A digital twin-based probability-interval hybrid uncertainty quantification method

The credibility of digital twin predictions is contingent upon their capacity to manage uncertainty. Conventional methods typically characterise all uncertainties via unified probability distributions; however, in scenarios involving sparse data or conflicting expert knowledge, imposing precise distributions upon epistemic uncertainty may yield misleading results. This article proposes a probability-interval mixed uncertainty digital twin framework (PI-DTF) that constructs a dual-channel representation and propagation model. It employs probability distributions to characterise inherent aleatory uncertainty, whilst introducing a ‘belief-plausibility’ structure to describe epistemic uncertainty. The framework computes response probability bounds or belief functions via a nested, double-loop strategy, combining outer-loop Monte Carlo sampling with inner-loop interval optimisation. Furthermore, a dual-track dynamic updating mechanism is implemented, wherein real-time observational data refine probabilistic parameters via Bayesian inference and interval parameters via evidence fusion rules. This article demonstrates the computational process of mixed propagation through a nonlinear analytical numerical case study and validates the framework via an engineering application concerning the structural integrity assessment of critical nuclear power plant piping. Results indicate that, compared with traditional purely probabilistic methods, PI-DTF yields failure probability intervals rather than isolated point estimates, thereby providing more robust risk metrics.