Dynamic Imaging: Real-Time Detection of Local Structural Damage with Blind Separation of Low-Rank Background and Sparse Innovation

Abstract

Real-time close-up imaging (filming or video surveillance) of structures is used to automate detection of local component-level damage by exploiting the spatiotemporal data structure of the multiple temporal frames of structures. Specifically, the multiple frames are decomposed into a superposition of a low-rank background component and a sparse innovation (dynamic) component by a technique called principal component pursuit (PCP, or robust principal component analysis). The low-rank component represents the irrelevant, temporally correlated background of the multiple frames, whereas the sparse innovation component indicates the salient, evolutionary damage-induced information. The sparse innovation component is then quantitatively measured for continuous alert and indication of the damage evolution. It is a data-driven and unsupervised (blind) approach that requires no parametric model or prior structural information for calibration. In addition, PCP has an overwhelming probability of success under broad conditions and can be implemented by an efficient convex optimization program without tuning parameters. Laboratory experiments on concrete structures demonstrate that the proposed dynamic imaging method can efficiently and effectively track and indicate the evolution of small or severe damage by the recovered outstanding sparse innovation component (with the low-rank background subtracted from the original images). The proposed method has the potential to benefit real-time automated local damage surveillance and diagnosis of structures where experts’ visual inspection is not needed or not possible.