This paper proposes an optimal sensor placement (OSP) framework for parameter estimation, virtual sensing, and condition monitoring using information theory. The framework uses a Bayesian OSP method combined with modal expansion to minimize the information entropy about quantities of interest (QoI), such as strain time histories at critical locations of the structure, without the knowledge of input excitation. The proposed optimization framework also accounts for variations in sensor installation cost at different locations on the monitored structure. The framework is evaluated numerically using a realistic model of an offshore wind turbine on a jacket support structure under installation cost assumptions and considering information entropy of the QoI. The QoI in this numerical study are considered to be the strain time history at one or more locations on the support structure in one problem and the parameters of the structure in the other. A correlation length is considered to account for the spatial correlation of data between adjacent sensors. Effects of the correlation length and input loads on the OSP results for parameter estimation are studied. The considered structural parameters for estimation in this study include (1) modulus of elasticity of tower elements (tower stiffness), (2) modulus of elasticity of jacket elements (jacket stiffness), and (3) vertical foundation spring (soil stiffness). The effect of a subjective weight between the information entropy and sensor configuration cost on the OSP results is also investigated. Different optimal designs are achieved for different weight factors, and the Pareto solutions for OSP are presented. It is found that the OSP framework is an effective tool for decision-makers considering the cost of instrumentation. The presented Pareto optimal solutions can give insight into the value of OSP given a limited budget.
Read full abstract