Keywords: genetic algorithms, damage detection, fault diagnosis, finite element.

The use of natural frequency as a diagnostic parameter in structural damage detection and vibration monitoring, has been discussed in the last decade for many authors in several works. Following this way, this paper deals with a genetic algorithm based methodology to detect faults through experimental measurements of the natural frequencies with assist of a parametric structural model. There are two methods that are investigated here: the first is based on frequency sensitivity to the damage and the second one is based on optimization techniques (genetic algorithm) and finite element parametric modelling. It is emphasized the advantages of this methodology due to the little amount of information necessary as well as robustness.

Genetic algorithms (GA) are optimization techniques based on the Darwin's Theory [2] of evolution and survival of the fittest. A GA simulates the evolutionary process numerically. They represent the parameters in a given problem by encoding them into a string or as stated in this paper by real coded numbers. A simple genetic algorithm consists of three basic operations, these being reproduction, crossover and mutation. The algorithm begins with a population of individuals each of them representing a possible solution of the problem.

Reference [1] has proposed a multiple damage location assurance criterion (MDLAC) to evaluate the correlation between experimental and numerical values of frequency variations due to damage. This index (MDLAC) assumes 1.0 for fully correlated numerical and experimental frequency changes measured, and 0.0 for non-correlated ones.

The main difference between the MDLAC proposed in [1] and the criteria proposed in this work is that the evaluation of the damage vector does not use approximations, but a structural finite element run for the corresponding damage vector. This ensures that non linear effects of large damage on structural natural frequencies are filtered out since an actual finite element run is accomplished.

A portal frame model that is proposed and analyzed in [3] is also used in this work to show the robustness of the both proposed methods. It was noticed from the analyzed examples that the symmetrical element was always identified together with the actual defective element as was expected. Therefore, both methods show robustness in the identification of the eight numerically created scenarios.

The evaluation results for the genetic algorithm in one site were worse than those from modal sensitivity analysis. Damage occurring on more than one site was not so successfull for the both methods. The genetic algorithm proposed here only uses numerically created scenarios for the identification and evaluation of the natural damaged and undamaged frequencies.

1

A. Messina, E.J. Williams, T. Contursi, "Structural damage detection by a sensitivity and statistical-based method", Journal of Sound and Vibration, No.216, pp.791-808, 1998. doi:10.1006/jsvi.1998.1728

2

C. Darwin. The Origin of Species, 1859.

3

J.E.F. Veizaga, "Identificação de Dano em Estruturas pela Variação das Características Modais", M.Sc. Dissertation (in Portuguese), PROMEC, Universidade Federal do Rio Grande do Sul, Porto Alegre, Brasil, 1993.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £62 +P&P)