Prediction and analysis of excitation sources of car booming noise through a Bayesian meta-model

Current approaches in the automotive domain to predict booming noise essentially target extreme loading conditions. This is useful when mechanical strength is of concern, but not representative of the actual vehicle usage. Usage is however important when addressing acoustic annoyance. One issue in this respect is the lack of databases representative of the diversity of client usages and, therefore, of the excitation forces applied to the vehicle in real usage conditions. This paper introduces a possible answer to this problem. First, it proposes a measurement protocol to estimate the excitation forces in real usage responsible of booming noise. Second, it provides an array of algorithms to analyze the large amount of data collected during the measurement step. In particular, an ad hoc Independent Component Analysis algorithm is introduced to extract excitation components specific to well-defined operating condition regions, thus providing insights in the excitation behaviour, as well as data reduction. Next, the excitation components are modelled as a function of the vehicle operating conditions with a Radial Basis Function network. A meta-model of the excitations in real usage conditions is thus obtained, composed of two levels: the analysis level and the modelling level. The proposed methodology is developed in the Bayesian framework. In addition to its advantages linked to the trivial introduction of prior knowledge, the Bayesian framework is found particularly useful for propagating uncertainties throughout the successive steps of the approach.