A probabilistic digital twin framework is developed for fatigue prognosis of aerospace structures by fusing passive sensing based on higher harmonic analysis, historical test data, and physics-based crack-growth models within a Dynamic Bayesian Network (DBN). Three progressive data assimilation schemes are evaluated: (i) purely online sensing, (ii) hybrid transition from historical-to-online data, and (iii) conditional progressive fusion of both sources. Transfer learning leverages offline experiments to constrain priors, enabling rapid convergence of material parameter posteriors and effective initial flaw size estimation while reducing the volume of online observations. Experimental validation on a metallic plate with a central hole under constant-amplitude cyclic loading demonstrates accurate whole-life prognosis, with DBN predictions closely tracking observed crack evolution while quantifying measurement, material, and model uncertainties via calibrated credible intervals. Benchmarking against Gaussian Process Regression (GPR) highlights the DBN’s superior sequential uncertainty quantification, while GPR offers computational efficiency for near-real-time extrapolation. The hybrid DBN framework thus provides robust, uncertainty-aware life prognosis, advancing digital twin methodologies for aerospace applications.
Naderi et al. (2026) studied this question.