Method based on dual-quadratic programming for frame structural optimization with large scale (Q944626): Difference between revisions

This useful paper deals with a method based on dual-quadratic programming for frame structural optimization with large scale. The present optimization model is established by taking the structural weight as objective function, one of main geometrical parameters of beam elements (a sectional moment of inertia) as design variable, and the stresses and nodal displacements as constraints. The proposed model is expressed as follows: find \(I_{yi}\) \((i=1,\dots,n)\), make \(w=\sum^n_{i=1} \rho_i l_i A_i (I_{yi})\to\min\), subject to \(u_{jl}\leq \overline{u}_{jl}\) \((j=1,\dots, J\), \(l=1,\dots, L)\), \(\sigma_{il} \leq \overline{\sigma}_i\) \((i=1,\dots, n\), \(l=1,\dots, L)\), \(\underline{I}_{yi} \leq I_{yi} \leq \overline{I}_{yi}\) \((i=1,\dots,n)\), where \(I_{yi}\) is the \(y\)-axial sectional moment of inertia of the i-th beam element and is considered as a design variable, \(n\) is the number of design variables, \(J\) is the number of displacement constraints, \(L\) is the number of loading cases. According to the full stress criterion, the maximal stress of every beam element is equal to the allowable stress at least under one loading case. This is expressed by: \(\max_{l=1,\dots,L} [\max_x \sigma_{il}(x)]= \overline{\sigma}_i\). The main result: Optimality criteria method and mathematical programming are combined to found the sectional optimization model of frame structures. To avoid the computations caused by first-order approximation, the stress constraints are approached in zeroth-order approximation and transformed into movable sectional lower limits with the full stress criterion. To avoid sensitivity analysis of displacements to design variables from the Taylor expansion, the displacement constraints are transformed into explicit expressions by the unit virtual load method. Using the dual-quadratic programming, the large-scale optimization problem can be solved easily and efficiently. Finally, the numerical results on two examples allow to conclude that the approximate scaling step and deletion of negative constraints can improve the solving efficiency. Note that the authors present program which is more efficient and accurate than Nastran software.