Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List

Full text release has been delayed at the author's request until December 18, 2024

ETD Abstract Container

Abstract Header

Uncertainty Quantification Using Simulation-based and Simulation-free methods with Active Learning Approaches

Zhang, Chi
http://orcid.org/0000-0003-1304-8209
http://rave.ohiolink.edu/etdc/view?acc_num=osu1660927235697858

Abstract Details

2022, Doctor of Philosophy, Ohio State University, Civil Engineering.
Uncertainty quantification is important in many engineering and scientific domains, as uncertainties, of both aleatory and epistemic types, are ubiquitous and inevitable since the complete knowledge cannot be achieved. The probability of failure quantifies the probability of a system failing to meet a specific performance requirement. It is a vital measurement of performance when uncertainties are considered, and it can facilitate the design optimization and decision making for critical infrastructure systems. The computational costs of uncertainty quantification are often prohibitive due to the nature of multi-query analysis and expensive numerical models. Surrogate models can be used to facilitate the reliability analysis. Kriging is the among the most popular surrogate models for reliability analysis due to its capability of providing uncertainty information. How to best utilize the simulation data to construct the Kriging model is a primary research topic in the reliability domain. This dissertation offers the following novel contributions to this research topic: • A novel methodology for adaptive Kriging reliability methods is proposed. It considers the global impact of adding new training points and focuses on reducing the error in the most effective manner. • An effective multi-fidelity reliability method is proposed. The information source and training points can be selected simultaneously to achieve optimal construction of the surrogate model. • A two-phase approach for reliability updating with adaptive Kriging is proposed. The error of posterior failure probability introduced by the Kriging model is quantified. • Adaptive Kriging method is integrated with value of information analysis, and a knowledge sharing scheme is developed to enhance the training efficiency. While surrogate models such as Kriging substantially reduce the computational cost of multi-query analyses, they still require costly simulations of complex computational models. In the past few years, a new paradigm has emerged for direct solution of partial differential equations that form the foundation of many computational models. This dissertation also developed a simulation-free reliability method based on Physics-Informed Neural Networks (PINNs) for uncertainty quantification purposes. The novel contribution regarding this is as follows: • An active learning scheme is proposed for sampling of training points for reliability analysis with PINNs with significant performance gain.
Abdollah Shafieezadeh (Advisor)
Halil Sezen (Committee Member)
Jieun Hur (Committee Member)
200 p.
Civil Engineering
Uncertainty quantification; reliability analysis; reliability updating; value of information analysis; surrogate model; Kriging; Gaussian progress model; multi-fidelity reliability analysis; Physics-Informed-Neural-Networks

Recommended Citations

Citations

  • Zhang, C. (2022). Uncertainty Quantification Using Simulation-based and Simulation-free methods with Active Learning Approaches [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1660927235697858

    APA Style (7th edition)

  • Zhang, Chi. Uncertainty Quantification Using Simulation-based and Simulation-free methods with Active Learning Approaches. 2022. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1660927235697858.

    MLA Style (8th edition)

  • Zhang, Chi. "Uncertainty Quantification Using Simulation-based and Simulation-free methods with Active Learning Approaches." Doctoral dissertation, Ohio State University, 2022. http://rave.ohiolink.edu/etdc/view?acc_num=osu1660927235697858

    Chicago Manual of Style (17th edition)

osu1660927235697858
© 2022, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.