B0515
Title: Confounder-adjusted covariances of sensor outputs and applications to structural health monitoring
Authors: Lizzie Neumann - Helmut Schmidt University (Germany) [presenting]
Philipp Wittenberg - Helmut Schmidt University (Germany)
Jan Gertheiss - Helmut Schmidt University (Germany)
Abstract: Covariances of sensor outputs or derived features such as natural frequencies are the basis or an important building block of many methods used for damage detection in structural health monitoring. This article discusses a nonparametric, kernel-based estimate of a conditional covariance matrix that considers confounder information such as temperature. The approach on both raw sensor measurements (acceleration, strain, inclination) and derived features (natural frequencies) are illustrated. In particular, conditional covariances allow correcting for known/available environmental and operational conditions in principal component analysis in an explicit fashion. Thus, the approach improves an output-only method for removing external influences by incorporating additional input information that would otherwise be ignored.
