Abstract
This paper proposes a method improving precision of sound source identification by inverse acoustic analysis. The inverse acoustic analysis identifies a vibrational state for mechanical structures by measuring the radiated sound in many points and solving the inverse problem. In this analysis, the uniqueness of solution is not always guaranteed, and the problem is to avoid the instability of the inverse matrix in sound source identification. A numerical simulation is performed about a plate model, on which many point sources are assumed that radiate sound pressure by their vibration. Sound pressures are measured by many microphones and the inverse matrix is calculated from acoustic transfer functions. This paper presents two kinds of methods for changing the coordinates of measurement points in order to decrease the instability of the inverse matrix. One is the singular value sensitivity analysis and the other is the statistic approach by using normal distribution. Both methods are applied for the noise source identification of the plate, and the results are compared, and their features are discussed.
