Data-driven multivariate algorithms for damage detection and identification: Evaluation and comparison

Abstract

This article is concerned with the experimental validation of a structural health monitoring methodology for damage detection and identification. Three different data-driven multivariate algorithms are considered here to obtain the baseline pattern. These are based on principal component analysis, independent component analysis and hierarchical non-linear principal component analysis. The contribution of this article is to examine and compare the three proposed algorithms that have been reported as reliable methods for damage detection and identification. The approach is based on a distributed piezoelectric active sensor network for the excitation and detection of structural dynamic responses. A woven multilayered composite plate and a simplified aircraft composite skin panel are used as examples to test the approaches. Data-driven baseline patterns are built when the structure is known to be healthy from wavelet coefficients of the structural dynamic responses. Damage is then simulated by adding masses at different positions of the structures. The data from the structure in different states (damaged or not) are then projected into the different models by each actuator in order to generate the input feature vectors of a self-organizing map from the computed components together with squared prediction error measures. All three methods are shown to be successful in detecting and classifying the simulated damages. At the end, a critical comparison is given in order to investigate the advantages and disadvantages of each method for the damage detection and identification tasks.