Active transfer learning for structural health monitoring

TL;DR

This paper addresses data scarcity and cross-structure distribution shifts in structural health monitoring (SHM) by proposing an online active transfer-learning framework that integrates a Bayesian domain-adaptation model (DA-RVM) with stream-based active learning. The approach uses a linear, interpretable target mapping (scale $\mathbf{s}$, translation $\mathbf{t}$, rotation $\boldsymbol{\theta}$) to align source and target domains while a relevance-vector machine (RVM) classifier learns shared boundaries, updating the mapping as target labels become available. The method is validated on a population of lab-scale bridges under varying temperatures and pseudo-damage states, showing data-efficient learning and the ability to classify health-states not yet observed in the target domain, often with substantially fewer labelled samples than fully supervised approaches. The results imply meaningful practical impact: inspections can be guided by informative queries that reduce operational costs while maintaining decision confidence, enabling data-informed operation and maintenance in PBSHM.

Abstract

Data for training structural health monitoring (SHM) systems are often expensive and/or impractical to obtain, particularly for labelled data. Population-based SHM (PBSHM) aims to address this limitation by leveraging data from multiple structures. However, data from different structures will follow distinct distributions, potentially leading to large generalisation errors for models learnt via conventional machine learning methods. To address this issue, transfer learning -- in the form of domain adaptation (DA) -- can be used to align the data distributions. Most previous approaches have only considered \emph{unsupervised} DA, where no labelled target data are available; they do not consider how to incorporate these technologies in an online framework -- updating as labels are obtained throughout the monitoring campaign. This paper proposes a Bayesian framework for DA in PBSHM, that can improve unsupervised DA mappings using a limited quantity of labelled target data. In addition, this model is integrated into an active sampling strategy to guide inspections to select the most informative observations to label -- leading to further reductions in the required labelled data to learn a target classifier. The effectiveness of this methodology is evaluated on a population of experimental bridges. Specifically, this population includes data corresponding to several damage states, as well as, a comprehensive set of environmental conditions. It is found that combining transfer learning and active learning can improve data efficiency when learning classification models in label-scarce scenarios. This result has implications for data-informed operation and maintenance of structures, suggesting a reduction in inspections over the operational lifetime of a structure -- and therefore a reduction in operational costs -- can be achieved.

Figures (22)

• Figure 1: A demonstration of the assumptions made by conventional unsupervised DA (top) and the active DA approach proposed in this paper (bottom).
• Figure 2: Flow chart showing the process of learning a predictive model with transfer learning (a) and stream-based active learning (b).
• Figure 3: Graphical model representation of the proposed DA-RVM. Nodes correspond to variables: shaded nodes denote observed variables, solid outlines indicate random variables, and dotted outlines represent deterministic nodes. Arrows without a connected parent node indicate prior distributions. Plates represent replicates over dimensions for the mapping variables and classes for classifier weights.
• Figure 4: Toy example with shaded regions showing the entropy in label predictions produced by a Bayesian logistic regression model (left) and an RVM (right).
• Figure 5: Flow chart to illustrate the active-learning process with DA.