An examination of some aspects of factor analysis in damage detection

Abstract

It is well known that dynamic properties can be affected by environmental factors and this complicates the identification of damage. Among procedures explored to mitigate the fluctuations in dynamic features from environmental effects factor analysis is one of the simplest and has the merit that it operates without the need to measure the environmental variables. The method has been examined by several researchers in recent years but experience is limited and questions on some technical details remain open. This paper presents an investigation on three aspects that may affect the performance. The most important is the number of factors used in the analysis. The second relates to whether the diagonal nature on the covariance of random errors is enforced or not, and the third examines the influence of the criterion used to solve the over-determined set of equations that arise in the method. Results are obtained by simulating a system where the stiffness is cubically related to changes in temperature and where the temperature field, in each simulation, is a realization of a random process with a prescribed spatial correlation.