The optimization techniques aim to reduce the used material in the structure design without violating the imposed design and manufacturing constraints; thus, the materials cost is decreased, and less material is consumed. In addition to the low-cost and positive sustainability impact of Structural Design Optimization (SDO), it promotes the engineers to develop innovative designs for several real-life structural applications.
The previous studies in the domain of SDO gainfully contributed in the way that expanded our knowledge in this specific field. However, this research theme still watches debates on various issues, not only how we could achieve the desired physical features of a structure, but also how we could make this is happening efficiently at a lowest possible computational cost. With this regard, a couple of potential research opportunities have been extracted from the literature, and in-context novel contributions were presented in the current thesis.
The literature of structural design optimization problems reveals that global optimization algorithms or Meta-heuristics (MHs) are the best available techniques that could be used to solve such hard optimization problems. Though, the main challenge confronted the engineer is which available MH fits much better to the structure design problem of his attention. Unfortunately, the literature of SDO experiences a lack of systematic assessment pattern that could help the engineers to overcome this issue. Currently, the commonly used measures of MHs performance by MHs developers are the statistical operators such as min, max, mean and standard deviation of the obtained solutions. However, such measures are not enough to reflect the actual MH performance when these measures are used alone. So, a comprehensive assessment criterion has been developed here to include more efficient measures like the practical reliability, price (computational cost), normalized price, performance rate, solution quality and Fitness-landscape analysis. Additionally, two different convergence rates were imposed to examine the MHs at slow and fast rates. As well as the reproducibility of the numerical experiments results considered within the procedure of MHs assessment. Lastly, the proposed criterion has been employed to compare five different Ant Colony Optimization (ACO) variants. The proposed measures demonstrated a comprehensive assessment of the compared ACOs performance.
Several studies were conducted to improve the MHs searching performance, and the results were promising in this direction. Unfortunately, the literature of SDO demonstrated an extreme tendency to develop new “metaphor” MHs instead of improving the performance of wellestablished MHs that have a remarkable history of solving Non-Polynomial (NP) -hard optimization problems. However, this thesis examines the possible improvements of two selected MHs searching features via integrating local search movements to MH's main structure to improve the intensification effort. The first selected MH is Cuckoo Search (CS) algorithm, which is intensively used to solve a variety of optimization problems such as weight minimization of the truss structure, Travelling Salesman Problem (TSP),…etc. CS is designed to solve unconstrained continuous optimization problems as most of MHs. Consequently, the original CS has been adapted and modified here to solve discrete SDO problems, and it is named Adapted CS Algorithm (ADCSA). The intensification effort of ADCSA improved through four different local search movements of permutation, swap, bit flip and insertion. ADCSA has been applied to solve two different SDO problems, and the obtained results of both case studies reveal that the proposed ADCSA has a considerable performance in solving SDO problems. The other improved MH was the ACO variant of the Hyper Cube Framework (HCFACO), which was selected based on the results of the comparison study of five different ACO variants that previously mentioned. The enhancement of the HCFACO intensification effort was carried out by integrating two local search movements of insertion and bit flip. The performance of improved version HCFACO, Enhanced HCFACO (EHCFACO), has been examined through solving a well-known SDO benchmarking problem. EHCFACO exhibited a significant performance compared to the original HCFACO and the other five ACO variants.
Regarding the structural design optimization frameworks, the literature review has shown that both deterministic and probabilistic SDO approaches are mostly used. Nevertheless, the scarcity of studies of uncertainty design using anti-optimization is understandable because of the associated expensive design analysis cost of the objective function. This approach has two levels, the top level devoted to the optimization phase, while the bottom level works to antioptimize the obtained optimal solution. During this process of optimization and antioptimization, a large number of objective function calls is taking place, and for those SDO problems with expensive functions, this approach becomes unfeasible. However, a costeffective uncertainty framework has been developed in the current study. The accompanied expensive cost of the objective function evaluation has been tackled by replacing the blackbox function (FEA software) by an Artificial Neural Network (ANN). The proposed procedure was applied to optimize a novel case study of a perforated composite laminated plate subjected to the uncertainty of loading conditions and the location of the cut-out center. The attained results reveal that using ANN techniques offers a cost-effective solution for SDO problems with expensive objective functions.
In addition to those specific research opportunities mentioned above, some others appeared during the research process, such as the effect of selection of the initial population, solution representation and adaptive generation of a new solution on the performance of MHs. Moreover, two novel SDO examples have been developed for customized I-beam overhead gantry crane and perforated composite laminated plate. In general, this thesis has gone some way towards enhancing our understanding of how to tackle the complexity of SDO problems.
| Date | 10 Feb 2021 |
|---|
| Original language | American English |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Van Ngan Lê (Supervisor) & Thien-My Dao (Co-supervisor) |
|---|