Real time structural damage detection using recursive singular spectrum analysis

Abstract

A baseline independent approach towards identifying structural damage in real time for dynamically vibrating systems using recursive singular spectral analysis (RSSA) is proposed. The principal characteristic of the method lies in its inherent ability to identify events using the inputs from a single sensor. The algorithm takes the input of a single channel in real time and produces a time lagged Hankel matrix from the series. Streaming acceleration data is used to obtain recursive proper orthogonal components online using first order perturbation (FOP) method. This generates transformed responses at each time stamp, which provide information regarding the current state of the structure. Subsequently, the need for offline post processing is eliminated by facilitating a continuous real time monitoring, without resorting to baseline data. As time series models work extremely well in discerning the key patterns of such responses, time-varying autoregressive (TVAR) modeling is adopted. Numerical simulations performed on a nonlinear base isolated system excited using white noise provide successful detection results. The applicability of the proposed algorithm towards experimental studies is described in detail using a cantilever beam setup subjected to earthquake excitation. Successful detection results indicating a change of state in real time validate the robustness of the algorithm towards catering a wide range of application specific problems.