Knowledge sharing for improving damage identification across a population of heterogeneous laboratory-scale aircraft models

Advances in machine learning and pattern recognition have led to the development of data-driven approaches for Structural Health Monitoring (SHM). However, their application in real case studies is often limited by the lack or incompleteness of experimental data. Thus, Population-Based Structural Health Monitoring (PBSHM) addresses these issues by promoting knowledge-sharing between similar structures. The PBSHM theory distinguishes between homogeneous and heterogeneous populations. Structures in a heterogeneous population include different sources of variability, which affect their dynamic response and could reduce the effectiveness of knowledge-sharing performance, leading to so-called negative transfer. This study investigates how attribute variations influence knowledge transfer in a population of heterogeneous laboratory-scale aircraft. The transfer-learning problem is solved via a domain-adaptation algorithm, i.e., the Joint Distribution Adaptation (JDA), considering damage-detection and localisation tasks.