Optimal placement of non-redundant sensors for structural health monitoring under model uncertainty and measurement noise

Abstract

Dense distribution of sensors for structural health monitoring can supply sufficient – yet redundant – information, especially for sensors with high reliability and excellent quality. The design space is thus classified into several clusters conveying equivalent information to assist in a non-redundant sensor layout and avoid dense distributions. Further, practical issues of model uncertainty and measurement noise should also be considered. Based on the effective independence method, the sensor design problem in this work is formulated with two optimization objectives under model uncertainty and measurement noise. Gaussian process regression model is employed to relieve the computation burden when evaluating two objectives. Accordingly, a methodology for robust design of non-redundant sensors is newly developed for the first time, and demonstrated via application to case studies. Optimized designs disperse sensors in the space and tend to place sensors where small amplitudes of dynamic information are exhibited to be robust with respect to uncertainties.