Environmental Effect Compensation in SHM through Conditional Covariances: Application to the Icebreaker "Polarstern"

In structural health monitoring (SHM) systems, data is collected from a multitude of sensors measuring, for example, vibration, the strain on the structure, and potentially multiple different, additional features holding environmental or operational information. Notably, changes in the measured output do not necessarily originate from damage to the structure but are often induced by changes in the environment. Accurately determining these confounding factors - and subsequently removing their influence - is therefore of high importance for reliable diagnostics. Many current methods for compensating environmental or operational effects rely on various forms of so-called response surface modeling, which, however, overlook possible variations in higher-order statistical moments. To address this issue, changes in the (conditional) covariance can be analyzed and quantified using a suitable non-parametric, kernel-based estimator. In prior work, this methodology has been successfully applied to multiple bridges and output quantities such as natural frequencies as well as strain or inclination data. In this work, we consider data gathered during the 2019/2020 MOSAiC arctic expedition, where the icebreaker "Polarstern" had been equipped with 36 strain sensors positioned at various spots across the vessel's deck and frame, as well as ambient temperature sensors. It is shown that both the strain data itself and the various correlations between different strain measurements are strongly impacted by the temperature, highlighting its confounding effect and potential clouding of diagnostically relevant signals, which cannot be easily explained or removed via response surface modeling alone.