The need to reliably solve large structural design optimization problems in a reasonable time frame naturally leads to the investigation of the scope of parallel processing in optimization. Parallel optimization may be approached in two general ways. First, the analysis could be the target of parallel processing specially if it is significantly large compared with the overall computation. The second approach is the parallel implementation of the optimization algorithm, which can then be easily adapted for solving any appropriately formulated optimization problem. In this work the basics of parallel computer architectures and some popular parallelization strategies are described. Simulated annealing (SA), a stochastic, discrete optimization technique, is chosen for its global capability, robustness, and suitability for parallel processing. The concept of “shakeup” in SA, which simulates re-annealing, is introduced. The beneficial effects of shakeup for escaping local optima is demonstrated by solving a two-dimensional multimodal optimization problem. A pin-jointed 10-storied multibay plane truss structure is considered as an example optimization problem and has been solved using both serial as well as parallel versions of the SA algorithm. The parallel result, achieved with a relatively small parallel configuration of the computer, indicates that very large structural designs can be optimized in much shorter times along with high probability of achieving global solutions even on moderately sized parallel computers.
ASJC Scopus subject areas
- Aerospace Engineering