A Comparative Study of Compressive Sensing Approaches for a Structural Damage Diagnosis

Damage detection and localization from dense sensing networks have been studied in structural health monitoring research community over the past few decades. However, with inevitable rapid growth of monitoring data, data transmission and processing from densely located sensors has become challenging due to the power constraints and computational costs. To address this problem, an image-based damage detection framework with compressive sensing approach is proposed in this paper. This method implements compressive sensing by exploiting the sparsity of the samples in discrete cosine transform (DCT). The suggested methodology divides the sensor network space into blocks and selects subsets of sensors from each block. The selected sensors are regularized through a greedy algorithm called orthogonal matching pursuit (OMP) to obtain the sparsest representation of the samples in DCT domain. Avoiding the recovery of the original signal, the change statistics of DCT coefficients are employed to estimate damage existence and location. Performance of this method with a method previously proposed by the authors based on hierarchical sampling in the sensor network, change point analysis, and Bayesian probability estimation is compared through damage localization of a cracked steel gusset connection model.