Real time damage detection using recursive principal components and time varying auto-regressive modeling

• Published in: Mechanical systems and signal processing Vol. 101; pp. 549 - 574
• Main Authors: Krishnan, M., Bhowmik, B., Hazra, B., Pakrashi, V.
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 15.02.2018; Elsevier BV
• Subjects: Algorithms; Autoregressive models; Cantilever beams; Change detection; Computer simulation; Covariance matrix; Damage detection; Damage Sensitive Features (DSF); Degrees of freedom; Earthquake construction; Earthquake damage; Earthquakes; Excitation; Matrix methods; Nonlinear systems; Perturbation methods; Principal components analysis; Real time; Recursive methods; Recursive Principal Component Analysis (RPCA); Regression analysis; Robustness (mathematics); Seismic engineering; Seismic response; Structural damage; Time-Varying Autoregressive Modeling (TVAR); White noise; Recursive Principal Component Analysis (RPCA); Damage Sensitive Features (DSF); Time-Varying Autoregressive Modeling (TVAR)
• ISSN: 0888-3270; 1096-1216; 1096-1216
• DOI: 10.1016/j.ymssp.2017.08.037

•Real time damage detection framework is proposed.•RPCA in conjunction with TVAR modeling is used for online damage detection.•Recursive damage sensitive features are proposed.•Damage detection for an underdetermined system addressed using this framework.•Online spatio-temporal damage detection. In this paper, a novel baseline free approach for continuous online damage detection of multi degree of freedom vibrating structures using Recursive Principal Component Analysis (RPCA) in conjunction with Time Varying Auto-Regressive Modeling (TVAR) is proposed. In this method, the acceleration data is used to obtain recursive proper orthogonal components online using rank-one perturbation method, followed by TVAR modeling of the first transformed response, to detect the change in the dynamic behavior of the vibrating system from its pristine state to contiguous linear/non-linear-states that indicate damage. Most of the works available in the literature deal with algorithms that require windowing of the gathered data owing to their data-driven nature which renders them ineffective for online implementation. Algorithms focussed on mathematically consistent recursive techniques in a rigorous theoretical framework of structural damage detection is missing, which motivates the development of the present framework that is amenable for online implementation which could be utilized along with suite experimental and numerical investigations. The RPCA algorithm iterates the eigenvector and eigenvalue estimates for sample covariance matrices and new data point at each successive time instants, using the rank-one perturbation method. TVAR modeling on the principal component explaining maximum variance is utilized and the damage is identified by tracking the TVAR coefficients. This eliminates the need for offline post processing and facilitates online damage detection especially when applied to streaming data without requiring any baseline data. Numerical simulations performed on a 5-dof nonlinear system under white noise excitation and El Centro (also known as 1940 Imperial Valley earthquake) excitation, for different damage scenarios, demonstrate the robustness of the proposed algorithm. The method is further validated on results obtained from case studies involving experiments performed on a cantilever beam subjected to earthquake excitation; a two-storey benchscale model with a TMD and, data from recorded responses of UCLA factor building demonstrate the efficacy of the proposed methodology as an ideal candidate for real time, reference free structural health monitoring.