The Efficient Multi-Objective Design of Air-Vehicle Configurations using ModelCenter
Stephen J. Leary, Philip Birtwell, Harriet Holden, Graham Johnson · 11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference · 2006
In this paper one approach adopted at BAE SYSTEMS Advanced Technology Centre (ATC) to the aerodynamic design of air vehicles is described. The process utilised makes use of Phoenix Integration’s process integration tool ModelCenter to manage the overall procedure. Design decisions in this work are made using BAE SYSTEMS in house optimisation tool DECIFER. This is a suite of optimisation algorithms including gradient search, multiobjective stochastic search, design of experiments and response surface methods that have been integrated into the ModelCenter environment. Results demonstrate that design decisions can be made in an efficient manner with optimal or near optimal designs found with a low number of calls to the objective function. This paper describes the application of multi-objective design optimisation algorithms to the aerodynamic design of air vehicles. The design of a full system involves consideration of criteria spanning multiple disciplines such as aerodynamic performance, structural integrity, stealth and manufacturing cost to name but a few. Even within one discipline there may be several goals to meet. For instance, within aerodynamics there will be a requirement to design for both lift and drag, multiple points in the flight envelope and so on. These lead to the definition of objective functions which need to be extremised and constraints that need to be satisfied. It is the design for aerodynamic performance alone that is considered in this paper. However the problems solved involve the use of multiple objectives and the algorithms employed are thus readily applicable in a truly multidisciplinary setting. The problems encountered when applying optimisation algorithms in aerodynamic design are well known. One example is noise in objective or constraint functions which can be an artefact of re-meshing or the level of convergence in a Computational Fluid Dynamics (CFD) solver. Hill climbing methods provide a means of making local changes to a design but they can often be confused by noisy landscapes. Stochastic searches tend to be robust to noise as they are able to search over multimodal landscapes but this tends to come at the cost of an increased number of function evaluations. Another problem that is often encountered is the cost of high-fidelity computer simulations that are used. These can provide bottlenecks to the application of direct optimisation algorithms, particularly the aforementioned stochastic searches. The situation becomes even more serious when multiple objectives are present and trade-off designs need to be highlighted.