Recovery of vibration signal based on a super-exponential algorithm

Vibration-based analysis is an important technique in machine fault diagnosis. In complex machines, the vibration generated by a component is easily affected by the vibration of other components or is corrupted by noise from other sources. Hence, the fault-related vibration must be recovered from among those sources for accurate diagnosis. In this paper, a super-exponential algorithm (SEA) based on the single-input single-output (SISO) convolution mixing model is investigated to deconvolve the vibration signal of interest. Based on simulated vibration signals generated by a test bench in a laboratory and real signals generated by industrial machines, the characteristics of the SEA with skewness and kurtosis schemes are studied extensively for the recovery of signals with different statistical distributions. It is shown that the SEA with the skewness scheme is more suitable for the isolation of a vibration signal with an asymmetric distribution, whereas the SEA with the kurtosis scheme is more efficient for detecting a vibration with abrupt changes in its distribution. Furthermore, the effect of the equalizer length on the characteristics of the equalized signal is discussed. A performance curve that presents the equalization performance is defined, and a method that uses this curve to select the optimal equalizer length for successfully recovering the defect signal from the mixed signal is then proposed. The study presented in this paper thus shows that the SEA is a promising method for the recovery of vibrations in machine fault diagnosis.