Mechanical and Civil Engineering Seminar
Engineers increasingly rely on computational models to predict adverse events influencing engineering systems and to make critical decisions that impact the safety and well-being of our society. Examples include but are not limited to the safety of structural and infrastructural systems subjected to natural and man-made hazards and the performance and resilience of such systems. Given such high expectations, risk and safety predictions must be formulated based on robust analysis and proper quantification of epistemic and aleatory uncertainties. Furthermore, in the century of digitalization, ubiquitous sensing, data acquisition and statistical learning are changing the nature of risk and safety predictions into dynamic assessments. The advantage of Bayesian learning in answering these calls is unparalleled. In the first part of this talk, I will discuss the challenges associated with rare-event estimation in the presence of Bayesian learning, and I will introduce algorithmic solutions based on Hamiltonian Monte Carlo, subset simulations, and active learning meta-modelling. In the second part of the talk, I will present an application of Bayesian learning and prediction in the context of fluid-induced seismicity associated with subsurface exploitation for energy production. In particular, I formulate a framework based on a hierarchical nonhomogeneous Poisson process, including both epistemic and aleatory uncertainties. I will then show how Bayesian inference can be used to develop a predictive model for the number and magnitudes of fluid-induced seismic events. I will conclude the talk with my vision of future challenges and research opportunities that are emerging on the boundaries of civil engineering, data science, geoscience, and energy systems.