The Digital Twin Paradigm suggests an ideal scenario where some virtual model exactly reflects its physical asset counterpart. The reality, as always, may not be that simple. A constrained cost / time / resource budget generally does not allow the engineer to know everything about the physical asset. Data may be discretely sampled, incomplete, noisy, and messy. How can we target an exact virtual representation when our knowledge is uncertain?  

This fundamental mismatch between the ideals of the Digital Twin Paradigm, and the reality of the engineer’s experience, may be navigated with principled treatment of uncertainty in analysis. Probabilistic approaches allow engineers to quantify and propagate uncertainty. Surrogate models, computationally cheap emulations of complex virtual models, may be leveraged in to unlock such probabilistic approaches in practice. The goal moves from "What is the exact state of my physical asset?" towards "What is the risk my asset is unsafe?"