Structural Health Monitoring Sensor Network Optimization through Bayesian Experimental Design

Abstract

Structural health monitoring (SHM) may be exploited to estimate the mechanical properties of existing structures and check for potential damage. Among commonly used methodologies for property characterization, the Bayesian approach holds the lead because it is endowed with the particular advantage of quantifying associated uncertainties. These uncertainties arise owing to diverse factors including (1) sensor accuracy and positioning, (2) environmental influences, and (3) modeling errors. In minimizing the influence of sensor-related uncertainties, an optimal design may be adopted for the SHM campaign to maximize the information content of the measurements. Here, a procedure based on Bayesian experimental design is proposed to quantify the expected utility of the sensor network. The positions of the used sensors are selected in a way that maximizes the Shannon information gain between the prior and posterior probability distributions of the parameters to be estimated. In order to numerically solve the resulting optimization problem, surrogate models based on polynomial chaos expansion (PCE) and stochastic optimization methods are used. The use of surrogates allows one to reduce the computational cost of the associated model runs. The method is applied to a large-scale example, namely the Pirelli Tower in Milan.