PRINCIPAL COMPONENT ANALYSIS FOR NOISE SOURCE IDENTIFICATION

摘要

One of the tools used to estimate the contributions of different potential noise sources in automotive vehicles consists of establishing causal relations between several measured vibration and acoustical signals. Classical techniques are based on ordinary, partial and multiple coherence. This paper discusses a multivariate statistical analysis technique, principal component analysis, that overcomes many disadvantages revealed, by the currently used coherence methods. The proposed technique calculates in a linear least squares sense a minimum set of uncorrelated signals (principal components) that by linear combinations describe the measured signals. The information provided by the number of uncorrelated signals and by the linear combinations can lead to a clearer insight into the correlation patterns between the signals (virtual coherence). The technique also allows viewing the operating vibration behaviour of the tested structure, which can be helpful to locate force inputs as well as to identify the energy transmission paths. Some examples, numerical simulations as well as real life experiments, illustrate some interesting features of the proposed technique.