On Partitioning of an SHM Problem and Parallels with Transfer Learning

Abstract

In the current work, a problem-splitting approach and a scheme motivated by transfer learning is applied to a structural health monitoring problem. The specific problem in this case is that of localising damage on an aircraft wing. The original experiment is described, together with the initial approach, in which a neural network was trained to localise damage. The results were not ideal, partly because of a scarcity of training data, and partly because of the difficulty in resolving two of the damage cases. In the current paper, the problem is split into two sub-problems and an increase in classification accuracy is obtained. The sub-problems are obtained by separating out the most difficult-to-classify damage cases. A second approach to the problem is considered by adopting ideas from transfer learning (usually applied in much deeper) networks to see if a network trained on the simpler damage cases can help with feature extraction in the more difficult cases. The transfer of a fixed trained batch of layers between the networks is found to improve classification by making the classes more separable in the feature space and to speed up convergence.

On the application of kernelised Bayesian transfer learning to population-based structural health monitoring

“…The Gnat dataset has been well-studied [40,41,[44][45][46][47], and is an experimental dataset in which an aircraft wing was excited with a band-limited Gaussian noise and the response measured at several locations via uni-axial accelerometers. Inspection panels were subsequently removed, giving pseudo-damage scenarios.…”

Challenges for SHM from Structural Repairs: An Outlier-Informed Domain Adaptation Approach

“…In previous work, it was identified that the smaller panels caused confusion during localisation [26]. Furthermore, recent analysis on the dataset has demonstrated that a hierarchy of classifiers can be used to classify whether the class belongs to the large or small panel subsets, before localisation to each specific panel, with the approach showing boosted classification performance when compared to the complete problem [27]. For these reasons this paper considers the localisation problem considering only the large panels, i.e.…”

On the Application of Heterogeneous Transfer Learning to Population-Based Structural Health Monitoring

“…The Gnat aircraft dataset is an experimental dataset that has been well-studied within the field of SHM [36,37,38,39,40,41]. Although the dataset is from one structure, namely an aircraft wing from a Gnat trainer aircraft [37,38], the dataset does form two distinct data domains due to data shift that occurs from changes caused by reattaching inspection panels (even when the torque of the fasteners is monitored [38]).…”