Human–Machine Collaboration Framework for Bridge Health Monitoring

Abstract

In bridge health monitoring (BHM), a prominent goal is to rapidly deliver assessment metrics for these essential and aging urban lifelines when subjected to natural hazard. A vibration-based machine learning (ML) BHM paradigm has been established over the past three decades to allow near-real-time automated health state classification, with a particular focus on the tasks of feature engineering and ML damage identification. This paper presents the human–machine collaboration (H-MC) framework to address challenges of this paradigm, especially in the context of reinforced concrete highway BHM. These challenges include specification of strong motion events, data multidimensionality, and ML model interpretability. The H-MC framework for BHM employs the techniques of multivariate novelty detection and probability of exceedance envelope models with ordinal filter-based feature selection to maximize the use of available data from both recorded and simulated events while maintaining the statistical and physical significance of the results. The framework is demonstrated using a numerical example and two case studies. The findings show the effectiveness of the proposed method for efficient damage assessment to facilitate rapid decision-making.