Multi-Objective Design Optimisation of a Diffuser-Ejector Exhaust Duct for Helicopter Engines

The paper demonstrates the successful application of an optimisation methodology for the design of a diffuser-ejector exhaust duct. Maximising simultaneously pressure recovery and the entrainment ratio are diverging objectives which could hardly be achieved by a conventional manual trial-and-error approach relying on the designer’s experience. This multi-objective design problem has been solved for the axis-symmetric exhaust duct with a given characteristic length, inlet section and minimal standoff distance by coupling a parametric method with 2D CFD analysis. Open cubic B-splines have been employed to generate the contoured duct shape, for which the control-point vertices have been defined by a total of 17 engineering parameters. A bell mouth inlet has been chosen for the ejector inlet. The parameter constraints result from weight and integration requirements. Three characteristic engine operating points have been chosen for the multi-point and multi-objective shape optimisation. The entire process of model building, meshing, performing the 2D CFD calculation and post-processing to extract the required metrics has been fully automated. A commercial process integration software package is used to link the different tools together in a unified environment. The design space exploration is carried out via a latin-hypercube sampling technique. This random space filling method has been chosen because of its considerable lower number of experiments compared to factorial sampling techniques. Parameter ranking is obtained by a weighted average of the correlation coefficients for each objective. The parameter hierarchy is slightly different for the engine operating points. However, there exists a clear threshold separating the influential parameters from the insignificant ones. A subsequent DOE is performed for the reduced parameter set for which the minimum number of experiments has been chosen as twice the number of experiments to generate a quadratic response surface. The Normal-Boundary Intersection method is applied to find the Pareto front based on the response surface model as surrogate model. The results show that a gain of 20% for the pressure recovery for a given entrainment ratio could be achieved compared to a configuration defined by a manual trial-and-error approach. The great benefit of the present method is its capability to handle easily geometrical constraints and the weight of the different design objectives which may change even during the detailed design phase.