Reliability-based Robust Design Optimization Using Polynomial Chaos Expansion for Aero-Engine Inlet Applications
CHALLENGE - As proven by recent studies, the application of Polynomial Chaos expansion is one of the most efficient methodologies to accurately manage uncertainties in industrial design. However, this methodology requires a minimum number of samples, which heavily increases in proportion with the number of uncertainties. Therefore, a typical industrial optimization case (for instance, at least 10 simultaneous uncertainties) can hardly be treated as a feasible task. For this reason, we propose some approaches to efficiently handle industrial problems of this kind, both in terms of Uncertainty Quantification (UQ) and in terms of Robust Design Optimization (RDO).
SOLUTION - For the UQ, we propose methodologies that allow to identify which Polynomial Chaos terms have higher statistical effects on the performances of the system, reducing the number of unknown coefficients and therefore the number of needed samples to complete the UQ. For the RDO, we propose a methodology based on the reliability formulation of objectives, which guarantees the reduction of objectives number compared to a classical RDO approach. The industrial validation case targeted in this study is a typical regional jet engine nacelle. Acoustic requirements are prescribed by customers and aviation agencies for three typical flight conditions: Approach; Sideline (or Take-Off); and Flyover (or Cut-Back). Since these requirements can be in contrast to each other and an acoustic liner providing the best attenuation in one flight condition does not necessarily perform well in the other conditions, liners shall be designed through a multiobjective optimization procedure. A campaign of experiments has been conducted by Leonardo Aircraft, which has performed the geometrical measurements of a series of acoustic panels. The database was therefore been imported in modeFRONTIER software from ESTECO.
BENEFITS - The optimization problem defined in previous section has been solved using two different algorithms independently, the SIMPLEX and a Genetic Algorithm (MOGAII) of modeFRONTIER, to determine which approach is the most efficient, both in terms of accuracy and in terms of the required number of design evaluations. At the end, the design RID #150 for MOGAII and design RID#54 have been selected as optimal ones.