25th INTERNATIONAL CONGRESS OF THE AERONAUTICAL SCIENCES OPTIMIZATION LOOP BASED ON A CFD RANS CODE.

The design of gas turbine engines is a com-plex and time consuming engineering task in-volving numerous iterative processes. Among all the devices composing an engine, the com-bustion chamber benefits from recent progress in High Performance Computing for Computational Fluid Dynamics (CFD) which is intensively used. Although CFD offers substantial improvements in flexibility and costs when compared to exper-imental approaches, the number of CFD compu-tations still remains large and engineering inten-sive. A strategy to alleviate the CFD specialist in-volvement in reaching an optimal solution is de-sired to fully capitalize on the method. One solu-tion aims at automating the optimization process performed with CFD computations. From the optimization point of view, the ex-pensive step is the CFD run necessary to evalu-ate the objective function value representing the performance of a set of control parameters. The promising method developed in this work con-sists in performing optimization with classical techniques based on an inexpensive metamodel-ing [1] of the objective function. The idea is to reduce the number of CFD computations while proposing a rigorous framework to converge ro-bustly to an optimal design. Implementation issues such as efficiency, nu-merical stability of the model, techniques to ac-celerate the optimization process, exploration / exploitation conflict and the different levels of parallelism involved in CFD optimization are ad-dressed in this paper. Practicality of the optimiza-tion tool is here inherited from the powerful cou-pling device PALM [2]. Illustration of the approach when applied to a simple engine cooling system is presented: Val-idations steps and comparisons of the proposed surrogate approach with the very well known op-timization method Simplex [3] are shown.