoIs

        6. Import input/output samples from data files

  3. Reliability Analysis

    1. Dakota

      1. Local Reliability [← minor bug fix]

      2. Global Reliability

      3. Importance Sampling

  4. Bayesian Calibration [← minor bug fix]

    1. Dakota

      1. DREAM

    2. UCSD_UQ

      1. Transitional Markov Chain Monte Carlo (TMCMC) for Bayesian estimation

        1. Override default log-likelihood function

        2. Override default error covariance structure

        3. Calibrate multipliers on error covariance structure

  5. Deterministic Calibration

    1. Dakota

      1. NL2SOL

      2. OPT++GaussNewton

      3. Gradient-free optimization

  6. Surrogate Modeling

    1. SimCenterUQ

      1. Train Gaussian Process (GP) Surrogate Model

        1. Multifidelity surrogate modeling

        2. Adaptive design of experiments options for surrogate modeling

        3. Nugget optimization options for surrogate modeling

        4. Stochastic Kriging [← new option for ‘no replications’]

      2. Surrogate modeling using Probabilistic Learning on Manifolds (PLoM) *

  7. CustomUQ

    1. Configure UQ analysis using JSON file

Note

Support for the running computation to be preformed on TACC’s high performance computer, Frontera, is provided through DesignSafe for all but the methods indicated with a star (*).

5.2. FEM (Computational Model Specification)

  1. OpenSees

  2. FEAPpv

  3. Python

  4. Custom

  5. SurrogateGP [← simplified user interface]

  6. None

  7. Multi model

5.3. RV (Random Variable Options)

  1. Inspect PDF of RV

  2. Distributions available: 1

    1. Normal

    2. Lognormal

    3. Beta

    4. Uniform

    5. Weibull

    6. Gumbel

    7. Continuous 2

    8. Exponential 3

    9. Discrete 3

    10. Gamma 3

    11. Chi-squared 3

    12. Truncated exponential 3

Note

1: For SimCentreUQ and UCSD algorithms only, the RVs can be defined by any of parameters, moments, or a dataset. 2: Available for Optimization routines in Dakota only. 3: Available in SimCenterUQ and UCSD only.

5.4. EDP (Outputs from Computational Models)

  1. Scalar quantities of interest

  2. Vector quantities of interest

5.5. RES (Summary and Visualization of UQ Analysis Results)

  1. Summary statistics of outputs displayed

    1. Mean

    2. Standard deviation

  2. All output values presented in spreadsheet

    1. Update chart by clicking on spreadsheet columns

  3. Output values visualized in interactive chart

    1. Scatter plot [← display of correlation coefficient]

    2. Histogram

    3. Cumulative distribution

    4. Inspect points on chart

  4. Spreadsheet save options

    1. Save Table

    2. Save Columns Separately (Useful after Bayesian updating, the posterior samples can later be directly loaded in quoFEM)

    3. Save RVs (Useful for surrogate model training)

    4. Save QoIs (Useful for surrogate model training)

    5. Save Surrogate Predictions (Only for the surrogate model results)

  5. Visualization of surrogate modeling (GP) results

    1. Goodness-of-fit measures

    2. 90% confidence interval and prediction interval

    3. Save GP model [← simplified surrogate model file]

  6. Visualization of PLoM training results

    1. PCA representation error plot

    2. Diffusion maps eigenvalue plot