As the result of vibration emission in air, machine sound signal carries affluent information about the working condition of machine and it can be used to make mechanical fault diagnosis. The fundamental problems with fault diagnosis are the estimation of the number of sound sources and the localization of sound sources. The wave superposition can be employed to identify and locate sound sources, which is based on the idea that an acoustic radiator can be approximated and represented by the sum of the fields due to a finite number of interior point sources. But, in practice, a large number of measurements must be used in order to achieve a desired resolution, which makes the reconstruction process very time-consuming and expensive. In this paper, a combined wave superposition method has been developed reconstruct to acoustic radiation from machine acoustical signals. This method combines the advantages of both the wave superposition and Helmholtz equationleast squares methods, and it allows for reconstruction of the acoustic field from an arbitrary object with relatively few measurements, thus significantly enhancing the reconstruction efficiency. After sound source localization, the blind source separation (BSS) is proposed to extract acoustical feature from the mixed measuring sound signals. In a semi-anechoic chamber, a cross-planar microphone array, which consists of 29 microphones, was successfully applied to obtain the two-dimensional mapping of the sound sources. The location, the sound pressure, and the properties in frequency domain of the sound sources can be found through this method precisely. The experimental results demonstrate that the methods presented can potentially become an acoustical diagnosis tool.
Using Blind Source Separation and a Compact Microphone Array to Improve the Error Rate of Speech Recognition
