Computational & Technology Resources
an online resource for computational,
engineering & technology publications 

CivilComp Proceedings
ISSN 17593433 CCP: 83
PROCEEDINGS OF THE EIGHTH INTERNATIONAL CONFERENCE ON COMPUTATIONAL STRUCTURES TECHNOLOGY Edited by: B.H.V. Topping, G. Montero and R. Montenegro
Paper 57
Robust Optimization for EarthquakeResistant Design J.E. Hurtado and N. Aguirre
National University of Colombia, Manizales, Colombia J.E. Hurtado, N. Aguirre, "Robust Optimization for EarthquakeResistant Design", in B.H.V. Topping, G. Montero, R. Montenegro, (Editors), "Proceedings of the Eighth International Conference on Computational Structures Technology", CivilComp Press, Stirlingshire, UK, Paper 57, 2006. doi:10.4203/ccp.83.57
Keywords: robust optimization, particle swarm, support vector machines, Monte Carlo simulation, stochastic equivalent linearization, earthquake engineering.
Summary
The design of earthquakeresistant structures is generally carried
out by means of code procedures that incorporate the large
uncertainties associated to the seismic phenomenon in a partial
manner, namely that associated to the peak ground acceleration.
The randomness of the time history and the uncertainties in
material properties are normally not considered. This is due to
the difficulties linked to a full probabilistic analysis in
earthquake engineering. The estimation of failure probabilities
can be performed using either random vibration theories or Monte
Carlo simulation [1]. In the former case, rough, usually
very inaccurate estimates of failure probabilities can be obtained
with methods that do not require large computational costs but a
sound knowledge of the involved techniques by the analyst; for
these reasons nonlinear random vibration methods are rarely
applied in the design practice. In the latter case it is
necessary to perform several nonlinear analysis using a different
set of acceleration history and structural parameters in each of
them. In this case the conceptual framework is not so involved
but the computational costs are very large.
Consequently, despite the clear code specification of the probability of the peak ground acceleration exceeding a threshold, codes usually do not require the design to meet similar thresholds for the probability of failure of the structural system in different failure modes. This implies that the introduction of the probabilistic considerations into earthquakeresistant design is normally limited only to the decision on the peak value of the ground motion. This restriction is in marked contrast to the large uncertainties associated with earthquake input and response. In this paper a method for advancing a step further into the consideration of probabilistic concepts in practical earthquake resistant design is proposed. To this end the desirable but difficult goal of estimating the failure probabilities, requiring a full probabilistic design, is abandoned and instead it is proposed to adopt the more practical goal of controlling the secondorder dynamic response. This means renouncing to the reliabilitybased design and assuming a robust design [2] for the earthquake load case. This is due to the fact that, whereas the estimation of failure probabilities in earthquake resistant design faces the difficulties mentioned above, the estimation of nonlinear second order responses can be easily obtained using well established nonlinear random vibration methods. Therefore, the robust design can be formulated by linking this technique to an optimization method yielding the best solution without significant computational effort. To this end the nonlinear analysis of the structure is carried out by means of the random vibration procedure known as equivalent linearization method [1], which has found some practical applications in earthquake engineering [3]. The optimal solution is sought by minimizing a cost function formulated in terms of probabilistic secondorder responses that incorporate the many uncertainty sources present in the model. The search space is explored with a combination of two relevant computational learning tools, which are the particle swarm optimization method [4] and the support vector machines [5]. The first is employed for searching the optimal solution while the latter is a classification (or pattern recognition) technique that is very useful for discriminating feasible from infeasible solutions while reducing the search space. In fact, at a difference with respect to other pattern classification methods the support vector machines provide the auxiliary tool of a margin band around the discrimination function, which allows a drastic simplification of the number of solver calls when using random methods of structural optimization and reliability [6]. The procedure is illustrated with an example concerning a fourstory building for which the maximal flexibility is sought under the constraint that the standard deviations of the drift displacements do not exceed specified threshold values. It is shown that the procedure converges very rapidly. References
purchase the fulltext of this paper (price £20)
go to the previous paper 
