Impact of parameters on damage indicators of structures:

Symbolic data analysis and principal components

Abstract

Several methods have been proposed for detecting damage to structures. Some of these methods are based
on vibrational data, where it is known that changes in physical properties can be noted through changes in
vibrational characteristics. Most techniques are based on variations of modal parameters. The direct use of
accelerations offers significant advantages in damage detection, making the process faster and more
straightforward by eliminating the need for an identification stage. The scarce exploration of raw dynamic
data (accelerations) is due to the high number of data to be stored and manipulated. Some authors suggest
Symbolic Data Analysis (SDA) and Principal Components Analysis (PCA) as methods for reducing the volume
of data without losing fundamental characteristics. This work aims to analyze the number of categories used
in SDA and the number of principal components in PCA that achieve the best damage state classification
results applied to experimental data from a simply supported beam and numerical data from a 4-story
Benchmark. Two techniques for determining the number of SDA categories are proposed.