On a Population-Based Structural Health Monitoring Framework: An Aerospace Case Study

Abstract

One of the biggest problems facing the uptake of structural health monitoring (SHM) in industry is a general lack of available damage-state data. Population-based structural health monitoring (PBSHM) is one solution to this problem, expanding the available information and datasets for a monitoring campaign by leveraging measurements across populations of structures and computer models. The information from across a population can be utilised in performing and improving inferences on particular target structures or the complete population as a whole. The challenge in PBSHM is in knowing which information should be shared between individual structures in a population, and how that information can be shared and transferred between members of the population. This paper demonstrates a methodology for performing PBSHM, which is experimentally validated on a population of aircraft wings, transferring label information from a Gnat trainer aircraft wing to an unlabelled Piper Tomahawk aircraft wing dataset. The proposed approach combines abstract representations of structures as graphs with transfer learning technologies, in order to produce a classifier that generalises across the population. The methodology is shown to be successful, with the labelled Gnat dataset diagnosing locations of damage on the (unlabelled) Piper Tomahawk aircraft with 100% accuracy.