Abstract

A digital twin is a computational model of a physical system that is kept aligned (a.k.a. calibrated) with that system as it evolves, and used to predict its behaviour and support decisions. Calibration is formulated as an inverse problem in which the mismatch between model and reality can only be estimated by sampling an expensive stochastic simulator, distinguishing the parameter that is true from the one that best matches model to system, and asking what “matching” should mean when the twin produces a distribution of possible futures rather than a single prediction. Because each evaluation is costly, calibration is then treated as a black-box stochastic optimization problem, with surrogate models and sequential design as the tools that make repeated recalibration affordable. The course closes with systems that change over time, where a twin must recognise that it has fallen out of alignment, and with the relationship between calibrating a twin and using it to decide. Example applications are drawn from epidemiology, queueing systems, and energy.

Syllabus

• Digital twins and stochastic simulators: states, key performance indicators, and what a twin must do to remain useful
• Model discrepancy, non-identifiability, and Bayesian calibration
• Distributional calibration: proper scoring rules, probability integral transforms, and distances between distributions
• Surrogate models for expensive simulators: Gaussian process regression and stochastic kriging under simulation noise
• Sequential experimental design: acquisition criteria and the allocation of simulation effort across the parameter space
• Nonstationarity: distribution-free prediction intervals and change-point detection
• From calibration to decision-making

Short Bio

Sara Shashaani is an Associate Professor and Bowman Faculty Scholar in the Fitts Department of Industrial and Systems Engineering at North Carolina State University. Her research lies at the intersection of stochastic optimization and Monte Carlo simulation, with a focus on zeroth-order methods for nonconvex problems and digital twin methodologies. She is a 2024 Goodnight Innovator and a 2025 MGB-SIAM Early Career Fellow. She received the Best Contributed Paper Award at the 2025 Winter Simulation Conference, and her doctoral student was recognized as a second-place winner of the 2025 IISE Pritsker Doctoral Dissertation Award. Dr. Shashaani has served the INFORMS Simulation Society as an elected board member and is a co-creator of SimOpt, an open-source library for simulation optimization. Her research has had impact across a range of application domains, including renewable energy, climate adaptation, advanced manufacturing, and public health.