Keywords: neural networks, simulation, inverse problems, structural reliability analysis, structural identification.

Advances in computational hardware and software resources since the early 1990s resulted in the development, of new, non-conventional data processing and simulation methods. Among these methods soft computing has to be mentioned as one of the most important approaches to the so-called intelligent methods of information processing. Artificial neural networks (ANNs), expert and fuzzy systems, evolutionary methods are the most popular soft computing techniques. Especially ANNs have been widely used in many fields of science and technology, as well as, an increasing number of problems in structural engineering. From among general problems that can be analyzed by means of ANNs the simulation and identification problems can be classified as follows [1]:

• simulation is related to direct methods of structural analysis, i.e., for known inputs (e.g., excitations of mechanical systems (MSs)) and characteristics of structures or materials outputs (responses of MS) are searched;
• inverse simulation (partial identification, for example, of an unknown excitation) takes place if inputs correspond to known responses of a MS and excitations are searched as outputs of ANNs; and
• identification is associated with the inverse analysis of structures and materials, i.e., excitations and responses are known and the MS characteristics are searched.

In this work the application of ANNs is focused on the simulation, i.e. structural reliability analysis, and identification, i.e. flaw detection, problems. Many sources of uncertainty (material, geometry, loads etc) are inherent in structural design and functioning. Reliability analysis leads to safety measures that a design engineer has to take into account due to the aforementioned uncertainties. Reliability analysis problems, especially when earthquake loadings are considered, are highly computationally intensive tasks since in order to calculate the structural behaviour under seismic loads a large number of dynamic analyses (such as multi-modal response spectrum analyses) are required [2]. Soft computing techniques are used in order to reduce the aforementioned computational cost [3]. An ANN is trained utilizing available information generated from selected multi-modal response spectrum analyses. The trained ANN is then used to predict the maximum inter-storey drift due to different sets of random variables. After the maximum inter-storey drift is predicted, the probability of failure is calculated by means of Monte Carlo Simulation (MCS). The results of the proposed methodology in the test examples show its efficiency and its potential for treating large-scale practical problems.

Another very promising field of soft computing applications in computational mechanics is flaw or damage detection, which basically can be considered as an inverse problem. For example, material or parameter identification problems, which can be formulated as output-error optimization problems, can be solved very efficiently with the proposed technique. For the classical formulation and solution with classical or less classical (e.g. filter-driven or genetic) optimization algorithms and neural networks details can be found in references [4,5,6] and the review article on inverse analysis [1]. In this work the methodology is extended by using a neural network technique for the replacement of the mechanical analysis modelling and, consequently, the genetic optimization for the solution of the inverse crack or defect optimization problem. The effectiveness is compared with the results of the previous, one-method techniques for characteristic test examples.

1

G.E. Stavroulakis, G. Bolzon, Z. Waszczyszyn, L. Ziemianski, "Inverse Analysis", in R. de Borst, H.A. Mang, (Eds.), "Numerical and Computational Methods", Elsevier Publishers, Chapter 13, 685-718, 2003.

2

Y. Tsompanakis, N.D. Lagaros, M. Papadrakakis, "Reliability analysis of structures under seismic loading", in H.A. Mang, F.G. Rammerstorfer, J. Eberhardsteiner (Eds.), Proc. 5th World Congress on Computational Mechanics WCCM-V, Vienna, Austria, July 7-12, 2002.

3

M. Papadrakakis, V. Papadopoulos, N.D. Lagaros, "Structural reliability analysis of elastic-plastic structures using neural networks and Monte Carlo simulation", Comp. Meth. Appl. Mech. Engrg., 136, 145-163, 1996. doi:10.1016/0045-7825(96)01011-0

4

G.E. Stavroulakis, H. Antes, "Flaw identification in elastomechanics: BEM simulation with local and genetic optimization", Structural Optimization, 16(2/3), 162-175, 1998. doi:10.1007/BF01202827

5

G.E. Stavroulakis, "Inverse and crack identification problems in engineering mechanics", Habilitation Thesis, Technical University Braunschweig, Kluwer Academic Publishers, Dordrecht, Boston, London, 2000.

6

M. Engelhardt, A. Likas, G.E. Stavroulakis, "Neural crack identification", in H.A. Mang, F.G. Rammerstorfer, J. Eberhardsteiner (Eds.), Proc. 5th World Congress on Computational Mechanics WCCM-V, Vienna, Austria, July 7-12, 2002.

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 £80 +P&P)