Optimization of complex multidisciplinary designs under uncertainty via simulated annealing

• Main Author: Michael Magnin
• Other Authors: Yongki Go Tiauw Hiong
• Format: Theses and Dissertations
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/41767
• Institution: Nanyang Technological University

The objective of this research is to develop a method that combines a multiobjective multidisciplinary design optimization algorithm with a method for propagating and mitigating uncertainties. Simulated annealing is chosen as the multidisciplinary design optimization algorithm and a probabilistic method is implemented to handle uncertainties via Monte Carlo simulation. A multi-objective simulated annealing algorithm is implemented and evaluated on a representative test problem. Separately, a probabilistic uncertainty method is implemented and applied to the Ship Tracking and Environmental Protection Satellite conceptual-design problem. The next step applies a deterministic simulated annealing algorithm to the design problem. Finally, the simulated annealing algorithm is combined with the probabilistic uncertainty method and is applied to the design problem. In addition to developing a novel simulated annealing under uncertainty method, this thesis provides three additional principal contributions. The first is an implementation of multi-objective optimization under uncertainty. This research also represents the first application of a method for quantifying and mitigating uncertainties to a system-level satellite model. Finally, a novel method for deriving a composite solution from a subset of leading solutions provided by the algorithm is also introduced. The composite solution is then compared to solutions obtained from a baseline study performed using concurrent engineering methods, a deterministic multiobjective simulated annealing algorithm, and the single best solution of the investigated method. The investigated method provides a better solution than the baseline at median confidence levels; however, at higher confidence levels the baseline solution is superior. The composite solution of the investigated method is ~ superior to the single best solution at all confidence levels. Similarly the composite I solution yields a superior result compared to the deterministic simulated annealing solution.