Sequential Bayesian diagnosis and prognosis of structural deterioration from measured structural responses

Abstract

Structural health monitoring of deteriorating structures requires the latent deterioration process to be inferred indirectly from measured structural responses and propagated to quantities of engineering interest. This paper presents a probabilistic framework that formalizes this process by distinguishing Bayesian inference of uncertain deterioration and model parameters from diagnosis of the current deterioration state, prognosis of its future evolution, and prediction of structural reliability. The deterioration model thereby provides the common probabilistic link between monitoring-based inference and these downstream tasks. To enable sequential inference as monitoring information accumulates, Bayesian Updating with Structural Reliability Methods (BUS) is extended through a nested formulation of the BUS events associated with successively growing monitoring datasets. The resulting Sequential BUS method allows Subset Simulation to continue from the augmented conditional sample population obtained at the preceding monitoring stage rather than restarting from the prior distribution. The posterior population can then be propagated to monitoring-informed reliability analysis. The framework is demonstrated for fatigue deterioration of a redundant jacket-type structure, where modal properties obtained from simulated vibration monitoring are used to infer a high-dimensional deterioration model of welded connections. The results show how successive monitoring observations constrain plausible deterioration histories, provide probabilistic system-level diagnosis, and update crack-growth prognosis and structural reliability while retaining ambiguity between deterioration scenarios that produce similar structural responses. The proposed framework thus provides a coherent probabilistic route from response-based monitoring data to deterioration diagnosis, prognosis, and reliability prediction.