Multi-Objective Matched Synchrosqueezing Chirplet Transform for Fault Feature Extraction From Marine Turbochargers

Abstract

Turbocharger is one of the vital parts of a diesel engine causing a high failure rate. Its surface vibration signal contains important time-varying features. To better process the nonstationary signals with time-varying features and perform the time-frequency transformation on the turbocharger surface vibration signal, a novel multi-objective matched synchrosqueezing chirplet transform method is proposed in this paper. The method is based on the Linear Chirplet Transform to optimize the selection of the demodulation rate. Parameters such as Rayleigh entropy and signal-to-noise ratio are used as targets to select the value of the optimal demodulation rate. Then the local maximum synchrosqueezing transform is used as a post-processing method for time-frequency rearrangement of the signal. This method improves the energy concentration of the transformation result while maintaining the ability of signal reconstruction. On the test stand, the turbocharger fault samples were obtained. The time-domain signals of the turbocharger at the $1\times $ , $2\times $ , and $9\times $ frequencies were reconstructed based on the time-frequency transformation results, and fault characteristic parameters were extracted from them. Then the effectiveness of the fault feature parameter identification ability was validated by Principal Component Analysis. The study showed that our proposed MOMSSCT method’s transformation results have high time-frequency energy aggregation, clear trajectories, and a 37.5% reduction in frequency spread width. The extracted fault characteristic parameters have good fault classifiability under various turbocharger operating conditions. Using the fault features extracted by MOMSSCT, the diagnostic accuracy rate can reach more than 85%.

Keywords

• Fault feature parameter extraction
• time-frequency transformation
• turbocharger