Selecting high-quality non-sensor locations from reconstructed dynamic responses of double-layer cylindrical reticulated shells considering step excitations

Abstract

Because long-span spatial structures have a large number of nodes and members, monitoring tasks such as model updating and damage detection necessitate a large amount of high-quality test information. More test information can be obtained by reconstructing dynamic responses for non-sensor areas. However, because reconstruction errors are unknown, current study cannot tell whether a non-sensor location has high reconstruction accuracy or not. Thus, a novel technique for selecting high-quality non-sensor locations from reconstructed dynamic responses obtained solely from finite sensor test information was suggested. Firstly, dynamic responses are reconstructed by linearly combining only a few fixed contributing modes. This neatly converts the issue of reconstructing dynamic responses at non-sensor locations to that of estimating contributing mode coefficients (i.e., generalized coordinates). An iterative strategy for determining optimal sensor locations was proposed in order to achieve unbiased estimation of contributing mode coefficients at each moment. The dynamic responses at non-sensor locations were naturally reconstructed using estimates of contributing mode coefficients. Secondly, following the reconstruction error analysis, a correlation coefficient index was proposed to select high-quality non-sensor locations with high reconstruction accuracy in a reliable manner. As a typical representative of spatial structures, a double-layer cylindrical reticulated shell was taken as the example to validate the proposed method. Numerical results reveal that by using the proposed method, dynamic responses at non-sensor locations can conveniently be reconstructed and many high-quality non-sensor locations can reliably be selected. This explorative work is conducive to increasing the quantity of high-quality test information in a single dynamic test by using very few sensors, with test efficiency greatly improved and test cost significantly saved.