A digital twin is an evolving virtual model of a specific system or physical asset, assimilating asset lifecycle data so that the digital twin becomes a dynamically updated asset-specific model that underpins intelligent automation and drives key decisions. Digital twins have potential impact across critical areas of national security, industrial development, and societal well-being. If made reliably predictive, digital twins could revolutionize key decision-making processes that depend on dynamically evolving estimates of the state of a complex system. This paper illustrates how a predictive digital twin – one that combines data-driven learning with predictive physics-based modelling – can contribute to improved mission readiness. The digital twin is represented mathematically as a probabilistic graphical model in which the key elements of state, control, observations, quantities of interest, and reward are modelled as random variables. The graphical model represents the relationships between these different elements, as well as their evolution in time and their uncertainties. The formulation is illustrated for the development of a structural digital twin for an unmanned aerial vehicle (UAV). The digital twin combines high-fidelity structural finite element models, computationally efficient reduced-order models, and observational data generated from onboard structural sensors. An illustrative example shows how the digital twin is updated as the UAV undergoes in-flight structural degradation and then used to optimally re-plan the mission trajectory.