Application of the polynomial chaos expansion method for forecasting structural response of two full-scale case studies

Abstract

Predicting the behavior of engineering structures with high accuracy remains a challenging task as a result of their continuous interaction with the immediate environment and varying operating conditions. In that context, forecasting tools are primarily focused on the creation of a model of a so-called baseline system. This established model serves as a foundation for identifying changes when new outputs deviate from the predictions made by the model. Physics-based numerical models, like the finite element method, often carry significant uncertainty stemming from assumptions regarding structural characteristics, environmental influences, and various loads affecting the system under study. Consequently, identifying the source of any existing discrepancies between obtained model results and measured data is difficult. This paper demonstrates a straightforward implementation of the polynomial chaos expansion method for the formulation of prognostic data-driven models targeted at tracking changes in continuously measured structural response. The method's effectiveness and positive features are showcased via practical application onto two full-scale engineering structures: a concrete arch dam and an industrial steel chimney. The models utilize environmental as well as response data collected over two years and two months of monitoring of these structures, respectively. The obtained results reveal the models' considerable potential as a long-term monitoring tool for autonomous assessment of structural behavior.

Keywords

• structural health monitoring
• polynomial chaos expansion
• data-driven models
• engineering structures
• sensitivity
• uncertainty
• forecasting