A multivariate, attractor-based approach to structural health monitoring

Abstract

In this work, recent advances in the use of nonlinear time-series analysis for structural health monitoring are extended to incorporate multivariate data. Structural response data recorded at multiple locations are combined using a multivariate time delay embedding in order to reconstruct the structure's dynamical attractor. Using this approach, a global phase-space representation of the dynamics may be realized for spatially extended systems. A new attractor-based metric, chaotic amplification of attractor distortion (CAAD), is then introduced as a damage sensitive feature. The approach is implemented using data acquired from a composite beam, bolted at either end to steel plates. Degradation to the system is introduced as a loosening of the bolts at one end of the structure. Results based on multivariate attractor reconstruction show a clear ability to detect both the presence and magnitude of damage to the connection. Comparisons are then drawn between this approach and one where the same feature is extracted from attractors reconstructed using data acquired from the individual sensor locations. These features are combined “post-extraction” using a linear discriminant coordinant analysis. Performing the analysis separately at the individual sensor locations results in a significant reduction in discriminating power.