Keywords: damage identification, artificial neural network, model updating, modal analysis, finite element method, natural frequencies.

In this paper the use of classical model updating methods and neural network methods for damage localization are presented. Both methods are based in modal parameters and have been applied to the same experimental problem. The main objective of this research is to establish the comparatives qualities of each method and explore the possibility of combining both methods. Model updating is a consolidated method for damage detection, but it is a long process that requires a considerable computational and experimental cost. Artificial neural networks (ANN) methods are faster and consequently they require less computational cost if the neural network is properly chosen and trained [1].

A laboratory modal test of a beam supported at four points has been developed in order to apply finite element model update and damage detection. In a first step model updating has been made by means of an iterative method [2]. In a second step a small damage is introduced to the beam and damage localization is applied with both iterative and neural network methods.

The full process of numerical model updating is developed. Guyan reduction and the independence effective method have been used to select the best transducer positions. Correlation, validation and adaptation of experimental and numerical data techniques have been used. The main adaptation technique used is the improved reduction system (IRS). The two principal correlation techniques are: the modal assurance criterion (MAC) and the orthogonal mode comparison method (OMCM). Also parameterization, damage localization and sensibility analyses have been completed before the final updating. Different iterative methods based in penalty functions have been applied depending of the amount of experimental data and the number of parameters to update.

The ANNs have been trained with the results obtained from the numerical simulations using the finite element method. These numerical simulations have been done introducing damage with different locations in the beam. Once the ANN is trained, it is evaluated as a tool to predict the position of the damage with the natural frequencies measured in the laboratory modal test.

The main conclusion is that the updating model is a really good technique in order to approximate the real boundary conditions of a structure. In the method based on the finite element method and ANNs, it is important to make a good calibration of the ANN structure that is appropriated for each problem, because only one of the six ANN structures provided good capabilities to detect the damage position.

1

A. Garcia-Gonzalez, A. Gonzalez Herrera, A. Garcia Cerezo, "Damage Localization based on Modal Parameters using the Finite Element Method and Neural Networks", in B.H.V. Topping, J.M. Adam, F.J. Pallarés, R. Bru and M.L. Romero, (Editors), "Proceedings of the Tenth International Conference on Computational Structures Technology", Civil-Comp Press, Stirlingshire, UK, 2010. doi:10.4203/ccp.93.48

2

M.I. Friswell, J.E. Mottershead, "Finite Element Model Updating in Structural Dynamics", Kluwer Academic Publishers, 1996.

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 £145 +P&P)