The SAMgram: a novel local enhancement and targeted extraction strategy for weak bearing fault characteristics

Abstract

Envelope analysis is one of the most commonly used methods in rolling bearing fault diagnosis. However, when a signal contains heavy noise, even if an appropriate frequency band is selected, the fault information can still be overwhelmed. Unlike traditional use of spectral amplitude modulation, a novel SAMgram is proposed in this article for the enhancement and extraction of weak bearing fault characteristics in a signal, where local spectral amplitude modulation (LSAM) is performed to highlight bearing fault components and improve the proportion of fault information. Meanwhile, a targeted indicator called normalized harmonic kurtosis is proposed to select an optimal modified filtered signal automatically by quantifying repetitive transient characteristics. To extend LSAM to practical applications, two spectrum segmentation strategies are provided based on scanning spectrum and trend component, named the scanning SAMgram and adaptive SAMgram, respectively, which aim at determining the optimal modified filtered signal for demodulation by searching for the combinations of frequency bands and magnitude orders. A simulated signal of bearing compound faults and experimental signals of bearing outer ring and inner ring faults indicate that the proposed method can not only select a frequency band related to bearing defect to eliminate interference of invalid components but also highlight fault characteristics in the selected frequency band and weaken the disturbance of noise, which is superior over traditional envelope analysis-based methods and more applicable for fault diagnosis of rolling bearings under complex environments.