Data-driven modeling of broadband trailing-edge noise

The broadband noise emitted at the trailing edge of an airfoil represents a significant contribution to the noise emission of several industrial components, in both energy and aeronautical fields. Several analytical models focus the attention on some features of the boundary layer close to the trailing edge and use this information to predict the emissions. However, the prediction capability of these models is limited since they are based on several simplifying assumptions. Recently, research efforts have been devoted to the development of machine learning techniques which make it possible to analyze large amounts of experimental data in order to automatically extract modeling knowledge. In this work, Artificial Neural Networks (ANNs) are proposed as empirical models to describe the wall pressure spectrum and the noise directivity. First of all, a study on the choice of the ANN architecture is performed. In order to accomplish this task, an artificial database is generated by using existing semi-empirical models for the prediction of the wall pressure spectrum at different angles of attack: this makes it possible to identify the minimum complexity that the ANN should have in order to accurately describe the spectrum. A second ANN is trained on the directivity distribution obtained by the Amiet analytical theory: both shallow and deep architectures are investigated. The motivation of the present work lies in the fact that the existing analytical models used for building the artificial database are fairly good approximations of the physical phenomena: this means that the chosen ANN architecture is sufficiently complex to accurately describe also a measured noise emission which should represent a perturbation with respect to the models. In this way it is possible to improve the prediction ability of the ANN model by enriching the database with experimental data: this would lead to a general model which is not limited by the simplifying assumptions on which the analytical theories are based.