THE MINIMIZATION OF MOMENTS AND REACTIVE FORCES IN GRILLAGES WITH A GENETIC ALGORITHM

The aim of the article is to report a technology for the optimization of grillage-type foundations seeking for the least possible reactive forces in the piles for a given number of piles and in the absolute value of the bending moments when connecting beams of the grillage. Mathematically, this seems to be the global optimization problem possessing a large number of local minima points. Both goals can be achieved choosing appropriate pile positions under connecting beams; however, these two problems contradict to each other and lead to diff erent schemes for pile placement. Therefore, we suggest using a compromise objective function (to be minimized) that consists of the largest reactive force arising in all piles and that occurring in the absolute value of the bending moment when connecting beams, both with the given weights. Bending moments are calculated at three points of each beam. The design parameters of the problem are positions of the piles. The feasible space of design parameters is determined by two constraints. First, during the optimization process, piles can move only along connecting beams. Therefore, the two-dimensional grillage is “unfolded” to the one-dimensional construct, and supports are allowed to range through this space freely. Second, the minimum allowable distance between two adjacent piles is introduced due to the specific capacities of a pile driver. Also, due to some considerations into the scheme of pile placement, the designer sometimes may introduce immovable supports (usually at the corners of the grillage) that do not participate in the optimization process and always retain their positions. However, such supports hinder to achieve a global solution to a problem and are not treated in this paper. The initial data for the problem are as follows: a geometrical scheme of the grillage, the given number of piles, a cross-section and material data on connecting beams, the minimum possible distance between adjacent supports and loading data given in the form of concentrated loads or trapezoidal distributed loadings. The results of the solution are the required positions of piles. This solution can serve as a pilot project for more detailed design. The entire optimization problem is solved in two steps. First, the grillage is transformed into the one-dimensional construct and the optimizer decides about a routine solution (i.e. the positions of piles in this construct). Second, backward transformation returns pile positions into the two-dimensional grillage and the “black-box” finite element program returns the corresponding objective function value. On the basis of this value, the optimizer predicts new positions of piles etc. The finite element program idealizes connecting beams as beam elements and piles – as mesh nodes of the finite element with a given boundary conditions in the form of vertical and rotational stiff ness. Since the problem may have several tens of design parameters, the only choice for optimization algorithms is using stochastic optimization algorithms. In our case, we use the original elitist real-number genetic algorithm and launch the program sufficient number of times in order to exclude large scattering of results. Three numerical examples are presented for the optimization of 10-pile grillage: when optimizing purely the largest reactive force, purely the largest in the absolute value of the bending moment and both parameters with equal weights.