Surveillance of Mechanical Systems on the Basis of Vibration Signature Analysis

Abstract

A comprehensive experimental study is presented to assess the utility of a proposed structural health monitoring and damage detection methodology based on vibration signature analysis of the test article. The approach uses a time-domain least-squares-based method to identify the reduced-order system matrices of an equivalent linear model whose order matches the number of available sensors. A quantification of the level of the system nonlinearity is obtained by determining the residual nonlinear forces involved in the system dynamics. The approach is applied to an intricate mechanical system about which virtually no information was available; i.e., the system was essentially a “black box.” By using similar measurements from a reference version of the test article and two subsequently modified versions, it is shown that through the use of higher-order statistics involving the probability density functions of key system parameters, a reliable measure of the extent of variation of the system influence functions may be obtained. The use of measures of the identified quantities’ dispersion offers a practical method for quantifying the reliability of the estimated changes involved in dealing with real-world (i.e., not noise-free) measurements that result in uncertain estimates of the physical changes in the article being monitored. [S0021-8936(00)03303-1]