Teaching
Undergraduate Courses
Probability Concepts and Civil Engineering (CE 119)
Introduces students to elements of probability theory (notion of random variable, PDF/CDF, moments, well-known distributions and theorems (CLT)) and statistical data analysis (maximum likelihood estimation, confidence intervals, hypothesis testing, regression). Intro to coding with Python (open and visualize a dataset, create functions to perform statistical analysis routines and use Python packages numpy, scipy, scikit-learn).
Offered every Fall semester.
Graduate Courses
Uncertainty Quantification Concepts for Civil and Environmental Engineering (CE 599)
New course introduced in 2021. Introduces students to various concepts in uncertainty quantification, starting from basics of probability theory (random events, descriptors of univariate and multivariate random variables, well-known distributions and theorems (CLT, Bayes' theorem)), and moving forward with an intro to random processes, techniques for forward propagation of uncertainties (Monte Carlo simulation, perturbation methods) and probabilistic inverse modeling (frequentist and Bayesian paradigms). Last lecture opens up on research UQ fields such as sensitivity analysis or surrogate modeling.
Grading includes homeworks, in-class exam and a project (can be designed by the student to meet their research needs).
Offered every Fall.
Structural Identification and Health Monitoring (CE 599)
New course, introduced in Spring 2023. Introduces students to the field of structural identification and structural health monitoring, with a focus on vibration-based techniques and identification of dynamical systems. Subjects covered include:
Introduction and core concepts: Hierarchy of damage identification (from identification to localization to prognosis). Importance of uncertainty quantification. Data-based vs. model-based methods.
Modal identification (includes a review of dynamics and frequency domain analysis), focus on the FDD (Frequency Domain Decomposition) method
Parameter estimation using frequentist (maximum likelihood estimation, link with OLS, normal equations with example to training ARX model for time-series modeling, parameter uncertainty via Fisher information matrix, AIC for model selection) and Bayesian methods (Bayesian updating, Laplace method and MCMC, discussion about parameter identifiability).
Follow-up to Bayesian methods: Bayesian filtering for on-line identification of dynamical systems
Machine learning for SHM: SHM as a statistical pattern recognition, unsupervised learning and outlier detection methods. Recent advances in SHM including deep learning for vision-based SHM.