Optimal sensor placement for state estimation of a thin double-curved shell structure

Thin double-curved shell structures are often found in architecture and engineering when large areas are spanned without intermediate supports. Due to their large span and low weight, the influence of external loads, like snow and wind, is significant and may lead to inefficiencies in the load bearing behavior as well as induce vibrations. Research has shown that such negative effects can be compensated by adding active components to the structure, allowing a load- and oscillation-dependent response to improve structural stability. A principal requirement of such systems is the determination of the current state through the implementation of appropriate sensors. This paper presents a method for optimal sensor placement on shell structures such that the state of oscillation of the system can be reconstructed and model-based methods for active vibration damping can be applied. The method uses the number of sensors as a surrogate for implementation cost and an observability measure as optimization objectives. The latter is derived from the observability gramian and considers the average observation energy. The influence of measurement noise and model uncertainties on the observability of the system is taken into account explicitly. The method is tested on a flexible thin shell structure that is modeled by Finite Element Methods using ANSYS. The equations of motion are transformed into modal space where model reduction methods are applied. The resulting model is used to optimize the sensor locations. The optimization is performed by the Multi-objective Simulated Annealing algorithm. The proposed concept is tested on an experimental plant and optimization results and exemplary optimal sensor configurations are presented.

► A method for optimal sensor placement for state estimation of flexible structures is developed.
► The optimization criterion includes an energy-based observability criterion and the installation cost.
► Measurement noise (sensors) and modeling errors (system) are explicitly taken into account for the optimization.
► The presented method is tested on an experimental plant at University of Stuttgart.