A Transfer Learning Application for Damage Identification Across a Population of Experimental Bridges

Labeled health-state data, especially associated to various environmental conditions and damage scenarios, are often limited and difficult to collect when dealing with Structural Health Monitoring (SHM) of real-scale infrastructures. Therefore, to enrich the available dataset, Population-based Structural Health Monitoring (PBSHM) represents an attractive solution, whose main goal is to transfer labels information across a population of similar structures. Despite its promises, applications of PBSHM in bridge monitoring are still missing. In order to contribute filling this research gap, the present paper investigates the transfer between different configurations of a laboratory-scale bridge model, subjected to multiple experimental tests under changing environmental conditions. Several damage scenarios are simulated to obtain realistic structural stiffness reductions along the spans and seizing of the bearings at the top of the supports. Frequency-domain features are extracted from the measured vibration data and afterwards exploited to perform damage assessment via domain adaptation algorithms. The application described in this paper demonstrates the possibility to exchange damage labels and address damage identification across bridge configurations, as well as to improve single asset performance with a multi-source approach utilizing the entire network, thereby overcoming the limitations of conventional Machine Learning-based SHM methods.