Abstract Summary
In the field of civil infrastructures, Structural Health Monitoring generally suffers from a scarcity of labelled data describing the damage classes and rare environmental conditions. To solve this issue, the proposed work adopts a Transfer Learning approach for leveraging information from a source structure, characterised by a rich class of damage labels, to im-prove inferences on a target structure with limited knowledge and sparse datasets. The goal is to train a machine learning algorithm on a bridge undergoing damage and to afterwards transfer these labelled damage-state data across the members of the investigated population. Given possible differences in the underlying distributions of each structure, Domain Adapta-tion techniques are applied to match the domains in a shared feature space. The methodolo-gy is validated on a heterogeneous population composed of two numerical bridges of differ-ent geometry and materials. Finite Element Models are built to simulate artificial data stemming from operational conditions and multiple damage scenarios. Results prove the ef-fectiveness of the presented approach to successfully exchange health-state information within a population of bridges, thus reducing the computational burden and enabling popu-lation-based SHM.
