Time-frequency feature extraction of a cracked shaft using an adaptive kernel

Adaptive time-frequency representations have many advantages compared with conventional methods. In this paper, a new method is proposed to adapt Smoothed Pseudo Wigner-Ville distribution to match signal's time-frequency content. It is based on maximizing a local time-frequency concentration measure for different time and frequency smoothing window lengths. Subsequently, the optimized values are used for constructing an adaptive kernel over time. The proposed transform is then applied to vibration signals of healthy and cracked shafts which are acquired through run-up, and the crack signature is obtained. Results show that enhanced improvement in resolution is obtained while the computational cost is not very high.