Foundations of population-based SHM, Part III: Heterogeneous populations – Mapping and transfer

• Published in: Mechanical systems and signal processing Vol. 149; p. 107142
• Main Authors: Gardner, P., Bull, L.A., Gosliga, J., Dervilis, N., Worden, K.
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 15-02-2021; Elsevier BV
• Subjects: Adaptation; Classifiers; Context; Domain adaptation; Domains; Foundations; Graphical representations; Knowledge management; Labels; Mapping; Population; Population-based structural health monitoring; Structural health monitoring; Topology; Transfer learning; Domain adaptation; Population-based structural health monitoring; Transfer learning
• ISSN: 0888-3270; 1096-1216
• DOI: 10.1016/j.ymssp.2020.107142

•For the first time a new population-based methodology is proposed that alleviates problems with sparsity of data in SHM.•The applicability of domain adaptation assumptions when considering each population type and SHM problem are outlined.•A novel MMD-based algorithm is proposed for an (L + 1)-problem.•A new approach for using physics-based models and domain adaption is proposed for labelling data from structures in PBSHM.•This is the third paper (of three parts) covering the foundations of population-based SHM.
This is the third and final paper in a series laying foundations for a theory/methodology of Population-Based Structural Health Monitoring (PBSHM). PBSHM involves utilising knowledge from one set of structures in a population and applying it to a different set, such that predictions about the health states of each member in the population can be performed and improved. Central ideas behind PBSHM are those of knowledge transfer and mapping. In the context of PBSHM, knowledge transfer involves using information from a source domain structure, where labels are known for given feature sets, and mapping these onto the unlabelled feature space of a different, target domain structure. This mapping means a classifier trained on the transformed source domain data will generalise to the unlabelled target domain data; i.e.a classifier built on one structure will generalise to another, making Structural Heath Monitoring (SHM) cost-effective and applicable to a wide range of challenging industrial scenarios. This process of mapping features and labels across source and target domains is defined here via domain adaptation, a subcategory of transfer learning. A mathematical underpinning for when domain adaptation is possible in a structural dynamics context is provided, with reference to topology within a graphical representation of structures. Subsequently, a novel procedure for performing domain adaptation on topologically different structures is outlined.