Data-driven analysis models for slender structures using explainable artificial intelligence

This project aims to study dynamic behavior of slender structures using structural health monitoring (SHM) data. An explainable artificial intelligence (XAI) approach, specifically an explainable machine learning approach, is proposed as a conceptual basis. A suite of machine learning techniques will be created that, compared to many “traditional” machine learning approaches, produce more explainable models, while maintaining high learning performance and prediction accuracy. As opposed to “black box” models inherent to many traditional machine learning approaches employed in civil and environmental engineering, so-called "glass box" models will be introduced that are explainable to engineers in practice. The machine learning techniques, validated in preliminary work, will focus on specific aspects of the dynamic behavior of slender structures, which have proven responsible for causing excessive oscillations. SHM data recorded from a relay tower and from a pedestrian bridge, studied by the German applicant and by the Greek collaborating partner, respectively, will be analyzed on a data-driven basis without underlying physical principles a priori considered, in an attempt to complement knowledge derived from theory on the dynamic behavior of slender structures, thus creating a holistic view on the performance of slender structures subjected to dynamic loads. For example, resonant phenomena between wind loads and the structural response of towers (vortex-induced vibrations), flutter, galloping, or the interaction between the dynamic loads and the structural response (“lock-in” effect) will be investigated. It is expected that the XAI approach will enable engineers to (better) understand, appropriately trust, and effectively manage the generation of artificially intelligent slender structure analyses in engineering practice.