Multidisciplinary design optimization under uncertainty

点击翻译按钮获取中文摘要

Multidisciplinary design optimization (MDO) is a new tool in solving design problems in many real-world engineering problems, such as, renewable energy systems, transportation systems, biomedical systems, among others. The challenges it tackles are: modelling systems from various disciplines, dealing with different kinds of optimization problems (linear, nonlinear, global, integer, etc) with multiobjectives, and dealing with the very large computational needs. MDO needs to use suitable optimization methods to solve problems in different disciplines, for example, problems that are linear but are extremely large scale might require sparse linear programming type models or methods that use large iterative sparse solvers, others may involve highly nonsmooth functions requiring global optimization methods, and most problems require modelling of probabilistic dependence, and so on. There have been many MDO methods suggested with the general principle of decomposing the problem into solving a sequence of sub problems and some of these methods decompose according to sub disciplines. There are many methods such as multidisciplinary feasible, individual discipline feasible, simultaneous analysis and design, collaborative, surrogate optimization, among others. While the issue of large scale is considered in these methods, the issue of uncertainty is still a major challenge. Uncertainty brings an order of complexity more to these problems and is an important issue in MDO problems. New probabilistic uncertainty models based on copulas (especially the Vine copulas) are suitable for such decomposed problems. In this proposal we propose to combine existing deterministic MDO methods with reliability and copula based models to solve real world multidisciplinary optimization problems under uncertainty.