Metamodeling strategies for value of information computation

The benefits of structural health monitoring (SHM) within the life-cycle of a structure include reducing risk, quantifying uncertainties and hence prevention of unnecessary maintenance actions. Value of information (VoI) is one such indicator that has been frequently used to formally quantify the benefit of SHM. Adopting a life-cycle perspective of the incurred costs typically due to failure and maintenance, VoI is defined as the difference between the costs with and without SHM information. Hence, VoI can also be used to compare different SHM strategies and instruments. However, the application of VoI to realistic problems has so far been restricted due to a number of issues, of which the computational complexity is the most significant. As the structural model and the maintenance strategy become more comprehensive (and realistic), the evaluation of VoI becomes intractable due to the increasing number of decision/event possibilities. Metamodeling or surrogate modelling strategies are widely adopted to approximate complex models into simpler functional forms using a limited number of simulations. In this paper, we investigate polynomial chaos expansion (PCE) and a kriging metamodelling framework for VoI computation and demonstrate it for a structural component undergoing linear degradation. The efficiency of the metamodelling strategies is evaluated against a crude Monte Carlo sampling based approach. The reduction in computation costs can enable the implementation of VoI in realistic scenarios. Options for further improvement in the metamodel efficiency are also discussed.