Author:

Keywords:

Damage Detection, Genetic Algorithms, Modal Data

Abstract:

This investigation addresses the problem of damage assessment in mechanical and civil engineering structures. The focus is on characterization with regard to detection, location, and quantification of structural damage using vibration data. The principle of vibration-based damage detection techniques is to detect damage using changes in the dynamic characteristics of a structure caused by the damage. Contrary to current nondestructive testing techniques, vibration-based techniques are global damage detection methods, thus they are able to locate damage throughout the whole structure. Therefore, it is not necessary to know an approximate location of the damage or that the location is accessible to be inspected. Particularly, this research focuses on structural damage detection using model-updating methods. Model updating is an inverse method to identify uncertain parameters of a numerical model, and it is usually formulated as an inverse optimization problem. In inverse damage detection, the algorithm uses the differences between models of the structure updated before and after the presence of damage to localize and determine the damage extend. The basic assumption is that damage can be directly related to a decrease of stiffness in the structure. Crucial factors in inverse damage detection are: the construction of an accurate enough numerical model, definition of an appropriate damage parameterization, setting up the objective function and using a robust optimization algorithm. This investigation comprehensively reviews and discusses these factors.The primary contribution of this research is the development of a robust damage detection algorithm. The algorithm uses forced or ambient vibration data to detect, locate, and quantify structural damage. This research presents a novel optimization procedure based upon Parallel Genetic Algorithms. The objective function is constructed after studying several correlation coefficients: the one that provides the best performance is selected. A damage penalization technique is developed. With this technique, the algorithm is capable to discard falsely detected damages due to experimental noise or numerical errors. In addition, the work proposes a methodology to distinguish between changes in the vibration data caused by environmental variability and changes caused by structural damage. Several laboratory and real-life applications are used to apply and validate the proposed algorithm with promising results.