Professional Documents
Culture Documents
Structural Optimization of Grid Shells - Design Parameters and Combined Strategies
Structural Optimization of Grid Shells - Design Parameters and Combined Strategies
Abstract: The optimization of structures is a tricky process that involves strategies and mathematical algorithms in which the design
parameters—introduced in terms of variables, constraint conditions, optimization functions, penalty conditions, and so forth—play very
important roles. Their selection and introduction in the optimization process could influence the characteristics and the level of optimi-
Downloaded from ascelibrary.org by Tufts University on 10/26/17. Copyright ASCE. For personal use only; all rights reserved.
zation of the derived solutions. This paper describes the roles of design parameters used in different optimization strategies. Moreover,
an efficient optimization approach, which combines form-finding (FF), sizing-optimization (SO), and topologic-optimization (TO)
strategies in a multilevel process for which design variables and constraint conditions are opportunely selected, is proposed. Numerical
analyses referring to some canopy case studies derived from the current literature are also presented in the paper. The comparisons
among optimization approaches, featuring FF, SO, and TO optimization strategies performed singularly or in combination throughout a
simple sequence of phases, emphasize the effectiveness of the proposed approach for obtaining light structural solutions for grid shells.
DOI: 10.1061/(ASCE)AE.1943-5568.0000286. © 2017 American Society of Civil Engineers.
Sizing Optimization
Taking into account the shapes of canopies derived from the FF, the
SO was performed by considering the different boundary restraint
schemes (R2, R3, and R4), the different grid densities (3 3, 6 6,
and 12 12), and moreover, the range of diameters (d1, d3) and the
criteria for assigning the diameter to canopy members during the
optimization process (d3a, d3b).
The numerical analyses were developed through the commer-
cial software Karamba, a finite-element plug-in developed for
Grasshopper and fully embedded in the three-dimensional (3D) Fig. 3. Flowchart of the optimization strategies
modeling software, Rhinoceros. Karamba allows for interactive
calculations on the responses of 3D structures while considering evidence is common to all the restraint schemes and to all the mesh
the parametric environment of Grasshopper, and then, it shows all densities.
the potentialities in terms of geometric modeling and structural All the schemes were characterized by a maximum displacement
optimization. Moreover, it introduces user subroutines for both value lower than the admissible one. This finding means that the
the preprocessing and postprocessing phases. constraint condition influencing the SO was the utilization ratio of
A mono-objective genetic algorithm strategy was used for the canopy members.
developing the SO process according to the following parameters: The results also reveal that both the mesh density and the
population size of 50, crossover rate of 0.8, and mutation rate of restraint scheme influenced the weight and the stiffness of the solu-
0.1. In particular, constraints in terms of maximum displacement tion derived at the end of the SO. In particular, although it was
(Dmax ≤ 0.12 m) and the maximum utilization ratio of members observed that an increase in the mesh refinement provided a reduc-
(Umax ≤ 1, which takes into account both the strength and local tion in the weight, the influence of the boundary restraints depended
buckling), were both introduced. In addition, the structural on both the number of restrained sides of the canopy and the config-
weight, W, was considered the objective function to minimize uration of the restraints. Indeed, the solutions for the restraint con-
during the SO (Fig. 3). figuration R3 (three sides restrained) were characterized by greater
The results derived from the SO strategy are presented in Figs. 4 values of structural weight than were the solutions for R2 (two sides
and 5 in terms of weight and of maximum displacement normalized restrained); however, the solutions corresponding to the restraint
with respect to the limit value (Dmax), respectively. The figures configuration R4 (four sides restrained) showed lower values of
show that the SO provides better results in terms of weight when the structural weight than either the R2 or R3 scheme configurations
three diameters were considered and assigned according to the crite- did. This evidence is common to both criteria selected for assigning
rion that was based on the utilization ratio (solutions in d3b). This the diameters to the members.
Fig. 4. Weight derived from application of only the SO strategy by di- Fig. 5. Displacements derived from application of only the SO strat-
ameter criteria: (a) d1; (b) d3a; (c) d3b egy by diameter criteria: (a) d1; (b) d3a; (c) d3b
diagonals than either the solution derived from the FF or from the
Sizing-Optimization and Topologic-Optimization SO. For the sake of brevity, the proposed strategy that was applied
Approach to the 6 6 mesh for the three restraint schemes is the only one
Considering the previously obtained solutions of the canopies presented.
derived through the FF and the subsequent SO strategy, an addi- The results obtained from the TO approach are reported in Fig.
tional optimization strategy was analyzed. In particular, a TO based 6 in terms of structural weight: The obtained solutions were com-
on the removal of some of the diagonals of the grid shell was per- pared against the corresponding solutions derived by applying
formed by considering the same computer code and parameters, only the SO. Moreover, Fig. 7 shows the grid configurations
objective function, and constraints of the mono-objective genetic obtained at the end of the optimization process (i.e., after the TO
algorithm as were used for the SO strategy. For this optimization, a phase).
sequence composed of a preliminary FF phase and two subsequent From Figs. 6 and 7, it can be observed that, despite a reduction in
optimization phases, where both SO and TO were based on the the number of diagonals composing the grid, which depended on
same objective function and constraint conditions, were used to both the restraint scheme and the range of diameters and criteria of
derive a final solution that is characterized by a reduced number of assignments used for the SO, the introduction of the TO led to a
was based only on the SO, another approach was used that com-
bines the FF, SO, and TO approaches and is referred to as mixed SO
and TO approach; it is composed of four phases (Fig. 3): a prelimi-
nary FF phase for deriving the shape; a SO phase based on the same
cross-section size for each member (d1), which differs from the
sizes used in the previous approach; a TO phase; and a final SO
phase developed by considering only the d3a and d3b criteria. In
particular, all phases of the process that were based on the SO and
the TO were developed by considering the same objective func-
tion (i.e., the structural weight, W) and the same constraint condi-
tions (i.e., the maximum displacement, Dmax, and the maximum
utilization ratio, Umax) as introduced in the mono-objective
genetic algorithm.
The results derived at the end of the SO phase, introduced after
the TO phase, are reported in Fig. 8 in terms of weight and in com-
parison to the other analyzed approaches. From the comparisons, it
Fig. 6. Comparison of solutions derived from the singular SO strategy
is clear that, despite some cases in which the mixed SO and TO
and from the SO strategy followed by the TO strategy by specific crite-
approach led to lower structural weights, in some cases, the mixed
ria: (a) same cross-section sizes for all the members; (b) orientation of
members; (c) utilization ratio of members
SO and TO approach led to solutions of greater structural weight
than the other approaches.
Fig. 11. Comparison of solutions derived from the described strategies and approaches