Probabilistic learning on manifolds for design optimisation of aero-acoustic liner impedance

Amritesh Sinha (1) , Christophe Desceliers (1) , Christian Soize (1) , Guilherme Cunha (2)

1 MSME - Laboratoire Modélisation et Simulation Multi-Echelle
2 AIRBUS Operations SAS

We address the problem of noise reduction in modern aircraft engines, targeting the low frequency tonal noises by means of tailored acoustic liners in the nacelle. Due to the prohibitive cost of high fidelity computational models for design optimisation, we use Probabilistic learning on Manifolds (PLoM) for constructing statistical meta-models (surrogate models) of the liner acoustic impedances. This allows a learned set to be generated from a given training set whose points are realisations of a non-Gaussian random vector whose support of its probability distribution is concentrated in a subset (a manifold). This approach preserves the concentration of the probability measure for the learned set and has been developed for the case of small training sets as opposed to big data. We then construct a probabilistic meta-model of the liner impedance for which the training set is constructed with CAA performed by Airbus. Conditional statistics of the random impedance are estimated. This allows a robust meta-model of the liner impedance to be constructed.

Domaines: Machine Learning [stat.ML]; Statistiques [math.ST]; Sciences de l'ingénieur [physics]

Amritesh Sinha, Christophe Desceliers, Christian Soize, Guilherme Cunha. Probabilistic learning on manifolds for design optimisation of aero-acoustic liner impedance. International Conference on Uncertainty in Structural Dynamics, USD 2022, KU Leuven, Sep 2022, Leuven, Belgium. pp.1-7. ⟨hal-03799254⟩