Iuliano, Emiliano (2011) Aerodynamic Shape Optimization with Physics-based Surrogate Models. [Tesi di dottorato] (Unpublished)
Download (18MB) | Preview
|Item Type:||Tesi di dottorato|
|Uncontrolled Keywords:||POD transonic optimization evolutionary surrogate meta-model design|
|Date Deposited:||09 Dec 2011 19:26|
|Last Modified:||30 Apr 2014 19:49|
Aircraft design, as many other engineering applications, are increasingly relying on computational power. The growing need for multi-disciplinarity and high-fidelity in design optimization and industrial applications implies a huge number of repeated simulations to find an optimal design candidate. Indeed, a strong eﬀort has been done in the recent past to introduce potentially highly accurate analysis methods both in geometry and physics modelling. The main drawback is that they are computationally expensive. The solution of non-linear steady or unsteady aerodynamic flows by numerically solving the Navier-Stokes equations implies an amount of data storage, data handling and processor costs that may result very intensive even when implemented on modern state-of-art computing platforms. This turns out to be an even bigger issue when used within parametric studies, automated search or optimization loops which typically may require thousands analysis evaluations. The core issue of a design optimization problem is the search process of an optimal solution. However, when facing complex problems, the high-dimensionality of the design space and the high-multi-modality of the target functions cannot be tackled with standard techniques. Surrogate and reduced order modelling can provide a valuable alternative at a much lower computational cost. A global surrogate model is generally referred to as a low-cost model able to provide an approximation of a selected objective function over the whole design space. A reduced order model is a surrogate which is further able to capture and reproduce the physics embedded in the high-fidelity model by using a low-dimensional basis. Hence, a reduced order modelling of high-fidelity data (e.g. coming from accurate numerical solvers) with limited computational cost is a highly desirable feature. This is particularly true in CFD-based aerodynamic optimization. Commonly used RANS solvers are still time-consuming when complex fluid dynamics cases have to be faced, e.g. a wing-body aircraft configuration including engine and tail-planes. In this perspective, the present research aims at making a step towards bridging the gap between design stages through the coupling and exploitation of advanced analysis methods, reduced order/meta-modelling, optimization techniques and CAD-based tools towards the aero- dynamic design of innovative aircraft configurations with reasonable computational resources. The introduction of physics-based surrogate models will allow to correctly drive the design process since the very early stages and, hence, to refine the evaluation of potentially cost/environment saving concepts.
Actions (login required)