Nonparametric probabilistic vibroacoustic analysis with Nastran : a computational tool for estimating the likelihood of automobiles experimental FRF measurements

International audience; Improvement of vibroacoustic models prediction capabilities in a probabilistic context requires a adapted metric to compare experimental results with stochasitic computations. The likelihood appears as the natural tool to compare experiments with probabilistic computations as soon as the probability of a given result may be computed. Since vibroacoustic analysis mainly rely on complex Frequency Response Functions ([FRF] = {ω → [FRF(ω)]}) matrices that can be easily measured and computed, the likelihood of such complex and frequency dependent matrices is investigated. A two stage statistical reduction, based on Indepen-dant Components Analysis, is proposed in order to separate statisticaly independent components with complex amplitudes which probability may be computed independently one from each others. Bi-dimensional probability density fonctions of the complex components amplitudes are deduced from a Monte-Carlo simulation of a non-parametric stochastic model, using MSC/NASTRAN. The proposed statistical reduction presents many interesting properties regarding the physical understanding of FRF matrices as well as a numerical aspects.