Robust Bayesian optimal sensor placement for model parameter estimation and response predictions

Abstract

Optimal sensor placement (OSP) in complex systems implies a configuration that maximizes the information gain by the sensors. This configuration is identified by maximizing, with respect to the location of the sensors, an expected utility function on information-based measures. The resulting multi-dimensional integrals are estimated using asymptotic approximations and sampling techniques. Herein we extend such formulations to be robust to uncertainties in nuisance parameters associated with system and prediction error model conditions, and environmental variabilities. The nuisance parameters are not inferred from the collected data but they affect the learning or prediction processes. Robust designs are achieved by maximizing the expected amount of information in the data and minimizing its fluctuation due to uncertainties in the nuisance parameters. We demonstrate, through a number of examples in structural dynamics, the capabilities of the proposed OSP framework for model parameter estimation and model response predictions. © 2020 European Association for Structural Dynamics. All rights reserved.