Railway Bridges Health Monitoring Supported by Artificial Intelligence

This chapter discusses the detection of damages in railway bridges based on vibration responses induced by traffic and using bridge health monitoring systems. To achieve this goal, an innovative data-driven Artificial Intelligence (AI)-based methodology is proposed, consisting of a combination of time series analysis and advanced multivariate statistical techniques for an unsupervised learning approach. Different combinations of techniques are implemented and tested to achieve the most robust, generic, and effective one. Damage sensitive features of train induced responses are extracted and allow taking advantage, not only of the repeatability of the loading, but also, and more importantly, of its great magnitude, thus enhancing the sensitivity to small-magnitude structural changes. A comparison between the performance obtained from AutoRegressive (AR) and AutoRegressive Exogenous (ARX) models as feature extractors is conducted. The use of a regression-based method such as Multivariate Linear Regression (MLR) or a latent variable method such as Principal Component Analysis (PCA) grants the strategy the ability to remove environmental and operational effects and proves the importance of feature modelling. Feature discrimination is addressed by evaluating the performance of outlier analysis and clustering algorithms. The efficiency of the proposed methodology is verified on a complex bowstring-arch railway bridge. A digital twin of the bridge is used to simulate baseline and damage conditions by performing finite element time-history analysis using as input measurements of real temperatures, noise effects, train speeds, and loads. The methodology proved to be highly robust to false detections and sensitive to early damage by automatically identifying small stiffness reductions in the concrete slab, diaphragms, and arches, as well as friction increase in the bearing devices.