Learning loads on in‐service underground infrastructure with a trans‐dimensional Bayesian inversion method

Monitoring external loads on underground structures is crucial for structural health assessment. Inverting earth pressures from observable structural responses, such as deformation data, holds promise. However, existing methods often rely on presumptions about pressure complexity, which can be infeasible for many poorly performing in-service infrastructures. This paper proposes a trans-dimensional Bayesian method that simultaneously infers both the complexity and magnitude of earth pressures by parameterizing a set of a priori unknown variables, where the exact number of parameters remains undetermined. A Bayesian framework is employed to represent the posterior distribution of these parameters, with a trans-dimensional Markov chain specifically designed for statistical inference of the distributed pressures. Case studies demonstrate that the proposed method outperforms traditional methods, which are limited by rigid presumptions. Furthermore, it is shown that the inferred pressures can reproduce comprehensive structural responses, such as internal forces, providing new tools and insights for structural health monitoring of underground infrastructures.