Identifiability based sensor configuration analysis for Bayesian structural identification

When performing structural health monitoring, the design of an optimal sensor configuration aims at improving the effectiveness of structural identification (st-id) and anomaly detection while reducing costs/redundant data. Since this activity takes place when only simulated data is available, model uncertainties have a great impact on the results. This paper studies the application of a hybrid Bayesian methodology for optimal sensor configuration improvement of st-id. It exhibits a comprehensive formulation for uncertainty quantification, aided with multiple response Gaussian processes, and a reduced computational effort relatively to previous works. Information of the optimal sensor position, amount and diversity of data that should be collected, before the actual monitoring phase, can be obtained. An aluminium bridge scale model instrumented with strain gauges/thermocouples, and supported at one end with springs, whose stiffness have to be identified, is used as a case-study. The methodology is applied to a finite element model of the actual structure, and the optimal location of strain gauges is obtained. Validation is carried out by comparison of these results against a subsequent st-id of the bridge supports stiffness, when measurements are made available. Results show that: the used performance criterion is a reliable index of how accurate the identification is (identifiability); the computational effort of the method can be three times less than previous works; and even with poorly informative responses it is possible to design a reliable and cost-effective sensor configuration.