Feature extraction based on the 3D spectrum analysis of acoustic signals to identify rotor malfunction

Abstract

The acoustic signals of an operating rotor contain plenty of information about its running condition and can be measured by non-contact devices. Therefore, it is necessary to base fault diagnosis on the acoustic signals of a rotor. In this study, the typical malfunctions of unbalance, looseness, rubbing, misalignment, and abrupt unbalance were simulated in experiments. Acoustic signals were received by BSWA microphones and a 16-channal Sony tape recorder, which were analyzed by 3D spectrum. The sequential spectra can be separated by uniform increments of time or speed. When data is acquired by a relatively slow acceleration with uniform increases in speed, one dimension (the ordinate) is equivalent to the speed of rotation. Thus, the plot demonstrates the frequency versus speed-of-rotation interference diagram. The plot allows the analyst to evaluate the various frequency components of vibration as the machine approaches its operating speed. The abrupt changes in surging sound signals of rotor malfunctions can be detected, and their fault spectrum features are shown in the 3D spectrum. The results demonstrate the potential and feasibility of this approach for the diagnosis of rotor malfunctions.