Underdetermined blind source separation based condition monitoring

A common technique of mechanical vibration measurement requires an operator to use vibrometer by attaching the accelerometer directly to a machine. The technique, however, poses a unsafe operation and involves significant man-power. This paper proposes a non-contact vibration measurement for machines condition monitoring by using acoustic emission information. It essentially uses the emitted signal to identify machinery condition. The measurement setup is an array of microphones is placed on critical operation at a power plant by assuming that the number of the sound sources are more than that of the sensors (underdetermined case). This approach is closely related to the sparse representation based on the signals in the time-frequency domain. These suggest that the proposed technique may accurately measure the vibration using acoustical emission. The paper also presents comparative results between baseline, estimated acoustic and vibration signals.