The paper presents a preliminary investigation on the applicability of stochastic Radial Basis Functions (RBF) in the development of dynamically adaptive meta-models for aeroacoustic applications. The analysis focuses on the influence of the RBF kernel and the chosen stochastic parameters on the modelling of target functions of interest in the aeroacoustics of aircraft. The rationale underlying the research is related to the key role that aeroacoustics plays in the establishment of the commercial aviation scenario foreseen for the next three decades. Indeed, the sustainable development for the airborne transportation system is strongly constrained by community noise, which, nowadays, limits the increase of the capacity of the existing airports and prevents the building of new ones. In such a situation, the design of the next generation of aircraft must take into account the impact of noise on the population since the early conceptual phase of the design. This causes a substantial increase of the required computational resources, especially for unconventional, breakthrough concepts for which simple semi—empirical models are not available and the only viable strategy is computational aeroacoustics. The availability of reliable meta-models can give a significant contribution in two ways: i) in a process of multiobjective, multidisciplinary design optimization a dynamic adaptive stochastic meta-model can reduce significantly the calls to the computationally expansive tools and enhance the effectiveness of the design space exploration; ii) the versatility and applicability range of end-user tools for the estimate community noise impact can be greatly improved by fast yet accurate models of the noise signature of novel concepts. Two target functions are analysed here: the total acoustic field induced by a point source co—moving with a scattering profile, and the shielding factor along a line of observation points below the scatterer. The performance of RBF meta-models based on tailored kernels is compared to the most commonly used kernels in terms of accuracy and convergence rate. The effectiveness of the dynamic meta-model update based on uncertainty quantification is assessed for different choices of the stochastic parameter.
展开▼
